Skip to contentSkip to conversation

ovpn-dco: introduce data-channel offload support

PATCH Superseded Antonio Quartulli 7 participants 162 messages Jun 24, 2022 – Aug 11, 2022 1 message not in the archive Page 1 of 2

v1 of 101 Superseded Latest is v101 →

lore
Share & tools
lore
#0 lore raw

cover ovpn-dco: introduce data-channel offload support

This patchset is *almost* the same as the previous patchset, but it got fragmented in multiple patches for easier review.

I am not setting any version on these patches as they do not match previous versions.

The idea behind splitting patches is that they should be mergeable and compilable one-by-one.

Ideally they should not introduce any regression in existing functionalities, however, it may not always be apossible to *test* the new code as it will require a subsequent patch to be used. Still we will be able to compile the new code and review it.

While splitting patches, some ugly things have been fixed/changed..so hopefully splitting patches has already paid off a bit..

I hope this will help with getting this new functionality merged soon.

This patchset reflects what is currently pushed to the "dco" branch.

Best Regards,

Antonio Quartulli (24):
dco: introduce low-level code for handling ovpn-dco in the Linux
kernel
dco: add helper function to detect if DCO is enabled or not
dco: use specific metric when installing routes
dco: create DCO interface using SITNL
dco: let open_tun_generic handle the DCO case
dco: initialize context and save pointer in TLS object
dco: add option check - disable DCO if conflict is detected
dco: allow user to disable it at runtime
dco: configure keys in DCO right after generating them
dco: periodically check and possibly rotate/delete keys
dco: split option parsing routines
dco: check that pulled options are compatible
dco: implement dco support for p2p/client code path
dco: implement dco support for p2mp/server code path
dco: add documentation for ovpn-dco-linux
GitHub Actions: add Linux DCO build (on Ubuntu 20.04)
dco: turn supported ciphers list into a function
dco-win: implement GetOverlappedResultEx for mingw32
dco-win: add platform dependant check on incompatible options
do_open_tun: restyle "can preserve TUN" check
dco-win: introduce low-level code for handling ovpn-dco-win in Windows
dco-win: implement ovpn-dco support in P2P Windows code path
dco-win: add documentation to README.dco.md
dco-win: update GH Actions config file

Arne Schwabe (1):
tun: extract close_tun_handle into its own fucntion and print correct
type

 .github/workflows/build.yaml                  |  16 +-
 Changes.rst                                   |   9 +
 README.dco.md                                 | 132 +++
 config-msvc.h                                 |   2 +
 configure.ac                                  |  37 +
 dev-tools/special-files.lst                   |   2 +
 doc/man-sections/advanced-options.rst         |  13 +
 doc/man-sections/server-options.rst           |   6 +
 src/compat/Makefile.am                        |   3 +-
 src/compat/compat-dco_get_overlapped_result.c |  46 +
 src/compat/compat.h                           |   8 +
 src/compat/compat.vcxproj                     |   1 +
 src/compat/compat.vcxproj.filters             |   3 +
 src/openvpn/Makefile.am                       |   7 +-
 src/openvpn/dco.c                             | 621 ++++++++++++
 src/openvpn/dco.h                             | 318 ++++++
 src/openvpn/dco_internal.h                    |  79 ++
 src/openvpn/dco_linux.c                       | 940 ++++++++++++++++++
 src/openvpn/dco_linux.h                       |  59 ++
 src/openvpn/dco_win.c                         | 402 ++++++++
 src/openvpn/dco_win.h                         |  58 ++
 src/openvpn/errlevel.h                        |   2 +
 src/openvpn/event.h                           |   3 +
 src/openvpn/forward.c                         |  89 +-
 src/openvpn/init.c                            | 456 ++++++---
 src/openvpn/init.h                            |   4 +-
 src/openvpn/mtcp.c                            |  59 +-
 src/openvpn/mudp.c                            |  13 +
 src/openvpn/multi.c                           | 221 +++-
 src/openvpn/multi.h                           |  14 +-
 src/openvpn/networking_sitnl.c                |  11 +
 src/openvpn/openvpn.vcxproj                   |  11 +-
 src/openvpn/openvpn.vcxproj.filters           |  27 +
 src/openvpn/options.c                         |  53 +-
 src/openvpn/options.h                         |  15 +
 src/openvpn/ovpn-dco-win.h                    | 108 ++
 src/openvpn/ovpn_dco_linux.h                  | 265 +++++
 src/openvpn/socket.c                          |  94 +-
 src/openvpn/socket.h                          |  26 +-
 src/openvpn/ssl.c                             |  74 +-
 src/openvpn/ssl.h                             |   7 +-
 src/openvpn/ssl_common.h                      |  23 +
 src/openvpn/tun.c                             | 250 +++--
 src/openvpn/tun.h                             |  64 +-
 44 files changed, 4292 insertions(+), 359 deletions(-)
 create mode 100644 README.dco.md
 create mode 100644 src/compat/compat-dco_get_overlapped_result.c
 create mode 100644 src/openvpn/dco.c
 create mode 100644 src/openvpn/dco.h
 create mode 100644 src/openvpn/dco_internal.h
 create mode 100644 src/openvpn/dco_linux.c
 create mode 100644 src/openvpn/dco_linux.h
 create mode 100644 src/openvpn/dco_win.c
 create mode 100644 src/openvpn/dco_win.h
 create mode 100644 src/openvpn/ovpn-dco-win.h
 create mode 100644 src/openvpn/ovpn_dco_linux.h
Signature · 1 line
2.35.1
#1 lore raw

PATCH 1/25 dco: introduce low-level code for handling ovpn-dco in the Linux kernel

---
 configure.ac                        |  34 +
 dev-tools/special-files.lst         |   1 +
 src/openvpn/Makefile.am             |   3 +
 src/openvpn/dco.h                   | 165 +++++
 src/openvpn/dco_internal.h          |  78 +++
 src/openvpn/dco_linux.c             | 934 ++++++++++++++++++++++++++++
 src/openvpn/dco_linux.h             |  60 ++
 src/openvpn/errlevel.h              |   2 +
 src/openvpn/openvpn.vcxproj         |   7 +-
 src/openvpn/openvpn.vcxproj.filters |  15 +
 src/openvpn/ovpn_dco_linux.h        | 265 ++++++++
 src/openvpn/tun.h                   |   3 +
 12 files changed, 1566 insertions(+), 1 deletion(-)
 create mode 100644 src/openvpn/dco.h
 create mode 100644 src/openvpn/dco_internal.h
 create mode 100644 src/openvpn/dco_linux.c
 create mode 100644 src/openvpn/dco_linux.h
 create mode 100644 src/openvpn/ovpn_dco_linux.h
configure.ac +34 -0
index 9c898718..353da08c 100644
@@ -142,6 +142,13 @@ AC_ARG_ENABLE(
 	[enable_small="no"]
 )
 
+AC_ARG_ENABLE(
+	[dco],
+	[AS_HELP_STRING([--enable-dco], [enable data channel offload support using ovpn-dco kernel module @<:@default=no@:>@])],
+	,
+	[enable_dco="no"]
+)
+
 AC_ARG_ENABLE(
 	[iproute2],
 	[AS_HELP_STRING([--enable-iproute2], [enable support for iproute2 @<:@default=no@:>@])],
@@ -760,6 +767,32 @@ PKG_CHECK_MODULES(
 	[]
 )
 
+
+if test "$enable_dco" = "yes"; then
+dnl
+dnl Include generic netlink library used to talk to ovpn-dco
+dnl
+
+	case "$host" in
+		*-*-linux*)
+			PKG_CHECK_MODULES([LIBNL_GENL],
+					  [libnl-genl-3.0 >= 3.4.0],
+					  [have_libnl="yes"],
+					  [AC_MSG_ERROR([libnl-genl-3.0 package not found or too old. Is the development package and pkg-config installed? Must be version 3.4.0 or newer])]
+			)
+
+			CFLAGS="${CFLAGS} ${LIBNL_GENL_CFLAGS}"
+			LIBS="${LIBS} ${LIBNL_GENL_LIBS}"
+
+			AC_DEFINE(ENABLE_DCO, 1, [Enable shared data channel offload])
+			AC_MSG_NOTICE([Enabled ovpn-dco support for Linux])
+			;;
+		*)
+			AC_MSG_NOTICE([Ignoring --enable-dco on non Linux platform])
+			;;
+	esac
+fi
+
 if test "${with_crypto_library}" = "openssl"; then
 	AC_ARG_VAR([OPENSSL_CFLAGS], [C compiler flags for OpenSSL])
 	AC_ARG_VAR([OPENSSL_LIBS], [linker flags for OpenSSL])
@@ -1196,6 +1229,7 @@ fi
 AM_CONDITIONAL([HAVE_SITNL], [false])
 
 if test "${enable_iproute2}" = "yes"; then
+	test "${enable_dco}" = "yes" && AC_MSG_ERROR([iproute2 support cannot be enabled when using DCO])
 	test -z "${IPROUTE}" && AC_MSG_ERROR([ip utility is required but missing])
 	AC_DEFINE([ENABLE_IPROUTE], [1], [enable iproute2 support])
 else if test "${have_sitnl}" = "yes"; then
dev-tools/special-files.lst +1 -0
index 64ee9e1a..33e830d7 100644
@@ -1,3 +1,4 @@
 E:doc/doxygen/doc_key_generation.h     # @verbatim section gets mistreated, exclude it
 E:src/compat/compat-lz4.c              # Preserve LZ4 upstream formatting
 E:src/compat/compat-lz4.h              # Preserve LZ4 upstream formatting
+E:src/openvpn/ovpn_dco_linux.h         # Preserve ovpn-dco upstream formatting
src/openvpn/Makefile.am +3 -0
index 57729480..91635b67 100644
@@ -53,6 +53,8 @@ openvpn_SOURCES = \
 	crypto.c crypto.h crypto_backend.h \
 	crypto_openssl.c crypto_openssl.h \
 	crypto_mbedtls.c crypto_mbedtls.h \
+	dco.h dco_internal.h \
+	dco_linux.c dco_linux.h \
 	dhcp.c dhcp.h \
 	dns.c dns.h \
 	env_set.c env_set.h \
@@ -75,6 +77,7 @@ openvpn_SOURCES = \
 	mbuf.c mbuf.h \
 	memdbg.h \
 	misc.c misc.h \
+	ovpn_dco_linux.h \
 	platform.c platform.h \
 	console.c console.h console_builtin.c console_systemd.c \
 	mroute.c mroute.h \
src/openvpn/dco.h new file +165 -0

165 lines not on this page · raw message

src/openvpn/dco_internal.h new file +78 -0

78 lines not on this page · raw message

src/openvpn/dco_linux.c new file +934 -0

934 lines not on this page · raw message

src/openvpn/dco_linux.h new file +60 -0

60 lines not on this page · raw message

src/openvpn/errlevel.h +2 -0
index e616a496..5bb1e65e 100644
@@ -91,6 +91,7 @@
 #define D_OSBUF              LOGLEV(3, 43, 0)        /* show socket/tun/tap buffer sizes */
 #define D_PS_PROXY           LOGLEV(3, 44, 0)        /* messages related to --port-share option */
 #define D_IFCONFIG           LOGLEV(3, 0,  0)        /* show ifconfig info (don't mute) */
+#define D_DCO                LOGLEV(3, 0, 0)         /* show DCO related messages */
 
 #define D_SHOW_PARMS         LOGLEV(4, 50, 0)        /* show all parameters on program initiation */
 #define D_SHOW_OCC           LOGLEV(4, 51, 0)        /* show options compatibility string */
@@ -114,6 +115,7 @@
 #define D_TAP_WIN_DEBUG      LOGLEV(6, 69, M_DEBUG)  /* show TAP-Windows driver debug info */
 #define D_CLIENT_NAT         LOGLEV(6, 69, M_DEBUG)  /* show client NAT debug info */
 #define D_XKEY               LOGLEV(6, 69, M_DEBUG)  /* show xkey-provider debug info */
+#define D_DCO_DEBUG          LOGLEV(6, 69, M_DEBUG)  /* show DCO related lowlevel debug messages */
 
 #define D_SHOW_KEYS          LOGLEV(7, 70, M_DEBUG)  /* show data channel encryption keys */
 #define D_SHOW_KEY_SOURCE    LOGLEV(7, 70, M_DEBUG)  /* show data channel key source entropy */
src/openvpn/openvpn.vcxproj +6 -1
index 860ef892..bc1a0300 100644
@@ -276,9 +276,10 @@
     <ClCompile Include="crypto.c" />
     <ClCompile Include="crypto_openssl.c" />
     <ClCompile Include="cryptoapi.c" />
-    <ClCompile Include="env_set.c" />
+    <ClCompile Include="dco_linux.c" />
     <ClCompile Include="dhcp.c" />
     <ClCompile Include="dns.c" />
+    <ClCompile Include="env_set.c" />
     <ClCompile Include="error.c" />
     <ClCompile Include="event.c" />
     <ClCompile Include="fdmisc.c" />
@@ -362,6 +363,9 @@
     <ClInclude Include="crypto_backend.h" />
     <ClInclude Include="crypto_openssl.h" />
     <ClInclude Include="cryptoapi.h" />
+    <ClInclude Include="dco.h" />
+    <ClInclude Include="dco_internal.h" />
+    <ClInclude Include="dco_linux.h" />
     <ClInclude Include="dhcp.h" />
     <ClInclude Include="dns.h" />
     <ClInclude Include="env_set.h" />
@@ -396,6 +400,7 @@
     <ClInclude Include="openvpn.h" />
     <ClInclude Include="options.h" />
     <ClInclude Include="otime.h" />
+    <ClInclude Include="ovpn_dco_linux.h" />
     <ClInclude Include="packet_id.h" />
     <ClInclude Include="perf.h" />
     <ClInclude Include="ping.h" />
src/openvpn/openvpn.vcxproj.filters +15 -0
index f76e5923..3c21a4c6 100644
@@ -36,6 +36,9 @@
     <ClCompile Include="cryptoapi.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="dco_linux.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="dhcp.c">
       <Filter>Source Files</Filter>
     </ClCompile>
@@ -299,6 +302,15 @@
     <ClInclude Include="cryptoapi.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClCompile Include="dco.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="dco_internal.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
+    <ClInclude Include="dco_linux.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
     <ClInclude Include="dhcp.h">
       <Filter>Header Files</Filter>
     </ClInclude>
@@ -398,6 +410,9 @@
     <ClInclude Include="otime.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="ovpn_dco_linux.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
     <ClInclude Include="packet_id.h">
       <Filter>Header Files</Filter>
     </ClInclude>
src/openvpn/ovpn_dco_linux.h new file +265 -0

265 lines not on this page · raw message

src/openvpn/tun.h +3 -0
index 4bc35916..60cd574d 100644
@@ -40,6 +40,7 @@
 #include "misc.h"
 #include "networking.h"
 #include "ring_buffer.h"
+#include "dco.h"
 
 #ifdef _WIN32
 #define WINTUN_COMPONENT_ID "wintun"
@@ -214,6 +215,8 @@ struct tuntap
 #endif
     /* used for printing status info only */
     unsigned int rwflags_debug;
+
+    dco_context_t dco;
 };
 
 static inline bool
Signature · 1 line
2.35.1
#3 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: introduce low-level code for handling ovpn-dco in the Linux kernel

In reply to Antonio Quartulli

I have stared at this code before, and it "seems to be reasonable".

What I *can* attest is that it does not break existing "master" functionality on client or server, Linux or FreeBSD or "make distcheck" (autoconf changes, new modules, ...). I did not expect anything, but this was sort of the risk at this point.

Calling configure with --enable-dco does compile the new stuff, so it is at least "compile safe" (and no warnings).

Some review comments that could be fixed in a followup patch:

- some of the functions (like "ovpn_dco_nlmsg_create()", "dco_new_peer()") could definitely use a bit of commenting so non-netlink-experts would understand what these functions do, and why... (and I'm sure I mentioned that before).

- dco_new_peer() treats "localaddr" and "remoteaddr" differently when packing, which looks weird -> I think a comment that this is to reduce kernel code ("we need full sockaddr for the remote, and only the local IP address for local") would avoid confusion.

- openvpn_dco_init_netlink() carries this TODO

+ /* TODO: Why are we setting this buffer size? */

... so, someone should know, and add an appropriate comment :-)

- a few words of explanation on "... multicast message sent by the ovpn-dco kernel module" might be good - I see you subscribe to it, and explain mcast_family_handler(), but I can not find any indication of "what sort of messages would we expect to see here?".

- ovpn_dco_nlmsg_create() has

+ if (!nl_msg)
+ {
+ msg(M_ERR, "cannot allocate netlink message");
+ return NULL;
+ }

... which never returns, as M_ERR is fatal. OTOH, *some* callers do

+    struct nl_msg *nl_msg = ovpn_dco_nlmsg_create(dco, OVPN_CMD_REGISTER_PACKET);
+    if (!nl_msg)
+    {

+ msg(M_ERR, "%s: cannot allocate message to register for control packets",

... but not *all* of them (dco_new_peer() does not check nl_msg at all).

So this should be made consistent - like, either ovpn_dco_nlmsg_create() will M_ERR on allocation failure, or return NULL (and never M_ERR). If it does M_ERR itself, no need to check that in the callers...

- open_tun_dco() reports an *error* top create DCO interface with D_DCO_DEBUG

+ msg(D_DCO_DEBUG, "Cannot create DCO interface %s: %d", dev, ret);

... and two lines down, an error on if_nametoindex() leads to a FATAL

+ msg(M_FATAL, "DCO: cannot retrieve ifindex for interface %s", dev);

... so what? Is this function allowed to fail, and return "ret < 0", or should it M_ERR/M_FATAL on errors? There is no pre-function comment that could clarify the intent...

- dco_new_key() has code to support "none"

+ if (dco_cipher != OVPN_CIPHER_ALG_NONE)
+ {

... but dco_cipher can never be OVPN_CIPHER_ALG_NONE anymore... and if that indent is removed, the NLA_PUT calls for _iv do not need to wrap :-)

- dco_set_peer() does not pass on the mss yet

- in ovpn_handle_msg(), can this be called uninitialized?

+ if (!attrs[OVPN_ATTR_IFINDEX])
+ {
+ msg(D_DCO, "ovpn-dco: Received message without ifindex");
+ return NL_SKIP;
+ }

(not sure if nla_parse() will ensure that there is enough attributes, or just "well-formed" - and whether *attrs actually gets zeroed by nla_parse(), as openvpn_handle_msg() has no CLEAR() )

- same function - under which conditions can this fire? Kernel getting confused?

+    uint32_t ifindex = nla_get_u32(attrs[OVPN_ATTR_IFINDEX]);
+    if (ifindex != dco->ifindex)
+    {

+ msg(D_DCO, "ovpn-dco: received message type %d with mismatched ifindex %d\n",

Uncrustify has complained at me when I merged the patch (because in that moment, ovpn_dco_linux.h was "newly modified" and the exclusion rule does not match on the pre-commit-hook) - but as discussed, this is a bit complicated due to "kernel style" vs "openvpn style", so I've left it alone for the moment.

Your patch has been applied to the master branch.

commit e34437c26b764851555e4acbe2ccca6bec235c7e

Author: Antonio Quartulli
Date:   Fri Jun 24 10:37:45 2022 +0200
     dco: introduce low-level code for handling ovpn-dco in the Linux kernel

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24512.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#4 lore raw

Re: [Openvpn-devel] [PATCH applied] Re: dco: introduce low-level code for handling ovpn-dco in the Linux kernel

Quoted 5 lines · On Dienstag, 28. Juni 2022 17:07:14 CEST Gert Doering wrote:

Uncrustify has complained at me when I merged the patch (because in that moment, ovpn_dco_linux.h was "newly modified" and the exclusion rule does not match on the pre-commit-hook) - but as discussed, this is a bit complicated due to "kernel style" vs "openvpn style", so I've left it alone for the moment.

That's something which needs to be improved. Who is dev-tools/special-files.lst intended for? Can uncrustify read it?

Heiko

#5 lore raw

Re: [Openvpn-devel] [PATCH applied] Re: dco: introduce low-level code for handling ovpn-dco in the Linux kernel

Hi,

Quoted 9 lines · Heiko Hund
Quoted 5 lines · On Dienstag, 28. Juni 2022 17:07:14 CEST Gert Doering wrote:

Uncrustify has complained at me when I merged the patch (because in that moment, ovpn_dco_linux.h was "newly modified" and the exclusion rule does not match on the pre-commit-hook) - but as discussed, this is a bit complicated due to "kernel style" vs "openvpn style", so I've left it alone for the moment.

That's something which needs to be improved. Who is dev-tools/special-files.lst intended for? Can uncrustify read it?

We have dev-tools/reformat-all.sh which was used for the "Grand Reformat" commits - that one reads the special-files.lst, which can contain "E: <exclude this file>" or some other magic stuff for "pre/post-patch because uncrustify is stupid".

Having the pre-commit-hook honour the E: exclusion lines would certainly make handling these files a bit easier - but then, I do not expect updates to them very often, so this is not super high priority.

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#6 lore raw

PATCH 2/25 dco: add helper function to detect if DCO is enabled or not

In reply to Antonio Quartulli
---
 src/openvpn/options.h | 20 ++++++++++++++++++++
 src/openvpn/tun.h     |  1 +
 2 files changed, 21 insertions(+)
src/openvpn/options.h +20 -0
index c2937dc3..8152e755 100644
@@ -879,4 +879,24 @@ void options_string_import(struct options *options,
 
 bool key_is_external(const struct options *options);
 
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+
+/**
+ * Returns whether the current configuration has dco enabled.
+ */
+static inline bool
+dco_enabled(const struct options *o)
+{
+    return !o->tuntap_options.disable_dco;
+}
+
+#else /* if defined(ENABLE_DCO) && defined(TARGET_LINUX) */
+
+static inline bool
+dco_enabled(const struct options *o)
+{
+    return false;
+}
+
+#endif
 #endif /* ifndef OPTIONS_H */
src/openvpn/tun.h +1 -0
index 60cd574d..5fcea590 100644
@@ -139,6 +139,7 @@ struct tuntap_options {
 
 struct tuntap_options {
     int txqueuelen;
+    bool disable_dco;
 };
 
 #else  /* if defined(_WIN32) || defined(TARGET_ANDROID) */
Signature · 1 line
2.35.1
#8 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: add helper function to detect if DCO is enabled or not

In reply to Antonio Quartulli

Trivial enough :-) - compile tested on Linux (with --enable-dco) and FreeBSD.

Your patch has been applied to the master branch.

commit 4f1671fabdce2df94f01e63ff97191325c5bfa8c

Author: Antonio Quartulli
Date:   Fri Jun 24 10:37:46 2022 +0200
     dco: add helper function to detect if DCO is enabled or not

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24513.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#9 lore raw

Re: [Openvpn-devel] [PATCH 02/25] dco: add helper function to detect if DCO is enabled or not

In reply to Antonio Quartulli

src/openvpn/options.h void options_string_import(struct options *options, · line 891

Quoted 8 lines · On Freitag, 24. Juni 2022 10:37:46 CEST Antonio Quartulli wrote:
src/openvpn/options.h +8 -0
+/**
+ * Returns whether the current configuration has dco enabled.
+ */
+static inline bool
+dco_enabled(const struct options *o)
+{
+    return !o->tuntap_options.disable_dco;
+}

I think it would be beneficial if this took a tun_options* instead, as it could be used more widely and replace the many "!tt->options.disable_dco" occurrences and alike.

#10 lore raw

PATCH 3/25 dco: use specific metric when installing routes

In reply to Antonio Quartulli

When using DCO iroutes and routes all live in the same routing table, However, the latter should always come after the former.

for this reason assign a default metric of 200 to routes. iroutes will later get a metric of 100.

---
 src/openvpn/dco.h  |  2 ++
 src/openvpn/init.c | 10 ++++++++++
 2 files changed, 12 insertions(+)
src/openvpn/dco.h +2 -0
index dcadba78..063e5028 100644
@@ -38,6 +38,8 @@ struct event_set;
 struct options;
 struct tuntap;
 
+#define DCO_DEFAULT_METRIC  200
+
 #if defined(ENABLE_DCO)
 
 /**
src/openvpn/init.c +10 -0
index 638fd10c..f586bfc7 100644
@@ -1385,6 +1385,11 @@ do_init_route_list(const struct options *options,
     int dev = dev_type_enum(options->dev, options->dev_type);
     int metric = 0;
 
+    if (dco_enabled(options))
+    {
+        metric = DCO_DEFAULT_METRIC;
+    }
+
     if (dev == DEV_TYPE_TUN && (options->topology == TOP_NET30 || options->topology == TOP_P2P))
     {
         gw = options->ifconfig_remote_netmask;
@@ -1421,6 +1426,11 @@ do_init_route_ipv6_list(const struct options *options,
     const char *gw = NULL;
     int metric = -1;            /* no metric set */
 
+    if (dco_enabled(options))
+    {
+        metric = DCO_DEFAULT_METRIC;
+    }
+
     gw = options->ifconfig_ipv6_remote;         /* default GW = remote end */
     if (options->route_ipv6_default_gateway)
     {
Signature · 1 line
2.35.1
Acked-by on v1 3/25
#11 lore raw

Re: [Openvpn-devel] [PATCH 03/25] dco: use specific metric when installing routes

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 5 lines

When using DCO iroutes and routes all live in the same routing table, However, the latter should always come after the former.

for this reason assign a default metric of 200 to routes. iroutes will later get a metric of 100.

Can you also add documentation of this? (if it is not coming in a later patch)

#12 lore raw

PATCH v2 3/25 dco: use specific metric when installing routes

In reply to Antonio Quartulli

When using DCO iroutes and routes all live in the same routing table, However, the latter should always come after the former.

for this reason assign a default metric of 200 to routes. iroutes will later get a metric of 100.

---

Changes from v1:
* added comments before "(if (dco_enabled(c))" to explain why we need a different metric

 src/openvpn/dco.h  |  2 ++
 src/openvpn/init.c | 15 +++++++++++++++
 2 files changed, 17 insertions(+)
src/openvpn/dco.h +2 -0
index dcadba78..063e5028 100644
@@ -38,6 +38,8 @@ struct event_set;
 struct options;
 struct tuntap;
 
+#define DCO_DEFAULT_METRIC  200
+
 #if defined(ENABLE_DCO)
 
 /**
src/openvpn/init.c +15 -0
index 638fd10c..03221cbb 100644
@@ -1385,6 +1385,15 @@ do_init_route_list(const struct options *options,
     int dev = dev_type_enum(options->dev, options->dev_type);
     int metric = 0;
 
+    /* if DCO is enabled we have both regular routes and iroutes in the system
+     * routing table, and normal routes must have a higher metric for that to
+     * work so that iroutes are always matched first
+     */
+    if (dco_enabled(options))
+    {
+        metric = DCO_DEFAULT_METRIC;
+    }
+
     if (dev == DEV_TYPE_TUN && (options->topology == TOP_NET30 || options->topology == TOP_P2P))
     {
         gw = options->ifconfig_remote_netmask;
@@ -1421,6 +1430,12 @@ do_init_route_ipv6_list(const struct options *options,
     const char *gw = NULL;
     int metric = -1;            /* no metric set */
 
+    /* see explanation in do_init_route_list() */
+    if (dco_enabled(options))
+    {
+        metric = DCO_DEFAULT_METRIC;
+    }
+
     gw = options->ifconfig_ipv6_remote;         /* default GW = remote end */
     if (options->route_ipv6_default_gateway)
     {
Signature · 1 line
2.35.1
Acked-by on v2 3/25
#13 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: use specific metric when installing routes

Arne has already ACKed the functionality (and I'm fine with that), and I ACK the extra comments. We could spend days on perfect wordsmithing here, but I think they are good enough.

Please do not forget to include the metric thing in the "overall documentation" patch that was mentioned :-)

Your patch has been applied to the master branch.

commit 90d9c38b116b8c816596d71a5cf6db7fd068d61a

Author: Antonio Quartulli
Date:   Tue Jun 28 20:56:23 2022 +0200
     dco: use specific metric when installing routes

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24599.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#14 lore raw

PATCH 4/25 dco: create DCO interface using SITNL

In reply to Antonio Quartulli
---
 src/openvpn/networking_sitnl.c | 11 +++++++++++
 1 file changed, 11 insertions(+)
src/openvpn/networking_sitnl.c +11 -0
index bffcb067..0944ad0a 100644
@@ -28,6 +28,7 @@
 
 #include "syshead.h"
 
+#include "dco.h"
 #include "errlevel.h"
 #include "buffer.h"
 #include "misc.h"
@@ -1344,6 +1345,16 @@ net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type,
 
     struct rtattr *linkinfo = SITNL_NEST(&req.n, sizeof(req), IFLA_LINKINFO);
     SITNL_ADDATTR(&req.n, sizeof(req), IFLA_INFO_KIND, type, strlen(type) + 1);
+#if defined(ENABLE_DCO)
+    if (arg && (strcmp(type, "ovpn-dco") == 0))
+    {
+        dco_context_t *dco = arg;
+        struct rtattr *data = SITNL_NEST(&req.n, sizeof(req), IFLA_INFO_DATA);
+        SITNL_ADDATTR(&req.n, sizeof(req), IFLA_OVPN_MODE, &dco->ifmode,
+                      sizeof(uint8_t));
+        SITNL_NEST_END(&req.n, data);
+    }
+#endif
     SITNL_NEST_END(&req.n, linkinfo);
 
     req.i.ifi_family = AF_PACKET;
Signature · 1 line
2.35.1
Acked-by on v1 4/25
#15 lore raw

Re: [Openvpn-devel] [PATCH 04/25] dco: create DCO interface using SITNL

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 24 lines
---
  src/openvpn/networking_sitnl.c | 11 +++++++++++
  1 file changed, 11 insertions(+)
src/openvpn/networking_sitnl.c +4 -0
index bffcb067..0944ad0a 100644
@@ -28,6 +28,7 @@
 
  #include "syshead.h"
 
+#include "dco.h"
  #include "errlevel.h"
  #include "buffer.h"
  #include "misc.h"
@@ -1344,6 +1345,16 @@ net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type,
 
      struct rtattr *linkinfo = SITNL_NEST(&req.n, sizeof(req), IFLA_LINKINFO);
      SITNL_ADDATTR(&req.n, sizeof(req), IFLA_INFO_KIND, type, strlen(type) + 1);
+#if defined(ENABLE_DCO)
+    if (arg && (strcmp(type, "ovpn-dco") == 0))
+    {

src/openvpn/networking_sitnl.c net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type, · line 1360

Quoted 10 lines
src/openvpn/networking_sitnl.c +7 -0
+        dco_context_t *dco = arg;
+        struct rtattr *data = SITNL_NEST(&req.n, sizeof(req), IFLA_INFO_DATA);
+        SITNL_ADDATTR(&req.n, sizeof(req), IFLA_OVPN_MODE, &dco->ifmode,
+                      sizeof(uint8_t));
+        SITNL_NEST_END(&req.n, data);
+    }
+#endif
      SITNL_NEST_END(&req.n, linkinfo);
 
      req.i.ifi_family = AF_PACKET;
#16 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: create DCO interface using SITNL

In reply to Antonio Quartulli

This does not really do anything yet, and it won't affect "non dco" builds. For DCO builds, it looks reasonable and compiles without warning (and all it really does is "if dco, then add IFLA_OVPN_MODE" to the netlink message, passing in p2p / p2mp mode)

Your patch has been applied to the master branch.

commit 2a02c90a4501b04a43a25824fe9f7b980e10fc40

Author: Antonio Quartulli
Date:   Fri Jun 24 10:37:48 2022 +0200
     dco: create DCO interface using SITNL

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24514.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#17 lore raw

PATCH 5/25 dco: let open_tun_generic handle the DCO case

In reply to Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---
 src/openvpn/init.c |   2 +-
 src/openvpn/tun.c  | 133 +++++++++++++++++++++++++++++++++------------
 src/openvpn/tun.h  |   2 +-
 3 files changed, 101 insertions(+), 36 deletions(-)
src/openvpn/init.c +1 -1
index f586bfc7..7099eba4 100644
@@ -1758,7 +1758,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/tun.c +99 -34

282 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1

A message that is not in the archive (sent off-list or not yet received): [email protected]...

#19 lore raw

Re: [Openvpn-devel] [PATCH 05/25] dco: let open_tun_generic handle the DCO case

Hi,

Quoted 11 lines · Arne Schwabe

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 8 lines

+                int ret = open_tun_dco(tt, ctx, dynamic_name); +                if (ret == -EEXIST)
+                {
+                    msg(M_INFO, "TUN/TAP device %s exists previously, keep at program end",
+                        dynamic_name);
+                    tt->persistent_if = true;
+                }

Quoted 1 line

I think here the error message is really confusing. What does "keep at program end" mean? And if we fail at opening DCO tun device why do we talk about a TUN/TAP device and not about DCO?

IIRC I simply took this message from "the same scenario but without DCO", so the text was already bad :-D

But I agree we can improve it.

Thanks!

Quoted 3 lines

Arne

Signature · 1 line
Antonio Quartulli
#20 lore raw

Re: [Openvpn-devel] [PATCH 05/25] dco: let open_tun_generic handle the DCO case

In reply to Antonio Quartulli

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

src/openvpn/tun.c open_tun_generic(const char *dev, const char *dev_type, const char *dev_node, · line 1830

Quoted 7 lines
src/openvpn/tun.c +7 -0
+                int ret = open_tun_dco(tt, ctx, dynamic_name);
+                if (ret == -EEXIST)
+                {
+                    msg(M_INFO, "TUN/TAP device %s exists previously, keep at program end",
+                        dynamic_name);
+                    tt->persistent_if = true;
+                }

I think here the error message is really confusing. What does "keep at program end" mean? And if we fail at opening DCO tun device why do we talk about a TUN/TAP device and not about DCO?

Arne

#21 lore raw

PATCH v2 5/25 dco: let open_tun_generic handle the DCO case

In reply to Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c |   2 +-
 src/openvpn/tun.c  | 133 +++++++++++++++++++++++++++++++++------------
 src/openvpn/tun.h  |   2 +-
 3 files changed, 101 insertions(+), 36 deletions(-)
src/openvpn/init.c +1 -1
index 03221cbb..c9d05c31 100644
@@ -1763,7 +1763,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/tun.c +99 -34

282 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#22 lore raw

PATCH v3 1/1 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |   2 +-
 src/openvpn/options.c |   7 +++
 src/openvpn/tun.c     | 137 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 4 files changed, 113 insertions(+), 35 deletions(-)
src/openvpn/init.c +1 -1
index 03221cbb..c9d05c31 100644
@@ -1763,7 +1763,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +104 -33

277 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#23 lore raw

PATCH v4 1/1 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |   2 +-
 src/openvpn/options.c |   7 +++
 src/openvpn/tun.c     | 141 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 4 files changed, 116 insertions(+), 36 deletions(-)
src/openvpn/init.c +1 -1
index 03221cbb..c9d05c31 100644
@@ -1763,7 +1763,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +107 -34

280 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#24 lore raw

Re: [Openvpn-devel] [PATCH v4] dco: let open_tun_generic handle the DCO case

Quoted 1 line · On Mittwoch, 29. Juni 2022 14:49:45 CEST Antonio Quartulli wrote:

name when not specified b the user. For this reason the DCO case can

nit: b -> by

src/openvpn/tun.c open_tun_generic(const char *dev, const char *dev_type, const char *dev_node, · line 1793

Quoted 2 lines
src/openvpn/tun.c +2 -0
+                            strncpynt(tunname, dynamic_name,
+                                      sizeof(dynamic_name));

This need to be sizeof(tunname).

#25 lore raw

Re: [Openvpn-devel] [PATCH v4] dco: let open_tun_generic handle the DCO case

Quoted 4 lines · Heiko Hund
Quoted 1 line · On Mittwoch, 29. Juni 2022 14:49:45 CEST Antonio Quartulli wrote:

name when not specified b the user. For this reason the DCO case can

nit: b -> by

thanks

Quoted 5 lines

src/openvpn/tun.c open_tun_generic(const char *dev, const char *dev_type, const char *dev_node, · line 1793

Quoted 3 lines
src/openvpn/tun.c +2 -0
+                            strncpynt(tunname, dynamic_name,
+                                      sizeof(dynamic_name));

This need to be sizeof(tunname).

ouch, good catch! Thanks!

Signature · 1 line
Antonio Quartulli
#26 lore raw

PATCH v5 5/25 dco: let open_tun_generic handle the DCO case

In reply to Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |   2 +-
 src/openvpn/options.c |   7 +++
 src/openvpn/tun.c     | 140 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 4 files changed, 115 insertions(+), 36 deletions(-)
src/openvpn/init.c +1 -1
index 03221cbb..c9d05c31 100644
@@ -1763,7 +1763,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +106 -34

279 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#27 lore raw

PATCH 1/1 networking: add net_iface_type API

This new API can be used to retrieve the type of a specific interface. It's mostly platform dependant, but right now expected values are "ovpn-dco", "tun" or "tap".

Other values are possible too, but they are not of interest to us.

---

This patch is required by 05/25 v6 as it introduces the networking API required to retrieve an interafce type. This is then used by 05/25 v6.

 src/openvpn/networking.h          |  14 ++++
 src/openvpn/networking_iproute2.c |   9 +++
 src/openvpn/networking_sitnl.c    | 109 ++++++++++++++++++++++++++++++
 3 files changed, 132 insertions(+)
src/openvpn/networking.h +14 -0
index 79963756..cf6d39ac 100644
@@ -23,6 +23,8 @@
 
 #include "syshead.h"
 
+#define IFACE_TYPE_LEN_MAX 64
+
 struct context;
 
 #ifdef ENABLE_SITNL
@@ -100,6 +102,18 @@ void net_ctx_free(openvpn_net_ctx_t *ctx);
 int net_iface_new(openvpn_net_ctx_t *ctx, const openvpn_net_iface_t *iface,
                   const char *type, void *arg);
 
+/**
+ * Retrieve the interface type
+ *
+ * @param ctx       the implementation specific context
+ * @param iface     interface to query
+ * @param type      buffer where the type will be stored
+ *
+ * @return          0 on success, a negative error code otherwise
+ */
+int net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+                   char type[IFACE_TYPE_LEN_MAX]);
+
 /**
  * Remove an interface
  *
src/openvpn/networking_iproute2.c +9 -0
index 4b220576..a81e6b33 100644
@@ -78,6 +78,15 @@ net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type,
     return 0;
 }
 
+int
+net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+               char type[IFACE_TYPE_LEN_MAX])
+{
+    /* not supported by iproute2 */
+    msg(M_WARN, "%s: operation not supported by iproute2 backend");
+    return -1;
+}
+
 int
 net_iface_del(openvpn_net_ctx_t *ctx, const char *iface)
 {
src/openvpn/networking_sitnl.c +109 -0

115 lines not on this page · raw message

Signature · 1 line
2.35.1
#28 lore raw

PATCH v6 5/25 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v5:
* create TUN device when invoking --mktun with DCO enabled. Current code segfaults.
* fix message printing "..device created" for DCO cases
* print error string (and not just error code) when a DCO iface cannot be created
* when creating DCO device, if a TUN device with the same name is found, disable DCO and use the existing device
* don't use tunname as it contains the wrong string. Use dev or dynamic_name instead

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |  30 +++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 ++
 src/openvpn/tun.c     | 159 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 5 files changed, 161 insertions(+), 39 deletions(-)
src/openvpn/init.c +28 -2
index 03221cbb..99fcda04 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,13 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
+
+    /* DCO may have been disabled by open_tun(). propagate change */
+    c->options.tuntap_options = c->c1.tuntap->options;
+#if defined(ENABLE_DCO)
+    c->c2.tls_multi->opt.disable_dco = !dco_enabled(&c->options);
+#endif
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +124 -35

303 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#29 lore raw

PATCH v7 1/1 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v6:
* do not touch tls_multi in do_open_tun as it's still NULL (caused crash)
* re-assign disable_dco field in pre_connect object, if DCO is disabled by open_tun() in do_open_tun()

Changes from v5:
* create TUN device when invoking --mktun with DCO enabled. Current code segfaults.
* fix message printing "..device created" for DCO cases
* print error string (and not just error code) when a DCO iface cannot be created
* when creating DCO device, if a TUN device with the same name is found, disable DCO and use the existing device
* don't use tunname as it contains the wrong string. Use dev or dynamic_name instead

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |  31 +++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 ++
 src/openvpn/tun.c     | 159 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 5 files changed, 162 insertions(+), 39 deletions(-)
src/openvpn/init.c +29 -2
index 03221cbb..1adc06c9 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,14 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
+
+    /* DCO may have been disabled by open_tun(). propagate change */
+    c->options.tuntap_options.disable_dco = c->c1.tuntap->options.disable_dco;
+    if (c->options.pre_connect->tuntap_options_defined)
+    {
+        c->options.pre_connect->tuntap_options.disable_dco = c->c1.tuntap->options.disable_dco;
+    }
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +124 -35

303 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#30 lore raw

PATCH v8 5/25 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v7:
* enclose setting 'disable_dco' field within ifdef (in do_open_tun) to ensure code compiles also on platforms with no DCO support at all

Changes from v6:
* do not touch tls_multi in do_open_tun as it's still NULL (caused crash)
* re-assign disable_dco field in pre_connect object, if DCO is disabled by open_tun() in do_open_tun()

Changes from v5:
* create TUN device when invoking --mktun with DCO enabled. Current code segfaults.
* fix message printing "..device created" for DCO cases
* print error string (and not just error code) when a DCO iface cannot be created
* when creating DCO device, if a TUN device with the same name is found, disable DCO and use the existing device
* don't use tunname as it contains the wrong string. Use dev or dynamic_name instead

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |  33 ++++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 ++
 src/openvpn/tun.c     | 159 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 5 files changed, 164 insertions(+), 39 deletions(-)
src/openvpn/init.c +31 -2
index 03221cbb..8b952552 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,16 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
+
+#if defined(ENABLE_DCO)
+    /* DCO may have been disabled by open_tun(). propagate change */
+    c->options.tuntap_options.disable_dco = c->c1.tuntap->options.disable_dco;
+    if (c->options.pre_connect->tuntap_options_defined)
+    {
+        c->options.pre_connect->tuntap_options.disable_dco = c->c1.tuntap->options.disable_dco;
+    }
+#endif
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +124 -35

303 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#31 lore raw

PATCH v9 5/25 dco: let open_tun_generic handle the DCO case

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v8:
* remove disabling DCO from within do_open_tun(). Logic has been moved to dco_check_option* (in a future patch).

Changes from v7:
* enclose setting 'disable_dco' field within ifdef (in do_open_tun) to ensure code compiles also on platforms with no DCO support at all

Changes from v6:
* do not touch tls_multi in do_open_tun as it's still NULL (caused crash)
* re-assign disable_dco field in pre_connect object, if DCO is disabled by open_tun() in do_open_tun()

Changes from v5:
* create TUN device when invoking --mktun with DCO enabled. Current code segfaults.
* fix message printing "..device created" for DCO cases
* print error string (and not just error code) when a DCO iface cannot be created
* when creating DCO device, if a TUN device with the same name is found, disable DCO and use the existing device
* don't use tunname as it contains the wrong string. Use dev or dynamic_name instead

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |  24 ++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 ++
 src/openvpn/tun.c     | 144 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 5 files changed, 140 insertions(+), 39 deletions(-)
src/openvpn/init.c +22 -2
index 03221cbb..4feb578a 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index 233c02e0..705bb79a 100644
@@ -3528,6 +3528,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +109 -35

288 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index 5fcea590..cf02bf43 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#32 lore raw

Re: [Openvpn-devel] [PATCH v9 05/25] dco: let open_tun_generic handle the DCO case

Hi,

Quoted 5 lines · Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

NAK... I'm not sure what is happening, but if I apply this on top of current master, and build without(!) DCO, it will kill linux/sitnl operation hard for the "--dev tun30" case.

Running "openvpn --client ... --dev tun30" leads to

2022-07-14 15:59:43 Incoming Data Channel: Cipher 'AES-256-GCM' initialized with 256 bit key
2022-07-14 15:59:43 net_route_v4_best_gw query: dst 0.0.0.0
2022-07-14 15:59:43 net_route_v4_best_gw result: via 194.97.140.30 dev eno33554952
2022-07-14 15:59:43 GDG6: remote_host_ipv6=2607:fc50:1001:5200::4
2022-07-14 15:59:43 net_route_v6_best_gw query: dst 2607:fc50:1001:5200::4
2022-07-14 15:59:43 net_route_v6_best_gw result: via 2001:608:0:814::ffff dev eno33554952
2022-07-14 15:59:43 DCO device tun30 opened
2022-07-14 15:59:43 net_iface_mtu_set: rtnl: cannot get ifindex for tun30: No such device (errno=19)
2022-07-14 15:59:43 Linux can't set mtu (1500) on tun30
2022-07-14 15:59:43 Exiting due to fatal error

Building with --enable-dco breaks all operations on linux (because it gets confused between DCO and non DCO operation modes)...

2022-07-14 16:04:05 Assertion failed at dco_linux.c:453 (tt->type == DEV_TYPE_TUN)

and sometimes

2022-07-14 16:06:26 sitnl_send: rtnl: generic error (-95): Operation not supported
2022-07-14 16:06:26 net_iface_new: add tun255 type ovpn-dco
2022-07-14 16:06:26 sitnl_send: rtnl: generic error (-95): Operation not supported
2022-07-14 16:06:26 Cannot allocate TUN/TAP dev dynamically

(isn't "-95" an indication of "no DCO module available" and we should not try to iterate up to tun255 in this case, but give up with a clear error indication instead?)

thus

Test sets succeeded: none.
Test sets failed: 1 1a 1b 1c 1d 1e 2 2a 2b 2c 2d 2e 2f 3 4 4a 4b 5 6 8 8a 9 9a 9b 9x 11 11a.

... but since this is not default, I do not consider this a "this patch breaks things" problem. It compiles.

There is no --disable-dco option yet, so hard to test "would it work in that case?" - so I consider this part not really interesting yet.

But the "not built with DCO, --dev tun30" part should really not fail.

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#33 lore raw

PATCH v10 5/25 dco: let open_tun_generic handle the DCO case

In reply to Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

---

Changes from v9:
* rebased on top of latest master

Changes from v8:
* remove disabling DCO from within do_open_tun(). Logic has been moved to dco_check_option* (in a future patch).

