dhcp: use included bind version
PATCH 1/1 dhcp: use included bind version
From: Vyacheslav Yurkov <[email protected]>
ISC DHCP should link against BIND9 libraries, which are tailored specifically for DHCP. BIND9 package in Yocto core layer has different configuraiton, in particular it has threads and epoll enabled.
ISC DHCP isn't a multi-threaded application, running it with bind9 libraries compiled in with threading enabled is not something ISC can vouch for.
BIND9 libraries support a lot of options specifically geared towards optimizing DNS operations, many of which do not play nicely with ISC DHCP's "architecture". It isn't necessarily practical to build those libraries for both purposes.
If threading is enabled, then DHCP sporadically fails/crashes with messages like:
lib/isc/unix/socket.c:1