Changes from v7:
* enclose setting 'disable_dco' field within ifdef (in do_open_tun) to ensure code compiles also on platforms with no DCO support at all

Changes from v6:
* do not touch tls_multi in do_open_tun as it's still NULL (caused crash)
* re-assign disable_dco field in pre_connect object, if DCO is disabled by open_tun() in do_open_tun()

Changes from v5:
* create TUN device when invoking --mktun with DCO enabled. Current code segfaults.
* fix message printing "..device created" for DCO cases
* print error string (and not just error code) when a DCO iface cannot be created
* when creating DCO device, if a TUN device with the same name is found, disable DCO and use the existing device
* don't use tunname as it contains the wrong string. Use dev or dynamic_name instead

Changes from v4:
* in open_tun_generic() use sizeof(tunname) when copying to tunname

Changes from v3:
* explicitly mention "DCO" in message when DCO device is successfully opened, as per Arne's request

Changes from v2:
* do not abuse the dynamic_name variable. Rather use 'dev' when no dynamic name is requested
* ignore dev-node when using DCO, to avoid messing up naming logic in open_tun_generic. dev-node has sense when using DCO
* add comment as to why we need to fill tunname

Changes from v1:
* improved INFO message when device already exists as per Arne's request

 src/openvpn/init.c    |  24 ++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 ++
 src/openvpn/tun.c     | 144 ++++++++++++++++++++++++++++++++----------
 src/openvpn/tun.h     |   2 +-
 5 files changed, 140 insertions(+), 39 deletions(-)
src/openvpn/init.c +22 -2
index 1bfbf4eb..779fc4a5 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index b00acf7e..87d6fc31 100644
@@ -3667,6 +3667,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
     o->tuntap_options.disable_dco = !dco_check_option_conflict(D_DCO, o);
 #endif
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +109 -35

288 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index b7786f46..8ec8f51f 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.35.1
#34 lore raw

Re: [Openvpn-devel] [PATCH v10 05/25] dco: let open_tun_generic handle the DCO case

Hi,

Quoted 5 lines · Antonio Quartulli

open_tun_generic already contains the logic required to find a device name when not specified b the user. For this reason the DCO case can easily leverage on function and avoid code duplication.

*sigh*

First of all, my apologies for letting you jump through all these hoops - I'm NOT doing this to annoy you or drag my feet, but really to keep the code readable and maintainable.

That said, I've spent two days now pondering the latest version of this patch, and I can not ACK it - it's ugly beyond openvpn standards.

My main gripe is that we have this open_tun_generic() function which is used for "all but linux (and Windows)". So you add a call to it now for the DCO call (this is okay-ish) but then it's not actually *using* the code paths in open_tun_generic(), but adding new and linux-exclusive code paths #ifdef TARGET_LINUX.

So, on Linux with DCO, very little of that function is actually used (half of it is #ifdef TARGET_LINUX, and the non-#ifdef parts are "else" branches for "not DCO" - but "for not DCO", we do not call this in the first place) - all that remains is a for() loop to iterate device names, and a string_alloc() call...

That said, I've cooked up a v11 of 05, which introduces a much leaner open_tun_dco_generic() function - to be shared by Linux and FreeBSD - which does "only DCO things", leaving open_tun_generic() alone. I've tested this with the usual test cases

--dev tap (works, because options.c will already disable DCO)
--dev tun (iterates to "tun7" on my DCO test system)
--mktun --dev tun471 ; use --dev tun471 (works, falls back to tun) --dev tun472 (works as DCO device))

Right now this is sitting on top of my clone of your "dco" tree - I'll see that I can apply this to the right parts in the middle and send a 05 v11 out later, then we can haggle on the least ugly way forward.

gert

PS1: I'm fairly sure *I* did the existing #ifdef TARGET_NETBSD in
open_tun_generic() and I should not have done that. Bad precedence.

PS2: 05 v10 works, and does not break anything, But It Is Too Ugly.

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#35 lore raw

PATCH v11 5/25 dco: introduce open_tun_dco_generic() to open dynamic or fixed-name DCO devices

From: Antonio Quartulli <[email protected]...>

This function is similar to the essence of open_tun_generic(), but calling open_tun_dco() instead of trying to do a file open on "/dev/%s"

Previous attempts to save code duplication by including this into open_tun_generic() created additional #ifdef plus confusing call paths. So this is a clean new function, leaving the door open for a cleanup of open_tun_generic().

Also, introduce tun_dco_enabled(tt) to avoid the negative "!tt->options.disable_dco" calls.

--
v11:
- add new function open_tun_dco_generic() for Linux (and FreeBSD, later) instead of lumping this into open_tun_generic()
- pick up tun_dco_enabled() from a later patch in the series (easier to bring this in right now than to convert the code back and then patch it again later)

---
 src/openvpn/init.c    |  24 +++++++-
 src/openvpn/init.h    |   2 +-
 src/openvpn/options.c |   7 +++
 src/openvpn/tun.c     | 129 +++++++++++++++++++++++++++++++++++++-----
 src/openvpn/tun.h     |   2 +-
 5 files changed, 147 insertions(+), 17 deletions(-)
src/openvpn/init.c +22 -2
index 1bfbf4eb..779fc4a5 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
  * Persistent TUN/TAP device management mode?
  */
 bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
 {
     if (options->persist_config)
     {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
             msg(M_FATAL|M_OPTERR,
                 "options --mktun or --rmtun should only be used together with --dev");
         }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
 #ifdef ENABLE_FEATURE_TUN_PERSIST
         tuncfg(options->dev, options->dev_type, options->dev_node,
                options->persist_mode,
@@ -1763,7 +1783,7 @@ do_open_tun(struct context *c)
 #endif
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
     /* set the hardware address */
     if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
 bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
 bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index b00acf7e..87d6fc31 100644
@@ -3667,6 +3667,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
     o->tuntap_options.disable_dco = !dco_check_option_conflict(D_DCO, o);
 #endif
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
src/openvpn/tun.c +116 -13

237 lines not on this page · raw message

src/openvpn/tun.h +1 -1
index b7786f46..8ec8f51f 100644
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
  */
 
 void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
 void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
Signature · 1 line
2.25.1
author Acked-by on v11 5/25
#36 lore raw

Re: [Openvpn-devel] [PATCH v11 05/25] dco: introduce open_tun_dco_generic() to open dynamic or fixed-name DCO devices

Hi,

Quoted 11 lines · Gert Doering

From: Antonio Quartulli <[email protected]...>

This function is similar to the essence of open_tun_generic(), but calling open_tun_dco() instead of trying to do a file open on "/dev/%s"

Previous attempts to save code duplication by including this into open_tun_generic() created additional #ifdef plus confusing call paths. So this is a clean new function, leaving the door open for a cleanup of open_tun_generic().

Quoted 5 lines

Also, introduce tun_dco_enabled(tt) to avoid the negative "!tt->options.disable_dco" calls.

Thanks for providing a new revision of this tremendous patch :-D

Quoted 135 lines

--
v11:
- add new function open_tun_dco_generic() for Linux (and FreeBSD, later) instead of lumping this into open_tun_generic()
- pick up tun_dco_enabled() from a later patch in the series (easier to bring this in right now than to convert the code back and then patch it again later)

---
  src/openvpn/init.c    |  24 +++++++-
  src/openvpn/init.h    |   2 +-
  src/openvpn/options.c |   7 +++
  src/openvpn/tun.c     | 129 +++++++++++++++++++++++++++++++++++++-----
  src/openvpn/tun.h     |   2 +-
  5 files changed, 147 insertions(+), 17 deletions(-)
src/openvpn/init.c +22 -2
index 1bfbf4eb..779fc4a5 100644
@@ -1054,7 +1054,7 @@ do_genkey(const struct options *options)
   * Persistent TUN/TAP device management mode?
   */
  bool
-do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
+do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx)
  {
      if (options->persist_config)
      {
@@ -1069,6 +1069,26 @@ do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx)
              msg(M_FATAL|M_OPTERR,
                  "options --mktun or --rmtun should only be used together with --dev");
          }
+
+#if defined(ENABLE_DCO)
+        if (dco_enabled(options))
+        {
+            /* creating a DCO interface via --mktun is not supported as it does not
+             * make much sense. Since DCO is enabled by default, people may run into
+             * this without knowing, therefore this case should be properly handled.
+             *
+             * Disable DCO if --mktun was provided and print a message to let
+             * user know.
+             */
+            if (dev_type_enum(options->dev, options->dev_type) == DEV_TYPE_TUN)
+            {
+                msg(M_WARN, "Note: --mktun does not support DCO. Creating TUN interface.");
+            }
+
+            options->tuntap_options.disable_dco = true;
+        }
+#endif
+
  #ifdef ENABLE_FEATURE_TUN_PERSIST
          tuncfg(options->dev, options->dev_type, options->dev_node,
                 options->persist_mode,
@@ -1763,7 +1783,7 @@ do_open_tun(struct context *c)
  #endif
      /* open the tun device */
      open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap);
+             c->c1.tuntap, &c->net_ctx);
 
      /* set the hardware address */
      if (c->options.lladdr)
src/openvpn/init.h +1 -1
index 2b8c2dcc..5f412a33 100644
@@ -56,7 +56,7 @@ bool print_openssl_info(const struct options *options);
 
  bool do_genkey(const struct options *options);
 
-bool do_persist_tuntap(const struct options *options, openvpn_net_ctx_t *ctx);
+bool do_persist_tuntap(struct options *options, openvpn_net_ctx_t *ctx);
 
  bool possibly_become_daemon(const struct options *options);
 
src/openvpn/options.c +7 -0
index b00acf7e..87d6fc31 100644
@@ -3667,6 +3667,13 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
      o->tuntap_options.disable_dco = !dco_check_option_conflict(D_DCO, o);
  #endif
 
+    if (dco_enabled(o) && o->dev_node)
+    {
+        msg(M_WARN, "Note: ignoring --dev-node as it has no effect when using "
+            "data channel offload");
+        o->dev_node = NULL;
+    }
+
      /*
       * Save certain parms before modifying options during connect, especially
       * when using --pull
src/openvpn/tun.c +17 -0
index 3f9aa6ae..30abfdd1 100644
@@ -37,6 +37,7 @@
 
  #include "syshead.h"
 
+#include "openvpn.h"
  #include "tun.h"
  #include "fdmisc.h"
  #include "common.h"
@@ -1723,6 +1724,15 @@ tun_name_is_fixed(const char *dev)
      return has_digit(dev);
  }
 
+#if defined(TARGET_LINUX)
+static bool
+tun_dco_enabled(struct tuntap *tt)
+{
+    return !tt->options.disable_dco;
+}
+#endif
+
+
  #if !(defined(_WIN32) || defined(TARGET_LINUX))
  static void
  open_tun_generic(const char *dev, const char *dev_type, const char *dev_node,
@@ -1831,6 +1841,76 @@ open_tun_generic(const char *dev, const char *dev_type, const char *dev_node,
  }
  #endif /* !_WIN32 && !TARGET_LINUX */
 
+#if defined(TARGET_LINUX)
+static void
+open_tun_dco_generic(const char *dev, const char *dev_type,
+                     struct tuntap *tt, openvpn_net_ctx_t *ctx)
+{
+    char dynamic_name[256];
+    bool dynamic_opened = false;

src/openvpn/tun.c open_tun_generic(const char *dev, const char *dev_type, const char *dev_node, · line 1860

Quoted 10 lines
src/openvpn/tun.c +10 -0
+
+    if (tt->type == DEV_TYPE_NULL)
+    {
+        open_null(tt);
+        return;
+    }
+
+    /*
+     * dynamic open is indicated by --dev specified without
+     * explicit unit number.  Try opening /dev/[dev]n

I guess "/dev/[dev]n" should just be "[dev]n" ?

Quoted 215 lines
src/openvpn/tun.c +89 -13
+     * where n = [0, 255].
+     */
+
+    if (!tun_name_is_fixed(dev))
+    {
+        for (int i = 0; i < 256; ++i)
+        {
+            openvpn_snprintf(dynamic_name, sizeof(dynamic_name),
+                             "%s%d", dev, i);
+            if (open_tun_dco(tt, ctx, dynamic_name) == 0)
+            {
+                dynamic_opened = true;
+                msg(M_INFO, "DCO device %s opened", dynamic_name);
+                break;
+            }
+            msg(D_READ_WRITE | M_ERRNO, "Tried opening %s (failed)", dynamic_name);
+        }
+        if (!dynamic_opened)
+        {
+            msg(M_FATAL, "Cannot allocate DCO dev dynamically");
+        }
+        /* tt->actual_name is passed to up and down scripts and used as
+         * the ifconfig dev name */
+        tt->actual_name = string_alloc(dynamic_name, NULL);
+    }
+    /*
+     * explicit unit number specified
+     */
+    else
+    {
+        int ret = open_tun_dco(tt, ctx, dev);
+        if (ret == -EEXIST)
+        {
+            msg(M_INFO, "DCO device %s already exists, won't be destroyed at shutdown",
+                dev);
+            tt->persistent_if = true;
+        }
+        else if (ret < 0)
+        {
+            msg(M_ERR, "Cannot open DCO device %s: %s (%d)", dev,
+                strerror(-ret), ret);
+        }
+        else
+        {
+            msg(M_INFO, "DCO device %s opened", dev);
+        }
+
+        /* tt->actual_name is passed to up and down scripts and used as the ifconfig dev name */
+        tt->actual_name = string_alloc(dev, NULL);
+    }
+}
+#endif /* TARGET_LINUX */
+
  #if !defined(_WIN32)
  static void
  close_tun_generic(struct tuntap *tt)
@@ -1847,7 +1927,8 @@ close_tun_generic(struct tuntap *tt)
 
  #if defined (TARGET_ANDROID)
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
  #define ANDROID_TUNNAME "vpnservice-tun"
      struct user_pass up;
@@ -1944,7 +2025,8 @@ read_tun(struct tuntap *tt, uint8_t *buf, int len)
  #if !PEDANTIC
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      struct ifreq ifr;
 
@@ -1955,6 +2037,10 @@ open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tun
      {
          open_null(tt);
      }
+    else if (tun_dco_enabled(tt))
+    {
+        open_tun_dco_generic(dev, dev_type, tt, ctx);
+    }
      else
      {
          /*
@@ -2061,7 +2147,8 @@ open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tun
  #else  /* if !PEDANTIC */
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      ASSERT(0);
  }
@@ -2086,7 +2173,8 @@ tuncfg(const char *dev, const char *dev_type, const char *dev_node,
      clear_tuntap(tt);
      tt->type = dev_type_enum(dev, dev_type);
      tt->options = *options;
-    open_tun(dev, dev_type, dev_node, tt);
+
+    open_tun(dev, dev_type, dev_node, tt, ctx);
      if (ioctl(tt->fd, TUNSETPERSIST, persist_mode) < 0)
      {
          msg(M_ERR, "Cannot ioctl TUNSETPERSIST(%d) %s", persist_mode, dev);
@@ -2204,6 +2292,12 @@ close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx)
          net_ctx_reset(ctx);
      }
 
+#ifdef TARGET_LINUX
+    if (tun_dco_enabled(tt))
+    {
+        close_tun_dco(tt, ctx);
+    }
+#endif
      close_tun_generic(tt);
      free(tt);
  }
@@ -2227,7 +2321,8 @@ read_tun(struct tuntap *tt, uint8_t *buf, int len)
  #endif
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      int if_fd, ip_muxid, arp_muxid, arp_fd, ppa = -1;
      struct lifreq ifr;
@@ -2579,7 +2674,8 @@ read_tun(struct tuntap *tt, uint8_t *buf, int len)
  #elif defined(TARGET_OPENBSD)
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      open_tun_generic(dev, dev_type, dev_node, true, tt);
 
@@ -2673,7 +2769,8 @@ read_tun(struct tuntap *tt, uint8_t *buf, int len)
   */
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      open_tun_generic(dev, dev_type, dev_node, true, tt);
 
@@ -2813,7 +2910,8 @@ freebsd_modify_read_write_return(int len)
  }
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      open_tun_generic(dev, dev_type, dev_node, true, tt);
 
@@ -2941,7 +3039,8 @@ dragonfly_modify_read_write_return(int len)
  }
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      open_tun_generic(dev, dev_type, dev_node, true, tt);
 
@@ -3169,7 +3268,8 @@ open_darwin_utun(const char *dev, const char *dev_type, const char *dev_node, st
  #endif /* ifdef HAVE_NET_IF_UTUN_H */
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
  #ifdef HAVE_NET_IF_UTUN_H
      /* If dev_node does not start start with utun assume regular tun/tap */
@@ -3276,7 +3376,8 @@ read_tun(struct tuntap *tt, uint8_t *buf, int len)
  #elif defined(TARGET_AIX)
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      char tunname[256];
      char dynamic_name[20];
@@ -6585,7 +6686,8 @@ tuntap_post_open(struct tuntap *tt, const char *device_guid)
  }
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      const char *device_guid = NULL;
 
@@ -6894,7 +6996,8 @@ ipset2ascii_all(struct gc_arena *gc)
  #else /* generic */
 
  void
-open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt)
+open_tun(const char *dev, const char *dev_type, const char *dev_node, struct tuntap *tt,
+         openvpn_net_ctx_t *ctx)
  {
      open_tun_generic(dev, dev_type, dev_node, true, tt);
  }
src/openvpn/tun.h +0 -0
index b7786f46..8ec8f51f 100644

src/openvpn/tun.h tuntap_ring_empty(struct tuntap *tt) · line 255

Quoted 8 lines
src/openvpn/tun.h +1 -1
@@ -249,7 +249,7 @@ tuntap_ring_empty(struct tuntap *tt)
   */
 
  void open_tun(const char *dev, const char *dev_type, const char *dev_node,
-              struct tuntap *tt);
+              struct tuntap *tt, openvpn_net_ctx_t *ctx);
 
  void close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx);
 

The rest looks good!

Basically the code is the same but it was moved out to a separate functon in order to avoid more ifdefs ugliness. We are now duplicating the loop that creates the varios devX strings, but we can live with that.

Not sure my own ACK makes sense, but FWIW:

Signature · 1 line
Antonio Quartulli
Acked-by on v11 5/25
#37 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: introduce open_tun_dco_generic() to open dynamic or fixed-name DCO devices

In reply to Gert Doering

This is a joint work with many iterations... so we have Antonio's ACK on "the changes v10 to v11", and my ACK on the rest.

The patch has been tested on Linux with and without DCO enabled (passes everything in both cases, as long as the system does not have a DCO driver -> "no DCO" -> fallback).

With this patch, as it is, it will break on a DCO-in-kernel system if built with --enable-dco, as half the functionality is missing - but on any combination, it will fall back to non-DCO and that works. For "with DCO in kernel", I've tested this before sending the patch (on top of the DCO tree).

For good measure, since this is touching non-linux code as well (due to the extra argument to open_tun()) I've thrown it at the rejuvenated buildbot army - thanks, Dirk! - and that one was happy too.

I have *not* tested the --mktun branch (trivial enough to stare at) as we want to get rid of that functionality anyway. Until then, it's at least still working :-)

Your patch has been applied to the master branch.

commit 6a2fe47a482e270936bc1c9713fc55837b4551d5

Author: Antonio Quartulli
Date:   Thu Jul 21 20:24:25 2022 +0200
     dco: introduce open_tun_dco_generic() to open dynamic or fixed-name DCO devices

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Signed-off-by: Gert Doering <[email protected]...>
     Acked-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24717.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#38 lore raw

PATCH v2 1/1 networking: add net_iface_type API

In reply to Antonio Quartulli

This new API can be used to retrieve the type of a specific interface. It's mostly platform dependant, but right now expected values are "ovpn-dco", "tun" or "tap".

Other values are possible too, but they are not of interest to us.

This commit also extends the networking unit-test by using the newly introduced API in conjunction with iface_new and iface_del.

The t_next.sh script has been slightly adapted to allow running these tests in standalone (as they don't require any iproute2 counterpart).

---

Changes from v1:
* added unit-test for net_iface_type()
* adjusted t_net.sh script so that iface_new/type/del could be executed automatically

 src/openvpn/networking.h                   |  14 +++
 src/openvpn/networking_iproute2.c          |   9 ++
 src/openvpn/networking_sitnl.c             | 109 +++++++++++++++++++++
 tests/t_net.sh                             |  15 ++-
 tests/unit_tests/openvpn/test_networking.c |  25 +++--
 5 files changed, 163 insertions(+), 9 deletions(-)
src/openvpn/networking.h +14 -0
index 79963756..cf6d39ac 100644
@@ -23,6 +23,8 @@
 
 #include "syshead.h"
 
+#define IFACE_TYPE_LEN_MAX 64
+
 struct context;
 
 #ifdef ENABLE_SITNL
@@ -100,6 +102,18 @@ void net_ctx_free(openvpn_net_ctx_t *ctx);
 int net_iface_new(openvpn_net_ctx_t *ctx, const openvpn_net_iface_t *iface,
                   const char *type, void *arg);
 
+/**
+ * Retrieve the interface type
+ *
+ * @param ctx       the implementation specific context
+ * @param iface     interface to query
+ * @param type      buffer where the type will be stored
+ *
+ * @return          0 on success, a negative error code otherwise
+ */
+int net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+                   char type[IFACE_TYPE_LEN_MAX]);
+
 /**
  * Remove an interface
  *
src/openvpn/networking_iproute2.c +9 -0
index 4b220576..3062c1da 100644
@@ -78,6 +78,15 @@ net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type,
     return 0;
 }
 
+int
+net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+               char type[IFACE_TYPE_LEN_MAX])
+{
+    /* not supported by iproute2 */
+    msg(M_WARN, "%s: operation not supported by iproute2 backend", __func__);
+    return -1;
+}
+
 int
 net_iface_del(openvpn_net_ctx_t *ctx, const char *iface)
 {
src/openvpn/networking_sitnl.c +109 -0

115 lines not on this page · raw message

tests/t_net.sh +13 -2
index af78152c..bc91f825 100755
@@ -2,7 +2,8 @@
 
 IFACE="ovpn-dummy0"
 UNIT_TEST="./unit_tests/openvpn/networking_testdriver"
-MAX_TEST=${1:-7}
+LAST_AUTO_TEST=7
+LAST_TEST=8
 
 srcdir="${srcdir:-.}"
 top_builddir="${top_builddir:-..}"
@@ -128,7 +129,7 @@ else
     fi
 fi
 
-for i in $(seq 0 $MAX_TEST); do
+for i in $(seq 0 $LAST_AUTO_TEST); do
     # reload dummy module to cleanup state
     reload_dummy
     typeset -a STATE_TEST
@@ -168,4 +169,14 @@ done
 # remove interface for good
 $RUN_SUDO ip link del $IFACE
 
+for i in $(seq $(($LAST_AUTO_TEST + 1)) ${LAST_TEST}); do
+    $RUN_SUDO $UNIT_TEST $i
+    if [ $? -ne 0 ]; then
+        echo "unit-test $i errored out"
+        exit 1
+    fi
+
+    echo "Test $i: OK"
+done
+
 exit 0
tests/unit_tests/openvpn/test_networking.c +18 -7
index 10ed2cb5..c3baedb4 100644
@@ -2,6 +2,7 @@
 #include "syshead.h"
 #include "networking.h"
 
+#include <assert.h>
 
 static char *iface = "ovpn-dummy0";
 
@@ -16,14 +17,23 @@ net__iface_up(bool up)
 static int
 net__iface_new(const char *name, const char *type)
 {
-    printf("CMD: ip link add %s type %s\n", name, type);
     return net_iface_new(NULL, name, type, NULL);
 }
 
+static int
+net__iface_type(const char *name, const char *type)
+{
+    char ret_type[IFACE_TYPE_LEN_MAX];
+
+    assert(net_iface_type(NULL, name, ret_type) == 0);
+    assert(strcmp(type, ret_type) == 0);
+
+    return 0;
+}
+
 static int
 net__iface_del(const char *name)
 {
-    printf("CMD: ip link del %s\n", name);
     return net_iface_del(NULL, name);
 }
 
@@ -205,7 +215,7 @@ net__route_v6_add_gw(const char *dst_str, int prefixlen, const char *gw_str,
 static void
 usage(char *name)
 {
-    printf("Usage: %s <0-9>\n", name);
+    printf("Usage: %s <0-8>\n", name);
 }
 
 int
@@ -257,11 +267,12 @@ main(int argc, char *argv[])
         case 7:
             return net__route_v6_add_gw("2001:cafe:babe::", 48, "2001::2", 600);
 
+        /* following tests are standalone and do not print any CMD= */
         case 8:
-            return net__iface_new("dummy0815", "dummy");
-
-        case 9:
-            return net__iface_del("dummy0815");
+            assert(net__iface_new("dummy0815", "dummy") == 0);
+            assert(net__iface_type("dummy0815", "dummy") == 0);
+            assert(net__iface_del("dummy0815") == 0);
+            return 0;
 
         default:
             printf("invalid test: %d\n", test);
Signature · 1 line
2.35.1
#39 lore raw

Re: [Openvpn-devel] [PATCH v2 pre-05/25] networking: add net_iface_type API

Hi,

close, but NAK, due to...

Quoted 11 lines · Antonio Quartulli

This new API can be used to retrieve the type of a specific interface. It's mostly platform dependant, but right now expected values are "ovpn-dco", "tun" or "tap".

Other values are possible too, but they are not of interest to us.

This commit also extends the networking unit-test by using the newly introduced API in conjunction with iface_new and iface_del.

The t_next.sh script has been slightly adapted to allow running these tests in standalone (as they don't require any iproute2 counterpart).

Typo: t_net.sh (but this is not the crucial bit).

src/openvpn/networking_sitnl.c err: · line 1454

Quoted 11 lines
src/openvpn/networking_sitnl.c +11 -0
+int
+net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+               char type[IFACE_TYPE_LEN_MAX])
+{
+    struct sitnl_link_req req = { };
+    int ifindex = if_nametoindex(iface);
+
+    if (!ifindex)
+    {
+        return errno;
+    }

I think this violates the "errors are returned as negative errno" function contract (and this is a code change, I won't "fix this on the fly").

Quoted 1 line

+for i in $(seq $(($LAST_AUTO_TEST + 1)) ${LAST_TEST}); do

Does $(( $A + 1 )) work in POSIX shells?

t_net.sh currenty requires bash, but I wonder if this is really needed, and a POSIX /bin/sh would do the job.

Also, while at it - maybe add a unit test for "this interface does not exist" (as we need that functionality for the "--dev tun3" test)?

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#40 lore raw

Re: [Openvpn-devel] [PATCH v2 pre-05/25] networking: add net_iface_type API

Hi,

Quoted 19 lines · Gert Doering

Hi,

close, but NAK, due to...

Quoted 11 lines · Antonio Quartulli

This new API can be used to retrieve the type of a specific interface. It's mostly platform dependant, but right now expected values are "ovpn-dco", "tun" or "tap".

Other values are possible too, but they are not of interest to us.

This commit also extends the networking unit-test by using the newly introduced API in conjunction with iface_new and iface_del.

The t_next.sh script has been slightly adapted to allow running these tests in standalone (as they don't require any iproute2 counterpart).

Typo: t_net.sh (but this is not the crucial bit).

Quoted 14 lines

src/openvpn/networking_sitnl.c err: · line 1454

Quoted 11 lines
src/openvpn/networking_sitnl.c +11 -0
+int
+net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+               char type[IFACE_TYPE_LEN_MAX])
+{
+    struct sitnl_link_req req = { };
+    int ifindex = if_nametoindex(iface);
+
+    if (!ifindex)
+    {
+        return errno;
+    }

I think this violates the "errors are returned as negative errno" function contract (and this is a code change, I won't "fix this on the fly").

you're right. I thought I had fixed this...but I must have forgotten to add it to the commit.

Quoted 7 lines
Quoted 2 lines

+for i in $(seq $(($LAST_AUTO_TEST + 1)) ${LAST_TEST}); do

Does $(( $A + 1 )) work in POSIX shells?

t_net.sh currenty requires bash, but I wonder if this is really needed, and a POSIX /bin/sh would do the job.

I think there are other bashism in this script (check arrays and stuff like that). Back then we concluded that "since this script runs only on linux, it's ok like this".

Quoted 5 lines

Also, while at it - maybe add a unit test for "this interface does not exist" (as we need that functionality for the "--dev tun3" test)?

yap yap, can do!

Quoted 1 line

gert

Thanks!

Signature · 1 line
Antonio Quartulli
#41 lore raw

PATCH v3 1/1 networking: add net_iface_type API

In reply to Antonio Quartulli

This new API can be used to retrieve the type of a specific interface. It's mostly platform dependant, but right now expected values are "ovpn-dco", "tun" or "tap".

Other values are possible too, but they are not of interest to us.

This commit also extends the networking unit-test by using the newly introduced API in conjunction with iface_new and iface_del.

The t_net.sh script has been slightly adapted to allow running these tests in standalone (as they don't require any iproute2 counterpart).

---

Changes from v2:
* extended unit-test to check for "no device exists case"
* fixed typ0 in commit message
* properly return -errno instead of errno

Changes from v1:
* added unit-test for net_iface_type()
* adjusted t_net.sh script so that iface_new/type/del could be executed automatically

 src/openvpn/networking.h                   |  14 +++
 src/openvpn/networking_iproute2.c          |   9 ++
 src/openvpn/networking_sitnl.c             | 109 +++++++++++++++++++++
 tests/t_net.sh                             |  15 ++-
 tests/unit_tests/openvpn/test_networking.c |  28 ++++--
 5 files changed, 166 insertions(+), 9 deletions(-)
src/openvpn/networking.h +14 -0
index 79963756..cf6d39ac 100644
@@ -23,6 +23,8 @@
 
 #include "syshead.h"
 
+#define IFACE_TYPE_LEN_MAX 64
+
 struct context;
 
 #ifdef ENABLE_SITNL
@@ -100,6 +102,18 @@ void net_ctx_free(openvpn_net_ctx_t *ctx);
 int net_iface_new(openvpn_net_ctx_t *ctx, const openvpn_net_iface_t *iface,
                   const char *type, void *arg);
 
+/**
+ * Retrieve the interface type
+ *
+ * @param ctx       the implementation specific context
+ * @param iface     interface to query
+ * @param type      buffer where the type will be stored
+ *
+ * @return          0 on success, a negative error code otherwise
+ */
+int net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+                   char type[IFACE_TYPE_LEN_MAX]);
+
 /**
  * Remove an interface
  *
src/openvpn/networking_iproute2.c +9 -0
index 4b220576..3062c1da 100644
@@ -78,6 +78,15 @@ net_iface_new(openvpn_net_ctx_t *ctx, const char *iface, const char *type,
     return 0;
 }
 
+int
+net_iface_type(openvpn_net_ctx_t *ctx, const char *iface,
+               char type[IFACE_TYPE_LEN_MAX])
+{
+    /* not supported by iproute2 */
+    msg(M_WARN, "%s: operation not supported by iproute2 backend", __func__);
+    return -1;
+}
+
 int
 net_iface_del(openvpn_net_ctx_t *ctx, const char *iface)
 {
src/openvpn/networking_sitnl.c +109 -0

115 lines not on this page · raw message

tests/t_net.sh +13 -2
index af78152c..bc91f825 100755
@@ -2,7 +2,8 @@
 
 IFACE="ovpn-dummy0"
 UNIT_TEST="./unit_tests/openvpn/networking_testdriver"
-MAX_TEST=${1:-7}
+LAST_AUTO_TEST=7
+LAST_TEST=8
 
 srcdir="${srcdir:-.}"
 top_builddir="${top_builddir:-..}"
@@ -128,7 +129,7 @@ else
     fi
 fi
 
-for i in $(seq 0 $MAX_TEST); do
+for i in $(seq 0 $LAST_AUTO_TEST); do
     # reload dummy module to cleanup state
     reload_dummy
     typeset -a STATE_TEST
@@ -168,4 +169,14 @@ done
 # remove interface for good
 $RUN_SUDO ip link del $IFACE
 
+for i in $(seq $(($LAST_AUTO_TEST + 1)) ${LAST_TEST}); do
+    $RUN_SUDO $UNIT_TEST $i
+    if [ $? -ne 0 ]; then
+        echo "unit-test $i errored out"
+        exit 1
+    fi
+
+    echo "Test $i: OK"
+done
+
 exit 0
tests/unit_tests/openvpn/test_networking.c +21 -7
index 10ed2cb5..04347076 100644
@@ -2,6 +2,7 @@
 #include "syshead.h"
 #include "networking.h"
 
+#include <assert.h>
 
 static char *iface = "ovpn-dummy0";
 
@@ -16,14 +17,25 @@ net__iface_up(bool up)
 static int
 net__iface_new(const char *name, const char *type)
 {
-    printf("CMD: ip link add %s type %s\n", name, type);
     return net_iface_new(NULL, name, type, NULL);
 }
 
+static int
+net__iface_type(const char *name, const char *type)
+{
+    char ret_type[IFACE_TYPE_LEN_MAX];
+    int ret = net_iface_type(NULL, name, ret_type);
+    if (ret == 0)
+    {
+        assert(strcmp(type, ret_type) == 0);
+    }
+
+    return ret;
+}
+
 static int
 net__iface_del(const char *name)
 {
-    printf("CMD: ip link del %s\n", name);
     return net_iface_del(NULL, name);
 }
 
@@ -205,7 +217,7 @@ net__route_v6_add_gw(const char *dst_str, int prefixlen, const char *gw_str,
 static void
 usage(char *name)
 {
-    printf("Usage: %s <0-9>\n", name);
+    printf("Usage: %s <0-8>\n", name);
 }
 
 int
@@ -257,11 +269,13 @@ main(int argc, char *argv[])
         case 7:
             return net__route_v6_add_gw("2001:cafe:babe::", 48, "2001::2", 600);
 
+        /* following tests are standalone and do not print any CMD= */
         case 8:
-            return net__iface_new("dummy0815", "dummy");
-
-        case 9:
-            return net__iface_del("dummy0815");
+            assert(net__iface_new("dummy0815", "dummy") == 0);
+            assert(net__iface_type("dummy0815", "dummy") == 0);
+            assert(net__iface_del("dummy0815") == 0);
+            assert(net__iface_type("dummy0815", NULL) == -ENODEV);
+            return 0;
 
         default:
             printf("invalid test: %d\n", test);
Signature · 1 line
2.35.1
Acked-by on v3 1/1
#42 lore raw

[Openvpn-devel] [PATCH applied] Re: networking: add net_iface_type API

Thanks for the v3 changes.

I tested v2 very thoroughly on "non sitnl" builds (FreeBSD, Linux with --enable-iproute2) to make sure this really only affects sitnl - and that has not changed with v3. What has changed is "-errno", typo fix in the commit message, and the extra test for "what happens if the interface does not exist?" - these have been tested in a sitnl build, and pass.

OpenVPN itself does not yet use that new function, only the unit test.

As agree on IRC, we do not touch the bash-nature of t_net.sh today - we might want to move to "more POSIX" one day, but today Linux systems are assumed to have bash, and this is really linux (+sitnl) only.

Your patch has been applied to the master branch.

commit 26b6b6f0673fa1f3806c447a18bb0f705b231b35

Author: Antonio Quartulli
Date:   Wed Jul 13 14:43:32 2022 +0200
     networking: add net_iface_type API

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24688.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#43 lore raw

PATCH 6/25 dco: initialize context and save pointer in TLS object

In reply to Antonio Quartulli
---
 src/openvpn/init.c       | 49 ++++++++++++++++++++++++++++++++--------
 src/openvpn/ssl_common.h | 23 +++++++++++++++++++
 2 files changed, 63 insertions(+), 9 deletions(-)
src/openvpn/init.c +40 -9
index 7099eba4..7ab2c9a2 100644
@@ -55,6 +55,7 @@
 #include "auth_token.h"
 #include "mss.h"
 #include "mudp.h"
+#include "dco.h"
 
 #include "memdbg.h"
 
@@ -1295,15 +1296,23 @@ do_init_timers(struct context *c, bool deferred)
     }
 
     /* initialize pings */
-
-    if (c->options.ping_send_timeout)
+    if (dco_enabled(&c->options))
     {
-        event_timeout_init(&c->c2.ping_send_interval, c->options.ping_send_timeout, 0);
+        /* The DCO kernel module will send the pings instead of user space */
+        event_timeout_clear(&c->c2.ping_rec_interval);
+        event_timeout_clear(&c->c2.ping_send_interval);
     }
-
-    if (c->options.ping_rec_timeout)
+    else
     {
-        event_timeout_init(&c->c2.ping_rec_interval, c->options.ping_rec_timeout, now);
+        if (c->options.ping_send_timeout)
+        {
+            event_timeout_init(&c->c2.ping_send_interval, c->options.ping_send_timeout, 0);
+        }
+
+        if (c->options.ping_rec_timeout)
+        {
+            event_timeout_init(&c->c2.ping_rec_interval, c->options.ping_rec_timeout, now);
+        }
     }
 
     if (!deferred)
@@ -1708,6 +1717,12 @@ do_open_tun(struct context *c)
     /* initialize (but do not open) tun/tap object */
     do_init_tun(c);
 
+    /* inherit the dco context from the tuntap object */
+    if (c->c2.tls_multi)
+    {
+        c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+    }
+
 #ifdef _WIN32
     /* store (hide) interactive service handle in tuntap_options */
     c->c1.tuntap->options.msg_channel = c->options.msg_channel;
@@ -1756,6 +1771,11 @@ do_open_tun(struct context *c)
     /* Store the old fd inside the fd so open_tun can use it */
     c->c1.tuntap->fd = oldtunfd;
 #endif
+    if (dco_enabled(&c->options))
+    {
+        ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
+    }
+
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
              c->c1.tuntap, &c->net_ctx);
@@ -2979,12 +2999,20 @@ do_init_crypto_tls(struct context *c, const unsigned int flags)
         }
     }
 
+    /* let the TLS engine know if keys have to be installed in DCO or not */
+    to.disable_dco = !dco_enabled(options);
+
     /*
      * Initialize OpenVPN's master TLS-mode object.
      */
     if (flags & CF_INIT_TLS_MULTI)
     {
         c->c2.tls_multi = tls_multi_init(&to);
+        /* inherit the dco context from the tuntap object */
+        if (c->c1.tuntap)
+        {
+            c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+        }
     }
 
     if (flags & CF_INIT_TLS_AUTH_STANDALONE)
@@ -4365,15 +4393,18 @@ inherit_context_child(struct context *dest,
 #endif
 
     /* context init */
+
+    /* inherit tun/tap interface object now as it may be required
+     * to initialize the DCO context in init_instance()
+     */
+    dest->c1.tuntap = src->c1.tuntap;
+
     init_instance(dest, src->c2.es, CC_NO_CLOSE | CC_USR1_TO_HUP);
     if (IS_SIG(dest))
     {
         return;
     }
 
-    /* inherit tun/tap interface object */
-    dest->c1.tuntap = src->c1.tuntap;
-
     /* UDP inherits some extra things which TCP does not */
     if (dest->mode == CM_CHILD_UDP)
     {
src/openvpn/ssl_common.h +23 -0
index cef2611b..83373a97 100644
@@ -167,6 +167,12 @@ enum auth_deferred_result {
     ACF_FAILED        /**< deferred auth has failed */
 };
 
+enum dco_key_status {
+    DCO_NOT_INSTALLED,
+    DCO_INSTALLED_PRIMARY,
+    DCO_INSTALLED_SECONDARY
+};
+
 /**
  * Security parameter state of one TLS and data channel %key session.
  * @ingroup control_processor
@@ -197,6 +203,12 @@ struct key_state
      */
     int key_id;
 
+    /**
+     * Key id for this key_state, inherited from struct tls_session.
+     * @see tls_multi::peer_id.
+     */
+    uint32_t peer_id;
+
     struct key_state_ssl ks_ssl; /* contains SSL object and BIOs for the control channel */
 
     time_t initial;             /* when we created this session */
@@ -241,6 +253,8 @@ struct key_state
 
     struct auth_deferred_status plugin_auth;
     struct auth_deferred_status script_auth;
+
+    enum dco_key_status dco_status;
 };
 
 /** Control channel wrapping (--tls-auth/--tls-crypt) context */
@@ -404,6 +418,8 @@ struct tls_options
     const char *ekm_label;
     size_t ekm_label_size;
     size_t ekm_size;
+
+    bool disable_dco; /**< Whether keys have to be installed in DCO or not */
 };
 
 /** @addtogroup control_processor
@@ -636,6 +652,13 @@ struct tls_multi
     /**< Array of \c tls_session objects
      *   representing control channel
      *   sessions with the remote peer. */
+
+    /* Only used when DCO is used to remember how many keys we installed
+     * for this session */
+    int dco_keys_installed;
+    bool dco_peer_added;
+
+    dco_context_t *dco;
 };
 
 /**  gets an item  of \c key_state objects in the
Signature · 1 line
2.35.1
#45 lore raw

Re: [Openvpn-devel] [PATCH 06/25] dco: initialize context and save pointer in TLS object

In reply to Antonio Quartulli

Hi,

Quoted 5 lines · Antonio Quartulli
---
 src/openvpn/init.c       | 49 ++++++++++++++++++++++++++++++++--------
 src/openvpn/ssl_common.h | 23 +++++++++++++++++++
 2 files changed, 63 insertions(+), 9 deletions(-)

Without trying to understand the code flow in init.c, is there a deeper reason why this happens twice?

Quoted 10 lines
src/openvpn/init.c +6 -0
@@ -1708,6 +1717,12 @@ do_open_tun(struct context *c)
     /* initialize (but do not open) tun/tap object */
     do_init_tun(c);
 
+    /* inherit the dco context from the tuntap object */
+    if (c->c2.tls_multi)
+    {
+        c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+    }
+

[..]

Quoted 1 line
(quoted diff) +0 -0
@@ -2979,12 +2999,20 @@ do_init_crypto_tls(struct context *c, const unsigned int flags)

[..]

src/openvpn/init.c do_init_crypto_tls(struct context *c, const unsigned int flags) · line 3016

Quoted 12 lines
     /*
      * Initialize OpenVPN's master TLS-mode object.
      */
     if (flags & CF_INIT_TLS_MULTI)
     {
         c->c2.tls_multi = tls_multi_init(&to);
src/openvpn/init.c +5 -0
+        /* inherit the dco context from the tuntap object */
+        if (c->c1.tuntap)
+        {
+            c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+        }
     }

... and here again, but only sometimes?

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#46 lore raw

Re: [Openvpn-devel] [PATCH 06/25] dco: initialize context and save pointer in TLS object

Hi,

Quoted 16 lines · Gert Doering

Hi,

Quoted 5 lines · Antonio Quartulli
---
  src/openvpn/init.c       | 49 ++++++++++++++++++++++++++++++++--------
  src/openvpn/ssl_common.h | 23 +++++++++++++++++++
  2 files changed, 63 insertions(+), 9 deletions(-)

Without trying to understand the code flow in init.c, is there a deeper reason why this happens twice?

Quoted 4 lines
src/openvpn/init.c +0 -0
@@ -1708,6 +1717,12 @@ do_open_tun(struct context *c)
      /* initialize (but do not open) tun/tap object */
      do_init_tun(c);
 
Quoted 6 lines

src/openvpn/init.c do_open_tun(struct context *c) · line 1725

Quoted 6 lines
src/openvpn/init.c +6 -0
+    /* inherit the dco context from the tuntap object */
+    if (c->c2.tls_multi)
+    {
+        c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+    }
+

client code path. The server has no tls-multi object when passing here.

Quoted 17 lines

[..]

Quoted 1 line
(quoted diff) +0 -0
@@ -2979,12 +2999,20 @@ do_init_crypto_tls(struct context *c, const unsigned int flags)

[..]

src/openvpn/init.c do_init_crypto_tls(struct context *c, const unsigned int flags) · line 3016

Quoted 12 lines
      /*
       * Initialize OpenVPN's master TLS-mode object.
       */
      if (flags & CF_INIT_TLS_MULTI)
      {
          c->c2.tls_multi = tls_multi_init(&to);
src/openvpn/init.c +5 -0
+        /* inherit the dco context from the tuntap object */
+        if (c->c1.tuntap)
+        {
+            c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+        }
      }
Quoted 1 line

... and here again, but only sometimes?

server code path. The client has no tuntap object when passing here (it's done quite early, so we need the previous snippet). The server executes this code for each new client.

I hope this clarifies.

Cheers,

Signature · 1 line
Antonio Quartulli
#47 lore raw

PATCH v2 6/25 dco: initialize context and save pointer in TLS object

In reply to Antonio Quartulli

---

Changes from v1:
* renamed disable_dco to dco_enabled

 src/openvpn/init.c       | 49 ++++++++++++++++++++++++++++++++--------
 src/openvpn/ssl_common.h | 23 +++++++++++++++++++
 2 files changed, 63 insertions(+), 9 deletions(-)
src/openvpn/init.c +40 -9
index 779fc4a5..b6e1707f 100644
@@ -55,6 +55,7 @@
 #include "auth_token.h"
 #include "mss.h"
 #include "mudp.h"
+#include "dco.h"
 
 #include "memdbg.h"
 
@@ -1315,15 +1316,23 @@ do_init_timers(struct context *c, bool deferred)
     }
 
     /* initialize pings */
-
-    if (c->options.ping_send_timeout)
+    if (dco_enabled(&c->options))
     {
-        event_timeout_init(&c->c2.ping_send_interval, c->options.ping_send_timeout, 0);
+        /* The DCO kernel module will send the pings instead of user space */
+        event_timeout_clear(&c->c2.ping_rec_interval);
+        event_timeout_clear(&c->c2.ping_send_interval);
     }
-
-    if (c->options.ping_rec_timeout)
+    else
     {
-        event_timeout_init(&c->c2.ping_rec_interval, c->options.ping_rec_timeout, now);
+        if (c->options.ping_send_timeout)
+        {
+            event_timeout_init(&c->c2.ping_send_interval, c->options.ping_send_timeout, 0);
+        }
+
+        if (c->options.ping_rec_timeout)
+        {
+            event_timeout_init(&c->c2.ping_rec_interval, c->options.ping_rec_timeout, now);
+        }
     }
 
     if (!deferred)
@@ -1733,6 +1742,12 @@ do_open_tun(struct context *c)
     /* initialize (but do not open) tun/tap object */
     do_init_tun(c);
 
+    /* inherit the dco context from the tuntap object */
+    if (c->c2.tls_multi)
+    {
+        c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+    }
+
 #ifdef _WIN32
     /* store (hide) interactive service handle in tuntap_options */
     c->c1.tuntap->options.msg_channel = c->options.msg_channel;
@@ -1781,6 +1796,11 @@ do_open_tun(struct context *c)
     /* Store the old fd inside the fd so open_tun can use it */
     c->c1.tuntap->fd = oldtunfd;
 #endif
+    if (dco_enabled(&c->options))
+    {
+        ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
+    }
+
     /* open the tun device */
     open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
              c->c1.tuntap, &c->net_ctx);
@@ -3004,12 +3024,20 @@ do_init_crypto_tls(struct context *c, const unsigned int flags)
         }
     }
 
+    /* let the TLS engine know if keys have to be installed in DCO or not */
+    to.dco_enabled = dco_enabled(options);
+
     /*
      * Initialize OpenVPN's master TLS-mode object.
      */
     if (flags & CF_INIT_TLS_MULTI)
     {
         c->c2.tls_multi = tls_multi_init(&to);
+        /* inherit the dco context from the tuntap object */
+        if (c->c1.tuntap)
+        {
+            c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+        }
     }
 
     if (flags & CF_INIT_TLS_AUTH_STANDALONE)
@@ -4389,15 +4417,18 @@ inherit_context_child(struct context *dest,
 #endif
 
     /* context init */
+
+    /* inherit tun/tap interface object now as it may be required
+     * to initialize the DCO context in init_instance()
+     */
+    dest->c1.tuntap = src->c1.tuntap;
+
     init_instance(dest, src->c2.es, CC_NO_CLOSE | CC_USR1_TO_HUP);
     if (IS_SIG(dest))
     {
         return;
     }
 
-    /* inherit tun/tap interface object */
-    dest->c1.tuntap = src->c1.tuntap;
-
     /* UDP inherits some extra things which TCP does not */
     if (dest->mode == CM_CHILD_UDP)
     {
src/openvpn/ssl_common.h +23 -0
index cef2611b..c565d78c 100644
@@ -167,6 +167,12 @@ enum auth_deferred_result {
     ACF_FAILED        /**< deferred auth has failed */
 };
 
+enum dco_key_status {
+    DCO_NOT_INSTALLED,
+    DCO_INSTALLED_PRIMARY,
+    DCO_INSTALLED_SECONDARY
+};
+
 /**
  * Security parameter state of one TLS and data channel %key session.
  * @ingroup control_processor
@@ -197,6 +203,12 @@ struct key_state
      */
     int key_id;
 
+    /**
+     * Key id for this key_state, inherited from struct tls_session.
+     * @see tls_multi::peer_id.
+     */
+    uint32_t peer_id;
+
     struct key_state_ssl ks_ssl; /* contains SSL object and BIOs for the control channel */
 
     time_t initial;             /* when we created this session */
@@ -241,6 +253,8 @@ struct key_state
 
     struct auth_deferred_status plugin_auth;
     struct auth_deferred_status script_auth;
+
+    enum dco_key_status dco_status;
 };
 
 /** Control channel wrapping (--tls-auth/--tls-crypt) context */
@@ -404,6 +418,8 @@ struct tls_options
     const char *ekm_label;
     size_t ekm_label_size;
     size_t ekm_size;
+
+    bool dco_enabled; /**< Whether keys have to be installed in DCO or not */
 };
 
 /** @addtogroup control_processor
@@ -636,6 +652,13 @@ struct tls_multi
     /**< Array of \c tls_session objects
      *   representing control channel
      *   sessions with the remote peer. */
+
+    /* Only used when DCO is used to remember how many keys we installed
+     * for this session */
+    int dco_keys_installed;
+    bool dco_peer_added;
+
+    dco_context_t *dco;
 };
 
 /**  gets an item  of \c key_state objects in the
Signature · 1 line
2.35.1
Acked-by on v2 6/25
#48 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: initialize context and save pointer in TLS object

The change to init.c is easier viewed with "-w" - wrapping the existing code to initialize the ping intervals into "if (!dco_enabled)".

The remainder does not really do much yet - introduce structure elements and enums, but they are not used yet. Also, record the ACK from Arne on v1.

Tested on linux without --enable-dco (works), and with --enable-dco but without kernel DCO (works as well).

Your patch has been applied to the master branch.

commit a2d851d4fa69792f2bcde568e43ec0879d757788

Author: Antonio Quartulli
Date:   Wed Jul 20 14:30:21 2022 +0200
     dco: initialize context and save pointer in TLS object

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-By: Arne Schwabe <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24714.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#49 lore raw

PATCH 7/25 dco: add option check - disable DCO if conflict is detected

In reply to Antonio Quartulli
---
 src/openvpn/Makefile.am             |   2 +-
 src/openvpn/dco.c                   | 149 ++++++++++++++++++++++++++++
 src/openvpn/openvpn.vcxproj         |   1 +
 src/openvpn/openvpn.vcxproj.filters |   3 +
 4 files changed, 154 insertions(+), 1 deletion(-)
 create mode 100644 src/openvpn/dco.c
src/openvpn/Makefile.am +1 -1
index 91635b67..aaa1dbce 100644
@@ -53,7 +53,7 @@ openvpn_SOURCES = \
 	crypto.c crypto.h crypto_backend.h \
 	crypto_openssl.c crypto_openssl.h \
 	crypto_mbedtls.c crypto_mbedtls.h \
-	dco.h dco_internal.h \
+	dco.c dco.h dco_internal.h \
 	dco_linux.c dco_linux.h \
 	dhcp.c dhcp.h \
 	dns.c dns.h \
src/openvpn/dco.c new file +149 -0

149 lines not on this page · raw message

src/openvpn/openvpn.vcxproj +1 -0
index bc1a0300..0b3db7c7 100644
@@ -276,6 +276,7 @@
     <ClCompile Include="crypto.c" />
     <ClCompile Include="crypto_openssl.c" />
     <ClCompile Include="cryptoapi.c" />
+    <ClCompile Include="dco.c" />
     <ClCompile Include="dco_linux.c" />
     <ClCompile Include="dhcp.c" />
     <ClCompile Include="dns.c" />
src/openvpn/openvpn.vcxproj.filters +3 -0
index 3c21a4c6..16905079 100644
@@ -36,6 +36,9 @@
     <ClCompile Include="cryptoapi.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="dco.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="dco_linux.c">
       <Filter>Source Files</Filter>
     </ClCompile>
Signature · 1 line
2.35.1
#51 lore raw

PATCH v2 7/25 dco: add option check - disable DCO if conflict is detected

In reply to Antonio Quartulli

---

Changes from v1:
* add 'already existing device check' to dco_check_option_conflict_platform() so that DCO can be pre-emptively disabled if the following are true:
- an iface with the same name as provided by the user exists
- the iface is non-DCO

 src/openvpn/Makefile.am             |   2 +-
 src/openvpn/dco.c                   | 189 ++++++++++++++++++++++++++++
 src/openvpn/openvpn.vcxproj         |   1 +
 src/openvpn/openvpn.vcxproj.filters |   3 +
 4 files changed, 194 insertions(+), 1 deletion(-)
 create mode 100644 src/openvpn/dco.c
src/openvpn/Makefile.am +1 -1
index 91635b67..aaa1dbce 100644
@@ -53,7 +53,7 @@ openvpn_SOURCES = \
 	crypto.c crypto.h crypto_backend.h \
 	crypto_openssl.c crypto_openssl.h \
 	crypto_mbedtls.c crypto_mbedtls.h \
-	dco.h dco_internal.h \
+	dco.c dco.h dco_internal.h \
 	dco_linux.c dco_linux.h \
 	dhcp.c dhcp.h \
 	dns.c dns.h \
src/openvpn/dco.c new file +189 -0

189 lines not on this page · raw message

src/openvpn/openvpn.vcxproj +1 -0
index bc1a0300..0b3db7c7 100644
@@ -276,6 +276,7 @@
     <ClCompile Include="crypto.c" />
     <ClCompile Include="crypto_openssl.c" />
     <ClCompile Include="cryptoapi.c" />
+    <ClCompile Include="dco.c" />
     <ClCompile Include="dco_linux.c" />
     <ClCompile Include="dhcp.c" />
     <ClCompile Include="dns.c" />
src/openvpn/openvpn.vcxproj.filters +3 -0
index 3c21a4c6..16905079 100644
@@ -36,6 +36,9 @@
     <ClCompile Include="cryptoapi.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="dco.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="dco_linux.c">
       <Filter>Source Files</Filter>
     </ClCompile>
Signature · 1 line
2.35.1
#52 lore raw

PATCH v3 7/25 dco: add option check - disable DCO if conflict is detected

---

** this patch should be applied before 05/27 **

Changes from v2:
* add actual invocation to dco_check_option_conflict() in options.c
* add missing '}' in dco_check_option_conflict_ce()

Changes from v1:
* add 'already existing device check' to dco_check_option_conflict_platform() so that DCO can be pre-emptively disabled if the following are true:
- an iface with the same name as provided by the user exists
- the iface is non-DCO

 src/openvpn/Makefile.am             |   2 +-
 src/openvpn/dco.c                   | 190 ++++++++++++++++++++++++++++
 src/openvpn/openvpn.vcxproj         |   1 +
 src/openvpn/openvpn.vcxproj.filters |   3 +
 src/openvpn/options.c               |   5 +
 5 files changed, 200 insertions(+), 1 deletion(-)
 create mode 100644 src/openvpn/dco.c
src/openvpn/Makefile.am +1 -1
index 91635b67..aaa1dbce 100644
@@ -53,7 +53,7 @@ openvpn_SOURCES = \
 	crypto.c crypto.h crypto_backend.h \
 	crypto_openssl.c crypto_openssl.h \
 	crypto_mbedtls.c crypto_mbedtls.h \
-	dco.h dco_internal.h \
+	dco.c dco.h dco_internal.h \
 	dco_linux.c dco_linux.h \
 	dhcp.c dhcp.h \
 	dns.c dns.h \
src/openvpn/dco.c new file +190 -0

190 lines not on this page · raw message

src/openvpn/openvpn.vcxproj +1 -0
index bc1a0300..0b3db7c7 100644
@@ -276,6 +276,7 @@
     <ClCompile Include="crypto.c" />
     <ClCompile Include="crypto_openssl.c" />
     <ClCompile Include="cryptoapi.c" />
+    <ClCompile Include="dco.c" />
     <ClCompile Include="dco_linux.c" />
     <ClCompile Include="dhcp.c" />
     <ClCompile Include="dns.c" />
src/openvpn/openvpn.vcxproj.filters +3 -0
index 3c21a4c6..16905079 100644
@@ -36,6 +36,9 @@
     <ClCompile Include="cryptoapi.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="dco.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="dco_linux.c">
       <Filter>Source Files</Filter>
     </ClCompile>
src/openvpn/options.c +5 -0
index 95d4008a..7b919a1e 100644
@@ -3645,6 +3645,11 @@ options_postprocess_mutate(struct options *o, struct env_set *es)
         o->verify_hash_no_ca = true;
     }
 
+    /* check if any option should force disabling DCO */
+#if defined(TARGET_LINUX)
+    o->tuntap_options.disable_dco = !dco_check_option_conflict(D_DCO, o);
+#endif
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
Signature · 1 line
2.35.1
Acked-by on v3 7/25
#53 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: add option check - disable DCO if conflict is detected

Discussed this at length on IRC. We put this in first (+08), so *05* can go in without breaking non-DCO builds - because this one will ensure "disable_dco" is set there.

Understanding the DCO and non-DCO code is a bit confusing - the whole dco.c is #ifdef ENABLE_DCO, and dco_check_option_conflict() will be a "static inline ... return false;" in that case - and the dco.h bits of this came in via e34437c26b, so not obvious from this patch how the pieces work together.

This said, I've thrown 07 v3 at the t_server test machinery (Gentoo, with no "--enable-dco" at configure times) and it passed everything - unsurprisingly, as the new code is not compiled in.

Testing with --enable-dco also caused no problems - this test basically is a "will these parts compile correctly?" sanity check - the code will detect "there is no DCO kernel support here" on my machine and auto-disable DCO...

2022-07-19 11:06:31 Cannot find ovpn_dco netlink component: Object not found
2022-07-19 11:06:31 Note: Kernel support for ovpn-dco missing, disabling data channel offload.

On a machine *with* kernel support *and* --enable-dco, I assume this will also do nothing (as open_tun() is not yet DCO'ed - that's 05).

I have only skimmed the actual option check - seems to make sense :-) (and has an ACK from Arne, on 07 v1).

Whether the linux "fixed" check will work, we'll see after 05 is merged.

Your patch has been applied to the master branch.

commit 8989b0f2833d25c97654c25fa6a49d8fc0ef903d

Author: Antonio Quartulli
Date:   Tue Jul 19 00:17:57 2022 +0200
     dco: add option check - disable DCO if conflict is detected

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24701.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#54 lore raw

PATCH 1/1 tun: create tun_name_is_fixed helper

In reply to Antonio Quartulli

This helper encloses the (simple) logic used by OpenVPN to determine if the name passed to --dev has to be considered a fixed interface name or just a pattern.

Having a helper is useful because when this logic is required elsewhere, we can just re-use this logic without duplicating the code (which may mean introducing bugs if a future logic change should not update all spots).

The logic is actually fairly simple: check if the name contains a number (i.e. tun0). If so, consider the name a fixed device name.

While at it make has_digit() accept a signed argument because strings are normally signed (also isdigit() accepts a signed argument).

---
 src/openvpn/buffer.h | 4 ++--
 src/openvpn/tun.c    | 7 ++++++-
 src/openvpn/tun.h    | 1 +
 3 files changed, 9 insertions(+), 3 deletions(-)
src/openvpn/buffer.h +2 -2
index 231f1b0d..fece6336 100644
@@ -356,9 +356,9 @@ strncpynt(char *dest, const char *src, size_t maxlen)
 
 /* return true if string contains at least one numerical digit */
 static inline bool
-has_digit(const unsigned char *src)
+has_digit(const char *src)
 {
-    unsigned char c;
+    char c;
     while ((c = *src++))
     {
         if (isdigit(c))
src/openvpn/tun.c +6 -1
index f17db280..108090d0 100644
@@ -1718,6 +1718,11 @@ read_tun_header(struct tuntap *tt, uint8_t *buf, int len)
 }
 #endif /* if defined (TARGET_OPENBSD) || (defined(TARGET_DARWIN) && HAVE_NET_IF_UTUN_H) */
 
+bool
+tun_name_is_fixed(const char *dev)
+{
+    return has_digit(dev);
+}
 
 #if !defined(_WIN32)
 static void
@@ -1772,7 +1777,7 @@ open_tun_generic(const char *dev, const char *dev_type, const char *dev_node,
             else
 #endif
 
-            if (dynamic && !has_digit((unsigned char *)dev))
+            if (dynamic && !tun_name_is_fixed(dev))
             {
                 int i;
                 for (i = 0; i < 256; ++i)
src/openvpn/tun.h +1 -0
index cf02bf43..8ec8f51f 100644
@@ -694,5 +694,6 @@ tun_set(struct tuntap *tt,
 }
 
 const char *tun_stat(const struct tuntap *tt, unsigned int rwflags, struct gc_arena *gc);
+bool tun_name_is_fixed(const char *dev);
 
 #endif /* TUN_H */
Signature · 1 line
2.35.1
Acked-by on v1 1/1
#55 lore raw

[Openvpn-devel] [PATCH applied] Re: tun: create tun_name_is_fixed helper

Trivial enough, and confirmed that it does not break anything.

(Having the "has_digit()" in buffer.*h* when it is only ever called from tun.c smells like "we might eventually include this in tun_name_is_fixed() and get rid of the inline helper" - but this is a patch for another day. Also, we might do it the other way round, "everything that is not exactly 'tun', 'tap' or 'null' is considered fixed" - this whole open_tun() thing is a hairy mess of historic insanity).

Your patch has been applied to the master branch.

commit 4a88d2fbe73a64f8f3861089a2b6e3b8e583e71b

Author: Antonio Quartulli
Date:   Wed Jul 13 00:16:55 2022 +0200
     tun: create tun_name_is_fixed helper

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24676.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#56 lore raw

PATCH 8/25 dco: allow user to disable it at runtime

In reply to Antonio Quartulli
---
 src/openvpn/options.c | 29 +++++++++++++++++++++++++++++
 1 file changed, 29 insertions(+)
src/openvpn/options.c +29 -0
index 9a0634a5..7b450296 100644
@@ -61,6 +61,7 @@
 #include "ssl_verify.h"
 #include "platform.h"
 #include "xkey_common.h"
+#include "dco.h"
 #include <ctype.h>
 
 #include "memdbg.h"
@@ -106,6 +107,9 @@ const char title_string[] =
 #endif
 #endif
     " [AEAD]"
+#ifdef ENABLE_DCO
+    " [DCO]"
+#endif
     " built on " __DATE__
 ;
 
@@ -177,6 +181,9 @@ static const char usage_message[] =
     "                  does not begin with \"tun\" or \"tap\".\n"
     "--dev-node node : Explicitly set the device node rather than using\n"
     "                  /dev/net/tun, /dev/tun, /dev/tap, etc.\n"
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    "--disable-dco   : Do not attempt using Data Channel Offload.\n"
+#endif
     "--lladdr hw     : Set the link layer address of the tap device.\n"
     "--topology t    : Set --dev tun topology: 'net30', 'p2p', or 'subnet'.\n"
 #ifdef ENABLE_IPROUTE
@@ -1711,6 +1718,9 @@ show_settings(const struct options *o)
     SHOW_STR(dev);
     SHOW_STR(dev_type);
     SHOW_STR(dev_node);
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    SHOW_BOOL(tuntap_options.disable_dco);
+#endif
     SHOW_STR(lladdr);
     SHOW_INT(topology);
     SHOW_STR(ifconfig_local);
@@ -3210,6 +3220,14 @@ options_postprocess_verify(const struct options *o)
     }
 
     dns_options_verify(M_FATAL, &o->dns_options);
+
+    if (dco_enabled(o) && o->enable_c2c)
+    {
+        msg(M_WARN, "Note: --client-to-client has no effect when using data "
+            "channel offload: packets are always sent to the VPN "
+            "interface and then routed based on the system routing "
+            "table");
+    }
 }
 
 /**
@@ -3454,6 +3472,11 @@ options_postprocess_mutate(struct options *o)
         o->verify_hash_no_ca = true;
     }
 
+    /* check if any option should force disabling DCO */
+#if defined(TARGET_LINUX)
+    o->tuntap_options.disable_dco = !dco_check_option_conflict(D_DCO, o);
+#endif
+
     /*
      * Save certain parms before modifying options during connect, especially
      * when using --pull
@@ -5759,6 +5782,12 @@ add_option(struct options *options,
         options->windows_driver = parse_windows_driver(p[1], M_FATAL);
     }
 #endif
+    else if (streq(p[0], "disable-dco") || streq(p[0], "dco-disable"))
+    {
+#if defined(TARGET_LINUX)
+        options->tuntap_options.disable_dco = true;
+#endif
+    }
     else if (streq(p[0], "dev-node") && p[1] && !p[2])
     {
         VERIFY_PERMISSION(OPT_P_GENERAL);
Signature · 1 line
2.35.1
#57 lore raw

Re: [Openvpn-devel] [PATCH 08/25] dco: allow user to disable it at runtime

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 49 lines
---
  src/openvpn/options.c | 29 +++++++++++++++++++++++++++++
  1 file changed, 29 insertions(+)
src/openvpn/options.c +10 -0
index 9a0634a5..7b450296 100644
@@ -61,6 +61,7 @@
  #include "ssl_verify.h"
  #include "platform.h"
  #include "xkey_common.h"
+#include "dco.h"
  #include <ctype.h>
 
  #include "memdbg.h"
@@ -106,6 +107,9 @@ const char title_string[] =
  #endif
  #endif
      " [AEAD]"
+#ifdef ENABLE_DCO
+    " [DCO]"
+#endif
      " built on " __DATE__
  ;
 
@@ -177,6 +181,9 @@ static const char usage_message[] =
      "                  does not begin with \"tun\" or \"tap\".\n"
      "--dev-node node : Explicitly set the device node rather than using\n"
      "                  /dev/net/tun, /dev/tun, /dev/tap, etc.\n"
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    "--disable-dco   : Do not attempt using Data Channel Offload.\n"
+#endif
      "--lladdr hw     : Set the link layer address of the tap device.\n"
      "--topology t    : Set --dev tun topology: 'net30', 'p2p', or 'subnet'.\n"
  #ifdef ENABLE_IPROUTE
@@ -1711,6 +1718,9 @@ show_settings(const struct options *o)
      SHOW_STR(dev);
      SHOW_STR(dev_type);
      SHOW_STR(dev_node);
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    SHOW_BOOL(tuntap_options.disable_dco);
+#endif
      SHOW_STR(lladdr);
      SHOW_INT(topology);
      SHOW_STR(ifconfig_local);
@@ -3210,6 +3220,14 @@ options_postprocess_verify(const struct options *o)
      }

src/openvpn/options.c options_postprocess_verify(const struct options *o) · line 3230

Quoted 10 lines
src/openvpn/options.c +8 -0
 
      dns_options_verify(M_FATAL, &o->dns_options);
+
+    if (dco_enabled(o) && o->enable_c2c)
+    {
+        msg(M_WARN, "Note: --client-to-client has no effect when using data "
+            "channel offload: packets are always sent to the VPN "
+            "interface and then routed based on the system routing "
+            "table");
+    }

You should add at least one of them to the man page and document it.

Arne

#58 lore raw

Re: [Openvpn-devel] [PATCH 08/25] dco: allow user to disable it at runtime

In reply to Antonio Quartulli
Quoted 1 line · On Freitag, 24. Juni 2022 10:37:52 CEST Antonio Quartulli wrote:

+ else if (streq(p[0], "disable-dco") || streq(p[0], "dco-disable"))

Don't think we need to be backwards compatible here, or do we?

#59 lore raw

Re: [Openvpn-devel] [PATCH 08/25] dco: allow user to disable it at runtime

Hi,

Quoted 4 lines · Heiko Hund
Quoted 1 line · On Freitag, 24. Juni 2022 10:37:52 CEST Antonio Quartulli wrote:

+ else if (streq(p[0], "disable-dco") || streq(p[0], "dco-disable"))

Don't think we need to be backwards compatible here, or do we?

There's nothing to be backwards compatible with.
Thanks for highlighting this.

I am removing "dco-disable" in v2.

Cheers,

Quoted 10 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#60 lore raw

PATCH v2 8/25 dco: allow user to disable it at runtime

In reply to Antonio Quartulli

---

Changes from v1:
* removed "--dco-disable" option: we just need "--disable-dco"
* added text to manpage about --client-to-client being no-op
* added text to manpage about --disable-dco
* rebased on top of master+"dco: add option check - disable DCO if conflict is detected"

 doc/man-sections/generic-options.rst |  9 +++++++++
 doc/man-sections/server-options.rst  |  4 ++++
 src/openvpn/options.c                | 24 ++++++++++++++++++++++++
 3 files changed, 37 insertions(+)
doc/man-sections/generic-options.rst +9 -0
index 9060a235..394c2186 100644
@@ -171,6 +171,15 @@ which mode OpenVPN is configured as.
   on console) and ``--auth-nocache`` will fail as soon as key
   renegotiation (and reauthentication) occurs.
 
+--disable-dco
+  Disable "data channel offload" (DCO).
+
+  On Linux don't use the ovpn-dco device driver, but rather rely on the
+  legacy tun module.
+
+  You may want to use this option if your server needs to allow clients
+  older than version 2.4 to connect.
+
 --disable-occ
   Disable "options consistency check" (OCC).
 
doc/man-sections/server-options.rst +4 -0
index 08ee7bd3..04f4b4fb 100644
@@ -146,6 +146,10 @@ fast hardware. SSL/TLS authentication must be used in this mode.
   server. Don't use this option if you want to firewall tunnel traffic
   using custom, per-client rules.
 
+  Please note that when using data channel offload this option has no
+  effect. Packets are always sent to the tunnel interface and then
+  routed based on the system routing table.
+
 --disable
   Disable a particular client (based on the common name) from connecting.
   Don't use this option to disable a client due to key or password
src/openvpn/options.c +24 -0
index 7b919a1e..d864c6e2 100644
@@ -61,6 +61,7 @@
 #include "ssl_verify.h"
 #include "platform.h"
 #include "xkey_common.h"
+#include "dco.h"
 #include <ctype.h>
 
 #include "memdbg.h"
@@ -106,6 +107,9 @@ const char title_string[] =
 #endif
 #endif
     " [AEAD]"
+#ifdef ENABLE_DCO
+    " [DCO]"
+#endif
     " built on " __DATE__
 ;
 
@@ -177,6 +181,9 @@ static const char usage_message[] =
     "                  does not begin with \"tun\" or \"tap\".\n"
     "--dev-node node : Explicitly set the device node rather than using\n"
     "                  /dev/net/tun, /dev/tun, /dev/tap, etc.\n"
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    "--disable-dco   : Do not attempt using Data Channel Offload.\n"
+#endif
     "--lladdr hw     : Set the link layer address of the tap device.\n"
     "--topology t    : Set --dev tun topology: 'net30', 'p2p', or 'subnet'.\n"
 #ifdef ENABLE_IPROUTE
@@ -1785,6 +1792,9 @@ show_settings(const struct options *o)
     SHOW_STR(dev);
     SHOW_STR(dev_type);
     SHOW_STR(dev_node);
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    SHOW_BOOL(tuntap_options.disable_dco);
+#endif
     SHOW_STR(lladdr);
     SHOW_INT(topology);
     SHOW_STR(ifconfig_local);
@@ -3401,6 +3411,14 @@ options_postprocess_verify(const struct options *o)
     }
 
     dns_options_verify(M_FATAL, &o->dns_options);
+
+    if (dco_enabled(o) && o->enable_c2c)
+    {
+        msg(M_WARN, "Note: --client-to-client has no effect when using data "
+            "channel offload: packets are always sent to the VPN "
+            "interface and then routed based on the system routing "
+            "table");
+    }
 }
 
 /**
@@ -5839,6 +5857,12 @@ add_option(struct options *options,
         options->windows_driver = parse_windows_driver(p[1], M_FATAL);
     }
 #endif
+    else if (streq(p[0], "disable-dco"))
+    {
+#if defined(TARGET_LINUX)
+        options->tuntap_options.disable_dco = true;
+#endif
+    }
     else if (streq(p[0], "dev-node") && p[1] && !p[2])
     {
         VERIFY_PERMISSION(OPT_P_GENERAL);
Signature · 1 line
2.35.1
Acked-by on v2 8/25
#61 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: allow user to disable it at runtime

This does not really *do* much yet, but now you can do an --enable-dco build, and suppress the "no kernel support!! ey!" message with --disable-dco again ;-) - also, a build will now show [DCO] on --version

As agreed on IRC, rewrapped the p2p message ("table" on a separate line).

Your patch has been applied to the master branch.

commit 7a4c75927109acc2cd455140db312042781a3949

Author: Antonio Quartulli
Date:   Tue Jul 19 00:19:23 2022 +0200
     dco: allow user to disable it at runtime

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24702.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#62 lore raw

PATCH 9/25 dco: configure keys in DCO right after generating them

In reply to Antonio Quartulli

The ovpn-dco kernel module needs to be informed about the keys to be used to encrypt/decrypt data traffic to/from a peer.

Configure keys in DCO right afte they are generated by the SSL code, to avoid keeping them in memory longer than needed.

---
 src/openvpn/dco.c   | 57 ++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h   | 27 +++++++++++++++++
 src/openvpn/init.c  | 11 ++++---
 src/openvpn/multi.c |  2 +-
 src/openvpn/ssl.c   | 74 +++++++++++++++++++++++++++++++++------------
 src/openvpn/ssl.h   |  7 +++--
 6 files changed, 152 insertions(+), 26 deletions(-)
src/openvpn/dco.c +57 -0
index 1e45130a..e38614fa 100644
@@ -33,7 +33,64 @@
 #if defined(ENABLE_DCO)
 
 #include "syshead.h"
+#include "crypto.h"
 #include "dco.h"
+#include "errlevel.h"
+#include "openvpn.h"
+#include "ssl_common.h"
+#include "ssl_ncp.h"
+#include "tun.h"
+
+static int
+dco_install_key(struct tls_multi *multi, struct key_state *ks,
+                const uint8_t *encrypt_key, const uint8_t *encrypt_iv,
+                const uint8_t *decrypt_key, const uint8_t *decrypt_iv,
+                const char *ciphername)
+
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d keyid=%d", __func__, multi->peer_id,
+        ks->key_id);
+
+    /* Install a key in the PRIMARY slot only when no other key exist.
+     * From that moment on, any new key will be installed in the SECONDARY
+     * slot and will be promoted to PRIMARY when userspace says so (a swap
+     * will be performed in that case)
+     */
+    dco_key_slot_t slot = OVPN_KEY_SLOT_PRIMARY;
+    if (multi->dco_keys_installed > 0)
+    {
+        slot = OVPN_KEY_SLOT_SECONDARY;
+    }
+
+    int ret = dco_new_key(multi->dco, multi->peer_id, ks->key_id, slot,
+                          encrypt_key, encrypt_iv,
+                          decrypt_key, decrypt_iv,
+                          ciphername);
+    if ((ret == 0) && (multi->dco_keys_installed < 2))
+    {
+        multi->dco_keys_installed++;
+        ks->dco_status = (slot == OVPN_KEY_SLOT_PRIMARY) ? DCO_INSTALLED_PRIMARY :
+                         DCO_INSTALLED_SECONDARY;
+    }
+
+    return ret;
+}
+
+int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    struct key_direction_state kds;
+    key_direction_state_init(&kds, key_direction);
+
+    return dco_install_key(multi, ks,
+                           key2->keys[kds.out_key].cipher,
+                           key2->keys[(int)server].hmac,
+                           key2->keys[kds.in_key].cipher,
+                           key2->keys[1 - (int)server].hmac,
+                           ciphername);
+}
 
 static bool
 dco_check_option_conflict_ce(const struct connection_entry *ce, int msglevel)
src/openvpn/dco.h +27 -0
index 063e5028..b081c6fa 100644
@@ -35,7 +35,10 @@
  * order problems)
  */
 struct event_set;
+struct key2;
+struct key_state;
 struct options;
+struct tls_multi;
 struct tuntap;
 
 #define DCO_DEFAULT_METRIC  200
@@ -111,6 +114,22 @@ int dco_do_write(dco_context_t *dco, int peer_id, struct buffer *buf);
  */
 void dco_event_set(dco_context_t *dco, struct event_set *es, void *arg);
 
+/**
+ * Install the key material in DCO for the specified peer, at the specified slot
+ *
+ * @param multi     the TLS context of the current instance
+ * @param ks        the state of the key being installed
+ * @param key2      the container for the raw key material
+ * @param key_direction the key direction to be used to extract the material
+ * @param ciphername    the name of the cipher to use the key with
+ * @param server    whether we are running on a server instance or not
+ *
+ * @return          0 on success or a negative error code otherwise
+ */
+int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                    const struct key2 *key2, int key_direction,
+                    const char *ciphername, bool server);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -163,5 +182,13 @@ dco_event_set(dco_context_t *dco, struct event_set *es, void *arg)
 {
 }
 
+static inline int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    return 0;
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/init.c +7 -4
index 7ab2c9a2..06911cd0 100644
@@ -2201,8 +2201,9 @@ do_deferred_p2p_ncp(struct context *c)
     }
 #endif
 
-    if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                          frame_fragment, get_link_socket_info(c)))
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
+                                          &c->c2.frame, frame_fragment,
+                                          get_link_socket_info(c)))
     {
         msg(D_TLS_ERRORS, "ERROR: failed to set crypto cipher");
         return false;
@@ -2315,8 +2316,10 @@ do_deferred_options(struct context *c, const unsigned int found)
 #endif
 
         struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                              frame_fragment, get_link_socket_info(c)))
+        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                              &c->options, &c->c2.frame,
+                                              frame_fragment,
+                                              get_link_socket_info(c)))
         {
             msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
             return false;
src/openvpn/multi.c +1 -1
index ba2f6d58..c72575ae 100644
@@ -2286,7 +2286,7 @@ multi_client_generate_tls_keys(struct context *c)
     }
 #endif
     struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-    if (!tls_session_update_crypto_params(session, &c->options,
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
                                           &c->c2.frame, frame_fragment,
                                           get_link_socket_info(c)))
     {
src/openvpn/ssl.c +55 -19

146 lines not on this page · raw message

src/openvpn/ssl.h +5 -2
index 0ba86d3e..ba271971 100644
@@ -423,6 +423,7 @@ void tls_update_remote_addr(struct tls_multi *multi,
  * channel keys based on the supplied options. Does nothing if keys are already
  * generated.
  *
+ * @param multi           The TLS object for this instance.
  * @param session         The TLS session to update.
  * @param options         The options to use when updating session.
  * @param frame           The frame options for this session (frame overhead is
@@ -433,7 +434,8 @@ void tls_update_remote_addr(struct tls_multi *multi,
  *
  * @return true if updating succeeded or keys are already generated, false otherwise.
  */
-bool tls_session_update_crypto_params(struct tls_session *session,
+bool tls_session_update_crypto_params(struct tls_multi *multi,
+                                      struct tls_session *session,
                                       struct options *options,
                                       struct frame *frame,
                                       struct frame *frame_fragment,
@@ -548,7 +550,8 @@ show_available_tls_ciphers(const char *cipher_list,
  * can thus be called only once per session.
  */
 bool
-tls_session_generate_data_channel_keys(struct tls_session *session);
+tls_session_generate_data_channel_keys(struct tls_multi *multi,
+                                       struct tls_session *session);
 
 /**
  * Load ovpn.xkey provider used for external key signing
Signature · 1 line
2.35.1
#63 lore raw

Re: [Openvpn-devel] [PATCH 09/25] dco: configure keys in DCO right after generating them

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 6 lines
src/openvpn/dco.h +5 -0
+/**
+ * Install the key material in DCO for the specified peer, at the specified slot
+ *
+ * @param multi     the TLS context of the current instance
+ * @param ks        the state of the key being installed

src/openvpn/dco.h int dco_do_write(dco_context_t *dco, int peer_id, struct buffer *buf); · line 131

Quoted 10 lines
src/openvpn/dco.h +10 -0
+ * @param key2      the container for the raw key material
+ * @param key_direction the key direction to be used to extract the material
+ * @param ciphername    the name of the cipher to use the key with
+ * @param server    whether we are running on a server instance or not
+ *
+ * @return          0 on success or a negative error code otherwise
+ */
+int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                    const struct key2 *key2, int key_direction,
+                    const char *ciphername, bool server);

I think here the description might be outdated. Your method does not have a specified slot anymore. It would be good to document that this method instead has a hidden that it install the primary key on the first call and otherwards installs/overwrites the secondary key.

src/openvpn/ssl.c openvpn_PRF(const uint8_t *secret, · line 1457

Quoted 13 lines
src/openvpn/ssl.c +13 -0
+    if (dco_disabled)
+    {
+        init_key_ctx_bi(key, key2, key_direction, key_type, "Data Channel");
+        /* Initialize implicit IVs */
+        key_ctx_update_implicit_iv(&key->encrypt, key2->keys[(int)server].hmac,
+                                   MAX_HMAC_KEY_LENGTH);
+        key_ctx_update_implicit_iv(&key->decrypt,
+                                   key2->keys[1 - (int)server].hmac,
+                                   MAX_HMAC_KEY_LENGTH);
+    }
+
+    if (!dco_disabled)
+    {

I think an else here would be better.

Quoted 1 line

+ /* encrypt/decrypt context are unused with DCO */

Do we have actually checked this? IIrc the generation of the explicit exit notification might still try to generate a data channel key and use these contexts.

Arne

#64 lore raw

PATCH v2 9/25 dco: configure keys in DCO right after generating them

In reply to Antonio Quartulli

The ovpn-dco kernel module needs to be informed about the keys to be used to encrypt/decrypt data traffic to/from a peer.

Configure keys in DCO right afte they are generated by the SSL code, to avoid keeping them in memory longer than needed.

---

Changes from v1:
* adapt to new member name dco_enabled
* invert if blocks and condition in init_key_contexts() [and use 'else']
* fix comment for init_key_contexts()
* disable explicit-exit-notification in mutate_ce() when DCO is enabled

 src/openvpn/dco.c     | 55 ++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     | 29 +++++++++++++++++
 src/openvpn/init.c    | 11 ++++---
 src/openvpn/multi.c   |  2 +-
 src/openvpn/options.c | 10 ++++++
 src/openvpn/ssl.c     | 73 ++++++++++++++++++++++++++++++++-----------
 src/openvpn/ssl.h     |  7 +++--
 7 files changed, 161 insertions(+), 26 deletions(-)
src/openvpn/dco.c +55 -0
index b3fd135f..0471e4d0 100644
@@ -33,12 +33,67 @@
 #if defined(ENABLE_DCO)
 
 #include "syshead.h"
+#include "crypto.h"
 #include "dco.h"
+#include "errlevel.h"
 #include "networking.h"
+#include "openvpn.h"
 #include "options.h"
+#include "ssl_common.h"
 #include "ssl_ncp.h"
 #include "tun.h"
 
+static int
+dco_install_key(struct tls_multi *multi, struct key_state *ks,
+                const uint8_t *encrypt_key, const uint8_t *encrypt_iv,
+                const uint8_t *decrypt_key, const uint8_t *decrypt_iv,
+                const char *ciphername)
+
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d keyid=%d", __func__, multi->peer_id,
+        ks->key_id);
+
+    /* Install a key in the PRIMARY slot only when no other key exist.
+     * From that moment on, any new key will be installed in the SECONDARY
+     * slot and will be promoted to PRIMARY when userspace says so (a swap
+     * will be performed in that case)
+     */
+    dco_key_slot_t slot = OVPN_KEY_SLOT_PRIMARY;
+    if (multi->dco_keys_installed > 0)
+    {
+        slot = OVPN_KEY_SLOT_SECONDARY;
+    }
+
+    int ret = dco_new_key(multi->dco, multi->peer_id, ks->key_id, slot,
+                          encrypt_key, encrypt_iv,
+                          decrypt_key, decrypt_iv,
+                          ciphername);
+    if ((ret == 0) && (multi->dco_keys_installed < 2))
+    {
+        multi->dco_keys_installed++;
+        ks->dco_status = (slot == OVPN_KEY_SLOT_PRIMARY) ? DCO_INSTALLED_PRIMARY :
+                         DCO_INSTALLED_SECONDARY;
+    }
+
+    return ret;
+}
+
+int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    struct key_direction_state kds;
+    key_direction_state_init(&kds, key_direction);
+
+    return dco_install_key(multi, ks,
+                           key2->keys[kds.out_key].cipher,
+                           key2->keys[(int)server].hmac,
+                           key2->keys[kds.in_key].cipher,
+                           key2->keys[1 - (int)server].hmac,
+                           ciphername);
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {
src/openvpn/dco.h +29 -0
index 063e5028..1692f5c3 100644
@@ -35,7 +35,10 @@
  * order problems)
  */
 struct event_set;
+struct key2;
+struct key_state;
 struct options;
+struct tls_multi;
 struct tuntap;
 
 #define DCO_DEFAULT_METRIC  200
@@ -111,6 +114,24 @@ int dco_do_write(dco_context_t *dco, int peer_id, struct buffer *buf);
  */
 void dco_event_set(dco_context_t *dco, struct event_set *es, void *arg);
 
+/**
+ * Install the key material in DCO for the specified peer.
+ * The key is installed in the primary slot when no other key was yet installed.
+ * Any subsequent invocation will install the key in the secondary slot.
+ *
+ * @param multi     the TLS context of the current instance
+ * @param ks        the state of the key being installed
+ * @param key2      the container for the raw key material
+ * @param key_direction the key direction to be used to extract the material
+ * @param ciphername    the name of the cipher to use the key with
+ * @param server    whether we are running on a server instance or not
+ *
+ * @return          0 on success or a negative error code otherwise
+ */
+int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                    const struct key2 *key2, int key_direction,
+                    const char *ciphername, bool server);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -163,5 +184,13 @@ dco_event_set(dco_context_t *dco, struct event_set *es, void *arg)
 {
 }
 
+static inline int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    return 0;
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/init.c +7 -4
index b6e1707f..338d797b 100644
@@ -2226,8 +2226,9 @@ do_deferred_p2p_ncp(struct context *c)
     }
 #endif
 
-    if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                          frame_fragment, get_link_socket_info(c)))
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
+                                          &c->c2.frame, frame_fragment,
+                                          get_link_socket_info(c)))
     {
         msg(D_TLS_ERRORS, "ERROR: failed to set crypto cipher");
         return false;
@@ -2340,8 +2341,10 @@ do_deferred_options(struct context *c, const unsigned int found)
 #endif
 
         struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                              frame_fragment, get_link_socket_info(c)))
+        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                              &c->options, &c->c2.frame,
+                                              frame_fragment,
+                                              get_link_socket_info(c)))
         {
             msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
             return false;
src/openvpn/multi.c +1 -1
index ba2f6d58..c72575ae 100644
@@ -2286,7 +2286,7 @@ multi_client_generate_tls_keys(struct context *c)
     }
 #endif
     struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-    if (!tls_session_update_crypto_params(session, &c->options,
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
                                           &c->c2.frame, frame_fragment,
                                           get_link_socket_info(c)))
     {
src/openvpn/options.c +10 -0
index 87d6fc31..dba9d02c 100644
@@ -3194,6 +3194,16 @@ options_postprocess_mutate_ce(struct options *o, struct connection_entry *ce)
         ce->explicit_exit_notification = 0;
     }
 
+    /* when DCO is in use we can't send data channel packets.
+     * EEN needs to be re-implemented over the control channel in order
+     * to work.
+     */
+    if (dco_enabled(o) && ce->explicit_exit_notification)
+    {
+        msg(M_WARN, "NOTICE: --explicit-exit-notify ignored when "
+            "data channel offload is in use");
+        ce->explicit_exit_notification = 0;
+    }
 }
 
 #ifdef _WIN32
src/openvpn/ssl.c +54 -19

145 lines not on this page · raw message

src/openvpn/ssl.h +5 -2
index c8802707..76b1b674 100644
@@ -426,6 +426,7 @@ void tls_update_remote_addr(struct tls_multi *multi,
  * channel keys based on the supplied options. Does nothing if keys are already
  * generated.
  *
+ * @param multi           The TLS object for this instance.
  * @param session         The TLS session to update.
  * @param options         The options to use when updating session.
  * @param frame           The frame options for this session (frame overhead is
@@ -436,7 +437,8 @@ void tls_update_remote_addr(struct tls_multi *multi,
  *
  * @return true if updating succeeded or keys are already generated, false otherwise.
  */
-bool tls_session_update_crypto_params(struct tls_session *session,
+bool tls_session_update_crypto_params(struct tls_multi *multi,
+                                      struct tls_session *session,
                                       struct options *options,
                                       struct frame *frame,
                                       struct frame *frame_fragment,
@@ -551,7 +553,8 @@ show_available_tls_ciphers(const char *cipher_list,
  * can thus be called only once per session.
  */
 bool
-tls_session_generate_data_channel_keys(struct tls_session *session);
+tls_session_generate_data_channel_keys(struct tls_multi *multi,
+                                       struct tls_session *session);
 
 /**
  * Load ovpn.xkey provider used for external key signing
Signature · 1 line
2.35.1
#65 lore raw

Re: [Openvpn-devel] [PATCH v2 09/25] dco: configure keys in DCO right after generating them

Quoted 8 lines

index 87d6fc31..dba9d02c 100644

src/openvpn/options.c +0 -0
@@ -3194,6 +3194,16 @@ options_postprocess_mutate_ce(struct options *o, struct connection_entry *ce)
          ce->explicit_exit_notification = 0;
      }
 

src/openvpn/options.c options_postprocess_mutate_ce(struct options *o, struct connection_entry *ce) · line 3206

Quoted 10 lines
src/openvpn/options.c +10 -0
+    /* when DCO is in use we can't send data channel packets.
+     * EEN needs to be re-implemented over the control channel in order
+     * to work.
+     */
+    if (dco_enabled(o) && ce->explicit_exit_notification)
+    {
+        msg(M_WARN, "NOTICE: --explicit-exit-notify ignored when "
+            "data channel offload is in use");
+        ce->explicit_exit_notification = 0;
+    }

I don't like this. There is already the patch from me that allows this on the control channel. So we need a better solution than disabling it when dco is enabled. Did you check what happens if data packets are tried being sent when DCO is enabled? Maybe that just has a warning now that no key is active.

Arne

#66 lore raw

Re: [Openvpn-devel] [PATCH v2 09/25] dco: configure keys in DCO right after generating them

Quoted 12 lines · Arne Schwabe
Quoted 12 lines

index 87d6fc31..dba9d02c 100644

src/openvpn/options.c +4 -0
@@ -3194,6 +3194,16 @@ options_postprocess_mutate_ce(struct options 
 *o, struct connection_entry *ce)
           ce->explicit_exit_notification = 0;
       }
+    /* when DCO is in use we can't send data channel packets.
+     * EEN needs to be re-implemented over the control channel in order
+     * to work.
+     */
Quoted 8 lines

src/openvpn/options.c options_postprocess_mutate_ce(struct options *o, struct connection_entry *ce) · line 3206

Quoted 6 lines
src/openvpn/options.c +6 -0
+    if (dco_enabled(o) && ce->explicit_exit_notification)
+    {
+        msg(M_WARN, "NOTICE: --explicit-exit-notify ignored when "
+            "data channel offload is in use");
+        ce->explicit_exit_notification = 0;
+    }

I don't like this. There is already the patch from me that allows this on the control channel. So we need a better solution than disabling it when dco is enabled. Did you check what happens if data packets are tried being sent when DCO is enabled? Maybe that just has a warning now that no key is active.

Haven't tried, but by looking at the code it seems OpenVPN will assert out because it checks for the key context being not-null.

So, since it cannot work the way it is right now, I wanted to at least disable it.

Once implemented on the control channel it could then be re-enabled again.

Cheers,

Signature · 1 line
Antonio Quartulli
#67 lore raw

PATCH v3 9/25 dco: configure keys in DCO right after generating them

In reply to Antonio Quartulli

The ovpn-dco kernel module needs to be informed about the keys to be used to encrypt/decrypt data traffic to/from a peer.

Configure keys in DCO right afte they are generated by the SSL code, to avoid keeping them in memory longer than needed.

---

Changes from v2:
* re-enable explicit-exit-notification in every case
* add check to drop packet when attempting to send data packet and DCO is enabled (print warning as well)

Changes from v1:
* adapt to new member name dco_enabled
* invert if blocks and condition in init_key_contexts() [and use 'else']
* fix comment for init_key_contexts()
* disable explicit-exit-notification in mutate_ce() when DCO is enabled

---
 src/openvpn/dco.c     | 55 ++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     | 29 +++++++++++++++++
 src/openvpn/forward.c |  7 +++++
 src/openvpn/init.c    | 11 ++++---
 src/openvpn/multi.c   |  2 +-
 src/openvpn/options.c |  1 -
 src/openvpn/ssl.c     | 73 ++++++++++++++++++++++++++++++++-----------
 src/openvpn/ssl.h     |  7 +++--
 8 files changed, 158 insertions(+), 27 deletions(-)
src/openvpn/dco.c +55 -0
index b3fd135f..0471e4d0 100644
@@ -33,12 +33,67 @@
 #if defined(ENABLE_DCO)
 
 #include "syshead.h"
+#include "crypto.h"
 #include "dco.h"
+#include "errlevel.h"
 #include "networking.h"
+#include "openvpn.h"
 #include "options.h"
+#include "ssl_common.h"
 #include "ssl_ncp.h"
 #include "tun.h"
 
+static int
+dco_install_key(struct tls_multi *multi, struct key_state *ks,
+                const uint8_t *encrypt_key, const uint8_t *encrypt_iv,
+                const uint8_t *decrypt_key, const uint8_t *decrypt_iv,
+                const char *ciphername)
+
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d keyid=%d", __func__, multi->peer_id,
+        ks->key_id);
+
+    /* Install a key in the PRIMARY slot only when no other key exist.
+     * From that moment on, any new key will be installed in the SECONDARY
+     * slot and will be promoted to PRIMARY when userspace says so (a swap
+     * will be performed in that case)
+     */
+    dco_key_slot_t slot = OVPN_KEY_SLOT_PRIMARY;
+    if (multi->dco_keys_installed > 0)
+    {
+        slot = OVPN_KEY_SLOT_SECONDARY;
+    }
+
+    int ret = dco_new_key(multi->dco, multi->peer_id, ks->key_id, slot,
+                          encrypt_key, encrypt_iv,
+                          decrypt_key, decrypt_iv,
+                          ciphername);
+    if ((ret == 0) && (multi->dco_keys_installed < 2))
+    {
+        multi->dco_keys_installed++;
+        ks->dco_status = (slot == OVPN_KEY_SLOT_PRIMARY) ? DCO_INSTALLED_PRIMARY :
+                         DCO_INSTALLED_SECONDARY;
+    }
+
+    return ret;
+}
+
+int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    struct key_direction_state kds;
+    key_direction_state_init(&kds, key_direction);
+
+    return dco_install_key(multi, ks,
+                           key2->keys[kds.out_key].cipher,
+                           key2->keys[(int)server].hmac,
+                           key2->keys[kds.in_key].cipher,
+                           key2->keys[1 - (int)server].hmac,
+                           ciphername);
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {
src/openvpn/dco.h +29 -0
index 063e5028..1692f5c3 100644
@@ -35,7 +35,10 @@
  * order problems)
  */
 struct event_set;
+struct key2;
+struct key_state;
 struct options;
+struct tls_multi;
 struct tuntap;
 
 #define DCO_DEFAULT_METRIC  200
@@ -111,6 +114,24 @@ int dco_do_write(dco_context_t *dco, int peer_id, struct buffer *buf);
  */
 void dco_event_set(dco_context_t *dco, struct event_set *es, void *arg);
 
+/**
+ * Install the key material in DCO for the specified peer.
+ * The key is installed in the primary slot when no other key was yet installed.
+ * Any subsequent invocation will install the key in the secondary slot.
+ *
+ * @param multi     the TLS context of the current instance
+ * @param ks        the state of the key being installed
+ * @param key2      the container for the raw key material
+ * @param key_direction the key direction to be used to extract the material
+ * @param ciphername    the name of the cipher to use the key with
+ * @param server    whether we are running on a server instance or not
+ *
+ * @return          0 on success or a negative error code otherwise
+ */
+int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                    const struct key2 *key2, int key_direction,
+                    const char *ciphername, bool server);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -163,5 +184,13 @@ dco_event_set(dco_context_t *dco, struct event_set *es, void *arg)
 {
 }
 
+static inline int
+init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
+                const struct key2 *key2, int key_direction,
+                const char *ciphername, bool server)
+{
+    return 0;
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/forward.c +7 -0
index 6afe152b..28f3c088 100644
@@ -530,6 +530,13 @@ encrypt_sign(struct context *c, bool comp_frag)
     const uint8_t *orig_buf = c->c2.buf.data;
     struct crypto_options *co = NULL;
 
+    if (dco_enabled(&c->options))
+    {
+        msg(M_WARN, "Attempting to send data packet while data channel offload is in use. "
+            "Dropping packet");
+        c->c2.buf.len = 0;
+    }
+
     /*
      * Drop non-TLS outgoing packet if client-connect script/plugin
      * has not yet succeeded. In non-TLS tls_multi mode is not defined
src/openvpn/init.c +7 -4
index b6e1707f..338d797b 100644
@@ -2226,8 +2226,9 @@ do_deferred_p2p_ncp(struct context *c)
     }
 #endif
 
-    if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                          frame_fragment, get_link_socket_info(c)))
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
+                                          &c->c2.frame, frame_fragment,
+                                          get_link_socket_info(c)))
     {
         msg(D_TLS_ERRORS, "ERROR: failed to set crypto cipher");
         return false;
@@ -2340,8 +2341,10 @@ do_deferred_options(struct context *c, const unsigned int found)
 #endif
 
         struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(session, &c->options, &c->c2.frame,
-                                              frame_fragment, get_link_socket_info(c)))
+        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                              &c->options, &c->c2.frame,
+                                              frame_fragment,
+                                              get_link_socket_info(c)))
         {
             msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
             return false;
src/openvpn/multi.c +1 -1
index ba2f6d58..c72575ae 100644
@@ -2286,7 +2286,7 @@ multi_client_generate_tls_keys(struct context *c)
     }
 #endif
     struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-    if (!tls_session_update_crypto_params(session, &c->options,
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session, &c->options,
                                           &c->c2.frame, frame_fragment,
                                           get_link_socket_info(c)))
     {
src/openvpn/options.c +0 -1
index 11919f6d..3eae14a7 100644
@@ -3193,7 +3193,6 @@ options_postprocess_mutate_ce(struct options *o, struct connection_entry *ce)
         msg(M_WARN, "NOTICE: --explicit-exit-notify ignored for --proto tcp");
         ce->explicit_exit_notification = 0;
     }
-
 }
 
 #ifdef _WIN32
src/openvpn/ssl.c +54 -19

145 lines not on this page · raw message

src/openvpn/ssl.h +5 -2
index c8802707..76b1b674 100644
@@ -426,6 +426,7 @@ void tls_update_remote_addr(struct tls_multi *multi,
  * channel keys based on the supplied options. Does nothing if keys are already
  * generated.
  *
+ * @param multi           The TLS object for this instance.
  * @param session         The TLS session to update.
  * @param options         The options to use when updating session.
  * @param frame           The frame options for this session (frame overhead is
@@ -436,7 +437,8 @@ void tls_update_remote_addr(struct tls_multi *multi,
  *
  * @return true if updating succeeded or keys are already generated, false otherwise.
  */
-bool tls_session_update_crypto_params(struct tls_session *session,
+bool tls_session_update_crypto_params(struct tls_multi *multi,
+                                      struct tls_session *session,
                                       struct options *options,
                                       struct frame *frame,
                                       struct frame *frame_fragment,
@@ -551,7 +553,8 @@ show_available_tls_ciphers(const char *cipher_list,
  * can thus be called only once per session.
  */
 bool
-tls_session_generate_data_channel_keys(struct tls_session *session);
+tls_session_generate_data_channel_keys(struct tls_multi *multi,
+                                       struct tls_session *session);
 
 /**
  * Load ovpn.xkey provider used for external key signing
Signature · 1 line
2.35.1
Acked-by on v3 9/25
#68 lore raw

Re: [Openvpn-devel] [PATCH v3 09/25] dco: configure keys in DCO right after generating them

Am 28.07.22 um 17:20 schrieb Antonio Quartulli:

Quoted 14 lines

The ovpn-dco kernel module needs to be informed about the keys to be used to encrypt/decrypt data traffic to/from a peer.

Configure keys in DCO right afte they are generated by the SSL code, to avoid keeping them in memory longer than needed.

---

Changes from v2:
* re-enable explicit-exit-notification in every case
* add check to drop packet when attempting to send data packet and DCO is enabled (print warning as well)

Quoted 5 lines

Changes from v1:
* adapt to new member name dco_enabled
* invert if blocks and condition in init_key_contexts() [and use 'else']
* fix comment for init_key_contexts()
* disable explicit-exit-notification in mutate_ce() when DCO is enabled

#69 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: configure keys in DCO right after generating them

In reply to Antonio Quartulli

Tests without --enable-dco (full server side test) - passes everything.

Test with --enable-dco but no Kernel support (client side only) - also passes everything (spurious failure on one of the p2p tests, but that was likely related to "too many tests running in parallel").

Did not test on a system with DCO kernel support, as we do not have all bits and pices integrated yet.

I have not tested the "drop packet" case in forward.c (as that needs a DCO enabled kernel). Putting that on my "test with full DCO!" list
- the code certainly looks good.

Stared at code for a bit (even though it has the ACK).

Not sure I like the call chain ssl.c->dco.c->crypto.c for init_key_dco_bi() -> key_direction_state_init()... but changing that would require a bit more ssl.c/crypto.c refactoring.

For the non-DCO cases, the _bi stuff has "key_ctx_update_implicit_iv()" calls - are these done by the DCO kernel side? Can't find anything about IVs in the init_key_dco_bi()->... call chain...

Do the calls to "tls_session_update_crypto_params()" really need to get a "session" parameter passed in now? Since they get c->c2.tls_multi now, "sesion" is just one pointer deref away... so this might warrant a cleanup patch later on.

Your patch has been applied to the master branch.

commit 6a5612fe82453915755aca945ff4e876a25f582a

Author: Antonio Quartulli
Date:   Thu Jul 28 17:20:12 2022 +0200
     dco: configure keys in DCO right after generating them

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24758.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#70 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: configure keys in DCO right after generating them

In reply to Antonio Quartulli

Tests without --enable-dco (full server side test) - passes everything.

Test with --enable-dco but no Kernel support (client side only) - also passes everything (spurious failure on one of the p2p tests, but that was likely related to "too many tests running in parallel").

Did not test on a system with DCO kernel support, as we do not have all bits and pices integrated yet.

I have not tested the "drop packet" case in forward.c (as that needs a DCO enabled kernel). Putting that on my "test with full DCO!" list
- the code certainly looks good.

Stared at code for a bit (even though it has the ACK).

Not sure I like the call chain ssl.c->dco.c->crypto.c for init_key_dco_bi() -> key_direction_state_init()... but changing that would require a bit more ssl.c/crypto.c refactoring.

For the non-DCO cases, the _bi stuff has "key_ctx_update_implicit_iv()" calls - are these done by the DCO kernel side? Can't find anything about IVs in the init_key_dco_bi()->... call chain...

Do the calls to "tls_session_update_crypto_params()" really need to get a "session" parameter passed in now? Since they get c->c2.tls_multi now, "sesion" is just one pointer deref away... so this might warrant a cleanup patch later on.

Your patch has been applied to the master branch.

commit 6a5612fe82453915755aca945ff4e876a25f582a

Author: Antonio Quartulli
Date:   Thu Jul 28 17:20:12 2022 +0200
     dco: configure keys in DCO right after generating them

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Arne Schwabe <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24758.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#71 lore raw

PATCH 10/25 dco: periodically check and possibly rotate/delete keys

In reply to Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

---
 src/openvpn/dco.c     | 97 +++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     | 14 +++++++
 src/openvpn/forward.c | 19 +++++++++
 3 files changed, 130 insertions(+)
src/openvpn/dco.c +97 -0
index e38614fa..473eb564 100644
@@ -92,6 +92,103 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                            ciphername);
 }
 
+/**
+ * Find a usable key that is not the primary (i.e. the secondary key)
+ *
+ * @param multi     The TLS struct to retrieve keys from
+ * @param primary   The primary key that should be skipped during the scan
+ *
+ * @return          The secondary key or NULL if none could be found
+ */
+static struct key_state *
+dco_get_secondary_key(struct tls_multi *multi, const struct key_state *primary)
+{
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        struct key_ctx_bi *key = &ks->crypto_options.key_ctx_bi;
+
+        if (ks == primary)
+        {
+            continue;
+        }
+
+        if (ks->state >= S_GENERATED_KEYS && ks->authenticated == KS_AUTH_TRUE)
+        {
+            ASSERT(key->initialized);
+            return ks;
+        }
+    }
+
+    return NULL;
+}
+
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
 static bool
 dco_check_option_conflict_ce(const struct connection_entry *ce, int msglevel)
 {
src/openvpn/dco.h +14 -0
index b081c6fa..cb7f7e4f 100644
@@ -130,6 +130,14 @@ int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                     const struct key2 *key2, int key_direction,
                     const char *ciphername, bool server);
 
+/**
+ * Possibly swap or wipe keys from DCO
+ *
+ * @param dco           DCO device context
+ * @param multi         TLS multi instance
+ */
+void dco_update_keys(dco_context_t *dco, struct tls_multi *multi);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -190,5 +198,11 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
     return 0;
 }
 
+static inline void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    ASSERT(false);
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/forward.c +19 -0
index 6afe152b..99898e01 100644
@@ -41,6 +41,7 @@
 #include "dhcp.h"
 #include "common.h"
 #include "ssl_verify.h"
+#include "dco.h"
 
 #include "memdbg.h"
 
@@ -140,6 +141,18 @@ context_reschedule_sec(struct context *c, int sec)
     }
 }
 
+void
+check_dco_key_status(struct context *c)
+{
+    /* DCO context is not yet initialised or enabled */
+    if (!dco_enabled(&c->options))
+    {
+        return;
+    }
+
+    dco_update_keys(&c->c1.tuntap->dco, c->c2.tls_multi);
+}
+
 /*
  * In TLS mode, let TLS level respond to any control-channel
  * packets which were received, or prepare any packets for
@@ -182,6 +195,12 @@ check_tls(struct context *c)
 
     interval_schedule_wakeup(&c->c2.tmp_int, &wakeup);
 
+    /* Our current code has no good hooks in the TLS machinery to update
+     * DCO keys. So we check the key status after the whole TLS machinery
+     * has been completed and potentially update them
+     */
+    check_dco_key_status(c);
+
     if (wakeup)
     {
         context_reschedule_sec(c, wakeup);
Signature · 1 line
2.35.1
#72 lore raw

Re: [Openvpn-devel] [PATCH 10/25] dco: periodically check and possibly rotate/delete keys

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 1 line

+ ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);

It would to be good to explain this assertion. I just spend too long understanding it and I understand the code. Something along the lines that we expect that primary key has been installed before here.

This code might also be subject to very unlikely (maybe not even possible) race condition.

When deferred auth is active and the deferred auth just happens at the edge of the window before timing out tls_select_encryption_key /might/ already return the new key without it having been a valid secondary key before. But maybe we fail for the timeout check before.

Arne

#73 lore raw

PATCH v2 10/25 dco: periodically check and possibly rotate/delete keys

In reply to Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

---

Changes from v1:
* added comments to ASSERT() in dco_update_keys()

---
 src/openvpn/dco.c     | 101 ++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     |  14 ++++++
 src/openvpn/forward.c |  19 ++++++++
 3 files changed, 134 insertions(+)
src/openvpn/dco.c +101 -0
index 0471e4d0..07c16de1 100644
@@ -94,6 +94,107 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                            ciphername);
 }
 
+/**
+ * Find a usable key that is not the primary (i.e. the secondary key)
+ *
+ * @param multi     The TLS struct to retrieve keys from
+ * @param primary   The primary key that should be skipped during the scan
+ *
+ * @return          The secondary key or NULL if none could be found
+ */
+static struct key_state *
+dco_get_secondary_key(struct tls_multi *multi, const struct key_state *primary)
+{
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        struct key_ctx_bi *key = &ks->crypto_options.key_ctx_bi;
+
+        if (ks == primary)
+        {
+            continue;
+        }
+
+        if (ks->state >= S_GENERATED_KEYS && ks->authenticated == KS_AUTH_TRUE)
+        {
+            ASSERT(key->initialized);
+            return ks;
+        }
+    }
+
+    return NULL;
+}
+
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* either we have no primary key at all or, if we do, it must have been
+     * installed already (keys are installed upon generation in the TLS code)
+     */
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    /* same reason as the primary key above */
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {
src/openvpn/dco.h +14 -0
index 1692f5c3..b926e236 100644
@@ -132,6 +132,14 @@ int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                     const struct key2 *key2, int key_direction,
                     const char *ciphername, bool server);
 
+/**
+ * Possibly swap or wipe keys from DCO
+ *
+ * @param dco           DCO device context
+ * @param multi         TLS multi instance
+ */
+void dco_update_keys(dco_context_t *dco, struct tls_multi *multi);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -192,5 +200,11 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
     return 0;
 }
 
+static inline void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    ASSERT(false);
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/forward.c +19 -0
index 28f3c088..38d2683c 100644
@@ -41,6 +41,7 @@
 #include "dhcp.h"
 #include "common.h"
 #include "ssl_verify.h"
+#include "dco.h"
 
 #include "memdbg.h"
 
@@ -140,6 +141,18 @@ context_reschedule_sec(struct context *c, int sec)
     }
 }
 
+void
+check_dco_key_status(struct context *c)
+{
+    /* DCO context is not yet initialised or enabled */
+    if (!dco_enabled(&c->options))
+    {
+        return;
+    }
+
+    dco_update_keys(&c->c1.tuntap->dco, c->c2.tls_multi);
+}
+
 /*
  * In TLS mode, let TLS level respond to any control-channel
  * packets which were received, or prepare any packets for
@@ -182,6 +195,12 @@ check_tls(struct context *c)
 
     interval_schedule_wakeup(&c->c2.tmp_int, &wakeup);
 
+    /* Our current code has no good hooks in the TLS machinery to update
+     * DCO keys. So we check the key status after the whole TLS machinery
+     * has been completed and potentially update them
+     */
+    check_dco_key_status(c);
+
     if (wakeup)
     {
         context_reschedule_sec(c, wakeup);
Signature · 1 line
2.35.1
#74 lore raw

Re: [Openvpn-devel] [PATCH v2 10/25] dco: periodically check and possibly rotate/delete keys

Quoted 7 lines · Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

[...]

Small musings about the code comments:

Quoted 28 lines
src/openvpn/dco.c +28 -0
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* either we have no primary key at all or, if we do, it must have been
+     * installed already (keys are installed upon generation in the TLS code)
+     */
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 165

Quoted 10 lines
src/openvpn/dco.c +10 -0
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    /* same reason as the primary key above */
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys

I found the use of "userspace" here a bit confusing. This code is part of userspace, right? So wouldn't it be easier to say "we have promoted it"?

Quoted 40 lines
src/openvpn/dco.c +32 -0
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {
src/openvpn/dco.h +0 -0
index 1692f5c3..b926e236 100644

src/openvpn/dco.h int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 136

Quoted 5 lines
src/openvpn/dco.h +2 -0
@@ -132,6 +132,14 @@ int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                     const struct key2 *key2, int key_direction,
                     const char *ciphername, bool server);
 
+/**
+ * Possibly swap or wipe keys from DCO

I would actually prefer to be a bit more explicit here:
"Sync our key status to DCO, potentially swapping or wiping keys"

src/openvpn/dco.h int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 145

Quoted 9 lines
src/openvpn/dco.h +6 -0
+ *
+ * @param dco           DCO device context
+ * @param multi         TLS multi instance
+ */
+void dco_update_keys(dco_context_t *dco, struct tls_multi *multi);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;

[...]

Regards,

Signature · 1 line
  Frank Lichtenheld
#75 lore raw

Re: [Openvpn-devel] [PATCH v2 10/25] dco: periodically check and possibly rotate/delete keys

Hi,

Quoted 45 lines · Frank Lichtenheld
Quoted 7 lines · Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

[...]

Small musings about the code comments:

Quoted 33 lines
src/openvpn/dco.c +33 -0
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* either we have no primary key at all or, if we do, it must have been
+     * installed already (keys are installed upon generation in the TLS code)
+     */
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
Quoted 8 lines

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 165

Quoted 5 lines
src/openvpn/dco.c +5 -0
+    /* same reason as the primary key above */
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys

I found the use of "userspace" here a bit confusing. This code is part of userspace, right? So wouldn't it be easier to say "we have promoted it"?

Sure - whatever makes it easier to understand.

Quoted 44 lines
Quoted 44 lines
src/openvpn/dco.c +32 -0
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
  static bool
  dco_check_option_conflict_platform(int msglevel, const struct options *o)
  {
src/openvpn/dco.h +0 -0
index 1692f5c3..b926e236 100644
@@ -132,6 +132,14 @@ int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                      const struct key2 *key2, int key_direction,
                      const char *ciphername, bool server);
 
Quoted 5 lines

src/openvpn/dco.h int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 136

Quoted 2 lines
src/openvpn/dco.h +2 -0
+/**
+ * Possibly swap or wipe keys from DCO

I would actually prefer to be a bit more explicit here:
"Sync our key status to DCO, potentially swapping or wiping keys"

Makes sense. Thanks

Quoted 14 lines

src/openvpn/dco.h int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 145

Quoted 10 lines
src/openvpn/dco.h +6 -0
+ *
+ * @param dco           DCO device context
+ * @param multi         TLS multi instance
+ */
+void dco_update_keys(dco_context_t *dco, struct tls_multi *multi);
+
  #else /* if defined(ENABLE_DCO) */
 
  typedef void *dco_context_t;

[...]

Regards,

Signature · 1 line
Antonio Quartulli
#76 lore raw

Re: [Openvpn-devel] [PATCH v2 10/25] dco: periodically check and possibly rotate/delete keys

In reply to Antonio Quartulli
Quoted 7 lines · Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

Thinking about the actual code now.

[...]

Quoted 40 lines
src/openvpn/dco.c +40 -0
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* either we have no primary key at all or, if we do, it must have been
+     * installed already (keys are installed upon generation in the TLS code)
+     */
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    /* same reason as the primary key above */
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 177

Quoted 10 lines
src/openvpn/dco.c +10 -0
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }

Why do we have no error handling? We just assume that DCO swapped the keys but if dco_swap_keys returns an error that is probably not true?

Also can we really be sure that the kernel state is correct? I.e. we have determined we have swapped the keys, but can we be sure that the kernel actually has the two same keys and so a swap will replicate our state?

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 183

Quoted 6 lines
src/openvpn/dco.c +6 -0
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);

Would error handling make sense here? It is clearly not as critical as for dco_swap_keys because a delete is idempotent.

Quoted 7 lines
src/openvpn/dco.c +7 -0
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 200

Quoted 10 lines
src/openvpn/dco.c +7 -0
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {

Regards,
Frank

#77 lore raw

Re: [Openvpn-devel] [PATCH v2 10/25] dco: periodically check and possibly rotate/delete keys

Hi,

Quoted 57 lines · Frank Lichtenheld
Quoted 7 lines · Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

Thinking about the actual code now.

[...]

Quoted 45 lines
src/openvpn/dco.c +45 -0
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* either we have no primary key at all or, if we do, it must have been
+     * installed already (keys are installed upon generation in the TLS code)
+     */
+    ASSERT(!primary || primary->dco_status != DCO_NOT_INSTALLED);
+
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    /* same reason as the primary key above */
+    ASSERT(!secondary || secondary->dco_status != DCO_NOT_INSTALLED);
+
+    /* the current primary key was installed as secondary in DCO, this means
+     * that userspace has promoted it and we should tell DCO to swap keys
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        dco_swap_keys(dco, multi->peer_id);
Quoted 8 lines

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 177

Quoted 5 lines
src/openvpn/dco.c +5 -0
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }

Why do we have no error handling? We just assume that DCO swapped the keys but if dco_swap_keys returns an error that is probably not true?

Right - the call may fail for various reasons and in that case we should assume that no swap happened.

However, I am not sure what OpenVPN should do at this point. Trigger SIGUSR1 for this instance? The other end won't notice anything anyway.

Quoted 5 lines

Also can we really be sure that the kernel state is correct? I.e. we have determined we have swapped the keys, but can we be sure that the kernel actually has the two same keys and so a swap will replicate our state?

We have no way to confirm that, because we have no way to identify the keys that have been installed. We can only trust that we injected the right key at the right place.

Quoted 10 lines

src/openvpn/dco.c init_key_dco_bi(struct tls_multi *multi, struct key_state *ks, · line 183

Quoted 7 lines
src/openvpn/dco.c +6 -0
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);

Would error handling make sense here? It is clearly not as critical as for dco_swap_keys because a delete is idempotent.

The question again is "what do we do?" Maybe in this case we simply print a message but we continue as if it was successful?

Quoted 14 lines
Quoted 14 lines
src/openvpn/dco.c +13 -0
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
Quoted 10 lines
Quoted 4 lines

+
static bool
dco_check_option_conflict_platform(int msglevel, const struct options *o) {

Regards,
Frank

Thanks!

Quoted 6 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#78 lore raw

PATCH v3 10/25 dco: periodically check and possibly rotate/delete keys

In reply to Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

---

Changes from v2:
* reworked dco_update_keys():
** removed existing ASSERTs on keys and converted into something more
logic and related to the flow
** fixed comment about userspace (use "we" for more clarity) ** add error code handling. The idea is to abort operations as soon as a
failure is detected, so that the next iteration can make another attempt.

Changes from v1:
* added comments to ASSERT() in dco_update_keys()

---
 src/openvpn/dco.c     | 122 ++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     |  14 +++++
 src/openvpn/forward.c |  19 +++++++
 3 files changed, 155 insertions(+)
src/openvpn/dco.c +122 -0
index b0dd922a..dfdf10ea 100644
@@ -94,6 +94,128 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                            ciphername);
 }
 
+/**
+ * Find a usable key that is not the primary (i.e. the secondary key)
+ *
+ * @param multi     The TLS struct to retrieve keys from
+ * @param primary   The primary key that should be skipped during the scan
+ *
+ * @return          The secondary key or NULL if none could be found
+ */
+static struct key_state *
+dco_get_secondary_key(struct tls_multi *multi, const struct key_state *primary)
+{
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        struct key_ctx_bi *key = &ks->crypto_options.key_ctx_bi;
+
+        if (ks == primary)
+        {
+            continue;
+        }
+
+        if (ks->state >= S_GENERATED_KEYS && ks->authenticated == KS_AUTH_TRUE)
+        {
+            ASSERT(key->initialized);
+            return ks;
+        }
+    }
+
+    return NULL;
+}
+
+void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    msg(D_DCO_DEBUG, "%s: peer_id=%d", __func__, multi->peer_id);
+
+    /* this function checks if keys have to be swapped or erased, therefore it
+     * can't do much if we don't have any key installed
+     */
+    if (multi->dco_keys_installed == 0)
+    {
+        return;
+    }
+
+    struct key_state *primary = tls_select_encryption_key(multi);
+    /* no primary key available -> no usable key exists, therefore we should
+     * tell DCO to simply wipe all keys
+     */
+    if (!primary)
+    {
+        msg(D_DCO, "No encryption key found. Purging data channel keys");
+
+        int ret = dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_PRIMARY);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot delete primary key during wipe: %s (%d)", strerror(-ret), ret);
+            return;
+        }
+
+        ret = dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot delete secondary key during wipe: %s (%d)", strerror(-ret), ret);
+            return;
+        }
+
+        multi->dco_keys_installed = 0;
+        return;
+    }
+
+    /* if we have a primary key, it must have been installed already (keys
+     * are installed upon generation in the TLS code)
+     */
+    ASSERT(primary->dco_status != DCO_NOT_INSTALLED);
+
+    struct key_state *secondary = dco_get_secondary_key(multi, primary);
+    /* the current primary key was installed as secondary in DCO, this means
+     * we have promoted it and we should tell DCO to swap keys
+     */
+    if (primary->dco_status == DCO_INSTALLED_SECONDARY)
+    {
+        msg(D_DCO_DEBUG, "Swapping primary and secondary keys, now: id1=%d id2=%d",
+            primary->key_id, secondary ? secondary->key_id : -1);
+
+        int ret = dco_swap_keys(dco, multi->peer_id);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot swap keys: %s (%d)", strerror(-ret), ret);
+            return;
+        }
+
+        primary->dco_status = DCO_INSTALLED_PRIMARY;
+        if (secondary)
+        {
+            ASSERT(secondary->dco_status == DCO_INSTALLED_PRIMARY);
+            secondary->dco_status = DCO_INSTALLED_SECONDARY;
+        }
+    }
+
+    /* if we have no secondary key anymore, inform DCO about it */
+    if (!secondary && multi->dco_keys_installed == 2)
+    {
+        int ret = dco_del_key(dco, multi->peer_id, OVPN_KEY_SLOT_SECONDARY);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot delete secondary key: %s (%d)", strerror(-ret), ret);
+            return;
+        }
+        multi->dco_keys_installed = 1;
+    }
+
+    /* all keys that are not installed are set to NOT installed */
+    for (int i = 0; i < KEY_SCAN_SIZE; ++i)
+    {
+        struct key_state *ks = get_key_scan(multi, i);
+        if (ks != primary && ks != secondary)
+        {
+            ks->dco_status = DCO_NOT_INSTALLED;
+        }
+    }
+}
+
 static bool
 dco_check_option_conflict_platform(int msglevel, const struct options *o)
 {
src/openvpn/dco.h +14 -0
index 1692f5c3..b926e236 100644
@@ -132,6 +132,14 @@ int init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
                     const struct key2 *key2, int key_direction,
                     const char *ciphername, bool server);
 
+/**
+ * Possibly swap or wipe keys from DCO
+ *
+ * @param dco           DCO device context
+ * @param multi         TLS multi instance
+ */
+void dco_update_keys(dco_context_t *dco, struct tls_multi *multi);
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -192,5 +200,11 @@ init_key_dco_bi(struct tls_multi *multi, struct key_state *ks,
     return 0;
 }
 
+static inline void
+dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
+{
+    ASSERT(false);
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/forward.c +19 -0
index 28f3c088..38d2683c 100644
@@ -41,6 +41,7 @@
 #include "dhcp.h"
 #include "common.h"
 #include "ssl_verify.h"
+#include "dco.h"
 
 #include "memdbg.h"
 
@@ -140,6 +141,18 @@ context_reschedule_sec(struct context *c, int sec)
     }
 }
 
+void
+check_dco_key_status(struct context *c)
+{
+    /* DCO context is not yet initialised or enabled */
+    if (!dco_enabled(&c->options))
+    {
+        return;
+    }
+
+    dco_update_keys(&c->c1.tuntap->dco, c->c2.tls_multi);
+}
+
 /*
  * In TLS mode, let TLS level respond to any control-channel
  * packets which were received, or prepare any packets for
@@ -182,6 +195,12 @@ check_tls(struct context *c)
 
     interval_schedule_wakeup(&c->c2.tmp_int, &wakeup);
 
+    /* Our current code has no good hooks in the TLS machinery to update
+     * DCO keys. So we check the key status after the whole TLS machinery
+     * has been completed and potentially update them
+     */
+    check_dco_key_status(c);
+
     if (wakeup)
     {
         context_reschedule_sec(c, wakeup);
Signature · 1 line
2.35.1
#79 lore raw

Re: [Openvpn-devel] [PATCH v3 10/25] dco: periodically check and possibly rotate/delete keys

Quoted 11 lines · Antonio Quartulli

Data channel keys are periodically regenarated and installed in ovpn-dco.
However, there is a certain moment when keys are rotated in order to elect the new primary one.

Check the key status in userspace so that kernelspace can be informed as well when rotations happen.

---

Quoted 8 lines

Changes from v2:
* reworked dco_update_keys():
** removed existing ASSERTs on keys and converted into something more
logic and related to the flow
** fixed comment about userspace (use "we" for more clarity) ** add error code handling. The idea is to abort operations as soon as a
failure is detected, so that the next iteration can make another attempt.

Looks to me like this addresses all my comments.

Regards,

Signature · 1 line
  Frank Lichtenheld
Acked-by on v3 10/25
#80 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: periodically check and possibly rotate/delete keys

In reply to Antonio Quartulli

The review of that was a combined effort, mostly Frank on mail and Antonio and I on IRC - but the v3 is now easier to understand, and "should work nicely".

I have slightly extended the comment before "swapping primary and secondary keys" to make a bit clearer how we end there (we discussed that comment on IRC).

This said, I have only tested the "no DCO in kernel" cases, as the rest is still missing functinality - and for "no DCO", this becomes fairly much "no change", as the only new code otherwise is guarded with "if (!dco_enabled(&c->options))".

Your patch has been applied to the master branch.

commit cd4ba9279d9c4cf13cdc3d8c49143c57eb0fe448

Author: Antonio Quartulli
Date:   Tue Aug 2 17:16:04 2022 +0200
     dco: periodically check and possibly rotate/delete keys

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24785.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#81 lore raw

PATCH 11/25 dco: split option parsing routines

In reply to Antonio Quartulli

DCO will try to install keys upon generating them, however, this happens when parsing pushed cipher options (due to NCP).

For this reason we need to postpone parsing pushed cipher options to *after* the tunnel interface has been opened, otherwise we would have no DCO netdev object to operate on.

At the same time we split the parsing code, so that we can ensure that the NEW_PEER call can happen after the received peer-id has been parsed (it is required by all DCO API calls).

---
 src/openvpn/init.c  | 59 ++++++++++++++++++++++++++++-----------------
 src/openvpn/init.h  |  2 ++
 src/openvpn/multi.c |  7 ++++++
 3 files changed, 46 insertions(+), 22 deletions(-)
src/openvpn/init.c +37 -22
index 06911cd0..b0a4b252 100644
@@ -2068,14 +2068,6 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
                 return false;
             }
         }
-        else if (c->mode == MODE_POINT_TO_POINT)
-        {
-            if (!do_deferred_p2p_ncp(c))
-            {
-                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
-                return false;
-            }
-        }
 
         /* if --up-delay specified, open tun, do ifconfig, and run up script now */
         if (c->options.up_delay || PULL_DEFINED(&c->options))
@@ -2102,6 +2094,22 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        if (!pulled_options && c->mode == MODE_POINT_TO_POINT)
+        {
+            if (!do_deferred_p2p_ncp(c))
+            {
+                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
+                return false;
+            }
+        }
+
+        if (!finish_options(c))
+        {
+            msg(D_TLS_ERRORS, "ERROR: Failed to finish option processing");
+            return false;
+        }
+
+
         if (c->c2.did_open_tun)
         {
             c->c1.pulled_options_digest_save = c->c2.pulled_options_digest;
@@ -2307,23 +2315,30 @@ do_deferred_options(struct context *c, const unsigned int found)
         {
             return false;
         }
-        struct frame *frame_fragment = NULL;
+    }
+
+    return true;
+}
+
+bool
+finish_options(struct context *c)
+{
+    struct frame *frame_fragment = NULL;
 #ifdef ENABLE_FRAGMENT
-        if (c->options.ce.fragment)
-        {
-            frame_fragment = &c->c2.frame_fragment;
-        }
+    if (c->options.ce.fragment)
+    {
+        frame_fragment = &c->c2.frame_fragment;
+    }
 #endif
 
-        struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
-                                              &c->options, &c->c2.frame,
-                                              frame_fragment,
-                                              get_link_socket_info(c)))
-        {
-            msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
-            return false;
-        }
+    struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                          &c->options, &c->c2.frame,
+                                          frame_fragment,
+                                          get_link_socket_info(c)))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
+        return false;
     }
 
     return true;
src/openvpn/init.h +2 -0
index 2b8c2dcc..98e71d3a 100644
@@ -97,6 +97,8 @@ void reset_coarse_timers(struct context *c);
 
 bool do_deferred_options(struct context *c, const unsigned int found);
 
+bool finish_options(struct context *c);
+
 void inherit_context_child(struct context *dest,
                            const struct context *src);
 
src/openvpn/multi.c +7 -0
index c72575ae..34ab90b4 100644
@@ -2405,6 +2405,13 @@ multi_client_connect_late_setup(struct multi_context *m,
     {
         mi->context.c2.tls_multi->multi_state = CAS_FAILED;
     }
+    /* Continue processing options only if authentication hasn't failed.
+     * Otherwise it does not make sense and we may operate on a non-configured
+     * client instance */
+    else
+    {
+        finish_options(&mi->context);
+    }
 
     /* send push reply if ready */
     if (mi->context.c2.push_request_received)
Signature · 1 line
2.35.1
#82 lore raw

Re: [Openvpn-devel] [PATCH 11/25] dco: split option parsing routines

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

src/openvpn/init.c do_up(struct context *c, bool pulled_options, unsigned int option_types_found) · line 2110

Quoted 5 lines
src/openvpn/init.c +5 -0
+        if (!finish_options(c))
+        {
+            msg(D_TLS_ERRORS, "ERROR: Failed to finish option processing");
+            return false;
+        }

This error is a bit too generic for my taste. Can we make it more specific? Like "Failed to apply options when finalising tun setup" or something?

Arne

#83 lore raw

PATCH v2 11/25 dco: split option parsing routines

In reply to Antonio Quartulli

DCO will try to install keys upon generating them, however, this happens when parsing pushed cipher options (due to NCP).

For this reason we need to postpone parsing pushed cipher options to *after* the tunnel interface has been opened, otherwise we would have no DCO netdev object to operate on.

At the same time we split the parsing code, so that we can ensure that the NEW_PEER call can happen after the received peer-id has been parsed (it is required by all DCO API calls).

---

Changes from v1:
* removed error message in case of failure of finish_options(). The latter already warns the user about the failure - no need to print another generic message.

---
 src/openvpn/init.c  | 57 ++++++++++++++++++++++++++++-----------------
 src/openvpn/init.h  |  2 ++
 src/openvpn/multi.c |  7 ++++++
 3 files changed, 44 insertions(+), 22 deletions(-)
src/openvpn/init.c +35 -22
index 338d797b..db42d540 100644
@@ -2093,14 +2093,6 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
                 return false;
             }
         }
-        else if (c->mode == MODE_POINT_TO_POINT)
-        {
-            if (!do_deferred_p2p_ncp(c))
-            {
-                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
-                return false;
-            }
-        }
 
         /* if --up-delay specified, open tun, do ifconfig, and run up script now */
         if (c->options.up_delay || PULL_DEFINED(&c->options))
@@ -2127,6 +2119,20 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        if (!pulled_options && c->mode == MODE_POINT_TO_POINT)
+        {
+            if (!do_deferred_p2p_ncp(c))
+            {
+                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
+                return false;
+            }
+        }
+
+        if (!finish_options(c))
+        {
+            return false;
+        }
+
         if (c->c2.did_open_tun)
         {
             c->c1.pulled_options_digest_save = c->c2.pulled_options_digest;
@@ -2332,23 +2338,30 @@ do_deferred_options(struct context *c, const unsigned int found)
         {
             return false;
         }
-        struct frame *frame_fragment = NULL;
+    }
+
+    return true;
+}
+
+bool
+finish_options(struct context *c)
+{
+    struct frame *frame_fragment = NULL;
 #ifdef ENABLE_FRAGMENT
-        if (c->options.ce.fragment)
-        {
-            frame_fragment = &c->c2.frame_fragment;
-        }
+    if (c->options.ce.fragment)
+    {
+        frame_fragment = &c->c2.frame_fragment;
+    }
 #endif
 
-        struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
-                                              &c->options, &c->c2.frame,
-                                              frame_fragment,
-                                              get_link_socket_info(c)))
-        {
-            msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
-            return false;
-        }
+    struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                          &c->options, &c->c2.frame,
+                                          frame_fragment,
+                                          get_link_socket_info(c)))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
+        return false;
     }
 
     return true;
src/openvpn/init.h +2 -0
index 5f412a33..9389e0db 100644
@@ -97,6 +97,8 @@ void reset_coarse_timers(struct context *c);
 
 bool do_deferred_options(struct context *c, const unsigned int found);
 
+bool finish_options(struct context *c);
+
 void inherit_context_child(struct context *dest,
                            const struct context *src);
 
src/openvpn/multi.c +7 -0
index c72575ae..34ab90b4 100644
@@ -2405,6 +2405,13 @@ multi_client_connect_late_setup(struct multi_context *m,
     {
         mi->context.c2.tls_multi->multi_state = CAS_FAILED;
     }
+    /* Continue processing options only if authentication hasn't failed.
+     * Otherwise it does not make sense and we may operate on a non-configured
+     * client instance */
+    else
+    {
+        finish_options(&mi->context);
+    }
 
     /* send push reply if ready */
     if (mi->context.c2.push_request_received)
Signature · 1 line
2.35.1
#84 lore raw

Re: [Openvpn-devel] [PATCH v2 11/25] dco: split option parsing routines

Hi,

Quoted 7 lines · Antonio Quartulli

DCO will try to install keys upon generating them, however, this happens when parsing pushed cipher options (due to NCP).

For this reason we need to postpone parsing pushed cipher options to *after* the tunnel interface has been opened, otherwise we would have no DCO netdev object to operate on.

Quoted 5 lines

At the same time we split the parsing code, so that we can ensure that the NEW_PEER call can happen after the received peer-id has been parsed (it is required by all DCO API calls).

Some aspects of this patch I do not like, and others are broken, unfortunately.

What I do not like is the naming of finish_options(), because I find it confusing in relation to the "first half" of that, do_deferred_options()
- so maybe name this do_deferred_options_crypto()? But this is somewhat minor.

What I'm not sure about is the order of things in multi.c - but that might be due to misunderstanding. But I wonder why you are calling "multi_client_generate_tls_keys()" *before* processing the incoming crypto options in "finish_options()", which call update_crypto_params() - is there an interdependency? Or does this not matter?

The "broken" bit is: it breaks udp p2p TLS connections - the server will segfault. I guess this is due to moving p2p_ncp around...

The crash manifests as follows:

gdb openvpn
gdb> run server.conf
...

2022-08-02 12:50:45 us=144084 net_iface_mtu_set: mtu 1500 for tun5
2022-08-02 12:50:45 us=144179 net_iface_up: set tun5 up
2022-08-02 12:50:45 us=144241 net_addr_v6_add: fd00:abcd:204:8::1/64 dev tun5
2022-08-02 12:50:45 us=144396 Data Channel MTU parms [ mss_fix:0 max_frag:0 tun_mtu:1500 headroom:136 payload:1736 tailroom:557 ET:0 ]
2022-08-02 12:50:45 us=144457 Local Options String (VER=V4): 'V4,dev-type tun,link-mtu 1545,tun-mtu 1500,proto UDPv4,tun-ipv6,ifconfig 10.204.8.2 10.204.8.1,comp-lzo,cipher BF-CBC,auth SHA1,keysize 128,secret'
2022-08-02 12:50:45 us=144480 Expected Remote Options String (VER=V4): 'V4,dev-type tun,link-mtu 1545,tun-mtu 1500,proto UDPv4,tun-ipv6,ifconfig 10.204.8.1 10.204.8.2,comp-lzo,cipher BF-CBC,auth SHA1,keysize 128,secret'
2022-08-02 12:50:45 us=144539 Socket Buffers: R=[212992->212992] S=[212992->212992]
2022-08-02 12:50:45 us=144568 setsockopt(IPV6_V6ONLY=0)
2022-08-02 12:50:45 us=144616 UDPv6 link local (bound): [AF_INET6][undef]:51204
2022-08-02 12:50:45 us=144644 UDPv6 link remote: [AF_UNSPEC]

<waiting, client connects>

2022-08-02 12:52:13 us=388007 Peer Connection Initiated with [AF_INET6]::ffff:194.97.140.21:19880

Program received signal SIGSEGV, Segmentation fault0x00005555555d38e1 in check_session_cipher ([email protected]...=0x448, [email protected]...=0x7fffffffd4d0) at ssl_ncp.c:499
499 && streq(options->ciphername, session->opt->config_ciphername);

(gdb) where
#0 0x00005555555d38e1 in check_session_cipher ([email protected]....=0x448,

    [email protected]...=0x7fffffffd4d0) at ssl_ncp.c:499
#1  0x00005555555cab8b in tls_session_update_crypto_params (multi=0x0, 
    session=0x448, options=0x7fffffffd4d0, frame=0x7fffffffe0a0, 
    frame_fragment=0x0, lsi=0x55555562bc80) at ssl.c:1716
#2  0x000055555557899f in finish_options (c=<optimized out>) at init.c:2358
#3  0x0000555555578a65 in do_up ([email protected]...=0x7fffffffd4d0, 
    [email protected]...=false, 
    [email protected]...=0) at init.c:2131
#4  0x0000555555571326 in check_connection_established (c=0x7fffffffd4d0)
    at forward.c:311
#5  process_coarse_timers (c=0x7fffffffd4d0) at forward.c:634
#6  check_coarse_timers (c=0x7fffffffd4d0) at forward.c:717
#7  pre_select ([email protected]...=0x7fffffffd4d0) at forward.c:1823
#8  0x0000555555597b36 in tunnel_point_to_point (c=0x7fffffffd4d0)
    at openvpn.c:79

#9 openvpn_main (argc=2, argv=0x7fffffffe638) at openvpn.c:311 #10 0x00007ffff7cc21ea in ?? () from /lib64/libc.so.6
#11 0x00007ffff7cc229c in __libc_start_main () from /lib64/libc.so.6 #12 0x000055555555c0b1 in _start ()

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#85 lore raw

PATCH v3 11/25 dco: split option parsing routines

In reply to Antonio Quartulli

DCO will try to install keys upon generating them, however, this happens when parsing pushed cipher options (due to NCP).

For this reason we need to postpone parsing pushed cipher options to *after* the tunnel interface has been opened, otherwise we would have no DCO netdev object to operate on.

At the same time we split the parsing code, so that we can ensure that the NEW_PEER call can happen after the received peer-id has been parsed (it is required by all DCO API calls).

---

Changes from v2:
* rename finish_options() to do_deferred_options_part2()
* add comments to explain why this new function is required
* remove invocation in multi.c: we already perform key generation as last step

Changes from v1:
* removed error message in case of failure of finish_options(). The latter already warns the user about the failure - no need to print another generic message.

---
 src/openvpn/init.c | 70 +++++++++++++++++++++++++++++-----------------
 1 file changed, 45 insertions(+), 25 deletions(-)
src/openvpn/init.c +45 -25
index 338d797b..7ded843e 100644
@@ -2078,6 +2078,37 @@ options_hash_changed_or_zero(const struct sha256_digest *a,
            || !memcmp(a, &zero, sizeof(struct sha256_digest));
 }
 
+/**
+ * This function is expected to be invoked after open_tun() was performed.
+ *
+ * This kind of behaviour is required by DCO, because the following operations
+ * can be done only after the DCO device was created and the new peer was
+ * properly added.
+ */
+static bool
+do_deferred_options_part2(struct context *c)
+{
+    struct frame *frame_fragment = NULL;
+#ifdef ENABLE_FRAGMENT
+    if (c->options.ce.fragment)
+    {
+        frame_fragment = &c->c2.frame_fragment;
+    }
+#endif
+
+    struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                          &c->options, &c->c2.frame,
+                                          frame_fragment,
+                                          get_link_socket_info(c)))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
+        return false;
+    }
+
+    return true;
+}
+
 bool
 do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
 {
@@ -2093,14 +2124,6 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
                 return false;
             }
         }
-        else if (c->mode == MODE_POINT_TO_POINT)
-        {
-            if (!do_deferred_p2p_ncp(c))
-            {
-                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
-                return false;
-            }
-        }
 
         /* if --up-delay specified, open tun, do ifconfig, and run up script now */
         if (c->options.up_delay || PULL_DEFINED(&c->options))
@@ -2127,6 +2150,20 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        if (!pulled_options && c->mode == MODE_POINT_TO_POINT)
+        {
+            if (!do_deferred_p2p_ncp(c))
+            {
+                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
+                return false;
+            }
+        }
+
+        if (!do_deferred_options_part2(c))
+        {
+            return false;
+        }
+
         if (c->c2.did_open_tun)
         {
             c->c1.pulled_options_digest_save = c->c2.pulled_options_digest;
@@ -2332,23 +2369,6 @@ do_deferred_options(struct context *c, const unsigned int found)
         {
             return false;
         }
-        struct frame *frame_fragment = NULL;
-#ifdef ENABLE_FRAGMENT
-        if (c->options.ce.fragment)
-        {
-            frame_fragment = &c->c2.frame_fragment;
-        }
-#endif
-
-        struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
-                                              &c->options, &c->c2.frame,
-                                              frame_fragment,
-                                              get_link_socket_info(c)))
-        {
-            msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
-            return false;
-        }
     }
 
     return true;
Signature · 1 line
2.35.1
#86 lore raw

PATCH v4 11/25 dco: split option parsing routines

DCO will try to install keys upon generating them, however, this happens when parsing pushed cipher options (due to NCP).

For this reason we need to postpone parsing pushed cipher options to *after* the tunnel interface has been opened, otherwise we would have no DCO netdev object to operate on.

At the same time we split the parsing code, so that we can ensure that the NEW_PEER call can happen after the received peer-id has been parsed (it is required by all DCO API calls).

---

Changes from v3:
* call do_deferred_options_part2() only if !pulled_options. This is the same condition that triggered the call to do_deferred_options() in the first place, therefore part2 must follow the same logic.
* add extra comment in do_up to explain "step2" after open_tun()

Changes from v2:
* rename finish_options() to do_deferred_options_part2()
* add comments to explain why this new function is required
* remove invocation in multi.c: we already perform key generation as last step

Changes from v1:
* removed error message in case of failure of finish_options(). The latter already warns the user about the failure - no need to print another generic message.

---
 src/openvpn/init.c | 81 ++++++++++++++++++++++++++++++++--------------
 1 file changed, 56 insertions(+), 25 deletions(-)
src/openvpn/init.c +56 -25
index 338d797b..de8faeb4 100644
@@ -2078,6 +2078,37 @@ options_hash_changed_or_zero(const struct sha256_digest *a,
            || !memcmp(a, &zero, sizeof(struct sha256_digest));
 }
 
+/**
+ * This function is expected to be invoked after open_tun() was performed.
+ *
+ * This kind of behaviour is required by DCO, because the following operations
+ * can be done only after the DCO device was created and the new peer was
+ * properly added.
+ */
+static bool
+do_deferred_options_part2(struct context *c)
+{
+    struct frame *frame_fragment = NULL;
+#ifdef ENABLE_FRAGMENT
+    if (c->options.ce.fragment)
+    {
+        frame_fragment = &c->c2.frame_fragment;
+    }
+#endif
+
+    struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
+    if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
+                                          &c->options, &c->c2.frame,
+                                          frame_fragment,
+                                          get_link_socket_info(c)))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
+        return false;
+    }
+
+    return true;
+}
+
 bool
 do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
 {
@@ -2093,14 +2124,6 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
                 return false;
             }
         }
-        else if (c->mode == MODE_POINT_TO_POINT)
-        {
-            if (!do_deferred_p2p_ncp(c))
-            {
-                msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
-                return false;
-            }
-        }
 
         /* if --up-delay specified, open tun, do ifconfig, and run up script now */
         if (c->options.up_delay || PULL_DEFINED(&c->options))
@@ -2127,6 +2150,31 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        /* do_deferred_options_part2() and do_deferred_p2p_ncp() *must* be
+         * invoked after open_tun().
+         * This is required by DCO because we must have created the interface
+         * and added the peer before we can fiddle with the keys or any other
+         * data channel per-peer setting.
+         */
+        if (pulled_options)
+        {
+            if (!do_deferred_options_part2(c))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if (c->mode == MODE_POINT_TO_POINT)
+            {
+                if (!do_deferred_p2p_ncp(c))
+                {
+                    msg(D_TLS_ERRORS, "ERROR: Failed to apply P2P negotiated protocol options");
+                    return false;
+                }
+            }
+        }
+
         if (c->c2.did_open_tun)
         {
             c->c1.pulled_options_digest_save = c->c2.pulled_options_digest;
@@ -2332,23 +2380,6 @@ do_deferred_options(struct context *c, const unsigned int found)
         {
             return false;
         }
-        struct frame *frame_fragment = NULL;
-#ifdef ENABLE_FRAGMENT
-        if (c->options.ce.fragment)
-        {
-            frame_fragment = &c->c2.frame_fragment;
-        }
-#endif
-
-        struct tls_session *session = &c->c2.tls_multi->session[TM_ACTIVE];
-        if (!tls_session_update_crypto_params(c->c2.tls_multi, session,
-                                              &c->options, &c->c2.frame,
-                                              frame_fragment,
-                                              get_link_socket_info(c)))
-        {
-            msg(D_TLS_ERRORS, "OPTIONS ERROR: failed to import crypto options");
-            return false;
-        }
     }
 
     return true;
Signature · 1 line
2.35.1
Acked-by on v4 11/25
#87 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: split option parsing routines

Subjected to the full set of server side tests that so nicely crashed 11/25 v2 :-) (passed) and to client Linux/FreeBSD t_client tests including P2P NCP (passes as well).

The code looks good, and more correct as well. Thanks for all the comments and for renaming the function - if it helps my brain, future contributors might benefit from it as well.

Just for reference: do_deferred_p2p_ncp() is all "part2" type stuff, so it gets moved as a whole, not split.

NOTE: there is a small change in behaviour, which I'm not sure if problematic or not. The "part2" behaviour was previously encapsulated in

if (c->options.pull)

 {
    ...

}

and is no longer checking this in "_part2()".

This makes (as far as I can see) a difference if and only if a client *without* --pull (-> c->options.pull) will receive a PUSH_REPLY from a server that is behaving different from what we do - we only send PUSH_REPLY in reply to a PUSH_REQUEST, or to IV_PROTO=IV_PROTO_REQUEST_PUSH (and that is only happening if --pull / --client is set on the client).

In the server instance - which is called from multi.c - this would make a difference *if* we called _part2() from multi.c - but we do not(!), as that code path calls multi_client_generate_tls_keys(), which does the tls_session_update_crypto_params() for us... so, server would not call it before (!options.pull) and doesn't call it now (no call to _part2()) -> no change here.

Your patch has been applied to the master branch.

commit 6b9f4d71d859c27a9a71699aa899c9a2c5c3b680

Author: Antonio Quartulli
Date:   Wed Aug 3 11:50:12 2022 +0200
     dco: split option parsing routines

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24789.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#88 lore raw

PATCH 12/25 dco: check that pulled options are compatible

In reply to Antonio Quartulli

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

---
 src/openvpn/init.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
src/openvpn/init.c +23 -0
index b0a4b252..091cbd24 100644
@@ -2219,6 +2219,19 @@ do_deferred_p2p_ncp(struct context *c)
     return true;
 }
 
+
+static bool
+check_dco_pull_options(struct options *o)
+{
+    if (!o->use_peer_id)
+    {
+        msg(D_TLS_ERRORS, "OPTIONS IMPORT: Server did not request DATA_V2 packet "
+            "format required for data channel offload");
+        return false;
+    }
+    return true;
+}
+
 /*
  * Handle non-tun-related pulled options.
  */
@@ -2341,6 +2354,16 @@ finish_options(struct context *c)
         return false;
     }
 
+    /* Check if the pushed options are compatible with DCO if we have
+     * DCO enabled */
+    if (dco_enabled(&c->options) && !check_dco_pull_options(&c->options))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: pushed options are incompatible with "
+            "data channel offload. Use --disable-dco to connect"
+            "to this server");
+        return false;
+    }
+
     return true;
 }
 
Signature · 1 line
2.35.1
Acked-by on v1 12/25
#89 lore raw

Re: [Openvpn-devel] [PATCH 12/25] dco: check that pulled options are compatible

Am 24.06.22 um 10:37 schrieb Antonio Quartulli:

Quoted 2 lines

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

#90 lore raw

Re: [Openvpn-devel] [PATCH 12/25] dco: check that pulled options are compatible

In reply to Antonio Quartulli

Hi,

src/openvpn/init.c do_deferred_p2p_ncp(struct context *c) · line 2225

Quoted 12 lines · Antonio Quartulli
src/openvpn/init.c +4 -0
index b0a4b252..091cbd24 100644
@@ -2219,6 +2219,19 @@ do_deferred_p2p_ncp(struct context *c)
     return true;
 }
 
+
+static bool
+check_dco_pull_options(struct options *o)
+{

Nitpicking a bit - the "other" DCO option check functions are in dco.c (not yet, but in the patches leading up to this), so why is this one in init.c?

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#91 lore raw

Re: [Openvpn-devel] [PATCH 12/25] dco: check that pulled options are compatible

Hi,

Quoted 11 lines · Gert Doering

Hi,

Quoted 8 lines · Antonio Quartulli
src/openvpn/init.c +0 -0
index b0a4b252..091cbd24 100644
@@ -2219,6 +2219,19 @@ do_deferred_p2p_ncp(struct context *c)
      return true;
  }
 
Quoted 8 lines

src/openvpn/init.c do_deferred_p2p_ncp(struct context *c) · line 2225

Quoted 4 lines
src/openvpn/init.c +4 -0
+
+static bool
+check_dco_pull_options(struct options *o)
+{

Nitpicking a bit - the "other" DCO option check functions are in dco.c (not yet, but in the patches leading up to this), so why is this one in init.c?

good point. will move it to dco.c and send v2

Cheers,

Quoted 2 lines

gert

Signature · 1 line
Antonio Quartulli
#92 lore raw

PATCH v2 12/25 dco: check that pulled options are compatible

In reply to Antonio Quartulli

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

---

Changes from v1:
* move check_dco_pull_options() to dco.c (renamed to dco_check_pull_options())
* make options argument const
* add msglevel as first argument

 src/openvpn/dco.c  | 12 ++++++++++++
 src/openvpn/dco.h  | 17 +++++++++++++++++
 src/openvpn/init.c | 11 +++++++++++
 3 files changed, 40 insertions(+)
src/openvpn/dco.c +12 -0
index b39759e1..fa4a8487 100644
@@ -339,4 +339,16 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return true;
 }
 
+bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    if (!o->use_peer_id)
+    {
+        msg(msglevel, "OPTIONS IMPORT: Server did not request DATA_V2 packet "
+            "format required for data channel offload");
+        return false;
+    }
+    return true;
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +17 -0
index cb7f7e4f..a4bf6e37 100644
@@ -65,6 +65,17 @@ bool dco_available(int msglevel);
  */
 bool dco_check_option_conflict(int msglevel, const struct options *o);
 
+/**
+ * Check whether any of the options pushed by the server is not supported by
+ * our current dco implementation. If so print a warning at warning level
+ * for the first conflicting option found and return false.
+ *
+ * @param msglevel  the msg level to use to print the warnings
+ * @param o         the options struct that hold the options
+ * @return          true if no conflict was detected, false otherwise
+ */
+bool dco_check_pull_options(int msglevel, const struct options *o);
+
 /**
  * Initialize the DCO context
  *
@@ -154,6 +165,12 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return false;
 }
 
+static inline bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    return false;
+}
+
 static inline bool
 ovpn_dco_init(int mode, dco_context_t *dco)
 {
src/openvpn/init.c +11 -0
index 8e1221dd..ff2787d7 100644
@@ -2366,6 +2366,17 @@ finish_options(struct context *c)
         return false;
     }
 
+    /* Check if the pushed options are compatible with DCO if we have
+     * DCO enabled */
+    if (dco_enabled(&c->options) && !dco_check_pull_options(D_TLS_ERRORS,
+                                                            &c->options))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: pushed options are incompatible with "
+            "data channel offload. Use --disable-dco to connect"
+            "to this server");
+        return false;
+    }
+
     return true;
 }
 
Signature · 1 line
2.35.1
#93 lore raw

PATCH v3 12/25 dco: check that pulled options are compatible

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

---

Changes from v2:
* split if condition on two lines

Changes from v1:
* move check_dco_pull_options() to dco.c (renamed to dco_check_pull_options())
* make options argument const
* add msglevel as first argument

 src/openvpn/dco.c  | 12 ++++++++++++
 src/openvpn/dco.h  | 17 +++++++++++++++++
 src/openvpn/init.c | 11 +++++++++++
 3 files changed, 40 insertions(+)
src/openvpn/dco.c +12 -0
index b39759e1..fa4a8487 100644
@@ -339,4 +339,16 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return true;
 }
 
+bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    if (!o->use_peer_id)
+    {
+        msg(msglevel, "OPTIONS IMPORT: Server did not request DATA_V2 packet "
+            "format required for data channel offload");
+        return false;
+    }
+    return true;
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +17 -0
index cb7f7e4f..a4bf6e37 100644
@@ -65,6 +65,17 @@ bool dco_available(int msglevel);
  */
 bool dco_check_option_conflict(int msglevel, const struct options *o);
 
+/**
+ * Check whether any of the options pushed by the server is not supported by
+ * our current dco implementation. If so print a warning at warning level
+ * for the first conflicting option found and return false.
+ *
+ * @param msglevel  the msg level to use to print the warnings
+ * @param o         the options struct that hold the options
+ * @return          true if no conflict was detected, false otherwise
+ */
+bool dco_check_pull_options(int msglevel, const struct options *o);
+
 /**
  * Initialize the DCO context
  *
@@ -154,6 +165,12 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return false;
 }
 
+static inline bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    return false;
+}
+
 static inline bool
 ovpn_dco_init(int mode, dco_context_t *dco)
 {
src/openvpn/init.c +11 -0
index 8e1221dd..81fd49fa 100644
@@ -2366,6 +2366,17 @@ finish_options(struct context *c)
         return false;
     }
 
+    /* Check if the pushed options are compatible with DCO if we have
+     * DCO enabled */
+    if (dco_enabled(&c->options)
+        && !dco_check_pull_options(D_TLS_ERRORS, &c->options))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: pushed options are incompatible with "
+            "data channel offload. Use --disable-dco to connect"
+            "to this server");
+        return false;
+    }
+
     return true;
 }
 
Signature · 1 line
2.35.1
#94 lore raw

Re: [Openvpn-devel] [PATCH v3 12/25] dco: check that pulled options are compatible

Hi,

Quoted 4 lines · Antonio Quartulli

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

This patch is fine as it is, but during discussion we found a missing case - a server pushing "compress <something>".

There's basically (at least) 3 ways to handle this - so bringing this up for discussion, and a future patch

- add "compress" to check_dco_pull_options()
- extend dco_check_option_conflict() to set "allow-compression no" (this will make openvpn refuse all incoming "compress" configs on its own)
- or extend dco_check_option_conflict() to make "allow-compression asym/yes" disable DCO

I tend to option 3 - default for allow-compression is "no" anyway with 2.6, so if you have this in your config, compression could show up in a ccd/ file or pushed -> no DCO for you.

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#95 lore raw

Re: [Openvpn-devel] [PATCH v3 12/25] dco: check that pulled options are compatible

In reply to Antonio Quartulli

Hi,

Quoted 4 lines · Antonio Quartulli

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

I tried to test this (by forcing the error with --pull-filter ignore peer-id) and do not like the result - it does the pull, it does the OPTIONS IMPORT

2022-08-03 20:06:50 Pushed option removed by filter: 'peer-id 1'
2022-08-03 20:06:50 OPTIONS IMPORT: timers and/or timeouts modified
2022-08-03 20:06:50 OPTIONS IMPORT: compression parms modified
2022-08-03 20:06:50 OPTIONS IMPORT: --ifconfig/up options modified
2022-08-03 20:06:50 OPTIONS IMPORT: route options modified
2022-08-03 20:06:50 OPTIONS IMPORT: data channel crypto options modified

... then it does all the routing and ifconfig and stuff...

2022-08-03 20:06:50 net_route_v4_best_gw query: dst 0.0.0.0
2022-08-03 20:06:50 net_route_v4_best_gw result: via 195.30.8.94 dev ens160
2022-08-03 20:06:50 ROUTE_GATEWAY 195.30.8.94/255.255.255.224 IFACE=ens160 HWADDR=00:50:56:9d:04:1e
2022-08-03 20:06:50 GDG6: remote_host_ipv6=n/a
2022-08-03 20:06:50 net_route_v6_best_gw query: dst ::
2022-08-03 20:06:50 net_route_v6_best_gw result: via 2001:608:1:995a::ffff dev ens160
2022-08-03 20:06:50 ROUTE6_GATEWAY 2001:608:1:995a::ffff IFACE=ens160
2022-08-03 20:06:50 net_iface_new: add tun0 type ovpn-dco
2022-08-03 20:06:50 sitnl_send: rtnl: generic error (-17): File exists
2022-08-03 20:06:50 net_iface_new: add tun1 type ovpn-dco
...
2022-08-03 20:06:50 net_iface_up: set tun7 up
2022-08-03 20:06:50 net_addr_ptp_v4_add: 10.194.2.186 peer 10.194.2.185 dev tun7
2022-08-03 20:06:50 net_iface_mtu_set: mtu 1500 for tun7
2022-08-03 20:06:50 net_iface_up: set tun7 up
2022-08-03 20:06:50 net_addr_v6_add: fd00:abcd:194:2::102d/64 dev tun7
2022-08-03 20:06:50 net_route_v4_add: 10.194.0.0/16 via 10.194.2.185 dev [NULL] table 0 metric 200
2022-08-03 20:06:50 net_route_v4_add: 10.194.2.1/32 via 10.194.2.185 dev [NULL] table 0 metric 200

...

and *then* it hits the new check

2022-08-03 20:06:52 OPTIONS IMPORT: Server did not request DATA_V2 packet format required for data channel offload
2022-08-03 20:06:52 OPTIONS ERROR: pushed options are incompatible with data channel offload. Use --disable-dco to connectto this server
2022-08-03 20:06:52 Failed to open tun/tap interface

... leading to a bogus error message (the tun/tap interface opened just fine).

So I'd say the check is working fine, but it is happening at the wrong place - if I move it to the other OPTIONS IMPORT stuff at the end of "part1" (do_deferred_options()) I get a much nicer report...

2022-08-03 20:13:29 Pushed option removed by filter: 'peer-id 1'
2022-08-03 20:13:29 OPTIONS IMPORT: timers and/or timeouts modified
2022-08-03 20:13:29 OPTIONS IMPORT: compression parms modified
2022-08-03 20:13:29 OPTIONS IMPORT: --ifconfig/up options modified
2022-08-03 20:13:29 OPTIONS IMPORT: route options modified
2022-08-03 20:13:29 OPTIONS IMPORT: data channel crypto options modified
2022-08-03 20:13:29 OPTIONS IMPORT: Server did not request DATA_V2 packet format required for data channel offload
2022-08-03 20:13:29 OPTIONS ERROR: pushed options are incompatible with data channel offload. Use --disable-dco to connectto this server
2022-08-03 20:13:46 ERROR: Failed to apply push options

(it still says "failed to open tun/tap interface", but at least it says "ERROR: failed to apply push options" before)

So - can I have a v4 please?

PS: that missing blank "to connectto this server" is part of the v3 patch and could be fixed in v4 as well :-)

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#96 lore raw

PATCH v4 12/35 dco: check that pulled options are compatible

In reply to Antonio Quartulli

A server may push options that are not compatible with DCO. In this case we should log a message and bail out.

---

Changes from v3:
* move pull-option-check to before opening the tun device, for earlier bail out
* fix typ0 in error message (missing blank)

Changes from v2:
* split if condition on two lines

Changes from v1:
* move check_dco_pull_options() to dco.c (renamed to dco_check_pull_options())
* make options argument const
* add msglevel as first argument

---
 src/openvpn/dco.c  | 12 ++++++++++++
 src/openvpn/dco.h  | 17 +++++++++++++++++
 src/openvpn/init.c |  9 +++++++++
 3 files changed, 38 insertions(+)
src/openvpn/dco.c +12 -0
index b92a0e9c..8c22b7ea 100644
@@ -370,4 +370,16 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return true;
 }
 
+bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    if (!o->use_peer_id)
+    {
+        msg(msglevel, "OPTIONS IMPORT: Server did not request DATA_V2 packet "
+            "format required for data channel offload");
+        return false;
+    }
+    return true;
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +17 -0
index b926e236..fbb35906 100644
@@ -65,6 +65,17 @@ bool dco_available(int msglevel);
  */
 bool dco_check_option_conflict(int msglevel, const struct options *o);
 
+/**
+ * Check whether any of the options pushed by the server is not supported by
+ * our current dco implementation. If so print a warning at warning level
+ * for the first conflicting option found and return false.
+ *
+ * @param msglevel  the msg level to use to print the warnings
+ * @param o         the options struct that hold the options
+ * @return          true if no conflict was detected, false otherwise
+ */
+bool dco_check_pull_options(int msglevel, const struct options *o);
+
 /**
  * Initialize the DCO context
  *
@@ -156,6 +167,12 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return false;
 }
 
+static inline bool
+dco_check_pull_options(int msglevel, const struct options *o)
+{
+    return false;
+}
+
 static inline bool
 ovpn_dco_init(int mode, dco_context_t *dco)
 {
src/openvpn/init.c +9 -0
index de8faeb4..4423e162 100644
@@ -2382,6 +2382,15 @@ do_deferred_options(struct context *c, const unsigned int found)
         }
     }
 
+    /* Check if pushed options are compatible with DCO, if enabled */
+    if (dco_enabled(&c->options)
+        && !dco_check_pull_options(D_TLS_ERRORS, &c->options))
+    {
+        msg(D_TLS_ERRORS, "OPTIONS ERROR: pushed options are incompatible with "
+            "data channel offload. Use --disable-dco to connect to this server");
+        return false;
+    }
+
     return true;
 }
 
Signature · 1 line
2.35.1
Acked-by on v4 12/35
#97 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: check that pulled options are compatible

Thanks for shuffling this around a bit :-)

Tis is the same code change as v3, with a bit of reformatting in the message and comment, and it's applied in "part1" now (do_deferred_options()).

Does nothing if DCO is not available (not compiled in / no kernel support), does the right thing on DCO active with "--pull-filter ignore peer-id":

2022-08-04 10:30:38 Pushed option removed by filter: 'peer-id 0'
2022-08-04 10:30:38 OPTIONS IMPORT: timers and/or timeouts modified
2022-08-04 10:30:38 OPTIONS IMPORT: compression parms modified
2022-08-04 10:30:38 OPTIONS IMPORT: --ifconfig/up options modified
2022-08-04 10:30:38 OPTIONS IMPORT: route options modified
2022-08-04 10:30:38 OPTIONS IMPORT: data channel crypto options modified
2022-08-04 10:30:38 OPTIONS IMPORT: Server did not request DATA_V2 packet format required for data channel offload
2022-08-04 10:30:38 OPTIONS ERROR: pushed options are incompatible with data channel offload. Use --disable-dco to connect to this server
2022-08-04 10:30:38 ERROR: Failed to apply push options

As discused on IRC, I have changed D_TLS_ERRORS to D_PUSH_ERRORS - which is also "level 1", but fits the intent better.

Your patch has been applied to the master branch.

commit 46f6a7e8b6daf02bebe4a46498665274f1673ac0

Author: Antonio Quartulli
Date:   Thu Aug 4 08:40:16 2022 +0200
     dco: check that pulled options are compatible

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24797.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#98 lore raw

PATCH 13/25 dco: implement dco support for p2p/client code path

In reply to Antonio Quartulli

With this change we introduce ovpn-dco support only along the p2p/client code path. Server codebase is still unchanged.

---
 src/openvpn/dco.c     | 90 +++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     | 48 +++++++++++++++++++++++
 src/openvpn/event.h   |  3 ++
 src/openvpn/forward.c | 63 ++++++++++++++++++++++++++++--
 src/openvpn/init.c    | 34 +++++++++++++++-
 src/openvpn/init.h    |  2 +-
 src/openvpn/socket.h  |  1 +
 7 files changed, 236 insertions(+), 5 deletions(-)
src/openvpn/dco.c +90 -0
index 473eb564..2919c46d 100644
@@ -300,4 +300,94 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     return true;
 }
 
+int
+dco_p2p_add_new_peer(struct context *c)
+{
+    if (!dco_enabled(&c->options))
+    {
+        return 0;
+    }
+
+
+    struct tls_multi *multi = c->c2.tls_multi;
+    struct link_socket *ls = c->c2.link_socket;
+
+    struct in6_addr remote_ip6 = { 0 };
+    struct in_addr remote_ip4 = { 0 };
+
+    struct in6_addr *remote_addr6 = NULL;
+    struct in_addr *remote_addr4 = NULL;
+
+    const char *gw = NULL;
+
+    ASSERT(ls->info.connection_established);
+
+    /* In client mode if a P2P style topology is used we assume the
+     * remote-gateway is the IP of the peer */
+    if (c->options.topology == TOP_NET30 || c->options.topology == TOP_P2P)
+    {
+        gw = c->options.ifconfig_remote_netmask;
+    }
+    if (c->options.route_default_gateway)
+    {
+        gw = c->options.route_default_gateway;
+    }
+
+    /* These inet_pton conversion are fatal since options.c already implements
+     * checks to have only valid addresses when setting the options */
+    if (c->options.ifconfig_ipv6_remote)
+    {
+        if (inet_pton(AF_INET6, c->options.ifconfig_ipv6_remote, &remote_ip6) != 1)
+        {
+            msg(M_FATAL,
+                "DCO peer init: problem converting IPv6 ifconfig remote address %s to binary",
+                c->options.ifconfig_ipv6_remote);
+        }
+        remote_addr6 = &remote_ip6;
+    }
+
+    if (gw)
+    {
+        if (inet_pton(AF_INET, gw, &remote_ip4) != 1)
+        {
+            msg(M_FATAL, "DCO peer init: problem converting IPv4 ifconfig gateway address %s to binary", gw);
+        }
+        remote_addr4 = &remote_ip4;
+    }
+    else if (c->options.ifconfig_local)
+    {
+        msg(M_INFO, "DCO peer init: Need a peer VPN addresss to setup IPv4 (set --route-gateway)");
+    }
+
+    struct sockaddr *remoteaddr = &ls->info.lsa->actual.dest.addr.sa;
+
+    int ret = dco_new_peer(&c->c1.tuntap->dco, multi->peer_id,
+                           c->c2.link_socket->sd, NULL, remoteaddr,
+                           remote_addr4, remote_addr6);
+    if (ret < 0)
+    {
+        return ret;
+    }
+
+    c->c2.tls_multi->dco_peer_added = true;
+    c->c2.link_socket->info.dco_installed = true;
+
+    return 0;
+}
+
+void
+dco_remove_peer(struct context *c)
+{
+    if (!dco_enabled(&c->options))
+    {
+        return;
+    }
+
+    if (c->c1.tuntap && c->c2.tls_multi && c->c2.tls_multi->dco_peer_added)
+    {
+        dco_del_peer(&c->c1.tuntap->dco, c->c2.tls_multi->peer_id);
+        c->c2.tls_multi->dco_peer_added = false;
+    }
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +48 -0

59 lines not on this page · raw message

src/openvpn/event.h +3 -0
index a472afbe..f2438f97 100644
@@ -72,6 +72,9 @@
 #define MANAGEMENT_WRITE    (1 << (MANAGEMENT_SHIFT + WRITE_SHIFT))
 #define FILE_SHIFT          8
 #define FILE_CLOSED         (1 << (FILE_SHIFT + READ_SHIFT))
+#define DCO_SHIFT           10
+#define DCO_READ            (1 << (DCO_SHIFT + READ_SHIFT))
+#define DCO_WRITE           (1 << (DCO_SHIFT + WRITE_SHIFT))
 
 /*
  * Initialization flags passed to event_set_init
src/openvpn/forward.c +60 -3

91 lines not on this page · raw message

src/openvpn/init.c +33 -1
index 091cbd24..bdd2ad96 100644
@@ -2094,6 +2094,19 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        if (c->mode == MODE_POINT_TO_POINT)
+        {
+            /* ovpn-dco requires adding the peer now, before any option can be set,
+             * but *after* having parsed the pushed peer-id in do_deferred_options()
+             */
+            int ret = dco_p2p_add_new_peer(c);
+            if (ret < 0)
+            {
+                msg(D_DCO, "Cannot add peer to DCO: %s", strerror(-ret));
+                return false;
+            }
+        }
+
         if (!pulled_options && c->mode == MODE_POINT_TO_POINT)
         {
             if (!do_deferred_p2p_ncp(c))
@@ -2109,7 +2122,6 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             return false;
         }
 
-
         if (c->c2.did_open_tun)
         {
             c->c1.pulled_options_digest_save = c->c2.pulled_options_digest;
@@ -2364,6 +2376,22 @@ finish_options(struct context *c)
         return false;
     }
 
+    if (dco_enabled(&c->options)
+        && (c->options.ping_send_timeout || c->c2.frame.mss_fix))
+    {
+        int ret = dco_set_peer(&c->c1.tuntap->dco,
+                               c->c2.tls_multi->peer_id,
+                               c->options.ping_send_timeout,
+                               c->options.ping_rec_timeout,
+                               c->c2.frame.mss_fix);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot set parameters for DCO peer (id=%u): %s",
+                c->c2.tls_multi->peer_id, strerror(-ret));
+            return false;
+        }
+    }
+
     return true;
 }
 
@@ -4337,6 +4365,10 @@ close_instance(struct context *c)
         /* free buffers */
         do_close_free_buf(c);
 
+        /* close peer for DCO if enabled, needs peer-id so must be done before
+         * closing TLS contexts */
+        dco_remove_peer(c);
+
         /* close TLS */
         do_close_tls(c);
 
src/openvpn/init.h +1 -1
index 98e71d3a..5cc2a990 100644
@@ -30,7 +30,7 @@
  * Baseline maximum number of events
  * to wait for.
  */
-#define BASE_N_EVENTS 4
+#define BASE_N_EVENTS 5
 
 void context_clear(struct context *c);
 
src/openvpn/socket.h +1 -0
index a75adb00..0d521d22 100644
@@ -120,6 +120,7 @@ struct link_socket_info
     sa_family_t af;                     /* Address family like AF_INET, AF_INET6 or AF_UNSPEC*/
     bool bind_ipv6_only;
     int mtu_changed;            /* Set to true when mtu value is changed */
+    bool dco_installed;
 };
 
 /*
Signature · 1 line
2.35.1
#99 lore raw

Re: [Openvpn-devel] [PATCH 13/25] dco: implement dco support for p2p/client code path

Quoted 11 lines · On Freitag, 24. Juni 2022 10:37:57 CEST Antonio Quartulli wrote:

+ /* These inet_pton conversion are fatal since options.c already
implements
+ * checks to have only valid addresses when setting the
options */

src/openvpn/dco.c +3 -0
+    if (c->options.ifconfig_ipv6_remote)
+    {
+        if (inet_pton(AF_INET6, c->options.ifconfig_ipv6_remote,

&remote_ip6) != 1) + {

(quoted diff) +2 -0
+            msg(M_FATAL,
+                "DCO peer init: problem converting IPv6 ifconfig remote

address %s to binary", + c->options.ifconfig_ipv6_remote);

src/openvpn/dco.c dco_check_option_conflict(int msglevel, const struct options *o) · line 347

Quoted 3 lines
src/openvpn/dco.c +3 -0
+        }
+        remote_addr6 = &remote_ip6;
+    }

I'm undecided if these fatal errors are justified with respect to defensive programming or overly paranoid, because they will never appear.

#100 lore raw

Re: [Openvpn-devel] [PATCH 13/25] dco: implement dco support for p2p/client code path

Hi,

Quoted 16 lines · Heiko Hund
Quoted 11 lines · On Freitag, 24. Juni 2022 10:37:57 CEST Antonio Quartulli wrote:

+ /* These inet_pton conversion are fatal since options.c already
implements
+ * checks to have only valid addresses when setting the
options */

src/openvpn/dco.c +3 -0
+    if (c->options.ifconfig_ipv6_remote)
+    {
+        if (inet_pton(AF_INET6, c->options.ifconfig_ipv6_remote,

&remote_ip6) != 1) + {

(quoted diff) +2 -0
+            msg(M_FATAL,
+                "DCO peer init: problem converting IPv6 ifconfig remote

address %s to binary", + c->options.ifconfig_ipv6_remote);

src/openvpn/dco.c dco_check_option_conflict(int msglevel, const struct options *o) · line 347

Quoted 3 lines
src/openvpn/dco.c +3 -0
+        }
+        remote_addr6 = &remote_ip6;
+    }
Quoted 2 lines

I'm undecided if these fatal errors are justified with respect to defensive programming or overly paranoid, because they will never appear.

I'd say they are simply ASSERTs in disguise :-)

When a function returns an error I think it is always good habit to check it..then why not printing something meaningful at this point?

Cheers,

Quoted 9 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#101 lore raw

PATCH v2 13/25 dco: implement dco support for p2p/client code path

In reply to Antonio Quartulli

With this change we introduce ovpn-dco support only along the p2p/client code path. Server codebase is still unchanged.

---

Changes from v1:
* rebased

---
 src/openvpn/dco.c     | 91 +++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h     | 48 +++++++++++++++++++++++
 src/openvpn/event.h   |  3 ++
 src/openvpn/forward.c | 63 ++++++++++++++++++++++++++++--
 src/openvpn/init.c    | 33 ++++++++++++++++
 src/openvpn/init.h    |  2 +-
 src/openvpn/socket.h  |  1 +
 7 files changed, 237 insertions(+), 4 deletions(-)
src/openvpn/dco.c +91 -0
index 8c22b7ea..b5cc8a70 100644
@@ -36,6 +36,7 @@
 #include "crypto.h"
 #include "dco.h"
 #include "errlevel.h"
+#include "multi.h"
 #include "networking.h"
 #include "openvpn.h"
 #include "options.h"
@@ -382,4 +383,94 @@ dco_check_pull_options(int msglevel, const struct options *o)
     return true;
 }
 
+int
+dco_p2p_add_new_peer(struct context *c)
+{
+    if (!dco_enabled(&c->options))
+    {
+        return 0;
+    }
+
+
+    struct tls_multi *multi = c->c2.tls_multi;
+    struct link_socket *ls = c->c2.link_socket;
+
+    struct in6_addr remote_ip6 = { 0 };
+    struct in_addr remote_ip4 = { 0 };
+
+    struct in6_addr *remote_addr6 = NULL;
+    struct in_addr *remote_addr4 = NULL;
+
+    const char *gw = NULL;
+
+    ASSERT(ls->info.connection_established);
+
+    /* In client mode if a P2P style topology is used we assume the
+     * remote-gateway is the IP of the peer */
+    if (c->options.topology == TOP_NET30 || c->options.topology == TOP_P2P)
+    {
+        gw = c->options.ifconfig_remote_netmask;
+    }
+    if (c->options.route_default_gateway)
+    {
+        gw = c->options.route_default_gateway;
+    }
+
+    /* These inet_pton conversion are fatal since options.c already implements
+     * checks to have only valid addresses when setting the options */
+    if (c->options.ifconfig_ipv6_remote)
+    {
+        if (inet_pton(AF_INET6, c->options.ifconfig_ipv6_remote, &remote_ip6) != 1)
+        {
+            msg(M_FATAL,
+                "DCO peer init: problem converting IPv6 ifconfig remote address %s to binary",
+                c->options.ifconfig_ipv6_remote);
+        }
+        remote_addr6 = &remote_ip6;
+    }
+
+    if (gw)
+    {
+        if (inet_pton(AF_INET, gw, &remote_ip4) != 1)
+        {
+            msg(M_FATAL, "DCO peer init: problem converting IPv4 ifconfig gateway address %s to binary", gw);
+        }
+        remote_addr4 = &remote_ip4;
+    }
+    else if (c->options.ifconfig_local)
+    {
+        msg(M_INFO, "DCO peer init: Need a peer VPN addresss to setup IPv4 (set --route-gateway)");
+    }
+
+    struct sockaddr *remoteaddr = &ls->info.lsa->actual.dest.addr.sa;
+
+    int ret = dco_new_peer(&c->c1.tuntap->dco, multi->peer_id,
+                           c->c2.link_socket->sd, NULL, remoteaddr,
+                           remote_addr4, remote_addr6);
+    if (ret < 0)
+    {
+        return ret;
+    }
+
+    c->c2.tls_multi->dco_peer_added = true;
+    c->c2.link_socket->info.dco_installed = true;
+
+    return 0;
+}
+
+void
+dco_remove_peer(struct context *c)
+{
+    if (!dco_enabled(&c->options))
+    {
+        return;
+    }
+
+    if (c->c1.tuntap && c->c2.tls_multi && c->c2.tls_multi->dco_peer_added)
+    {
+        dco_del_peer(&c->c1.tuntap->dco, c->c2.tls_multi->peer_id);
+        c->c2.tls_multi->dco_peer_added = false;
+    }
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +48 -0

59 lines not on this page · raw message

src/openvpn/event.h +3 -0
index a472afbe..f2438f97 100644
@@ -72,6 +72,9 @@
 #define MANAGEMENT_WRITE    (1 << (MANAGEMENT_SHIFT + WRITE_SHIFT))
 #define FILE_SHIFT          8
 #define FILE_CLOSED         (1 << (FILE_SHIFT + READ_SHIFT))
+#define DCO_SHIFT           10
+#define DCO_READ            (1 << (DCO_SHIFT + READ_SHIFT))
+#define DCO_WRITE           (1 << (DCO_SHIFT + WRITE_SHIFT))
 
 /*
  * Initialization flags passed to event_set_init
src/openvpn/forward.c +60 -3

91 lines not on this page · raw message

src/openvpn/init.c +33 -0
index 4423e162..340c75d9 100644
@@ -2106,6 +2106,22 @@ do_deferred_options_part2(struct context *c)
         return false;
     }
 
+    if (dco_enabled(&c->options)
+        && (c->options.ping_send_timeout || c->c2.frame.mss_fix))
+    {
+        int ret = dco_set_peer(&c->c1.tuntap->dco,
+                               c->c2.tls_multi->peer_id,
+                               c->options.ping_send_timeout,
+                               c->options.ping_rec_timeout,
+                               c->c2.frame.mss_fix);
+        if (ret < 0)
+        {
+            msg(D_DCO, "Cannot set parameters for DCO peer (id=%u): %s",
+                c->c2.tls_multi->peer_id, strerror(-ret));
+            return false;
+        }
+    }
+
     return true;
 }
 
@@ -2150,6 +2166,19 @@ do_up(struct context *c, bool pulled_options, unsigned int option_types_found)
             }
         }
 
+        if (c->mode == MODE_POINT_TO_POINT)
+        {
+            /* ovpn-dco requires adding the peer now, before any option can be set,
+             * but *after* having parsed the pushed peer-id in do_deferred_options()
+             */
+            int ret = dco_p2p_add_new_peer(c);
+            if (ret < 0)
+            {
+                msg(D_DCO, "Cannot add peer to DCO: %s (%d)", strerror(-ret), ret);
+                return false;
+            }
+        }
+
         /* do_deferred_options_part2() and do_deferred_p2p_ncp() *must* be
          * invoked after open_tun().
          * This is required by DCO because we must have created the interface
@@ -4363,6 +4392,10 @@ close_instance(struct context *c)
         /* free buffers */
         do_close_free_buf(c);
 
+        /* close peer for DCO if enabled, needs peer-id so must be done before
+         * closing TLS contexts */
+        dco_remove_peer(c);
+
         /* close TLS */
         do_close_tls(c);
 
src/openvpn/init.h +1 -1
index 5f412a33..f53b65ee 100644
@@ -30,7 +30,7 @@
  * Baseline maximum number of events
  * to wait for.
  */
-#define BASE_N_EVENTS 4
+#define BASE_N_EVENTS 5
 
 void context_clear(struct context *c);
 
src/openvpn/socket.h +1 -0
index a75adb00..0d521d22 100644
@@ -120,6 +120,7 @@ struct link_socket_info
     sa_family_t af;                     /* Address family like AF_INET, AF_INET6 or AF_UNSPEC*/
     bool bind_ipv6_only;
     int mtu_changed;            /* Set to true when mtu value is changed */
+    bool dco_installed;
 };
 
 /*
Signature · 1 line
2.35.1
Acked-by on v2 13/25
#102 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: implement dco support for p2p/client code path

After all the preliminary infrastructure building, *this* is the beginning of the real thing :-)

I have tested

- full set of server side tests, without --enable-dco (this system does not have kernel DCO, so it does not matter) --> all works

- full set of client side tests, with --enable-dco, but no kernel DCO ("existing setup")
--> all works

- full set of client side tests, with --enable-dco AND kernel DCO (wohoo!)
--> some test instances disable DCO (like, SOCKS or HTTP proxy, or
TAP mode), and the fallback works ("pings succeed")

1b:openvpn.log Note: --http-proxy disables data channel offload.
1c:openvpn.log Note: --http-proxy disables data channel offload.
1d:openvpn.log Note: --socks-proxy disables data channel offload.
1e:openvpn.log Note: --socks-proxy disables data channel offload.
1z:openvpn.log Note: Using compression disables data channel offload.
2a:openvpn.log Note: cipher 'BF-CBC' in --data-ciphers is not supported by ovpn-dco, disabling data channel offload. 2d:openvpn.log Note: --socks-proxy disables data channel offload.
2e:openvpn.log Note: --socks-proxy disables data channel offload.
2z:openvpn.log Note: Using compression disables data channel offload.
3z:openvpn.log Note: Using compression disables data channel offload.
4:openvpn.log Note: dev-type not tun, disabling data channel offload.
4a:openvpn.log Note: dev-type not tun, disabling data channel offload.
4b:openvpn.log Note: dev-type not tun, disabling data channel offload.
6:openvpn.log Note: --fragment disables data channel offloa .
8:openvpn.log Note: Using compression disables data channel offload.
9:openvpn.log Note: dev-type not tun, disabling data channel offload.
23:openvpn.log Note: --data-cipher-fallback with cipher 'BF-CBC' disables data channel offload. 23a:openvpn.log Note: Using compression disables data channel offload.
23s:openvpn.log Note: --data-cipher-fallback with cipher 'BF-CBC' disables data channel offload. 24:openvpn.log Note: Using compression disables data channel offload.
24a:openvpn.log Note: Using compression disables data channel offload.

--> other instances claim to are using DCO ("ip -d link show"
shows "ovpn-dco") *and* packets are moved, so I guess it's using DCO...

these tests include "normal --client clients", "p2p --secret", "p2p --tls-secret", and "p2p --tls-secret with P2P NCP", using varying ciphers (-> BF-CBC/none forcing non-DCO, etc.)

	Test sets succeeded: 1 1a 1b 1c 1d 1e 1z 2 2a 2d 2e 2z 3 3z 4 4a 4b 5 6 8 9 23 23a 23s 24 24a.
	Test sets failed: 2b 2c 2f.

The 3 failures (2b, 2c, 2f) are all "IPv6 UDP fragments" (ping -s 3000, encapsulated in IPv6 UDP), which needs closer investigation. This works on a "--disable-dco" build, but the whole topic of UDP fragmentation is "outside OpenVPN", so this is not something a patch to OpenVPN can affect. tcpdump on an intermediate host can see outgoing fragments in the DCO case, but no replies - different from the non-DCO case, so this is going to be an interesting root cause hunt...

- I have not done performance tests, because the current test environment is not really suited for it yet (server instances are all non-DCO)

Plus, stared at the code and discussed with Antonio on IRC :-) - (especially the process_outgoing_link() change confused me a bit - the obvious answer to this is "these are control channel packets, which are still created by userland, but the design requires to avoid accessing the socket directly, so send to DCO module, and that one forwards").

I removed one spurious blank line from dco_p2p_add_new_peer().

Your patch has been applied to the master branch.

commit b6f7b285767e66f5cbd3854cf0ff918e87b31202

Author: Antonio Quartulli
Date:   Thu Aug 4 09:14:01 2022 +0200
     dco: implement dco support for p2p/client code path

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24798.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#103 lore raw

Re: [Openvpn-devel] [PATCH applied] Re: dco: implement dco support for p2p/client code path

Quoted 1 line · Gert Doering

[]>

Quoted 1 line

Your patch has been applied to the master branch.

Wohooo! Great stuff!

Thanks to everybody who contributed to this first milestone! However, the party is not over yet :-)

Cheers,

Quoted 19 lines

commit b6f7b285767e66f5cbd3854cf0ff918e87b31202

Author: Antonio Quartulli
Date:   Thu Aug 4 09:14:01 2022 +0200
      dco: implement dco support for p2p/client code path

      Signed-off-by: Antonio Quartulli <[email protected]...>
      Acked-by: Gert Doering <[email protected]...>
      Message-Id: <[email protected]...>
      URL: https://www.mail-archive.com/[email protected]/msg24798.html
      Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

Signature · 1 line
Antonio Quartulli
#104 lore raw

Re: [Openvpn-devel] [PATCH applied] Re: dco: implement dco support for p2p/client code path

In reply to Gert Doering

Hi,

Quoted 1 line · Gert Doering

Test sets failed: 2b 2c 2f.

So, this is an interesting one.

To trigger this, you need to connect over UDP + IPv6 transport, and then you need to inject a packet into "openvpn with DCO" that is fragmented into two parts, which *both* create an UDP packet after encapsulation which is bigger than 1500 bytes.

$ ping -c1 -s 2900 10.194.2.1

does this.

Tcpdump on the "tun0" interface shows

15:49:16.433666 IP (tos 0x0, ttl 64, id 18639, offset 0, flags [+], proto ICMP (1), length 1500)

    ubuntu2004 > 10.194.2.1: ICMP echo request, id 170, seq 1, length 1480
15:49:16.433703 IP (tos 0x0, ttl 64, id 18639, offset 1480, flags [none], proto ICMP (1), length 1448)
    ubuntu2004 > 10.194.2.1: icmp

so this is one "full sized 1500 byte frame", containing the first fragment, and then a 1448 byte frame containing the rest.

Connecting with --disable-dco yields this on the outside:

17:09:14.763519 IP6 (flowlabel 0x0082a, hlim 62, next-header Fragment (44) payload length: 1456) ubuntu2004 > phillip: frag (0x277a4245:0|1448) 34646 > 51194: UDP, bad length 1524 > 1440 17:09:14.763548 IP6 (flowlabel 0x0082a, hlim 62, next-header Fragment (44) payload length: 92) ubuntu2004 > phillip: frag (0x277a4245:1448|84) 17:09:14.763576 IP6 (flowlabel 0x0082a, hlim 62, next-header Fragment (44) payload length: 1456) ubuntu2004 > phillip: frag (0x1dec474f:0|1448) 34646 > 51194: UDP, bad length 1472 > 1440 17:09:14.763593 IP6 (flowlabel 0x0082a, hlim 62, next-header Fragment (44) payload length: 40) ubuntu2004 > phillip: frag (0x1dec474f:1448|32)

this is "4 packets are sent" (and, subsequently, 4 packets come back).

Both "(close to) full inside packets" are too big for an external 1500 byte
packet, and get fragmented. Is this ugly? Yes. Is it according to IP
specs? Yes. Do I expect this to work? Yes! (t_client excercises this case with the 3000-byte-pings).

Now, *with* DCO, we get this:

17:10:43.771215 IP6 (hlim 62, next-header Fragment (44) payload length: 1456) ubuntu2004 > phillip: frag (0xe79d84cc:0|1448) 58755 > 51194: UDP, bad length 1524 > 1440 17:10:43.771246 IP6 (hlim 62, next-header Fragment (44) payload length: 92) ubuntu2004 > phillip: frag (0xe79d84cc:1448|84)

(only *two* packets, then only keepalives)

And at the same time, the client (!) complains

2022-08-04 17:10:38 net_iface_new: add tun0 type ovpn-dco
2022-08-04 17:10:38 DCO device tun0 opened
(so, we're using DCO)
2022-08-04 17:10:43 read UDPv6 [CMSG=50|EMSGSIZE Path-MTU=1500]: Message too long (fd=3,code=90)

Now... reducing the ping size to 2500...

[email protected]...:~$ ping -c1 -s 2500 10.194.2.1
2508 bytes from 10.194.2.1: icmp_seq=1 ttl=64 time=126 ms

... actually works.

What is sent out by DCO in this case is 3 packets, 2 fragments, and one regular UDP packet

17:12:21.512762 IP6 (hlim 62, next-header Fragment (44) payload length: 1456) ubuntu2004 > phillip: frag (0x12728a31:0|1448) 58755 > 51194: UDP, bad length 1524 > 1440 17:12:21.512806 IP6 (hlim 62, next-header Fragment (44) payload length: 92) ubuntu2004 > phillip: frag (0x12728a31:1448|84)
17:12:21.512864 IP6 (hlim 63, next-header UDP (17) payload length: 1080) ubuntu2004.58755 > phillip.51194: [udp sum ok] UDP, length 1072

... because the second half of the fragmented ICMP packet now fits into one UDP/1500 byte even after openvpn encapsulation.

So my current guess is that the linux kernel sets *some* flag on "non-primary ICMP fragments" (inside) that get carried over to the final encapsulated packet, preventing further fragmentation (outside), but eliciting an "PMTU fail" error instead - which the newly renovated extended socket error handler picks up. In userland :-)

To add a data point:

[email protected]...:~$ ping -c1 -s 3500 10.194.2.1

creates *3* packets in the DCO case

17:14:08.780733 IP6 (hlim 62, next-header Fragment (44) payload length: 1456) ubuntu2004 > phillip: frag (0x16acdc72:0|1448) 58755 > 51194: UDP, bad length 1524 > 1440 17:14:08.780757 IP6 (hlim 62, next-header Fragment (44) payload length: 92) ubuntu2004 > phillip: frag (0x16acdc72:1448|84)
17:14:08.780842 IP6 (hlim 63, next-header UDP (17) payload length: 600) ubuntu2004.58755 > phillip.51194: [udp sum ok] UDP, length 592

... and one error message.

In the non-DCO case, this creates *5* packets - 2 1500-inside packet that get outside fragmented to 2+2 and one smaller "leftover" packet.

So the 2nd fragment is, again, refusing to be fragmented again...

(ping -s 5000 yields 7 packet without DCO, and only 3 with DCO, so a pattern starts to form ;-) )

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#105 lore raw

PATCH 14/25 dco: implement dco support for p2mp/server code path

In reply to Antonio Quartulli

This change introduces ovpn-dco support along the p2mp/server code path. Some code seems to be duplicate of the p2p version, but details are different, so it couldn't be shared.

---
 src/openvpn/dco.c   | 203 ++++++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h   |  49 ++++++++++
 src/openvpn/mtcp.c  |  59 +++++++++---
 src/openvpn/mudp.c  |  13 +++
 src/openvpn/multi.c | 212 +++++++++++++++++++++++++++++++++++---------
 src/openvpn/multi.h |  14 ++-
 6 files changed, 494 insertions(+), 56 deletions(-)
src/openvpn/dco.c +203 -0
index 2919c46d..48e007ea 100644
@@ -36,6 +36,7 @@
 #include "crypto.h"
 #include "dco.h"
 #include "errlevel.h"
+#include "multi.h"
 #include "openvpn.h"
 #include "ssl_common.h"
 #include "ssl_ncp.h"
@@ -390,4 +391,206 @@ dco_remove_peer(struct context *c)
     }
 }
 
+static bool
+dco_multi_get_localaddr(struct multi_context *m, struct multi_instance *mi,
+                        struct sockaddr_storage *local)
+{
+#if ENABLE_IP_PKTINFO
+    struct context *c = &mi->context;
+
+    if (!(c->options.sockflags & SF_USE_IP_PKTINFO))
+    {
+        return false;
+    }
+
+    struct link_socket_actual *actual = &c->c2.link_socket_info->lsa->actual;
+
+    switch (actual->dest.addr.sa.sa_family)
+    {
+        case AF_INET:
+        {
+            struct sockaddr_in *sock_in4 = (struct sockaddr_in *)local;
+#if defined(HAVE_IN_PKTINFO) && defined(HAVE_IPI_SPEC_DST)
+            sock_in4->sin_addr = actual->pi.in4.ipi_addr;
+#elif defined(IP_RECVDSTADDR)
+            sock_in4->sin_addr = actual->pi.in4;
+#else
+            /* source IP not available on this platform */
+            return false;
+#endif
+            sock_in4->sin_family = AF_INET;
+            break;
+        }
+
+        case AF_INET6:
+        {
+            struct sockaddr_in6 *sock_in6 = (struct sockaddr_in6 *)local;
+            sock_in6->sin6_addr = actual->pi.in6.ipi6_addr;
+            sock_in6->sin6_family = AF_INET6;
+            break;
+        }
+
+        default:
+            ASSERT(false);
+    }
+
+    return true;
+#else  /* if ENABLE_IP_PKTINFO */
+    return false;
+#endif /* if ENABLE_IP_PKTINFO */
+}
+
+int
+dco_multi_add_new_peer(struct multi_context *m, struct multi_instance *mi)
+{
+    struct context *c = &mi->context;
+
+    int peer_id = mi->context.c2.tls_multi->peer_id;
+    struct sockaddr *remoteaddr, *localaddr = NULL;
+    struct sockaddr_storage local = { 0 };
+    int sd = c->c2.link_socket->sd;
+
+    if (c->mode == CM_CHILD_TCP)
+    {
+        /* the remote address will be inferred from the TCP socket endpoint */
+        remoteaddr = NULL;
+    }
+    else
+    {
+        ASSERT(c->c2.link_socket_info->connection_established);
+        remoteaddr = &c->c2.link_socket_info->lsa->actual.dest.addr.sa;
+    }
+
+    struct in_addr remote_ip4 = { 0 };
+    struct in6_addr *remote_addr6 = NULL;
+    struct in_addr *remote_addr4 = NULL;
+
+    /* In server mode we need to fetch the remote addresses from the push config */
+    if (c->c2.push_ifconfig_defined)
+    {
+        remote_ip4.s_addr =  htonl(c->c2.push_ifconfig_local);
+        remote_addr4 = &remote_ip4;
+    }
+    if (c->c2.push_ifconfig_ipv6_defined)
+    {
+        remote_addr6 = &c->c2.push_ifconfig_ipv6_local;
+    }
+
+    if (dco_multi_get_localaddr(m, mi, &local))
+    {
+        localaddr = (struct sockaddr *)&local;
+    }
+
+    int ret = dco_new_peer(&c->c1.tuntap->dco, peer_id, sd, localaddr,
+                           remoteaddr, remote_addr4, remote_addr6);
+    if (ret < 0)
+    {
+        return ret;
+    }
+
+    c->c2.tls_multi->dco_peer_added = true;
+
+    if (c->mode == CM_CHILD_TCP)
+    {
+        multi_tcp_dereference_instance(m->mtcp, mi);
+        if (close(sd))
+        {
+            msg(D_DCO|M_ERRNO, "error closing TCP socket after DCO handover");
+        }
+        c->c2.link_socket->info.dco_installed = true;
+        c->c2.link_socket->sd = SOCKET_UNDEFINED;
+    }
+
+    return 0;
+}
+
+void
+dco_install_iroute(struct multi_context *m, struct multi_instance *mi,
+                   struct mroute_addr *addr)
+{
+#if defined(TARGET_LINUX)
+    if (!dco_enabled(&m->top.options))
+    {
+        return;
+    }
+
+    int addrtype = (addr->type & MR_ADDR_MASK);
+
+    /* If we do not have local IP addr to install, skip the route */
+    if ((addrtype == MR_ADDR_IPV6 && !mi->context.c2.push_ifconfig_ipv6_defined)
+        || (addrtype == MR_ADDR_IPV4 && !mi->context.c2.push_ifconfig_defined))
+    {
+        return;
+    }
+
+    struct context *c = &mi->context;
+    const char *dev = c->c1.tuntap->actual_name;
+
+    if (addrtype == MR_ADDR_IPV6)
+    {
+        int netbits = 128;
+        if (addr->type & MR_WITH_NETBITS)
+        {
+            netbits = addr->netbits;
+        }
+
+        net_route_v6_add(&m->top.net_ctx, &addr->v6.addr, netbits,
+                         &mi->context.c2.push_ifconfig_ipv6_local, dev, 0,
+                         DCO_IROUTE_METRIC);
+    }
+    else if (addrtype == MR_ADDR_IPV4)
+    {
+        int netbits = 32;
+        if (addr->type & MR_WITH_NETBITS)
+        {
+            netbits = addr->netbits;
+        }
+
+        in_addr_t dest = htonl(addr->v4.addr);
+        net_route_v4_add(&m->top.net_ctx, &dest, netbits,
+                         &mi->context.c2.push_ifconfig_local, dev, 0,
+                         DCO_IROUTE_METRIC);
+    }
+#endif /* if defined(TARGET_LINUX) */
+}
+
+void
+dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi)
+{
+#if defined(TARGET_LINUX)
+    if (!dco_enabled(&m->top.options))
+    {
+        return;
+    }
+    ASSERT(TUNNEL_TYPE(mi->context.c1.tuntap) == DEV_TYPE_TUN);
+
+    struct context *c = &mi->context;
+    const char *dev = c->c1.tuntap->actual_name;
+
+    if (mi->context.c2.push_ifconfig_defined)
+    {
+        for (const struct iroute *ir = c->options.iroutes;
+             ir;
+             ir = ir->next)
+        {
+            net_route_v4_del(&m->top.net_ctx, &ir->network, ir->netbits,
+                             &mi->context.c2.push_ifconfig_local, dev,
+                             0, DCO_IROUTE_METRIC);
+        }
+    }
+
+    if (mi->context.c2.push_ifconfig_ipv6_defined)
+    {
+        for (const struct iroute_ipv6 *ir6 = c->options.iroutes_ipv6;
+             ir6;
+             ir6 = ir6->next)
+        {
+            net_route_v6_del(&m->top.net_ctx, &ir6->network, ir6->netbits,
+                             &mi->context.c2.push_ifconfig_ipv6_local, dev,
+                             0, DCO_IROUTE_METRIC);
+        }
+    }
+#endif /* if defined(TARGET_LINUX) */
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +49 -0

70 lines not on this page · raw message

src/openvpn/mtcp.c +45 -14

118 lines not on this page · raw message

src/openvpn/mudp.c +13 -0
index 0810fada..14aa7236 100644
@@ -380,6 +380,19 @@ multi_process_io_udp(struct multi_context *m)
         multi_process_file_closed(m, mpp_flags);
     }
 #endif
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    else if (status & DCO_READ)
+    {
+        if (!IS_SIG(&m->top))
+        {
+            bool ret = true;
+            while (ret)
+            {
+                ret = multi_process_incoming_dco(m);
+            }
+        }
+    }
+#endif
 }
 
 /*
src/openvpn/multi.c +171 -41

296 lines not on this page · raw message

src/openvpn/multi.h +13 -1
index f1e9ab91..370d795c 100644
@@ -98,7 +98,9 @@ struct client_connect_defer_state
  * server-mode.
  */
 struct multi_instance {
-    struct schedule_entry se;  /* this must be the first element of the structure */
+    struct schedule_entry se;  /* this must be the first element of the structure,
+                                * We cast between this and schedule_entry so the
+                                * beginning of the struct must be identical */
     struct gc_arena gc;
     bool halt;
     int refcount;
@@ -310,6 +312,16 @@ void multi_process_float(struct multi_context *m, struct multi_instance *mi);
  */
 bool multi_process_post(struct multi_context *m, struct multi_instance *mi, const unsigned int flags);
 
+/**
+ * Process an incoming DCO message (from kernel space).
+ *
+ * @param m            - The single \c multi_context structur.e
+ *
+ * @return
+ *  - True, if the message was received correctly.
+ *  - False, if there was an error while reading the message.
+ */
+bool multi_process_incoming_dco(struct multi_context *m);
 
 /**************************************************************************/
 /**
Signature · 1 line
2.35.1
#106 lore raw

Re: [Openvpn-devel] [PATCH 14/25] dco: implement dco support for p2mp/server code path

src/openvpn/multi.c done: · line 3173

Quoted 3 lines · On Freitag, 24. Juni 2022 10:37:58 CEST Antonio Quartulli wrote:
src/openvpn/multi.c +3 -0
+    uint8_t *ptr = BPTR(&dco->dco_packet_in);
+    uint8_t op = ptr[0] >> P_OPCODE_SHIFT;
+    if (op == P_DATA_V2 || op == P_DATA_V2)

This looks odd. Seems you wanted to check for a second opcode, or is it obsolete?

Quoted 1 line

+ const char *reason = "(unknown reason by ovpn-dco)";

Wouldn't "ovpn-dco: unknown reason" be better?

#107 lore raw

Re: [Openvpn-devel] [PATCH 14/25] dco: implement dco support for p2mp/server code path

Hi,

Quoted 7 lines · Heiko Hund

src/openvpn/multi.c done: · line 3173

Quoted 3 lines · On Freitag, 24. Juni 2022 10:37:58 CEST Antonio Quartulli wrote:
src/openvpn/multi.c +3 -0
+    uint8_t *ptr = BPTR(&dco->dco_packet_in);
+    uint8_t op = ptr[0] >> P_OPCODE_SHIFT;
+    if (op == P_DATA_V2 || op == P_DATA_V2)

This looks odd. Seems you wanted to check for a second opcode, or is it obsolete?

we wanted want to bail out on DATA packets - not sure why this was repeated twice. Removing one check.

Quoted 4 lines
Quoted 2 lines

+ const char *reason = "(unknown reason by ovpn-dco)";

Wouldn't "ovpn-dco: unknown reason" be better?

Yeah, I'll do as you suggested.

While looking at this function, I realized that these checks should be performed before calling multi_process_incoming_link().

Fixing this in v2...

Thanks!

Signature · 1 line
Antonio Quartulli
#108 lore raw

Re: [Openvpn-devel] [PATCH 14/25] dco: implement dco support for p2mp/server code path

Am 05.07.2022 um 16:53 schrieb Antonio Quartulli:

Quoted 4 lines

Hi,

Quoted 1 line · Heiko Hund

On Freitag, 24. Juni 2022 10:37:58 CEST Antonio Quartulli wrote:

Quoted 9 lines
Quoted 6 lines
Quoted 3 lines

+    uint8_t *ptr = BPTR(&dco->dco_packet_in);
+    uint8_t op = ptr[0] >> P_OPCODE_SHIFT;
+    if (op == P_DATA_V2 || op == P_DATA_V2)

This looks odd. Seems you wanted to check for a second opcode, or is it obsolete?

we wanted want to bail out on DATA packets - not sure why this was repeated twice. Removing one check.

The first one should be P_DATA_V1

Arne

#109 lore raw

PATCH v2 14/25 dco: implement dco support for p2mp/server code path

In reply to Antonio Quartulli

This change introduces ovpn-dco support along the p2mp/server code path. Some code seems to be duplicate of the p2p version, but details are different, so it couldn't be shared.

---

Changes from v1:
* fix if condition P_DATA_V2 -> P_DATA_V1
* fix unknown reason string

 src/openvpn/dco.c   |   1 +
 src/openvpn/dco.h   |  49 ++++++++++
 src/openvpn/mtcp.c  |  59 +++++++++---
 src/openvpn/mudp.c  |  13 +++
 src/openvpn/multi.c | 216 +++++++++++++++++++++++++++++++++++---------
 src/openvpn/multi.h |  14 ++-
 6 files changed, 296 insertions(+), 56 deletions(-)
src/openvpn/dco.c +1 -0
index b5717a77..633dd7fd 100644
@@ -36,6 +36,7 @@
 #include "crypto.h"
 #include "dco.h"
 #include "errlevel.h"
+#include "multi.h"
 #include "networking.h"
 #include "openvpn.h"
 #include "options.h"
src/openvpn/dco.h +49 -0

70 lines not on this page · raw message

src/openvpn/mtcp.c +45 -14

118 lines not on this page · raw message

src/openvpn/mudp.c +13 -0
index 0cbca1a9..ddb1efc9 100644
@@ -381,6 +381,19 @@ multi_process_io_udp(struct multi_context *m)
         multi_process_file_closed(m, mpp_flags);
     }
 #endif
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    else if (status & DCO_READ)
+    {
+        if (!IS_SIG(&m->top))
+        {
+            bool ret = true;
+            while (ret)
+            {
+                ret = multi_process_incoming_dco(m);
+            }
+        }
+    }
+#endif
 }
 
 /*
src/openvpn/multi.c +175 -41

300 lines not on this page · raw message

src/openvpn/multi.h +13 -1
index f1e9ab91..370d795c 100644
@@ -98,7 +98,9 @@ struct client_connect_defer_state
  * server-mode.
  */
 struct multi_instance {
-    struct schedule_entry se;  /* this must be the first element of the structure */
+    struct schedule_entry se;  /* this must be the first element of the structure,
+                                * We cast between this and schedule_entry so the
+                                * beginning of the struct must be identical */
     struct gc_arena gc;
     bool halt;
     int refcount;
@@ -310,6 +312,16 @@ void multi_process_float(struct multi_context *m, struct multi_instance *mi);
  */
 bool multi_process_post(struct multi_context *m, struct multi_instance *mi, const unsigned int flags);
 
+/**
+ * Process an incoming DCO message (from kernel space).
+ *
+ * @param m            - The single \c multi_context structur.e
+ *
+ * @return
+ *  - True, if the message was received correctly.
+ *  - False, if there was an error while reading the message.
+ */
+bool multi_process_incoming_dco(struct multi_context *m);
 
 /**************************************************************************/
 /**
Signature · 1 line
2.35.1
Acked-by on v2 14/25
#110 lore raw

Re: [Openvpn-devel] [PATCH v2 14/25] dco: implement dco support for p2mp/server code path

Quoted 10 lines · On Donnerstag, 28. Juli 2022 21:55:01 CEST Antonio Quartulli wrote:

This change introduces ovpn-dco support along the p2mp/server code path. Some code seems to be duplicate of the p2p version, but details are different, so it couldn't be shared.

---

Changes from v1:
* fix if condition P_DATA_V2 -> P_DATA_V1
* fix unknown reason string

#111 lore raw

PATCH v3 14/25 dco: implement dco support for p2mp/server code path

In reply to Antonio Quartulli

This change introduces ovpn-dco support along the p2mp/server code path. Some code seems to be duplicate of the p2p version, but details are different, so it couldn't be shared.

---

Changes from v2:
* rebased

Changes from v1:
* fix if condition P_DATA_V2 -> P_DATA_V1
* fix unknown reason string

---
 src/openvpn/dco.c   | 202 ++++++++++++++++++++++++++++++++++++++
 src/openvpn/dco.h   |  49 ++++++++++
 src/openvpn/mtcp.c  |  59 ++++++++---
 src/openvpn/mudp.c  |  13 +++
 src/openvpn/multi.c | 232 ++++++++++++++++++++++++++++++++++++--------
 src/openvpn/multi.h |  14 ++-
 6 files changed, 513 insertions(+), 56 deletions(-)
src/openvpn/dco.c +202 -0
index a8735e88..09855643 100644
@@ -472,4 +472,206 @@ dco_remove_peer(struct context *c)
     }
 }
 
+static bool
+dco_multi_get_localaddr(struct multi_context *m, struct multi_instance *mi,
+                        struct sockaddr_storage *local)
+{
+#if ENABLE_IP_PKTINFO
+    struct context *c = &mi->context;
+
+    if (!(c->options.sockflags & SF_USE_IP_PKTINFO))
+    {
+        return false;
+    }
+
+    struct link_socket_actual *actual = &c->c2.link_socket_info->lsa->actual;
+
+    switch (actual->dest.addr.sa.sa_family)
+    {
+        case AF_INET:
+        {
+            struct sockaddr_in *sock_in4 = (struct sockaddr_in *)local;
+#if defined(HAVE_IN_PKTINFO) && defined(HAVE_IPI_SPEC_DST)
+            sock_in4->sin_addr = actual->pi.in4.ipi_addr;
+#elif defined(IP_RECVDSTADDR)
+            sock_in4->sin_addr = actual->pi.in4;
+#else
+            /* source IP not available on this platform */
+            return false;
+#endif
+            sock_in4->sin_family = AF_INET;
+            break;
+        }
+
+        case AF_INET6:
+        {
+            struct sockaddr_in6 *sock_in6 = (struct sockaddr_in6 *)local;
+            sock_in6->sin6_addr = actual->pi.in6.ipi6_addr;
+            sock_in6->sin6_family = AF_INET6;
+            break;
+        }
+
+        default:
+            ASSERT(false);
+    }
+
+    return true;
+#else  /* if ENABLE_IP_PKTINFO */
+    return false;
+#endif /* if ENABLE_IP_PKTINFO */
+}
+
+int
+dco_multi_add_new_peer(struct multi_context *m, struct multi_instance *mi)
+{
+    struct context *c = &mi->context;
+
+    int peer_id = mi->context.c2.tls_multi->peer_id;
+    struct sockaddr *remoteaddr, *localaddr = NULL;
+    struct sockaddr_storage local = { 0 };
+    int sd = c->c2.link_socket->sd;
+
+    if (c->mode == CM_CHILD_TCP)
+    {
+        /* the remote address will be inferred from the TCP socket endpoint */
+        remoteaddr = NULL;
+    }
+    else
+    {
+        ASSERT(c->c2.link_socket_info->connection_established);
+        remoteaddr = &c->c2.link_socket_info->lsa->actual.dest.addr.sa;
+    }
+
+    struct in_addr remote_ip4 = { 0 };
+    struct in6_addr *remote_addr6 = NULL;
+    struct in_addr *remote_addr4 = NULL;
+
+    /* In server mode we need to fetch the remote addresses from the push config */
+    if (c->c2.push_ifconfig_defined)
+    {
+        remote_ip4.s_addr =  htonl(c->c2.push_ifconfig_local);
+        remote_addr4 = &remote_ip4;
+    }
+    if (c->c2.push_ifconfig_ipv6_defined)
+    {
+        remote_addr6 = &c->c2.push_ifconfig_ipv6_local;
+    }
+
+    if (dco_multi_get_localaddr(m, mi, &local))
+    {
+        localaddr = (struct sockaddr *)&local;
+    }
+
+    int ret = dco_new_peer(&c->c1.tuntap->dco, peer_id, sd, localaddr,
+                           remoteaddr, remote_addr4, remote_addr6);
+    if (ret < 0)
+    {
+        return ret;
+    }
+
+    c->c2.tls_multi->dco_peer_added = true;
+
+    if (c->mode == CM_CHILD_TCP)
+    {
+        multi_tcp_dereference_instance(m->mtcp, mi);
+        if (close(sd))
+        {
+            msg(D_DCO|M_ERRNO, "error closing TCP socket after DCO handover");
+        }
+        c->c2.link_socket->info.dco_installed = true;
+        c->c2.link_socket->sd = SOCKET_UNDEFINED;
+    }
+
+    return 0;
+}
+
+void
+dco_install_iroute(struct multi_context *m, struct multi_instance *mi,
+                   struct mroute_addr *addr)
+{
+#if defined(TARGET_LINUX)
+    if (!dco_enabled(&m->top.options))
+    {
+        return;
+    }
+
+    int addrtype = (addr->type & MR_ADDR_MASK);
+
+    /* If we do not have local IP addr to install, skip the route */
+    if ((addrtype == MR_ADDR_IPV6 && !mi->context.c2.push_ifconfig_ipv6_defined)
+        || (addrtype == MR_ADDR_IPV4 && !mi->context.c2.push_ifconfig_defined))
+    {
+        return;
+    }
+
+    struct context *c = &mi->context;
+    const char *dev = c->c1.tuntap->actual_name;
+
+    if (addrtype == MR_ADDR_IPV6)
+    {
+        int netbits = 128;
+        if (addr->type & MR_WITH_NETBITS)
+        {
+            netbits = addr->netbits;
+        }
+
+        net_route_v6_add(&m->top.net_ctx, &addr->v6.addr, netbits,
+                         &mi->context.c2.push_ifconfig_ipv6_local, dev, 0,
+                         DCO_IROUTE_METRIC);
+    }
+    else if (addrtype == MR_ADDR_IPV4)
+    {
+        int netbits = 32;
+        if (addr->type & MR_WITH_NETBITS)
+        {
+            netbits = addr->netbits;
+        }
+
+        in_addr_t dest = htonl(addr->v4.addr);
+        net_route_v4_add(&m->top.net_ctx, &dest, netbits,
+                         &mi->context.c2.push_ifconfig_local, dev, 0,
+                         DCO_IROUTE_METRIC);
+    }
+#endif /* if defined(TARGET_LINUX) */
+}
+
+void
+dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi)
+{
+#if defined(TARGET_LINUX)
+    if (!dco_enabled(&m->top.options))
+    {
+        return;
+    }
+    ASSERT(TUNNEL_TYPE(mi->context.c1.tuntap) == DEV_TYPE_TUN);
+
+    struct context *c = &mi->context;
+    const char *dev = c->c1.tuntap->actual_name;
+
+    if (mi->context.c2.push_ifconfig_defined)
+    {
+        for (const struct iroute *ir = c->options.iroutes;
+             ir;
+             ir = ir->next)
+        {
+            net_route_v4_del(&m->top.net_ctx, &ir->network, ir->netbits,
+                             &mi->context.c2.push_ifconfig_local, dev,
+                             0, DCO_IROUTE_METRIC);
+        }
+    }
+
+    if (mi->context.c2.push_ifconfig_ipv6_defined)
+    {
+        for (const struct iroute_ipv6 *ir6 = c->options.iroutes_ipv6;
+             ir6;
+             ir6 = ir6->next)
+        {
+            net_route_v6_del(&m->top.net_ctx, &ir6->network, ir6->netbits,
+                             &mi->context.c2.push_ifconfig_ipv6_local, dev,
+                             0, DCO_IROUTE_METRIC);
+        }
+    }
+#endif /* if defined(TARGET_LINUX) */
+}
+
 #endif /* defined(ENABLE_DCO) */
src/openvpn/dco.h +49 -0

70 lines not on this page · raw message

src/openvpn/mtcp.c +45 -14

118 lines not on this page · raw message

src/openvpn/mudp.c +13 -0
index 0cbca1a9..ddb1efc9 100644
@@ -381,6 +381,19 @@ multi_process_io_udp(struct multi_context *m)
         multi_process_file_closed(m, mpp_flags);
     }
 #endif
+#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
+    else if (status & DCO_READ)
+    {
+        if (!IS_SIG(&m->top))
+        {
+            bool ret = true;
+            while (ret)
+            {
+                ret = multi_process_incoming_dco(m);
+            }
+        }
+    }
+#endif
 }
 
 /*
src/openvpn/multi.c +191 -41

304 lines not on this page · raw message

src/openvpn/multi.h +13 -1
index f1e9ab91..370d795c 100644
@@ -98,7 +98,9 @@ struct client_connect_defer_state
  * server-mode.
  */
 struct multi_instance {
-    struct schedule_entry se;  /* this must be the first element of the structure */
+    struct schedule_entry se;  /* this must be the first element of the structure,
+                                * We cast between this and schedule_entry so the
+                                * beginning of the struct must be identical */
     struct gc_arena gc;
     bool halt;
     int refcount;
@@ -310,6 +312,16 @@ void multi_process_float(struct multi_context *m, struct multi_instance *mi);
  */
 bool multi_process_post(struct multi_context *m, struct multi_instance *mi, const unsigned int flags);
 
+/**
+ * Process an incoming DCO message (from kernel space).
+ *
+ * @param m            - The single \c multi_context structur.e
+ *
+ * @return
+ *  - True, if the message was received correctly.
+ *  - False, if there was an error while reading the message.
+ */
+bool multi_process_incoming_dco(struct multi_context *m);
 
 /**************************************************************************/
 /**
Signature · 1 line
2.35.1
Acked-by on v3 14/25
#112 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: implement dco support for p2mp/server code path

v2 has an ACK from Heiko, so recording that. OTOH v3 is substantially different (the dco.c hunk was missing from v1+v2 - hidden in 13/25 v1 - and the multi.c tls_keys stuff is quite different), so I gave this my own stare-at-code, and of course the full test set.

The findings are lengthy, the followup work is non-trivial, but we have come a long way.

- stare-at-code

- the new additions to dco.c all make sense - the iroute thing, we discussed a lot beforehand, and the variants (sockaddr/inaddr) in local/remote address are due to "kernel wants them ready-to-use in target format".

Not sure if we want to consider the "!ENABLE_IP_PKTINFO" case here, ever, though... DCO requires a recent Linux (or FreeBSD) system, and those have them. Followup patch, though.

*Naming* some of these bits "remoteaddr" and "remote_addr*" is a bit unlucky, though... I do understand that "remoteaddr" is the remote end of the (UDP) socket, and "remote_addr*" is the inside ifconfig IP for the tunnel peer - but I think this could be made more clear.

The code bits

+ int netbits = 128;
+ if (addr->type & MR_WITH_NETBITS)
+ {
+ netbits = addr->netbits;
+ }

are ugly, and take too much code space. I think we need to fix this, as in, ensure that addr->netbits is always set, and for "host route things", MR_WITH_NETBITS would just control whether ".../bits" is used on printing, not for lookup. Maybe the code already does that, but we don't know. Another followup patch...

I noticed that we don't do error handling on the iroute installation
- but that is traditional OpenVPN behaviour, for all sorts of route additions. We log, but we never fail...

- in mtcp.c, there is a new "mi->socket_set_called = false" call, which seems to be unrelated to DCO, just cleaning up references (but it looks reasonable).

- there is some nesting cleanup (-> show -w) around mtcp.c line 625

- there is more nesting cleanup in multi.c / multi_learn_in_addr_t() (but that code is changed anyway, so it makes sense)

The actual changes here are needed for "real route" iroute handling, and have been discussed and agreed-on months ago already.

- as agreed on IRC, I changed the comment

        /* We do not want to install IP -> IP dev ovpn-dco0 */

     to

        /* "primary" is the VPN ifconfig address of the peer and already 
         * known to DCO, so only install "extra" iroutes (primary = false)
         */

- I need to test connecting from a 2.3 client to a DCO-enabled server, to excercise the "Client does not support DATA_V2..." path (but it looks good).

- I also need to test if "on a DCO enabled server, a ccd/ script produces something which is not allowed for DCO" (like, cipher BF-CBC etc.) gets handled correctly. The code is there now (multi.c, around line 2710).

- multi_close_instance_on_signal() and multi_signal_instance() just move around (--color-moved=zebra) [#ifdef ENABLE_MANAGEMENT]

- FreeBSD client test (no DCO whatsoever yet)
(everything passes, unsurprisingly)

- Linux client test, --enable-dco, no kernel DCO
(everything passes, unsurprisingly, not hitting these new code paths)

- Linux server test, no --enable-dco
(everything passes, and this is good - so the new code paths in multi.c do not break existing functionality and "known corner cases")

- Linux client test, --enable-dco, kernel DCO
(everything except 2c, 2d, 2f passes - known kernel issue, unchanged from before)

- Linux server test, --enable-dco, kernel DCO

- many of my server test instances had DCO incompatible options in there (like, comp-lzo, or just being TAP instances)

Note: cipher 'BF-CBC' in --data-ciphers is not supported by ovpn-dco, disabling data channel offload.
Note: dev-type not tun, disabling data channel offload.
Note: NOT using '--topology subnet' disables data channel offload.
Note: --fragment disables data channel offload.
Note: Using compression disables data channel offload.
Note: dev-type not tun, disabling data channel offload.

NOTE: the wording should be made more similar (disabling/disables), I'd say.

- clients connecting will produce this very confusing sequence of

    messages...:

	Aug  5 16:27:33 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 OPTIONS IMPORT: reading client specific options from: ccd/freebsd-74-amd64
	Aug  5 16:27:33 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 OPTIONS IMPORT: Server did not request DATA_V2 packet format required for data channel offload
	Aug  5 16:27:33 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 OPTIONS ERROR: pushed options are incompatible with data channel offload. Use --disable-dco to connect to this server

    (we are the server, so what did "the Server" do wrong, and we are not
    trying to connect to "this server" either???)

    ... but proceeds!

	Aug  5 16:27:33 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 MULTI: Learn: 10.220.2.8 -> freebsd-74-amd64/194.97.140.3:65038
	...
	Aug  5 16:27:33 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 Data Channel: using negotiated cipher 'AES-256-GCM'
	...
	Aug  5 16:27:36 ubuntu2004 tun-udp-p2mp[1834587]: freebsd-74-amd64/194.97.140.3:65038 SENT CONTROL [freebsd-74-amd64]: 'PUSH_REPLY,route 10.220.0.0 255.255.0.0,route-ipv6 fd00:abcd:220::/48,tun-ipv6,route-gateway 10.220.2.1,topology subnet,ping 10,ping-restart 30,compress stub-v2,ifconfig-ipv6 fd00:abcd:220:2::1006/64 fd00:abcd:220:2::1,ifconfig 10.220.2.8 255.255.255.0,peer-id 0,cipher AES-256-GCM' (status=1)

    ... and pinging 10.220.2.8 works just fine.

    The mentioning of ccd/ here is a red herring, if I move away the
    ccd/ file, I get the same complaints about DATA_V2.

    Seems we shuffled around the option checking code a bit too often
    (this is the no-longer _part2() bits, dco_check_pull_options(),
     I'm afraid, so, yeah, blame me).

- the udp p2p instance wants to send something "from userland" right after starting (before a peer is connected, no --remote) - this used to trigger an ASSERT(), now it just logs this:

ubuntu2004 tun-udp-p2p[1834288]: Attempting to send data packet while data channel offload is in use. Dropping packet

- connecting and disconnecting in rapid sequence (t_client tests on the other end) triggers these messages

Aug 5 16:38:31 ubuntu2004 tun-tcp-p2mp[1834574]: Received packet for peer-id unknown to OpenVPN: 1 Aug 5 16:38:31 ubuntu2004 tun-udp-p2p-tls-sha256[1834670]: ovpn-dco: received message type 3 with mismatched ifindex 8966 Aug 5 16:38:31 ubuntu2004 tun-udp-p2mp[1834587]: ovpn-dco: received message type 3 with mismatched ifindex 8966

which is interesting, because I am only connecting to one of them at a time - so if there are multiple DCO interfaces active, messages seems to bleed over.

- for connections "UDP over IPv6", we bump into the same issue that we've seen on the p2p DCO test - that is, double-fragmented packets trigger "packet is drop and extended socket error"

Aug 5 16:42:53 ubuntu2004 tun-udp-p2mp[1834587]: read UDPv6 [EMSGSIZE Path-MTU=1500]: Message too long (fd=7,code=90)

- these issues aside, a good number of tests worked :-) - some that still fail are stuff like "cipher none" that need adjustment of the testbed.

Test sets succeeded: 1 1a 1b 1c 1d 1e 1x 2a 2b 2d 2e 2w 2z1 2z2 3 4 4a 4b 5 5a 5b 5c 5d 5v1 5v2 5v3 5w1 5w2 5w3 5w4 5x1 5x2 5x3 5x4 6 9 9a 10 10a 10u 10v 10w 10x 10z. Test sets failed: 2 2c 2g 2f 2x 2y.

I'll followup with "why is each of them failing, and is this a problem of (Linux) DCO or not?".

Your patch has been applied to the master branch.

commit a5b4bad46978a01162fb820ea25594d6333aa9db

Author: Antonio Quartulli
Date:   Fri Aug 5 08:45:55 2022 +0200
     dco: implement dco support for p2mp/server code path

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Heiko Hund <[email protected]...>
     Acked-by: Gert Doering <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24811.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#113 lore raw

PATCH 15/25 dco: add documentation for ovpn-dco-linux

In reply to Antonio Quartulli
---
 Changes.rst                           |   9 ++
 README.dco.md                         | 123 ++++++++++++++++++++++++++
 doc/man-sections/advanced-options.rst |  13 +++
 doc/man-sections/server-options.rst   |   6 ++
 4 files changed, 151 insertions(+)
 create mode 100644 README.dco.md
Changes.rst +9 -0
index 67a23c79..275f8d64 100644
@@ -79,6 +79,15 @@ Cookie based handshake for UDP server
     shake. The tls-crypt-v2 option allows controlling if older clients are
     accepted.
 
+Data channel offloading with ovpn-dco
+    2.6.0+ implements support for data-channel offloading where the data packets
+    are directly processed and forwarded in kernel space thanks to the ovpn-dco
+    kernel module. The userspace openvpn program acts purely as a control plane
+    application. Note that DCO will use DATA_V2 packets in P2P mode, therefore,
+    this implies that peers must be running 2.6.0+ in order to have P2P-NCP
+    which brings DATA_V2 packet support.
+
+
 Deprecated features
 -------------------
 ``inetd`` has been removed
README.dco.md new file +123 -0

123 lines not on this page · raw message

doc/man-sections/advanced-options.rst +13 -0
index 5157c561..d5a6b4f2 100644
@@ -91,3 +91,16 @@ used when debugging or testing out special usage scenarios.
   *(Linux only)* Set the TX queue length on the TUN/TAP interface.
   Currently defaults to operating system default.
 
+--disable-dco
+  Disables the opportunistic use of data channel offloading if available.
+  Without this option, OpenVPN will opportunistically use DCO mode if
+  the config options and the running kernel supports using DCO.
+
+  Data channel offload currently requires data-ciphers to only contain
+  AEAD ciphers (AES-GCM and Chacha20-Poly1305) and Linux with the
+  ovpn-dco module.
+
+  Note that some options have no effect or cannot be used when DCO mode
+  is enabled.
+
+  On platforms that do not support DCO ``disable-dco`` has no effect.
doc/man-sections/server-options.rst +6 -0
index 08ee7bd3..31992732 100644
@@ -321,6 +321,12 @@ fast hardware. SSL/TLS authentication must be used in this mode.
   from the kernel to OpenVPN. Once in OpenVPN, the ``--iroute`` directive
   routes to the specific client.
 
+  However, when using DCO, the ``--iroute`` directive is usually enough
+  for DCO to fully configure the routing table. The extra ``--route``
+  directive is required only if the expected behaviour is to route the
+  traffic for a specific network to the VPN interface also when the
+  responsible client is not connected (traffic will then be dropped).
+
   This option must be specified either in a client instance config file
   using ``--client-config-dir`` or dynamically generated using a
   ``--client-connect`` script.
Signature · 1 line
2.35.1
#114 lore raw

Re: [Openvpn-devel] [PATCH 15/25] dco: add documentation for ovpn-dco-linux

On Fri, Jun 24, 2022 at 10:37:59AM +0200, Antonio Quartulli wrote:
[...]

README.dco.md line 121

Quoted 8 lines
README.dco.md new file +8 -0
+Current implementation limitations
+-------------------
+- --persistent-tun not tested/supported
+- fallback to non-dco in client mode missing
+- IPv6 mapped IPv4 addresses need Linux 5.4.189+/5.10.110+/5.12+ to work
+- Some incompatible options may not properly fallback to non-dco
+- TCP performance with ovpn-dco can still exhibit bad behaviour and drop to a
+  few megabits per seconds

Any way of qualifying this? Should one not use TCP mode at all currently? Or is this rare? Any workarounds?

Quoted 2 lines
+- Not all incompatible options are currently identified
+- No per client statistics. Only total statistics available on the interface

Regards,

Signature · 1 line
  Frank Lichtenheld
#115 lore raw

Re: [Openvpn-devel] [PATCH 15/25] dco: add documentation for ovpn-dco-linux

Hi,

Quoted 11 lines · Frank Lichtenheld

On Fri, Jun 24, 2022 at 10:37:59AM +0200, Antonio Quartulli wrote:
[...]

README.dco.md line 121

Quoted 8 lines
README.dco.md new file +8 -0
+Current implementation limitations
+-------------------
+- --persistent-tun not tested/supported
+- fallback to non-dco in client mode missing
+- IPv6 mapped IPv4 addresses need Linux 5.4.189+/5.10.110+/5.12+ to work
+- Some incompatible options may not properly fallback to non-dco
+- TCP performance with ovpn-dco can still exhibit bad behaviour and drop to a
+  few megabits per seconds
Quoted 2 lines

Any way of qualifying this? Should one not use TCP mode at all currently? Or is this rare? Any workarounds?

Good point. It was fairly rare, but we never dug in too much.
However, no extra tests were performed after ovpn-dco got quite refactored, so the problem may have gone.

More tests would be needed to confirm the issue.

Cheers,

Signature · 1 line
Antonio Quartulli
#116 lore raw

Re: [Openvpn-devel] [PATCH 15/25] dco: add documentation for ovpn-dco-linux

In reply to Antonio Quartulli
Quoted 5 lines · On Freitag, 24. Juni 2022 10:37:59 CEST Antonio Quartulli wrote:

+ application. Note that DCO will use DATA_V2 packets
in P2P mode, therefore,
+ this implies that peers must be running 2.6.0+
in order to have P2P-NCP
+ which brings DATA_V2 packet support.

[...]

Quoted 1 line

+- OpenVPN 2.4.0 is the minimum peer version.

Now I'm not sure what the minimum version is. Maybe clarify.

#117 lore raw

Re: [Openvpn-devel] [PATCH 15/25] dco: add documentation for ovpn-dco-linux

Hi,

Quoted 10 lines · Heiko Hund
Quoted 5 lines · On Freitag, 24. Juni 2022 10:37:59 CEST Antonio Quartulli wrote:

+ application. Note that DCO will use DATA_V2 packets
in P2P mode, therefore,
+ this implies that peers must be running 2.6.0+
in order to have P2P-NCP
+ which brings DATA_V2 packet support.

[...]

Quoted 1 line

+- OpenVPN 2.4.0 is the minimum peer version.

Now I'm not sure what the minimum version is. Maybe clarify.

I think you removed the context which should clarify this.

2.6.0 is required for P2P-NCP
2.4.0 is required for standard client/server mode

Cheers,

Quoted 11 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#118 lore raw

PATCH v2 15/25 dco: add documentation for ovpn-dco-linux

In reply to Antonio Quartulli

---

Changes from v1:
* removed text about TCP drop (cannot reproduce at the moment)
* clarified version requirements for p2p and client mode

---
 Changes.rst                           |   9 ++
 README.dco.md                         | 121 ++++++++++++++++++++++++++
 doc/man-sections/advanced-options.rst |  13 +++
 doc/man-sections/server-options.rst   |   6 ++
 4 files changed, 149 insertions(+)
 create mode 100644 README.dco.md
Changes.rst +9 -0
index 67a23c79..275f8d64 100644
@@ -79,6 +79,15 @@ Cookie based handshake for UDP server
     shake. The tls-crypt-v2 option allows controlling if older clients are
     accepted.
 
+Data channel offloading with ovpn-dco
+    2.6.0+ implements support for data-channel offloading where the data packets
+    are directly processed and forwarded in kernel space thanks to the ovpn-dco
+    kernel module. The userspace openvpn program acts purely as a control plane
+    application. Note that DCO will use DATA_V2 packets in P2P mode, therefore,
+    this implies that peers must be running 2.6.0+ in order to have P2P-NCP
+    which brings DATA_V2 packet support.
+
+
 Deprecated features
 -------------------
 ``inetd`` has been removed
README.dco.md new file +121 -0

121 lines not on this page · raw message

doc/man-sections/advanced-options.rst +13 -0
index 5157c561..d5a6b4f2 100644
@@ -91,3 +91,16 @@ used when debugging or testing out special usage scenarios.
   *(Linux only)* Set the TX queue length on the TUN/TAP interface.
   Currently defaults to operating system default.
 
+--disable-dco
+  Disables the opportunistic use of data channel offloading if available.
+  Without this option, OpenVPN will opportunistically use DCO mode if
+  the config options and the running kernel supports using DCO.
+
+  Data channel offload currently requires data-ciphers to only contain
+  AEAD ciphers (AES-GCM and Chacha20-Poly1305) and Linux with the
+  ovpn-dco module.
+
+  Note that some options have no effect or cannot be used when DCO mode
+  is enabled.
+
+  On platforms that do not support DCO ``disable-dco`` has no effect.
doc/man-sections/server-options.rst +6 -0
index 04f4b4fb..54ea8b66 100644
@@ -325,6 +325,12 @@ fast hardware. SSL/TLS authentication must be used in this mode.
   from the kernel to OpenVPN. Once in OpenVPN, the ``--iroute`` directive
   routes to the specific client.
 
+  However, when using DCO, the ``--iroute`` directive is usually enough
+  for DCO to fully configure the routing table. The extra ``--route``
+  directive is required only if the expected behaviour is to route the
+  traffic for a specific network to the VPN interface also when the
+  responsible client is not connected (traffic will then be dropped).
+
   This option must be specified either in a client instance config file
   using ``--client-config-dir`` or dynamically generated using a
   ``--client-connect`` script.
Signature · 1 line
2.35.1
Acked-by on v2 15/25
#119 lore raw

Re: [Openvpn-devel] [PATCH v2 15/25] dco: add documentation for ovpn-dco-linux

Quoted 6 lines · Antonio Quartulli

---

Changes from v1:
* removed text about TCP drop (cannot reproduce at the moment)
* clarified version requirements for p2p and client mode

Quoted 7 lines
---
 Changes.rst                           |   9 ++
 README.dco.md                         | 121 ++++++++++++++++++++++++++
 doc/man-sections/advanced-options.rst |  13 +++
 doc/man-sections/server-options.rst   |   6 ++
 4 files changed, 149 insertions(+)
 create mode 100644 README.dco.md

Regards,

Signature · 1 line
  Frank Lichtenheld
#120 lore raw

PATCH v3 15/25 dco: add documentation for ovpn-dco-linux

In reply to Antonio Quartulli

---

Changes from v2:
* more adjustments

Changes from v1:
* removed text about TCP drop (cannot reproduce at the moment)
* clarified version requirements for p2p and client mode

---
 Changes.rst                           |   9 ++
 README.dco.md                         | 122 ++++++++++++++++++++++++++
 doc/man-sections/advanced-options.rst |  13 +++
 doc/man-sections/server-options.rst   |   6 ++
 4 files changed, 150 insertions(+)
 create mode 100644 README.dco.md
Changes.rst +9 -0
index 67a23c79..275f8d64 100644
@@ -79,6 +79,15 @@ Cookie based handshake for UDP server
     shake. The tls-crypt-v2 option allows controlling if older clients are
     accepted.
 
+Data channel offloading with ovpn-dco
+    2.6.0+ implements support for data-channel offloading where the data packets
+    are directly processed and forwarded in kernel space thanks to the ovpn-dco
+    kernel module. The userspace openvpn program acts purely as a control plane
+    application. Note that DCO will use DATA_V2 packets in P2P mode, therefore,
+    this implies that peers must be running 2.6.0+ in order to have P2P-NCP
+    which brings DATA_V2 packet support.
+
+
 Deprecated features
 -------------------
 ``inetd`` has been removed
README.dco.md new file +122 -0

122 lines not on this page · raw message

doc/man-sections/advanced-options.rst +13 -0
index 5157c561..d5a6b4f2 100644
@@ -91,3 +91,16 @@ used when debugging or testing out special usage scenarios.
   *(Linux only)* Set the TX queue length on the TUN/TAP interface.
   Currently defaults to operating system default.
 
+--disable-dco
+  Disables the opportunistic use of data channel offloading if available.
+  Without this option, OpenVPN will opportunistically use DCO mode if
+  the config options and the running kernel supports using DCO.
+
+  Data channel offload currently requires data-ciphers to only contain
+  AEAD ciphers (AES-GCM and Chacha20-Poly1305) and Linux with the
+  ovpn-dco module.
+
+  Note that some options have no effect or cannot be used when DCO mode
+  is enabled.
+
+  On platforms that do not support DCO ``disable-dco`` has no effect.
doc/man-sections/server-options.rst +6 -0
index 04f4b4fb..54ea8b66 100644
@@ -325,6 +325,12 @@ fast hardware. SSL/TLS authentication must be used in this mode.
   from the kernel to OpenVPN. Once in OpenVPN, the ``--iroute`` directive
   routes to the specific client.
 
+  However, when using DCO, the ``--iroute`` directive is usually enough
+  for DCO to fully configure the routing table. The extra ``--route``
+  directive is required only if the expected behaviour is to route the
+  traffic for a specific network to the VPN interface also when the
+  responsible client is not connected (traffic will then be dropped).
+
   This option must be specified either in a client instance config file
   using ``--client-config-dir`` or dynamically generated using a
   ``--client-connect`` script.
Signature · 1 line
2.35.1
#122 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: add documentation for ovpn-dco-linux

In reply to Antonio Quartulli

Taking an easy one out of sequence... (but it means I have to tackle 14/ really quick now, otherwise this is all wrong).

Your patch has been applied to the master branch.

commit fba724e3fc958b5704dc22dd14ec93f4bb4d9e34

Author: Antonio Quartulli
Date:   Fri Aug 5 11:37:03 2022 +0200
     dco: add documentation for ovpn-dco-linux

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Frank Lichtenheld <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24817.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#123 lore raw

PATCH 16/25 GitHub Actions: add Linux DCO build (on Ubuntu 20.04)

In reply to Antonio Quartulli
---
 .github/workflows/build.yaml | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
.github/workflows/build.yaml +8 -1
index 6c267a61..b905c0d2 100644
@@ -209,15 +209,22 @@ jobs:
             libname: OpenSSL 1.1.1
             ssllib: openssl
             extraconf: "--disable-lzo --disable-lz4"
+          - os: ubuntu-20.04
+            sslpkg: "libssl-dev"
+            libname: OpenSSL 1.1.1
+            ssllib: openssl
+            extraconf: "--enable-dco"
+            nlpkg: "libnl-genl-3-dev"
 
     name: "gcc - ${{matrix.os}} - ${{matrix.libname}} ${{matrix.extraconf}}"
     env:
       SSLPKG: "${{matrix.sslpkg}}"
+      NLPKG: "${{matrix.nlpkg}}"
 
     runs-on: ${{matrix.os}}
     steps:
       - name: Install dependencies
-        run: sudo apt update && sudo apt install -y liblzo2-dev libpam0g-dev liblz4-dev linux-libc-dev man2html libcmocka-dev python3-docutils libtool automake autoconf ${SSLPKG}
+        run: sudo apt update && sudo apt install -y liblzo2-dev libpam0g-dev liblz4-dev linux-libc-dev man2html libcmocka-dev python3-docutils libtool automake autoconf ${SSLPKG} ${NLPKG}
       - name: Checkout OpenVPN
         uses: actions/[email protected]...
       - name: autoconf
Signature · 1 line
2.35.1
Acked-by on v1 16/25
#124 lore raw

Re: [Openvpn-devel] [PATCH 16/25] GitHub Actions: add Linux DCO build (on Ubuntu 20.04)

Quoted 5 lines · Antonio Quartulli
---
 .github/workflows/build.yaml | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

Looks good to me. Also note that this can applied on current master without waiting for the rest of series, since --enable-dco is already available.

Regards,

Signature · 1 line
  Frank Lichtenheld
#125 lore raw

[Openvpn-devel] [PATCH applied] Re: GitHub Actions: add Linux DCO build (on Ubuntu 20.04)

In reply to Antonio Quartulli

Indeed, this one is independent from the rest *and* the current tree compiles fine with --enable-dco *and* we want to ensure it stays that way :-) -> in it goes!

Your patch has been applied to the master branch.

commit d2b0c011300c66055764b8f43c272342d2b82584

Author: Antonio Quartulli
Date:   Fri Jun 24 10:38:00 2022 +0200
     GitHub Actions: add Linux DCO build (on Ubuntu 20.04)

     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Frank Lichtenheld <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24523.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#126 lore raw

PATCH 17/25 tun: extract close_tun_handle into its own fucntion and print correct type

In reply to Antonio Quartulli

From: Arne Schwabe <[email protected]...>

This moves closing the tun handle into its own function and also prints the adapter type we are operating on, instead hardcoding it to tap-windows.

While at it, set the handle to NULL after closing, to prevent a double close due to multiple invocations of this helper.

---
 src/openvpn/tun.c | 76 ++++++++++++++++++++++++++---------------------
 1 file changed, 42 insertions(+), 34 deletions(-)
src/openvpn/tun.c +42 -34
index 5eefb0c6..f324ac91 100644
@@ -6767,6 +6767,47 @@ netsh_delete_address_dns(const struct tuntap *tt, bool ipv6, struct gc_arena *gc
     argv_free(&argv);
 }
 
+static void
+close_tun_handle(struct tuntap *tt)
+{
+    const char *adaptertype = print_windows_driver(tt->windows_driver);
+
+    if (tt->hand)
+    {
+        dmsg(D_WIN32_IO_LOW, "Attempting CancelIO on %s adapter", adaptertype);
+        if (!CancelIo(tt->hand))
+        {
+            msg(M_WARN | M_ERRNO, "Warning: CancelIO failed on %s adapter", adaptertype);
+        }
+    }
+
+    dmsg(D_WIN32_IO_LOW, "Attempting close of overlapped read event on %s adapter", adaptertype);
+    overlapped_io_close(&tt->reads);
+
+    dmsg(D_WIN32_IO_LOW, "Attempting close of overlapped write event on %s adapter", adaptertype);
+    overlapped_io_close(&tt->writes);
+
+    if (tt->hand)
+    {
+        dmsg(D_WIN32_IO_LOW, "Attempting CloseHandle on %s adapter", adaptertype);
+        if (!CloseHandle(tt->hand))
+        {
+            msg(M_WARN | M_ERRNO, "Warning: CloseHandle failed on %s adapter", adaptertype);
+        }
+        tt->hand = NULL;
+    }
+
+    if (tt->windows_driver == WINDOWS_DRIVER_WINTUN)
+    {
+        CloseHandle(tt->rw_handle.read);
+        CloseHandle(tt->rw_handle.write);
+        UnmapViewOfFile(tt->wintun_send_ring);
+        UnmapViewOfFile(tt->wintun_receive_ring);
+        CloseHandle(tt->wintun_send_ring_handle);
+        CloseHandle(tt->wintun_receive_ring_handle);
+    }
+}
+
 void
 close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx)
 {
@@ -6836,43 +6877,10 @@ close_tun(struct tuntap *tt, openvpn_net_ctx_t *ctx)
 
     dhcp_release(tt);
 
-    if (tt->hand != NULL)
-    {
-        dmsg(D_WIN32_IO_LOW, "Attempting CancelIO on TAP-Windows adapter");
-        if (!CancelIo(tt->hand))
-        {
-            msg(M_WARN | M_ERRNO, "Warning: CancelIO failed on TAP-Windows adapter");
-        }
-    }
-
-    dmsg(D_WIN32_IO_LOW, "Attempting close of overlapped read event on TAP-Windows adapter");
-    overlapped_io_close(&tt->reads);
-
-    dmsg(D_WIN32_IO_LOW, "Attempting close of overlapped write event on TAP-Windows adapter");
-    overlapped_io_close(&tt->writes);
-
-    if (tt->hand != NULL)
-    {
-        dmsg(D_WIN32_IO_LOW, "Attempting CloseHandle on TAP-Windows adapter");
-        if (!CloseHandle(tt->hand))
-        {
-            msg(M_WARN | M_ERRNO, "Warning: CloseHandle failed on TAP-Windows adapter");
-        }
-    }
+    close_tun_handle(tt);
 
     free(tt->actual_name);
 
-    if (tt->windows_driver == WINDOWS_DRIVER_WINTUN)
-    {
-        CloseHandle(tt->rw_handle.read);
-        CloseHandle(tt->rw_handle.write);
-        UnmapViewOfFile(tt->wintun_send_ring);
-        UnmapViewOfFile(tt->wintun_receive_ring);
-        CloseHandle(tt->wintun_send_ring_handle);
-        CloseHandle(tt->wintun_receive_ring_handle);
-    }
-
-
     clear_tuntap(tt);
     free(tt);
     gc_free(&gc);
Signature · 1 line
2.35.1
Acked-by on v1 17/25
#127 lore raw

Re: [Openvpn-devel] [PATCH 17/25] tun: extract close_tun_handle into its own fucntion and print correct type

Quoted 6 lines · Antonio Quartulli

From: Arne Schwabe <[email protected]...>

This moves closing the tun handle into its own function and also prints the adapter type we are operating on, instead hardcoding it to tap-windows.

Quoted 8 lines

While at it, set the handle to NULL after closing, to prevent a double close due to multiple invocations of this helper.

---
 src/openvpn/tun.c | 76 ++++++++++++++++++++++++++---------------------
 1 file changed, 42 insertions(+), 34 deletions(-)

Trivial enough. Also can be applied independently of all changes before it in the series!

Regards,

Signature · 1 line
  Frank Lichtenheld
#128 lore raw

[Openvpn-devel] [PATCH applied] Re: tun: extract close_tun_handle into its own fucntion and print correct type

In reply to Antonio Quartulli

Code change looks good. Did not actually test anything, but it *does* pass compilation on MinGW / Unbuntu 18.04 :-)

The change to print the actual adapter type confuses "--color-move=zebra", but is a nice enhancement.

Your patch has been applied to the master branch.

commit abceec65daed9aa6359314a0243246acc1cc25bf

Author: Arne Schwabe
Date:   Fri Jun 24 10:38:01 2022 +0200
     tun: extract close_tun_handle into its own fucntion and print correct type

     Signed-off-by: Arne Schwabe <[email protected]...>
     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Frank Lichtenheld <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24527.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#129 lore raw

PATCH 18/25 dco: turn supported ciphers list into a function

In reply to Antonio Quartulli

Other platforms may need more complex logic to decide whether a cipher is supported or not, therefore turn hardcoded list into a function that can be implemented by each platform independently.

---
 src/openvpn/dco.c       |  4 ++--
 src/openvpn/dco.h       | 13 +++++++++++++
 src/openvpn/dco_linux.c |  6 ++++++
 src/openvpn/dco_linux.h |  1 -
 4 files changed, 21 insertions(+), 3 deletions(-)
src/openvpn/dco.c +2 -2
index 48e007ea..1612a64e 100644
@@ -236,7 +236,7 @@ dco_check_option_conflict(int msglevel, const struct options *o)
 
     /* At this point the ciphers have already been normalised */
     if (o->enable_ncp_fallback
-        && !tls_item_in_cipher_list(o->ciphername, DCO_SUPPORTED_CIPHERS))
+        && !tls_item_in_cipher_list(o->ciphername, dco_get_supported_ciphers()))
     {
         msg(msglevel, "Note: --data-cipher-fallback with cipher '%s' "
             "disables data channel offload.", o->ciphername);
@@ -288,7 +288,7 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     const char *token;
     while ((token = strsep(&tmp_ciphers, ":")))
     {
-        if (!tls_item_in_cipher_list(token, DCO_SUPPORTED_CIPHERS))
+        if (!tls_item_in_cipher_list(token, dco_get_supported_ciphers()))
         {
             msg(msglevel, "Note: cipher '%s' in --data-ciphers is not supported "
                 "by ovpn-dco, disabling data channel offload.", token);
src/openvpn/dco.h +13 -0
index 4b945b50..66ae444f 100644
@@ -200,6 +200,13 @@ void dco_install_iroute(struct multi_context *m, struct multi_instance *mi,
  */
 void dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi);
 
+/**
+ * Retrieve the list of ciphers supported by the current platform
+ *
+ * @return                   list of colon-separated ciphers
+ */
+const char *dco_get_supported_ciphers();
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -301,5 +308,11 @@ dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi)
 {
 }
 
+static inline const char *
+dco_get_supported_ciphers()
+{
+    return "";
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/dco_linux.c +6 -0
index 5e77139a..f86ea819 100644
@@ -931,4 +931,10 @@ dco_event_set(dco_context_t *dco, struct event_set *es, void *arg)
     }
 }
 
+const char *
+dco_get_supported_ciphers()
+{
+    return "AES-128-GCM:AES-256-GCM:AES-192-GCM:CHACHA20-POLY1305";
+}
+
 #endif /* defined(ENABLE_DCO) && defined(TARGET_LINUX) */
src/openvpn/dco_linux.h +0 -1
index e0e59fa6..416ea30a 100644
@@ -34,7 +34,6 @@
 typedef enum ovpn_key_slot dco_key_slot_t;
 typedef enum ovpn_cipher_alg dco_cipher_t;
 
-#define DCO_SUPPORTED_CIPHERS "AES-128-GCM:AES-256-GCM:AES-192-GCM:CHACHA20-POLY1305"
 
 typedef struct
 {
Signature · 1 line
2.35.1
Acked-by on v1 18/25
#130 lore raw

Re: [Openvpn-devel] [PATCH 18/25] dco: turn supported ciphers list into a function

Quoted 6 lines · On Freitag, 24. Juni 2022 10:38:02 CEST Antonio Quartulli wrote:

Other platforms may need more complex logic to decide whether a cipher is supported or not, therefore turn hardcoded list into a function that can be implemented by each platform independently.

Trivial and compile-checked.

#131 lore raw

PATCH v2 18/18 dco: turn supported ciphers list into a function

In reply to Antonio Quartulli

Other platforms may need more complex logic to decide whether a cipher is supported or not, therefore turn hardcoded list into a function that can be implemented by each platform independently.

---

Changes from v1:
* rebased

---
 src/openvpn/dco.c       |  4 ++--
 src/openvpn/dco.h       | 13 +++++++++++++
 src/openvpn/dco_linux.c |  6 ++++++
 src/openvpn/dco_linux.h |  1 -
 4 files changed, 21 insertions(+), 3 deletions(-)
src/openvpn/dco.c +2 -2
index 09855643..889ae270 100644
@@ -306,7 +306,7 @@ dco_check_option_conflict(int msglevel, const struct options *o)
 
     /* At this point the ciphers have already been normalised */
     if (o->enable_ncp_fallback
-        && !tls_item_in_cipher_list(o->ciphername, DCO_SUPPORTED_CIPHERS))
+        && !tls_item_in_cipher_list(o->ciphername, dco_get_supported_ciphers()))
     {
         msg(msglevel, "Note: --data-cipher-fallback with cipher '%s' "
             "disables data channel offload.", o->ciphername);
@@ -358,7 +358,7 @@ dco_check_option_conflict(int msglevel, const struct options *o)
     const char *token;
     while ((token = strsep(&tmp_ciphers, ":")))
     {
-        if (!tls_item_in_cipher_list(token, DCO_SUPPORTED_CIPHERS))
+        if (!tls_item_in_cipher_list(token, dco_get_supported_ciphers()))
         {
             msg(msglevel, "Note: cipher '%s' in --data-ciphers is not supported "
                 "by ovpn-dco, disabling data channel offload.", token);
src/openvpn/dco.h +13 -0
index 72569083..6b5c016a 100644
@@ -213,6 +213,13 @@ void dco_install_iroute(struct multi_context *m, struct multi_instance *mi,
  */
 void dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi);
 
+/**
+ * Retrieve the list of ciphers supported by the current platform
+ *
+ * @return                   list of colon-separated ciphers
+ */
+const char *dco_get_supported_ciphers();
+
 #else /* if defined(ENABLE_DCO) */
 
 typedef void *dco_context_t;
@@ -320,5 +327,11 @@ dco_delete_iroutes(struct multi_context *m, struct multi_instance *mi)
 {
 }
 
+static inline const char *
+dco_get_supported_ciphers()
+{
+    return "";
+}
+
 #endif /* defined(ENABLE_DCO) */
 #endif /* ifndef DCO_H */
src/openvpn/dco_linux.c +6 -0
index 5e77139a..f86ea819 100644
@@ -931,4 +931,10 @@ dco_event_set(dco_context_t *dco, struct event_set *es, void *arg)
     }
 }
 
+const char *
+dco_get_supported_ciphers()
+{
+    return "AES-128-GCM:AES-256-GCM:AES-192-GCM:CHACHA20-POLY1305";
+}
+
 #endif /* defined(ENABLE_DCO) && defined(TARGET_LINUX) */
src/openvpn/dco_linux.h +0 -1
index e0e59fa6..416ea30a 100644
@@ -34,7 +34,6 @@
 typedef enum ovpn_key_slot dco_key_slot_t;
 typedef enum ovpn_cipher_alg dco_cipher_t;
 
-#define DCO_SUPPORTED_CIPHERS "AES-128-GCM:AES-256-GCM:AES-192-GCM:CHACHA20-POLY1305"
 
 typedef struct
 {
Signature · 1 line
2.35.1
#132 lore raw

[Openvpn-devel] [PATCH applied] Re: dco: turn supported ciphers list into a function

As this is really identical to 18/25 v1 (just offsets and context) I've taken the ACK from Heiko, and not done very much testing of my own, just one round of "client tests with DCO in kernel" (unchanged).

Your patch has been applied to the master branch.

commit f9aafa3580a84e2e74a220e6482170130f43bc81

Author: Antonio Quartulli
Date:   Sun Aug 7 12:04:04 2022 +0200
     dco: turn supported ciphers list into a function

     Signed-off-by: Lev Stipakov <[email protected]...>
     Signed-off-by: Antonio Quartulli <[email protected]...>
     Acked-by: Heiko Hund <[email protected]...>
     Message-Id: <[email protected]...>
     URL: https://www.mail-archive.com/[email protected]/msg24835.html
     Signed-off-by: Gert Doering <[email protected]...>

--
kind regards,

Gert Doering

#133 lore raw

PATCH 19/25 dco-win: implement GetOverlappedResultEx for mingw32

In reply to Antonio Quartulli

GetOverlappedResultEx is not available on ming32 therefore we must provide some compat layer before being able to use this function.

---
 src/compat/Makefile.am                        |  3 +-
 src/compat/compat-dco_get_overlapped_result.c | 46 +++++++++++++++++++
 src/compat/compat.h                           |  8 ++++
 src/compat/compat.vcxproj                     |  1 +
 src/compat/compat.vcxproj.filters             |  3 ++
 5 files changed, 60 insertions(+), 1 deletion(-)
 create mode 100644 src/compat/compat-dco_get_overlapped_result.c
src/compat/Makefile.am +2 -1
index 6eb991dc..6dba08aa 100644
@@ -28,4 +28,5 @@ libcompat_la_SOURCES = \
 	compat-gettimeofday.c \
 	compat-daemon.c \
 	compat-strsep.c \
-	compat-versionhelpers.h
+	compat-versionhelpers.h \
+	compat-dco_get_overlapped_result.c
src/compat/compat-dco_get_overlapped_result.c new file +46 -0

46 lines not on this page · raw message

src/compat/compat.h +8 -0
index 026974a8..2b1ad6a6 100644
@@ -62,4 +62,12 @@ char *strsep(char **stringp, const char *delim);
 
 #endif
 
+#if defined(__MINGW32__) && !defined(__MINGW64__)
+BOOL dco_get_overlapped_result(HANDLE handle, OVERLAPPED *ov, DWORD *transferred,
+                               DWORD delay_millisec, BOOL unused);
+
+#else
+#define dco_get_overlapped_result GetOverlappedResultEx
+#endif
+
 #endif /* COMPAT_H */
src/compat/compat.vcxproj +1 -0
index fe03a51a..1dacb503 100644
@@ -159,6 +159,7 @@
   </ItemDefinitionGroup>
   <ItemGroup>
     <ClCompile Include="compat-basename.c" />
+    <ClCompile Include="compat-dco_get_overlapped_result.c" />
     <ClCompile Include="compat-dirname.c" />
     <ClCompile Include="compat-gettimeofday.c" />
     <ClCompile Include="compat-daemon.c" />
src/compat/compat.vcxproj.filters +3 -0
index 96ca026a..73fc9f91 100644
@@ -30,6 +30,9 @@
     <ClCompile Include="compat-strsep.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="compat-dco_get_overlapped_result.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="compat.h">
Signature · 1 line
2.35.1
#134 lore raw

Re: [Openvpn-devel] [PATCH 19/25] dco-win: implement GetOverlappedResultEx for mingw32

Hi,

Quoted 3 lines · Antonio Quartulli

GetOverlappedResultEx is not available on ming32 therefore we must provide some compat layer before being able to use this function.

I suppose "mingw32" here refers to I mingw-w64 for 32 bit (i686) target.

This symbol has been exported in kernel32.lib since version 7.0 release of mingw-w64 (~ 2019). Current release is version 10. Do we need to support older versions given that mingw build is now used only for development by a few of us?

I use debian which probably is the slowest to get new versions -- bullseye has version 8 (gcc version 10), on buster I use unstable which gives version 8. Ubuntu 20.04 has version 7, 22.04 has version 8.

Even if we must support 6.0 and older, let's not unconditionally use the compat layer proposed here.

Quoted 67 lines
---
 src/compat/Makefile.am                        |  3 +-
 src/compat/compat-dco_get_overlapped_result.c | 46 +++++++++++++++++++
 src/compat/compat.h                           |  8 ++++
 src/compat/compat.vcxproj                     |  1 +
 src/compat/compat.vcxproj.filters             |  3 ++
 5 files changed, 60 insertions(+), 1 deletion(-)
 create mode 100644 src/compat/compat-dco_get_overlapped_result.c
src/compat/Makefile.am +2 -1
index 6eb991dc..6dba08aa 100644
@@ -28,4 +28,5 @@ libcompat_la_SOURCES = \
        compat-gettimeofday.c \
        compat-daemon.c \
        compat-strsep.c \
-       compat-versionhelpers.h
+       compat-versionhelpers.h \
+       compat-dco_get_overlapped_result.c
src/compat/compat-dco_get_overlapped_result.c new file +36 -0
new file mode 100644
index 00000000..e14ce976
@@ -0,0 +1,46 @@
+/*
+ *  OpenVPN -- An application to securely tunnel IP networks
+ *             over a single UDP port, with support for SSL/TLS-based
+ *             session authentication and key exchange,
+ *             packet encryption, packet authentication, and
+ *             packet compression.
+ *
+ *  Copyright (C) 2021-2022 Lev Stipakov <[email protected]...>
+ *  Copyright (C) 2021-2022 OpenVPN Inc <[email protected]...>
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License version 2
+ *  as published by the Free Software Foundation.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program (see the file COPYING included with this
+ *  distribution); if not, write to the Free Software Foundation, Inc.,
+ *  59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#elif defined(_MSC_VER)
+#include "config-msvc.h"
+#endif
+
+#include "compat.h"
+
+#if defined(__MINGW32__) && !defined(__MINGW64__)
+BOOL
+dco_get_overlapped_result(HANDLE handle, OVERLAPPED *ov, DWORD

*transferred,

src/compat/compat-dco_get_overlapped_result.c line 42

Quoted 6 lines
src/compat/compat-dco_get_overlapped_result.c new file +6 -0
+                          DWORD delay_millisec, BOOL unused)
+{
+    BOOL res = GetOverlappedResult(handle, ov, transferred, FALSE);
+    if ((res == 0) && (GetLastError() == ERROR_IO_INCOMPLETE))
+    {
+        Sleep(delay_millisec);

While this may be ok assuming it's not inside any performance critical loops, no need to force this on newer versions of mingw-w64.

Alternatively, we could use AC_CHECK_FUNC to make this conditional or find the symbol GetOverlappedResultEx at runtime. But, imo, we do not need any of this as supporting mingw-w64 version 7+ should be enough.

Regards,

Selva

#135 lore raw

Re: [Openvpn-devel] [PATCH 19/25] dco-win: implement GetOverlappedResultEx for mingw32

Hi,

The problem this patch solves appeared on mingw-w64 for 32bit version 6.0.0, which is the latest version of mingw on Windows. When this patch was written (April 2021), we still used travis/appveyor which did mingw builds on Windows so back then patch made sense.

Since we moved to GHa since the last summer and mingw builds are done on Linux with new enough mingw, this patch could be safely removed.

I dropped this commit and pushed changes to my fork and GHa is happy:
https://github.com/lstipakov/openvpn/actions/runs/2567640010

-Lev

#136 lore raw

Re: [Openvpn-devel] [PATCH 19/25] dco-win: implement GetOverlappedResultEx for mingw32

Hi,

Quoted 7 lines · Lev Stipakov

Hi,

The problem this patch solves appeared on mingw-w64 for 32bit version 6.0.0, which is the latest version of mingw on Windows. When this patch was written (April 2021), we still used travis/appveyor which did mingw builds on Windows so back then patch made sense.

Quoted 6 lines

Since we moved to GHa since the last summer and mingw builds are done on Linux with new enough mingw, this patch could be safely removed.

I dropped this commit and pushed changes to my fork and GHa is happy:
https://github.com/lstipakov/openvpn/actions/runs/2567640010

The 'dco' branch has been modified accordingly:
* compat patch 19/25 removed;
* code now uses GetOverlappedResultEx() directly.

Cheers,

Signature · 1 line
Antonio Quartulli
#137 lore raw

PATCH 20/25 dco-win: add platform dependant check on incompatible options

In reply to Antonio Quartulli

Some platforms may have different constraints in terms of incompatible opions, therefore we add a function that explicitly checks those.

Also, add generic option check for when ovpn-dco-win is in use.

---
 src/openvpn/dco.c     | 25 +++++++++++++++++++++++++
 src/openvpn/options.c |  5 +++++
 2 files changed, 30 insertions(+)
src/openvpn/dco.c +25 -0
index 1612a64e..448f29b5 100644
@@ -190,6 +190,26 @@ dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
     }
 }
 
+static bool
+dco_check_option_conflict_platform(int msglevel, const struct options *o)
+{
+#if defined(_WIN32)
+    if (o->mode == MODE_SERVER)
+    {
+        msg(msglevel, "Only client and p2p data channel offload is supported "
+            "with ovpn-dco-win.");
+        return false;
+    }
+
+    if (o->persist_tun)
+    {
+        msg(msglevel, "--persist-tun is not supported with ovpn-dco-win.");
+        return false;
+    }
+#endif
+    return true;
+}
+
 static bool
 dco_check_option_conflict_ce(const struct connection_entry *ce, int msglevel)
 {
@@ -228,6 +248,11 @@ dco_check_option_conflict(int msglevel, const struct options *o)
         return false;
     }
 
+    if (!dco_check_option_conflict_platform(msglevel, o))
+    {
+        return false;
+    }
+
     if (dev_type_enum(o->dev, o->dev_type) != DEV_TYPE_TUN)
     {
         msg(msglevel, "Note: dev-type not tun, disabling data channel offload.");
src/openvpn/options.c +5 -0
index 7b450296..d9da8b8b 100644
@@ -2376,6 +2376,11 @@ options_postprocess_verify_ce(const struct options *options,
     {
         msg(M_USAGE, "--windows-driver wintun requires --dev tun");
     }
+
+    if (options->windows_driver == WINDOWS_DRIVER_WINDCO)
+    {
+        dco_check_option_conflict(M_USAGE, options);
+    }
 #endif /* ifdef _WIN32 */
 
     /*
Signature · 1 line
2.35.1
Acked-by on v1 20/25
#138 lore raw

Re: [Openvpn-devel] [PATCH 20/25] dco-win: add platform dependant check on incompatible options

Quoted 7 lines · On Freitag, 24. Juni 2022 10:38:04 CEST Antonio Quartulli wrote:

Some platforms may have different constraints in terms of incompatible opions, therefore we add a function that explicitly checks those.

Also, add generic option check for when ovpn-dco-win is in use.

#139 lore raw

Re: [Openvpn-devel] [PATCH 20/25] dco-win: add platform dependant check on incompatible options

In reply to Antonio Quartulli

Hi,

this one has an ACK, but needs a v2 still, because...

Quoted 7 lines · Antonio Quartulli

Some platforms may have different constraints in terms of incompatible opions, therefore we add a function that explicitly checks those.

Also, add generic option check for when ovpn-dco-win is in use.

[..]

src/openvpn/options.c options_postprocess_verify_ce(const struct options *options, · line 2384

Quoted 12 lines
src/openvpn/options.c +5 -0
@@ -2376,6 +2376,11 @@ options_postprocess_verify_ce(const struct options *options,
     {
         msg(M_USAGE, "--windows-driver wintun requires --dev tun");
     }
+
+    if (options->windows_driver == WINDOWS_DRIVER_WINDCO)
+    {
+        dco_check_option_conflict(M_USAGE, options);
+    }
 #endif /* ifdef _WIN32 */

... the WINDOWS_DRIVER_WINDCO enum is only introduced in one of the later hunks.

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#140 lore raw

Re: [Openvpn-devel] [PATCH 20/25] dco-win: add platform dependant check on incompatible options

In reply to Antonio Quartulli

Hi,

Quoted 4 lines · Antonio Quartulli

Some platforms may have different constraints in terms of incompatible opions, therefore we add a function that explicitly checks those.

Also, add generic option check for when ovpn-dco-win is in use.

Looking more closely at this, this patch needs a carefuly rebase :-)

src/openvpn/dco.c dco_update_keys(dco_context_t *dco, struct tls_multi *multi) · line 198

Quoted 12 lines
src/openvpn/dco.c +6 -0
@@ -190,6 +190,26 @@ dco_update_keys(dco_context_t *dco, struct tls_multi *multi)
     }
 }
 
+static bool
+dco_check_option_conflict_platform(int msglevel, const struct options *o)
+{
+#if defined(_WIN32)
+    if (o->mode == MODE_SERVER)
+    {

... it *introduces* dco_check_option_conflict_platform() with "only windows" content, but in the current master tree, this function already exists, with Linux content...

gert

Signature · 6 lines
"If was one thing all people took for granted, was conviction that if you 
 feed honest figures into a computer, honest figures come out. Never doubted 
 it myself till I met a computer with a sense of humor."
                             Robert A. Heinlein, The Moon is a Harsh Mistress

Gert Doering - Munich, Germany                             [email protected]...
#141 lore raw

PATCH 21/25 do_open_tun: restyle "can preserve TUN" check

In reply to Antonio Quartulli

The current condition checking if the TUN interface was preserved is dependant on the platform being Android or not. This makes the code reasonably ugly, especially because uncrustify can't indent properly.

On top of that, we will require an extra condition only for windows+DCO, which will make the check even uglier.

For this reason, factor out the check in a separate function which can keep the ifdefs craziness well hidden, while do_open_tun becomes (a bit) cleaner.

---
 src/openvpn/init.c | 283 +++++++++++++++++++++++----------------------
 1 file changed, 146 insertions(+), 137 deletions(-)
src/openvpn/init.c +146 -137
index bdd2ad96..f95dcb93 100644
@@ -1690,161 +1690,123 @@ do_init_tun(struct context *c)
  * Open tun/tap device, ifconfig, call up script, etc.
  */
 
+
+static bool
+can_preserve_tun(struct tuntap *tt)
+{
+#ifdef TARGET_ANDROID
+    return false;
+#elif _WIN32
+    return tt && tt->dco.real_tun_init;
+#else
+    return tt;
+#endif
+}
+
 static bool
 do_open_tun(struct context *c)
 {
     struct gc_arena gc = gc_new();
     bool ret = false;
 
-#ifndef TARGET_ANDROID
-    if (!c->c1.tuntap)
+    if (!can_preserve_tun(c->c1.tuntap))
     {
-#endif
 
 #ifdef TARGET_ANDROID
-    /* If we emulate persist-tun on android we still have to open a new tun and
-     * then close the old */
-    int oldtunfd = -1;
-    if (c->c1.tuntap)
-    {
-        oldtunfd = c->c1.tuntap->fd;
-        free(c->c1.tuntap);
-        c->c1.tuntap = NULL;
-        c->c1.tuntap_owned = false;
-    }
+        /* If we emulate persist-tun on android we still have to open a new tun and
+         * then close the old */
+        int oldtunfd = -1;
+        if (c->c1.tuntap)
+        {
+            oldtunfd = c->c1.tuntap->fd;
+            free(c->c1.tuntap);
+            c->c1.tuntap = NULL;
+            c->c1.tuntap_owned = false;
+        }
 #endif
 
-    /* initialize (but do not open) tun/tap object */
-    do_init_tun(c);
+        /* initialize (but do not open) tun/tap object */
+        do_init_tun(c);
 
-    /* inherit the dco context from the tuntap object */
-    if (c->c2.tls_multi)
-    {
-        c->c2.tls_multi->dco = &c->c1.tuntap->dco;
-    }
+        /* inherit the dco context from the tuntap object */
+        if (c->c2.tls_multi)
+        {
+            c->c2.tls_multi->dco = &c->c1.tuntap->dco;
+        }
 
 #ifdef _WIN32
-    /* store (hide) interactive service handle in tuntap_options */
-    c->c1.tuntap->options.msg_channel = c->options.msg_channel;
-    msg(D_ROUTE, "interactive service msg_channel=%" PRIu64, (unsigned long long) c->options.msg_channel);
+        /* store (hide) interactive service handle in tuntap_options */
+        c->c1.tuntap->options.msg_channel = c->options.msg_channel;
+        msg(D_ROUTE, "interactive service msg_channel=%" PRIu64, (unsigned long long) c->options.msg_channel);
 #endif
 
-    /* allocate route list structure */
-    do_alloc_route_list(c);
+        /* allocate route list structure */
+        do_alloc_route_list(c);
 
-    /* parse and resolve the route option list */
-    ASSERT(c->c2.link_socket);
-    if (c->options.routes && c->c1.route_list)
-    {
-        do_init_route_list(&c->options, c->c1.route_list,
-                           &c->c2.link_socket->info, c->c2.es, &c->net_ctx);
-    }
-    if (c->options.routes_ipv6 && c->c1.route_ipv6_list)
-    {
-        do_init_route_ipv6_list(&c->options, c->c1.route_ipv6_list,
-                                &c->c2.link_socket->info, c->c2.es,
-                                &c->net_ctx);
-    }
+        /* parse and resolve the route option list */
+        ASSERT(c->c2.link_socket);
+        if (c->options.routes && c->c1.route_list)
+        {
+            do_init_route_list(&c->options, c->c1.route_list,
+                               &c->c2.link_socket->info, c->c2.es, &c->net_ctx);
+        }
+        if (c->options.routes_ipv6 && c->c1.route_ipv6_list)
+        {
+            do_init_route_ipv6_list(&c->options, c->c1.route_ipv6_list,
+                                    &c->c2.link_socket->info, c->c2.es,
+                                    &c->net_ctx);
+        }
 
-    /* do ifconfig */
-    if (!c->options.ifconfig_noexec
-        && ifconfig_order() == IFCONFIG_BEFORE_TUN_OPEN)
-    {
-        /* guess actual tun/tap unit number that will be returned
-         * by open_tun */
-        const char *guess = guess_tuntap_dev(c->options.dev,
-                                             c->options.dev_type,
-                                             c->options.dev_node,
-                                             &gc);
-        do_ifconfig(c->c1.tuntap, guess, c->c2.frame.tun_mtu, c->c2.es,
-                    &c->net_ctx);
-    }
+        /* do ifconfig */
+        if (!c->options.ifconfig_noexec
+            && ifconfig_order() == IFCONFIG_BEFORE_TUN_OPEN)
+        {
+            /* guess actual tun/tap unit number that will be returned
+             * by open_tun */
+            const char *guess = guess_tuntap_dev(c->options.dev,
+                                                 c->options.dev_type,
+                                                 c->options.dev_node,
+                                                 &gc);
+            do_ifconfig(c->c1.tuntap, guess, c->c2.frame.tun_mtu, c->c2.es,
+                        &c->net_ctx);
+        }
 
-    /* possibly add routes */
-    if (route_order() == ROUTE_BEFORE_TUN)
-    {
-        /* Ignore route_delay, would cause ROUTE_BEFORE_TUN to be ignored */
-        do_route(&c->options, c->c1.route_list, c->c1.route_ipv6_list,
-                 c->c1.tuntap, c->plugins, c->c2.es, &c->net_ctx);
-    }
+        /* possibly add routes */
+        if (route_order() == ROUTE_BEFORE_TUN)
+        {
+            /* Ignore route_delay, would cause ROUTE_BEFORE_TUN to be ignored */
+            do_route(&c->options, c->c1.route_list, c->c1.route_ipv6_list,
+                     c->c1.tuntap, c->plugins, c->c2.es, &c->net_ctx);
+        }
 #ifdef TARGET_ANDROID
-    /* Store the old fd inside the fd so open_tun can use it */
-    c->c1.tuntap->fd = oldtunfd;
-#endif
-    if (dco_enabled(&c->options))
-    {
-        ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
-    }
-
-    /* open the tun device */
-    open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-             c->c1.tuntap, &c->net_ctx);
-
-    /* set the hardware address */
-    if (c->options.lladdr)
-    {
-        set_lladdr(&c->net_ctx, c->c1.tuntap->actual_name, c->options.lladdr,
-                   c->c2.es);
-    }
-
-    /* do ifconfig */
-    if (!c->options.ifconfig_noexec
-        && ifconfig_order() == IFCONFIG_AFTER_TUN_OPEN)
-    {
-        do_ifconfig(c->c1.tuntap, c->c1.tuntap->actual_name,
-                    c->c2.frame.tun_mtu, c->c2.es, &c->net_ctx);
-    }
-
-    /* run the up script */
-    run_up_down(c->options.up_script,
-                c->plugins,
-                OPENVPN_PLUGIN_UP,
-                c->c1.tuntap->actual_name,
-#ifdef _WIN32
-                c->c1.tuntap->adapter_index,
+        /* Store the old fd inside the fd so open_tun can use it */
+        c->c1.tuntap->fd = oldtunfd;
 #endif
-                dev_type_string(c->options.dev, c->options.dev_type),
-                c->c2.frame.tun_mtu,
-                print_in_addr_t(c->c1.tuntap->local, IA_EMPTY_IF_UNDEF, &gc),
-                print_in_addr_t(c->c1.tuntap->remote_netmask, IA_EMPTY_IF_UNDEF, &gc),
-                "init",
-                NULL,
-                "up",
-                c->c2.es);
-
-#if defined(_WIN32)
-    if (c->options.block_outside_dns)
-    {
-        dmsg(D_LOW, "Blocking outside DNS");
-        if (!win_wfp_block_dns(c->c1.tuntap->adapter_index, c->options.msg_channel))
+        if (dco_enabled(&c->options))
         {
-            msg(M_FATAL, "Blocking DNS failed!");
+            ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
         }
-    }
-#endif
 
-    /* possibly add routes */
-    if ((route_order() == ROUTE_AFTER_TUN) && (!c->options.route_delay_defined))
-    {
-        do_route(&c->options, c->c1.route_list, c->c1.route_ipv6_list,
-                 c->c1.tuntap, c->plugins, c->c2.es, &c->net_ctx);
-    }
+        /* open the tun device */
+        open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
+                 c->c1.tuntap, &c->net_ctx);
 
-    ret = true;
-    static_context = c;
-#ifndef TARGET_ANDROID
-}
-else
-{
-    msg(M_INFO, "Preserving previous TUN/TAP instance: %s",
-        c->c1.tuntap->actual_name);
+        /* set the hardware address */
+        if (c->options.lladdr)
+        {
+            set_lladdr(&c->net_ctx, c->c1.tuntap->actual_name, c->options.lladdr,
+                       c->c2.es);
+        }
 
-    /* explicitly set the ifconfig_* env vars */
-    do_ifconfig_setenv(c->c1.tuntap, c->c2.es);
+        /* do ifconfig */
+        if (!c->options.ifconfig_noexec
+            && ifconfig_order() == IFCONFIG_AFTER_TUN_OPEN)
+        {
+            do_ifconfig(c->c1.tuntap, c->c1.tuntap->actual_name,
+                        c->c2.frame.tun_mtu, c->c2.es, &c->net_ctx);
+        }
 
-    /* run the up script if user specified --up-restart */
-    if (c->options.up_restart)
-    {
+        /* run the up script */
         run_up_down(c->options.up_script,
                     c->plugins,
                     OPENVPN_PLUGIN_UP,
@@ -1856,24 +1818,71 @@ else
                     c->c2.frame.tun_mtu,
                     print_in_addr_t(c->c1.tuntap->local, IA_EMPTY_IF_UNDEF, &gc),
                     print_in_addr_t(c->c1.tuntap->remote_netmask, IA_EMPTY_IF_UNDEF, &gc),
-                    "restart",
+                    "init",
                     NULL,
                     "up",
                     c->c2.es);
-    }
+
 #if defined(_WIN32)
-    if (c->options.block_outside_dns)
-    {
-        dmsg(D_LOW, "Blocking outside DNS");
-        if (!win_wfp_block_dns(c->c1.tuntap->adapter_index, c->options.msg_channel))
+        if (c->options.block_outside_dns)
         {
-            msg(M_FATAL, "Blocking DNS failed!");
+            dmsg(D_LOW, "Blocking outside DNS");
+            if (!win_wfp_block_dns(c->c1.tuntap->adapter_index, c->options.msg_channel))
+            {
+                msg(M_FATAL, "Blocking DNS failed!");
+            }
         }
+#endif
+
+        /* possibly add routes */
+        if ((route_order() == ROUTE_AFTER_TUN) && (!c->options.route_delay_defined))
+        {
+            do_route(&c->options, c->c1.route_list, c->c1.route_ipv6_list,
+                     c->c1.tuntap, c->plugins, c->c2.es, &c->net_ctx);
+        }
+
+        ret = true;
+        static_context = c;
     }
+    else
+    {
+        msg(M_INFO, "Preserving previous TUN/TAP instance: %s",
+            c->c1.tuntap->actual_name);
+
+        /* explicitly set the ifconfig_* env vars */
+        do_ifconfig_setenv(c->c1.tuntap, c->c2.es);
+
+        /* run the up script if user specified --up-restart */
+        if (c->options.up_restart)
+        {
+            run_up_down(c->options.up_script,
+                        c->plugins,
+                        OPENVPN_PLUGIN_UP,
+                        c->c1.tuntap->actual_name,
+#ifdef _WIN32
+                        c->c1.tuntap->adapter_index,
+#endif
+                        dev_type_string(c->options.dev, c->options.dev_type),
+                        c->c2.frame.tun_mtu,
+                        print_in_addr_t(c->c1.tuntap->local, IA_EMPTY_IF_UNDEF, &gc),
+                        print_in_addr_t(c->c1.tuntap->remote_netmask, IA_EMPTY_IF_UNDEF, &gc),
+                        "restart",
+                        NULL,
+                        "up",
+                        c->c2.es);
+        }
+#if defined(_WIN32)
+        if (c->options.block_outside_dns)
+        {
+            dmsg(D_LOW, "Blocking outside DNS");
+            if (!win_wfp_block_dns(c->c1.tuntap->adapter_index, c->options.msg_channel))
+            {
+                msg(M_FATAL, "Blocking DNS failed!");
+            }
+        }
 #endif
 
-}
-#endif /* ifndef TARGET_ANDROID */
+    }
     gc_free(&gc);
     return ret;
 }
Signature · 1 line
2.35.1
Acked-by on v1 21/25
#142 lore raw

Re: [Openvpn-devel] [PATCH 21/25] do_open_tun: restyle "can preserve TUN" check

Quoted 4 lines · On Freitag, 24. Juni 2022 10:38:05 CEST Antonio Quartulli wrote:

The current condition checking if the TUN interface was preserved is dependant on the platform being Android or not. This makes the code reasonably ugly, especially because uncrustify can't indent properly.

Quoted 8 lines

On top of that, we will require an extra condition only for windows+DCO, which will make the check even uglier.

For this reason, factor out the check in a separate function which can keep the ifdefs craziness well hidden, while do_open_tun becomes (a bit) cleaner.

Looks mighty, but is rather trivial. Two things:

I think the code could benefit from factoring out the call to run_up_down() and block_outside_dns into two simple functions. Currently it's pretty much duplicated in the if and else branch. Besides that:

#143 lore raw

Re: [Openvpn-devel] [PATCH 21/25] do_open_tun: restyle "can preserve TUN" check

Hi,

Quoted 8 lines · Heiko Hund
Quoted 7 lines · On Freitag, 24. Juni 2022 10:38:05 CEST Antonio Quartulli wrote:

The current condition checking if the TUN interface was preserved is dependant on the platform being Android or not. This makes the code reasonably ugly, especially because uncrustify can't indent properly.

On top of that, we will require an extra condition only for windows+DCO, which will make the check even uglier.

Quoted 7 lines
Quoted 5 lines

For this reason, factor out the check in a separate function which can keep the ifdefs craziness well hidden, while do_open_tun becomes (a bit) cleaner.

Looks mighty, but is rather trivial. Two things:

Can also be reviewed with "git show -w" to highlight the real changes. Most of the diff is reindenting.

Quoted 4 lines

I think the code could benefit from factoring out the call to run_up_down() and block_outside_dns into two simple functions. Currently it's pretty much duplicated in the if and else branch. Besides that:

meat for another patch ;)

Quoted 2 lines

Thanks!

Quoted 9 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#144 lore raw

Re: [Openvpn-devel] [PATCH 21/25] do_open_tun: restyle "can preserve TUN" check

In reply to Antonio Quartulli

Hi,

src/openvpn/init.c do_init_tun(struct context *c) · line 1700

Quoted 7 lines
src/openvpn/init.c +7 -0
+static bool
+can_preserve_tun(struct tuntap *tt)
+{
+#ifdef TARGET_ANDROID
+    return false;
+#elif _WIN32
+    return tt && tt->dco.real_tun_init;

Looks like dco.real_tun_init has been added in the very next patch, 22/25.

#145 lore raw

Re: [Openvpn-devel] [PATCH 21/25] do_open_tun: restyle "can preserve TUN" check

Hi,

Quoted 11 lines · Lev Stipakov

Hi,

src/openvpn/init.c do_init_tun(struct context *c) · line 1700

Quoted 7 lines
src/openvpn/init.c +7 -0
+static bool
+can_preserve_tun(struct tuntap *tt)
+{
+#ifdef TARGET_ANDROID
+    return false;
+#elif _WIN32
+    return tt && tt->dco.real_tun_init;

Looks like dco.real_tun_init has been added in the very next patch, 22/25.

Darn - must have missed this during the patch split up.

Thanks for pointing this out.

Cheers,

Quoted 1 line
Signature · 1 line
Antonio Quartulli
#146 lore raw

PATCH 22/25 dco-win: introduce low-level code for handling ovpn-dco-win in Windows

In reply to Antonio Quartulli
---
 config-msvc.h                       |   2 +
 configure.ac                        |   9 +-
 dev-tools/special-files.lst         |   1 +
 src/openvpn/Makefile.am             |   4 +-
 src/openvpn/dco_internal.h          |   1 +
 src/openvpn/dco_win.c               | 402 ++++++++++++++++++++++++++++
 src/openvpn/dco_win.h               |  58 ++++
 src/openvpn/openvpn.vcxproj         |   3 +
 src/openvpn/openvpn.vcxproj.filters |   9 +
 src/openvpn/ovpn-dco-win.h          | 108 ++++++++
 10 files changed, 593 insertions(+), 4 deletions(-)
 create mode 100644 src/openvpn/dco_win.c
 create mode 100644 src/openvpn/dco_win.h
 create mode 100644 src/openvpn/ovpn-dco-win.h
config-msvc.h +2 -0
index b08beb52..b621f3fb 100644
@@ -87,3 +87,5 @@ typedef uint16_t in_port_t;
 #ifdef HAVE_CONFIG_MSVC_LOCAL_H
 #include <config-msvc-local.h>
 #endif
+
+#define ENABLE_DCO 1
configure.ac +6 -3
index 353da08c..bebed1ac 100644
@@ -144,7 +144,7 @@ AC_ARG_ENABLE(
 
 AC_ARG_ENABLE(
 	[dco],
-	[AS_HELP_STRING([--enable-dco], [enable data channel offload support using ovpn-dco kernel module @<:@default=no@:>@])],
+	[AS_HELP_STRING([--enable-dco], [enable data channel offload support using the ovpn-dco kernel module (always enabled on Windows) @<:@default=no@:>@])],
 	,
 	[enable_dco="no"]
 )
@@ -328,6 +328,7 @@ case "$host" in
 		;;
 	*-mingw*)
 		AC_DEFINE([TARGET_WIN32], [1], [Are we running WIN32?])
+		AC_DEFINE([ENABLE_DCO], [1], [DCO is always enabled on Windows])
 		AC_DEFINE_UNQUOTED([TARGET_PREFIX], ["W"], [Target prefix])
 		CPPFLAGS="${CPPFLAGS} -DWIN32_LEAN_AND_MEAN"
 		CPPFLAGS="${CPPFLAGS} -DNTDDI_VERSION=NTDDI_VISTA -D_WIN32_WINNT=_WIN32_WINNT_VISTA"
@@ -772,7 +773,6 @@ if test "$enable_dco" = "yes"; then
 dnl
 dnl Include generic netlink library used to talk to ovpn-dco
 dnl
-
 	case "$host" in
 		*-*-linux*)
 			PKG_CHECK_MODULES([LIBNL_GENL],
@@ -787,8 +787,11 @@ dnl
 			AC_DEFINE(ENABLE_DCO, 1, [Enable shared data channel offload])
 			AC_MSG_NOTICE([Enabled ovpn-dco support for Linux])
 			;;
+		*-mingw*)
+			AC_MSG_NOTICE([NOTE: --enable-dco ignored on Windows because it's always enabled])
+			;;
 		*)
-			AC_MSG_NOTICE([Ignoring --enable-dco on non Linux platform])
+			AC_MSG_NOTICE([Ignoring --enable-dco on non supported platform])
 			;;
 	esac
 fi
dev-tools/special-files.lst +1 -0
index 33e830d7..6d0bc055 100644
@@ -2,3 +2,4 @@ E:doc/doxygen/doc_key_generation.h     # @verbatim section gets mistreated, excl
 E:src/compat/compat-lz4.c              # Preserve LZ4 upstream formatting
 E:src/compat/compat-lz4.h              # Preserve LZ4 upstream formatting
 E:src/openvpn/ovpn_dco_linux.h         # Preserve ovpn-dco upstream formatting
+E:src/openvpn/ovpn-dco-win.h           # Preserve ovpn-dco-win upstream formatting
src/openvpn/Makefile.am +3 -1
index aaa1dbce..8d0e66b4 100644
@@ -55,6 +55,7 @@ openvpn_SOURCES = \
 	crypto_mbedtls.c crypto_mbedtls.h \
 	dco.c dco.h dco_internal.h \
 	dco_linux.c dco_linux.h \
+	dco_win.c dco_win.h \
 	dhcp.c dhcp.h \
 	dns.c dns.h \
 	env_set.c env_set.h \
@@ -77,6 +78,7 @@ openvpn_SOURCES = \
 	mbuf.c mbuf.h \
 	memdbg.h \
 	misc.c misc.h \
+	ovpn-dco-win.h \
 	ovpn_dco_linux.h \
 	platform.c platform.h \
 	console.c console.h console_builtin.c console_systemd.c \
@@ -151,5 +153,5 @@ openvpn_LDADD = \
 	$(OPTIONAL_INOTIFY_LIBS)
 if WIN32
 openvpn_SOURCES += openvpn_win32_resources.rc block_dns.c block_dns.h ring_buffer.h
-openvpn_LDADD += -lgdi32 -lws2_32 -lwininet -lcrypt32 -liphlpapi -lwinmm -lfwpuclnt -lrpcrt4 -lncrypt -lsetupapi
+openvpn_LDADD += -lgdi32 -lws2_32 -lwininet -lcrypt32 -liphlpapi -lwinmm -lfwpuclnt -lrpcrt4 -lncrypt -lsetupapi -lbcrypt
 endif
src/openvpn/dco_internal.h +1 -0
index 3ceb26d6..968d4042 100644
@@ -28,6 +28,7 @@
 #if defined(ENABLE_DCO)
 
 #include "dco_linux.h"
+#include "dco_win.h"
 
 /**
  * This file contains the internal DCO API definition.
src/openvpn/dco_win.c new file +402 -0

402 lines not on this page · raw message

src/openvpn/dco_win.h new file +58 -0

58 lines not on this page · raw message

src/openvpn/openvpn.vcxproj +3 -0
index 0b3db7c7..51e8fe79 100644
@@ -278,6 +278,7 @@
     <ClCompile Include="cryptoapi.c" />
     <ClCompile Include="dco.c" />
     <ClCompile Include="dco_linux.c" />
+    <ClCompile Include="dco_win.c" />
     <ClCompile Include="dhcp.c" />
     <ClCompile Include="dns.c" />
     <ClCompile Include="env_set.c" />
@@ -367,6 +368,7 @@
     <ClInclude Include="dco.h" />
     <ClInclude Include="dco_internal.h" />
     <ClInclude Include="dco_linux.h" />
+    <ClInclude Include="dco_win.h" />
     <ClInclude Include="dhcp.h" />
     <ClInclude Include="dns.h" />
     <ClInclude Include="env_set.h" />
@@ -401,6 +403,7 @@
     <ClInclude Include="openvpn.h" />
     <ClInclude Include="options.h" />
     <ClInclude Include="otime.h" />
+    <ClInclude Include="ovpn-dco-win.h" />
     <ClInclude Include="ovpn_dco_linux.h" />
     <ClInclude Include="packet_id.h" />
     <ClInclude Include="perf.h" />
src/openvpn/openvpn.vcxproj.filters +9 -0
index 16905079..117270bb 100644
@@ -42,6 +42,9 @@
     <ClCompile Include="dco_linux.c">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="dco_win.c">
+      <Filter>Source Files</Filter>
+    </ClCompile>
     <ClCompile Include="dhcp.c">
       <Filter>Source Files</Filter>
     </ClCompile>
@@ -314,6 +317,9 @@
     <ClInclude Include="dco_linux.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="dco_win.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
     <ClInclude Include="dhcp.h">
       <Filter>Header Files</Filter>
     </ClInclude>
@@ -413,6 +419,9 @@
     <ClInclude Include="otime.h">
       <Filter>Header Files</Filter>
     </ClInclude>
+    <ClInclude Include="ovpn-dco-win.h">
+      <Filter>Header Files</Filter>
+    </ClInclude>
     <ClInclude Include="ovpn_dco_linux.h">
       <Filter>Header Files</Filter>
     </ClInclude>
src/openvpn/ovpn-dco-win.h new file +108 -0

108 lines not on this page · raw message

Signature · 1 line
2.35.1
#147 lore raw

Re: [Openvpn-devel] [PATCH 22/25] dco-win: introduce low-level code for handling ovpn-dco-win in Windows

Hi,

Quoted 1 line

+ if (dco_get_overlapped_result(handle, ov, &transferred, poll_interval_ms, FALSE) != 0)

With the removal of mingw compat patch ("dco-win: implement GetOverlappedResultEx for mingw32"), this needs to be changed to GetOverlappedResultEx (signature is the same).

-Lev

#148 lore raw

Re: [Openvpn-devel] [PATCH 22/25] dco-win: introduce low-level code for handling ovpn-dco-win in Windows

In reply to Antonio Quartulli

Hi,

I did some review / testing and suggest following changes to this patch:

src/openvpn/dco_win.c line 100

Quoted 6 lines
src/openvpn/dco_win.c new file +6 -0
+    if (!DeviceIoControl(tt->hand, OVPN_IOCTL_START_VPN, NULL, 0, NULL, 0,
+                         &bytes_returned, NULL))
+    {
+        msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_START_VPN) failed with code %lu",
+            GetLastError());
+    }

Since recently openvpn correctly handles Win32 errors (doesn't mix them with CRT errors)
so this could be simplified like:

  msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_START_VPN) failed");

M_ERR prints error message along with error code.

<snip>

src/openvpn/dco_win.c line 132

Quoted 5 lines
src/openvpn/dco_win.c new file +5 -0
+            /* dco reported connection error */
+            struct gc_arena gc = gc_new();
+            msg(M_NONFATAL, "%s: %s", __func__, strerror_win32(err, &gc));
+            *signal_received = SIGUSR1;
+            gc_free(&gc);

Ditto, so

  msg(M_NONFATAL | M_ERRNO, "dco connect error");

(no need to instantiate gc_arena)

<snip>

src/openvpn/dco_win.c line 146

Quoted 2 lines
src/openvpn/dco_win.c new file +2 -0
+    /* we end up here when timeout occurs in userspace */
+    msg(M_NONFATAL, "%s: dco connect timeout", __func__);

I would just use

  msg(M_NONFATAL, "dco connect timeout");

<snip>

src/openvpn/dco_win.c line 236

Quoted 4 lines
src/openvpn/dco_win.c new file +4 -0
+        if (err != ERROR_IO_PENDING)
+        {
+            msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_NEW_PEER) failed with code %lu", err);
+        }

Ditto:

  msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_NEW_PEER) failed");

<snip>

src/openvpn/dco_win.c line 283

Quoted 7 lines
src/openvpn/dco_win.c new file +7 -0
+    DWORD bytes_returned = 0;
+    if (!DeviceIoControl(dco->tt->hand, OVPN_IOCTL_SET_PEER, &peer,
+                         sizeof(peer), NULL, 0, &bytes_returned, NULL))
+    {
+        msg(M_WARN, "DeviceIoControl(OVPN_IOCTL_SET_PEER) failed with code %lu", GetLastError());
+        return -1;
+    }

Ditto:

  msg(M_WARN | M_ERRNO, "DeviceIoControl(OVPN_IOCTL_SET_PEER) failed");

<snip>

Review

Quoted 2 lines
(quoted diff) +2 -0
+        msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_NEW_KEY) failed with code %lu",
+            GetLastError());

Same as above:

  msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_NEW_KEY) failed");

<snip>

src/openvpn/dco_win.c line 348

Quoted 2 lines
src/openvpn/dco_win.c new file +2 -0
+        msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_SWAP_KEYS) failed with code %lu",
+            GetLastError());

Ditto:

  msg(M_ERR, "DeviceIoControl(OVPN_IOCTL_SWAP_KEYS) failed");

We have discussed all those changes in chat and a new version with the changes is pushed to the dco branch.

-Lev

#149 lore raw

Re: [Openvpn-devel] [PATCH 22/25] dco-win: introduce low-level code for handling ovpn-dco-win in Windows

In reply to Antonio Quartulli
Quoted 6 lines · On Freitag, 24. Juni 2022 10:38:06 CEST Antonio Quartulli wrote:
src/openvpn/dco_win.c new file +4 -0
+int
+dco_del_key(dco_context_t *dco, unsigned int peerid, dco_key_slot_t slot)
+{
+    msg(D_DCO, "%s: peer-id %d, slot %d called but ignored", __func__,

peerid, + slot);
+ /* FIXME: Implement in driver first */

Suppose this is not critical and the driver gets rid of keys itself?

#150 lore raw

Re: [Openvpn-devel] [PATCH 22/25] dco-win: introduce low-level code for handling ovpn-dco-win in Windows

Hi,

Quoted 9 lines · Heiko Hund
Quoted 6 lines · On Freitag, 24. Juni 2022 10:38:06 CEST Antonio Quartulli wrote:
src/openvpn/dco_win.c new file +4 -0
+int
+dco_del_key(dco_context_t *dco, unsigned int peerid, dco_key_slot_t slot)
+{
+    msg(D_DCO, "%s: peer-id %d, slot %d called but ignored", __func__,

peerid, + slot);
+ /* FIXME: Implement in driver first */

Suppose this is not critical and the driver gets rid of keys itself?

Yeah, the dco-win driver does not offer the same API granularity because it does not want/need to cover all the usecases that linux does.

So yeah, this simply happens in kernel directly upon disconnection.

Cheers,

Quoted 10 lines

_______________________________________________
Openvpn-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/openvpn-devel

Signature · 1 line
Antonio Quartulli
#151 lore raw

PATCH 23/25 dco-win: implement ovpn-dco support in P2P Windows code path

In reply to Antonio Quartulli

With this change it is possible to use ovpn-dco-win when running OpenVPN in client or P2P mode.

---
 src/openvpn/forward.c |  7 ++++
 src/openvpn/init.c    | 29 +++++++++++--
 src/openvpn/options.c | 19 ++++++---
 src/openvpn/options.h | 15 +++----
 src/openvpn/socket.c  | 94 ++++++++++++++++++++++++++++++++++++++++---
 src/openvpn/socket.h  | 25 ++++++++----
 src/openvpn/tun.c     | 43 +++++++++++++++-----
 src/openvpn/tun.h     | 58 ++++++++++++++++++--------
 8 files changed, 233 insertions(+), 57 deletions(-)
src/openvpn/forward.c +7 -0
index 15bdbbae..1fbad43f 100644
@@ -860,6 +860,13 @@ read_incoming_link(struct context *c)
     /* check recvfrom status */
     check_status(status, "read", c->c2.link_socket, NULL);
 
+#ifdef _WIN32
+    if (dco_enabled(&c->options) && (status < 0) && (openvpn_errno() == ERROR_NETNAME_DELETED))
+    {
+        trigger_ping_timeout_signal(c);
+    }
+#endif
+
     /* Remove socks header if applicable */
     socks_postprocess_incoming_link(c);
 
src/openvpn/init.c +25 -4
index f95dcb93..77f36a1a 100644
@@ -1673,7 +1673,8 @@ do_init_tun(struct context *c)
                             c->c1.link_socket_addr.remote_list,
                             !c->options.ifconfig_nowarn,
                             c->c2.es,
-                            &c->net_ctx);
+                            &c->net_ctx,
+                            c->c1.tuntap);
 
 #ifdef _WIN32
     c->c1.tuntap->windows_driver = c->options.windows_driver;
@@ -1787,9 +1788,12 @@ do_open_tun(struct context *c)
             ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
         }
 
-        /* open the tun device */
-        open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-                 c->c1.tuntap, &c->net_ctx);
+        /* open the tun device (ovpn-dco-win already opened the device for the socket) */
+        if (!is_windco(c->c1.tuntap))
+        {
+            open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
+                     c->c1.tuntap, &c->net_ctx);
+        }
 
         /* set the hardware address */
         if (c->options.lladdr)
@@ -3541,6 +3545,23 @@ do_close_free_key_schedule(struct context *c, bool free_ssl_ctx)
 static void
 do_close_link_socket(struct context *c)
 {
+#if defined(_WIN32)
+    if (is_windco(c->c1.tuntap) && c->c2.link_socket
+        && c->c2.link_socket->info.dco_installed)
+    {
+        ASSERT(c->c2.link_socket_owned);
+
+        /* We rely on the tun_close to close the handle if also setup
+         * routes etc, since they cannot be delete when the interface
+         * handle has been closed */
+        if (!c->c1.tuntap->dco.real_tun_init)
+        {
+            do_close_tun_simple(c);
+        }
+        c->c2.link_socket->sd = SOCKET_UNDEFINED;
+    }
+#endif
+
     if (c->c2.link_socket && c->c2.link_socket_owned)
     {
         link_socket_close(c->c2.link_socket);
src/openvpn/options.c +14 -5
index d9da8b8b..00823604 100644
@@ -3151,8 +3151,8 @@ options_postprocess_mutate_invariant(struct options *options)
 #ifdef _WIN32
     const int dev = dev_type_enum(options->dev, options->dev_type);
 
-    /* when using wintun, kernel doesn't send DHCP requests, so don't use it */
-    if (options->windows_driver == WINDOWS_DRIVER_WINTUN
+    /* when using wintun/ovpn-dco-win, kernel doesn't send DHCP requests, so don't use it */
+    if ((options->windows_driver == WINDOWS_DRIVER_WINTUN || options->windows_driver == WINDOWS_DRIVER_WINDCO)
         && (options->tuntap_options.ip_win32_type == IPW32_SET_DHCP_MASQ || options->tuntap_options.ip_win32_type == IPW32_SET_ADAPTIVE))
     {
         options->tuntap_options.ip_win32_type = IPW32_SET_NETSH;
@@ -3248,10 +3248,12 @@ options_postprocess_setdefault_ncpciphers(struct options *o)
         /* custom --data-ciphers set, keep list */
         return;
     }
+#if !defined(_WIN32)
     else if (cipher_valid("CHACHA20-POLY1305"))
     {
         o->ncp_ciphers = "AES-256-GCM:AES-128-GCM:CHACHA20-POLY1305";
     }
+#endif
     else
     {
         o->ncp_ciphers = "AES-256-GCM:AES-128-GCM";
@@ -4076,7 +4078,8 @@ options_string(const struct options *o,
                       NULL,
                       false,
                       NULL,
-                      ctx);
+                      ctx,
+                      NULL);
         if (tt)
         {
             tt_local = true;
@@ -4463,13 +4466,19 @@ parse_windows_driver(const char *str, const int msglevel)
     {
         return WINDOWS_DRIVER_WINTUN;
     }
+
+    else if (streq(str, "ovpn-dco-win"))
+    {
+        return WINDOWS_DRIVER_WINDCO;
+    }
     else
     {
-        msg(msglevel, "--windows-driver must be tap-windows6 or wintun");
+        msg(msglevel, "--windows-driver must be tap-windows6, wintun "
+            "or ovpn-dco-win");
         return WINDOWS_DRIVER_UNSPECIFIED;
     }
 }
-#endif
+#endif /* ifdef _WIN32 */
 
 /*
  * parse/print topology coding
src/openvpn/options.h +5 -10
index 8152e755..3a5b433e 100644
@@ -879,24 +879,19 @@ void options_string_import(struct options *options,
 
 bool key_is_external(const struct options *options);
 
-#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
-
 /**
  * Returns whether the current configuration has dco enabled.
  */
 static inline bool
 dco_enabled(const struct options *o)
 {
+#if defined(_WIN32)
+    return o->windows_driver == WINDOWS_DRIVER_WINDCO;
+#elif defined(ENABLE_DCO)
     return !o->tuntap_options.disable_dco;
-}
-
-#else /* if defined(ENABLE_DCO) && defined(TARGET_LINUX) */
-
-static inline bool
-dco_enabled(const struct options *o)
-{
+#else
     return false;
+#endif /* defined(_WIN32) */
 }
 
-#endif
 #endif /* ifndef OPTIONS_H */
src/openvpn/socket.c +88 -6

136 lines not on this page · raw message

src/openvpn/socket.h +18 -7
index 0d521d22..462afa31 100644
@@ -34,6 +34,7 @@
 #include "proxy.h"
 #include "socks.h"
 #include "misc.h"
+#include "tun.h"
 
 /*
  * OpenVPN's default port number as assigned by IANA.
@@ -937,7 +938,8 @@ socket_connection_reset(const struct link_socket *sock, int status)
         {
             const int err = openvpn_errno();
 #ifdef _WIN32
-            return err == WSAECONNRESET || err == WSAECONNABORTED;
+            return err == WSAECONNRESET || err == WSAECONNABORTED
+                   || err == ERROR_CONNECTION_ABORTED;
 #else
             return err == ECONNRESET;
 #endif
@@ -1048,6 +1050,11 @@ link_socket_read_udp_win32(struct link_socket *sock,
                            struct link_socket_actual *from)
 {
     sockethandle_t sh = { .s = sock->sd };
+    if (sock->info.dco_installed)
+    {
+        addr_copy_sa(&from->dest, &sock->info.lsa->actual.dest);
+        sh.is_handle = true;
+    }
     return sockethandle_finalize(sh, &sock->reads, buf, from);
 }
 
@@ -1057,7 +1064,7 @@ int link_socket_read_udp_posix(struct link_socket *sock,
                                struct buffer *buf,
                                struct link_socket_actual *from);
 
-#endif
+#endif /* ifdef _WIN32 */
 
 /* read a TCP or UDP packet from link */
 static inline int
@@ -1065,7 +1072,10 @@ link_socket_read(struct link_socket *sock,
                  struct buffer *buf,
                  struct link_socket_actual *from)
 {
-    if (proto_is_udp(sock->info.proto)) /* unified UDPv4 and UDPv6 */
+    if (proto_is_udp(sock->info.proto)
+        || sock->info.dco_installed)
+    /* unified UDPv4 and UDPv6, for DCO the kernel
+     * will strip the length header */
     {
         int res;
 
@@ -1106,19 +1116,19 @@ link_socket_write_win32(struct link_socket *sock,
 {
     int err = 0;
     int status = 0;
-    sockethandle_t sh = { .s = sock->sd };
+    sockethandle_t sh = { .s = sock->sd, .is_handle = sock->info.dco_installed };
     if (overlapped_io_active(&sock->writes))
     {
         status = sockethandle_finalize(sh, &sock->writes, NULL, NULL);
         if (status < 0)
         {
-            err = WSAGetLastError();
+            err = SocketHandleGetLastError(sh);
         }
     }
     socket_send_queue(sock, buf, to);
     if (status < 0)
     {
-        WSASetLastError(err);
+        SocketHandleSetLastError(sh, err);
         return status;
     }
     else
@@ -1180,8 +1190,9 @@ link_socket_write(struct link_socket *sock,
                   struct buffer *buf,
                   struct link_socket_actual *to)
 {
-    if (proto_is_udp(sock->info.proto)) /* unified UDPv4 and UDPv6 */
+    if (proto_is_udp(sock->info.proto) || sock->info.dco_installed)
     {
+        /* unified UDPv4 and UDPv6 and DCO (kernel adds size header) */
         return link_socket_write_udp(sock, buf, to);
     }
     else if (proto_is_tcp(sock->info.proto)) /* unified TCPv4 and TCPv6 */
src/openvpn/tun.c +34 -9

100 lines not on this page · raw message

src/openvpn/tun.h +42 -16

111 lines not on this page · raw message

Signature · 1 line
2.35.1
#152 lore raw

Re: [Openvpn-devel] [PATCH 23/25] dco-win: implement ovpn-dco support in P2P Windows code path

Hi,

I did some testing and found out that this breaks when establishing TCP connection to a server which cannot be reached:

src/openvpn/socket.c phase2_socks_client(struct link_socket *sock, struct signal_info *sig_info) · line 2146

Quoted 13 lines
src/openvpn/socket.c +13 -0
+    *tt = dco_create_socket(sock->info.lsa->current_remote,
+                            sock->bind_local,
+                            sock->info.lsa->bind_local,
+                            c->options.dev_node,
+                            &c->gc,
+                            get_server_poll_remaining_time(sock->server_poll_timeout),
+                            signal_received);
+    if (*signal_received)
+    {
+        return;
+    }
+    c->c1.tuntap = tt;
+    sock->info.dco_installed = true;

If TCP server cannot be reached, the kernel reports it to userspace which triggers a signal. Signal handler calls
teardown code which accesses c->c1.tuntap, which is not set (since return is above the assignment), causing access violation.

dco_installed flag should also be moved above return, since if not set, c->c1.tuntap won't be cleared
which triggers this assert after reconnect:

    /* In this case persist-tun is enabled, which we don't support yet */
    ASSERT(!c->c1.tuntap);

I suggest to change code like this:

    /* This state is used by signal handler which does teardown,
     * so it has to be set before return */
    c->c1.tuntap = tt;
    sock->info.dco_installed = true;

    if (*signal_received)
    {
        return;
    }

Again, we discussed this in chat and suggested changes are already pushed to the dco branch.

-Lev

#153 lore raw

Re: [Openvpn-devel] [PATCH 23/25] dco-win: implement ovpn-dco support in P2P Windows code path

In reply to Antonio Quartulli
Quoted 1 line · On Freitag, 24. Juni 2022 10:38:07 CEST Antonio Quartulli wrote:

+ if (!is_windco(c->c1.tuntap))

[...]

Quoted 2 lines

+ if ((options->windows_driver ==
WINDOWS_DRIVER_WINTUN || options->windows_driver == WINDOWS_DRIVER_WINDCO)

[...]

Quoted 1 line

+create_socket_windco(struct context *c, struct link_socket *sock,

Thinking "_dco_win" would be a better suffix for consistency.

src/openvpn/socket.c socket_recv_queue(struct link_socket *sock, int maxsize) · line 3578

Quoted 9 lines
src/openvpn/socket.c +8 -1
-            status = WSAGetLastError();
+            if (sock->info.dco_installed)
+            {
+                status = GetLastError();
+            }
+            else
+            {
+                status = WSAGetLastError();
+            }

Small wrapper would be good as this construct appears twice.

#154 lore raw

PATCH v2 23/23 dco-win: implement ovpn-dco support in P2P Windows code path

In reply to Antonio Quartulli

With this change it is possible to use ovpn-dco-win when running OpenVPN in client or P2P mode.

---

Changes from v1:
* use suffix _dco_win instead of _windco
* create helper function to retrieve last error from socket object

 src/openvpn/forward.c |  7 ++++
 src/openvpn/init.c    | 29 ++++++++++++--
 src/openvpn/options.c | 19 ++++++---
 src/openvpn/options.h | 15 +++----
 src/openvpn/socket.c  | 93 ++++++++++++++++++++++++++++++++++++++++---
 src/openvpn/socket.h  | 25 ++++++++----
 src/openvpn/tun.c     | 43 +++++++++++++++-----
 src/openvpn/tun.h     | 58 +++++++++++++++++++--------
 8 files changed, 232 insertions(+), 57 deletions(-)
src/openvpn/forward.c +7 -0
index 15bdbbae..1fbad43f 100644
@@ -860,6 +860,13 @@ read_incoming_link(struct context *c)
     /* check recvfrom status */
     check_status(status, "read", c->c2.link_socket, NULL);
 
+#ifdef _WIN32
+    if (dco_enabled(&c->options) && (status < 0) && (openvpn_errno() == ERROR_NETNAME_DELETED))
+    {
+        trigger_ping_timeout_signal(c);
+    }
+#endif
+
     /* Remove socks header if applicable */
     socks_postprocess_incoming_link(c);
 
src/openvpn/init.c +25 -4
index a6ee1fa8..b4276130 100644
@@ -1698,7 +1698,8 @@ do_init_tun(struct context *c)
                             c->c1.link_socket_addr.remote_list,
                             !c->options.ifconfig_nowarn,
                             c->c2.es,
-                            &c->net_ctx);
+                            &c->net_ctx,
+                            c->c1.tuntap);
 
 #ifdef _WIN32
     c->c1.tuntap->windows_driver = c->options.windows_driver;
@@ -1811,9 +1812,12 @@ do_open_tun(struct context *c)
             ovpn_dco_init(c->mode, &c->c1.tuntap->dco);
         }
 
-        /* open the tun device */
-        open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
-                 c->c1.tuntap, &c->net_ctx);
+        /* open the tun device (ovpn-dco-win already opened the device for the socket) */
+        if (!is_dco_win(c->c1.tuntap))
+        {
+            open_tun(c->options.dev, c->options.dev_type, c->options.dev_node,
+                     c->c1.tuntap, &c->net_ctx);
+        }
 
 #if defined(ENABLE_DCO)
         /* DCO may have been disabled by open_tun(). propagate change */
@@ -3574,6 +3578,23 @@ do_close_free_key_schedule(struct context *c, bool free_ssl_ctx)
 static void
 do_close_link_socket(struct context *c)
 {
+#if defined(_WIN32)
+    if (is_dco_win(c->c1.tuntap) && c->c2.link_socket
+        && c->c2.link_socket->info.dco_installed)
+    {
+        ASSERT(c->c2.link_socket_owned);
+
+        /* We rely on the tun_close to close the handle if also setup
+         * routes etc, since they cannot be delete when the interface
+         * handle has been closed */
+        if (!c->c1.tuntap->dco.real_tun_init)
+        {
+            do_close_tun_simple(c);
+        }
+        c->c2.link_socket->sd = SOCKET_UNDEFINED;
+    }
+#endif
+
     if (c->c2.link_socket && c->c2.link_socket_owned)
     {
         link_socket_close(c->c2.link_socket);
src/openvpn/options.c +14 -5
index 85bdd3f6..fa66b513 100644
@@ -3225,8 +3225,8 @@ options_postprocess_mutate_invariant(struct options *options)
 #ifdef _WIN32
     const int dev = dev_type_enum(options->dev, options->dev_type);
 
-    /* when using wintun, kernel doesn't send DHCP requests, so don't use it */
-    if (options->windows_driver == WINDOWS_DRIVER_WINTUN
+    /* when using wintun/ovpn-dco-win, kernel doesn't send DHCP requests, so don't use it */
+    if ((options->windows_driver == WINDOWS_DRIVER_WINTUN || options->windows_driver == WINDOWS_DRIVER_WINDCO)
         && (options->tuntap_options.ip_win32_type == IPW32_SET_DHCP_MASQ || options->tuntap_options.ip_win32_type == IPW32_SET_ADAPTIVE))
     {
         options->tuntap_options.ip_win32_type = IPW32_SET_NETSH;
@@ -3322,10 +3322,12 @@ options_postprocess_setdefault_ncpciphers(struct options *o)
         /* custom --data-ciphers set, keep list */
         return;
     }
+#if !defined(_WIN32)
     else if (cipher_valid("CHACHA20-POLY1305"))
     {
         o->ncp_ciphers = "AES-256-GCM:AES-128-GCM:CHACHA20-POLY1305";
     }
+#endif
     else
     {
         o->ncp_ciphers = "AES-256-GCM:AES-128-GCM";
@@ -4161,7 +4163,8 @@ options_string(const struct options *o,
                       NULL,
                       false,
                       NULL,
-                      ctx);
+                      ctx,
+                      NULL);
         if (tt)
         {
             tt_local = true;
@@ -4548,13 +4551,19 @@ parse_windows_driver(const char *str, const int msglevel)
     {
         return WINDOWS_DRIVER_WINTUN;
     }
+
+    else if (streq(str, "ovpn-dco-win"))
+    {
+        return WINDOWS_DRIVER_WINDCO;
+    }
     else
     {
-        msg(msglevel, "--windows-driver must be tap-windows6 or wintun");
+        msg(msglevel, "--windows-driver must be tap-windows6, wintun "
+            "or ovpn-dco-win");
         return WINDOWS_DRIVER_UNSPECIFIED;
     }
 }
-#endif
+#endif /* ifdef _WIN32 */
 
 /*
  * parse/print topology coding
src/openvpn/options.h +5 -10
index f7e3f265..1b30cedd 100644
@@ -879,24 +879,19 @@ void options_string_import(struct options *options,
 
 bool key_is_external(const struct options *options);
 
-#if defined(ENABLE_DCO) && defined(TARGET_LINUX)
-
 /**
  * Returns whether the current configuration has dco enabled.
  */
 static inline bool
 dco_enabled(const struct options *o)
 {
+#if defined(_WIN32)
+    return o->windows_driver == WINDOWS_DRIVER_WINDCO;
+#elif defined(ENABLE_DCO)
     return !o->tuntap_options.disable_dco;
-}
-
-#else /* if defined(ENABLE_DCO) && defined(TARGET_LINUX) */
-
-static inline bool
-dco_enabled(const struct options *o)
-{
+#else
     return false;
+#endif /* defined(_WIN32) */
 }
 
-#endif
 #endif /* ifndef OPTIONS_H */
src/openvpn/socket.c +87 -6

141 lines not on this page · raw message

src/openvpn/socket.h +18 -7
index 0d521d22..462afa31 100644
@@ -34,6 +34,7 @@
 #include "proxy.h"
 #include "socks.h"
 #include "misc.h"
+#include "tun.h"
 
 /*
  * OpenVPN's default port number as assigned by IANA.
@@ -937,7 +938,8 @@ socket_connection_reset(const struct link_socket *sock, int status)
         {
             const int err = openvpn_errno();
 #ifdef _WIN32
-            return err == WSAECONNRESET || err == WSAECONNABORTED;
+            return err == WSAECONNRESET || err == WSAECONNABORTED
+                   || err == ERROR_CONNECTION_ABORTED;
 #else
             return err == ECONNRESET;
 #endif
@@ -1048,6 +1050,11 @@ link_socket_read_udp_win32(struct link_socket *sock,
                            struct link_socket_actual *from)
 {
     sockethandle_t sh = { .s = sock->sd };
+    if (sock->info.dco_installed)
+    {
+        addr_copy_sa(&from->dest, &sock->info.lsa->actual.dest);
+        sh.is_handle = true;
+    }
     return sockethandle_finalize(sh, &sock->reads, buf, from);
 }
 
@@ -1057,7 +1064,7 @@ int link_socket_read_udp_posix(struct link_socket *sock,
                                struct buffer *buf,
                                struct link_socket_actual *from);
 
-#endif
+#endif /* ifdef _WIN32 */
 
 /* read a TCP or UDP packet from link */
 static inline int
@@ -1065,7 +1072,10 @@ link_socket_read(struct link_socket *sock,
                  struct buffer *buf,
                  struct link_socket_actual *from)
 {
-    if (proto_is_udp(sock->info.proto)) /* unified UDPv4 and UDPv6 */
+    if (proto_is_udp(sock->info.proto)
+        || sock->info.dco_installed)
+    /* unified UDPv4 and UDPv6, for DCO the kernel
+     * will strip the length header */
     {
         int res;
 
@@ -1106,19 +1116,19 @@ link_socket_write_win32(struct link_socket *sock,
 {
     int err = 0;
     int status = 0;
-    sockethandle_t sh = { .s = sock->sd };
+    sockethandle_t sh = { .s = sock->sd, .is_handle = sock->info.dco_installed };
     if (overlapped_io_active(&sock->writes))
     {
         status = sockethandle_finalize(sh, &sock->writes, NULL, NULL);
         if (status < 0)
         {
-            err = WSAGetLastError();
+            err = SocketHandleGetLastError(sh);
         }
     }
     socket_send_queue(sock, buf, to);
     if (status < 0)
     {
-        WSASetLastError(err);
+        SocketHandleSetLastError(sh, err);
         return status;
     }
     else
@@ -1180,8 +1190,9 @@ link_socket_write(struct link_socket *sock,
                   struct buffer *buf,
                   struct link_socket_actual *to)
 {
-    if (proto_is_udp(sock->info.proto)) /* unified UDPv4 and UDPv6 */
+    if (proto_is_udp(sock->info.proto) || sock->info.dco_installed)
     {
+        /* unified UDPv4 and UDPv6 and DCO (kernel adds size header) */
         return link_socket_write_udp(sock, buf, to);
     }
     else if (proto_is_tcp(sock->info.proto)) /* unified TCPv4 and TCPv6 */
src/openvpn/tun.c +34 -9

100 lines not on this page · raw message

src/openvpn/tun.h +42 -16

111 lines not on this page · raw message

Signature · 1 line
2.35.1
Acked-by on v2 23/23
#155 lore raw

Re: [Openvpn-devel] [PATCH v2 23/23] dco-win: implement ovpn-dco support in P2P Windows code path

Quoted 11 lines · On Montag, 11. Juli 2022 16:19:44 CEST Antonio Quartulli wrote:

With this change it is possible to use ovpn-dco-win when running OpenVPN in client or P2P mode.

---

Changes from v1:
* use suffix _dco_win instead of _windco
* create helper function to retrieve last error from socket object

#156 lore raw

PATCH 24/25 dco-win: add documentation to README.dco.md

In reply to Antonio Quartulli
---
 README.dco.md | 9 +++++++++
 1 file changed, 9 insertions(+)
README.dco.md +9 -0
index e73e0fc2..ef56f0fe 100644
@@ -58,6 +58,13 @@ see a message like
 in your log.
 
 
+Getting started (Windows)
+-------------------------
+Getting started under windows is currently for brave people having experience
+with windows development. You need to compile openvpn yourself and also need
+to get the test driver installed on your system.
+
+
 DCO and P2P mode
 ----------------
 DCO is also available when running OpenVPN in P2P mode without --pull/--client option.
@@ -109,6 +116,8 @@ Limitations by design
 - topology subnet is the only supported `--topology` for servers
 - iroute directives install routes on the host operating system, see also
   Routing with ovpn-dco
+- (ovpn-dco-win) client and p2p mode only
+- (ovpn-dco-win) Chacha20-Poly1305 support available starting with Windows 11
 
 
 Current implementation limitations
Signature · 1 line
2.35.1
#157 lore raw

Re: [Openvpn-devel] [PATCH 24/25] dco-win: add documentation to README.dco.md

Quoted 7 lines · On Freitag, 24. Juni 2022 10:38:08 CEST Antonio Quartulli wrote:
README.dco.md +3 -0
+Getting started (Windows)
+-------------------------
+Getting started under windows is currently for brave people having

experience
+with windows development. You need to compile openvpn yourself and also need
+to get the test driver installed on your system.

Not so brave anymore, maybe rephrase with more concrete steps to perform.

#158 lore raw

PATCH 25/25 dco-win: update GH Actions config file

In reply to Antonio Quartulli
---
 .github/workflows/build.yaml | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
.github/workflows/build.yaml +1 -6
index b905c0d2..536dd9d6 100644
@@ -59,11 +59,6 @@ jobs:
     steps:
       - name: Install dependencies
         run: sudo apt update && sudo apt install -y mingw-w64 libtool automake autoconf man2html unzip
-      - name: Checkout ovpn-dco-win
-        uses: actions/[email protected]...
-        with:
-          repository: OpenVPN/ovpn-dco-win
-          path: ovpn-dco-win
       - name: Checkout OpenVPN
         uses: actions/[email protected]...
         with:
@@ -151,7 +146,7 @@ jobs:
         run: cp ./tap-windows-${TAP_WINDOWS_VERSION}/include/tap-windows.h ${HOME}/mingw/opt/include/
 
       - name: configure OpenVPN
-        run: PKG_CONFIG_PATH=${HOME}/mingw/opt/lib/pkgconfig DCO_SOURCEDIR=$(realpath ../ovpn-dco-win) LDFLAGS=-L$HOME/mingw/opt/lib CFLAGS=-I$HOME/mingw/opt/include OPENSSL_LIBS="-L${HOME}/opt/lib -lssl -lcrypto" OPENSSL_CFLAGS=-I$HOME/mingw/opt/include PREFIX=$HOME/mingw/opt LZO_CFLAGS=-I$HOME/mingw/opt/include LZO_LIBS="-L${HOME}/mingw/opt/lib -llzo2" ./configure  --host=${CHOST} --disable-lz4 --enable-dco
+        run: PKG_CONFIG_PATH=${HOME}/mingw/opt/lib/pkgconfig LDFLAGS=-L$HOME/mingw/opt/lib CFLAGS=-I$HOME/mingw/opt/include OPENSSL_LIBS="-L${HOME}/opt/lib -lssl -lcrypto" OPENSSL_CFLAGS=-I$HOME/mingw/opt/include PREFIX=$HOME/mingw/opt LZO_CFLAGS=-I$HOME/mingw/opt/include LZO_LIBS="-L${HOME}/mingw/opt/lib -llzo2" ./configure  --host=${CHOST} --disable-lz4
         working-directory: openvpn
 
       - name: build OpenVPN
Signature · 1 line
2.35.1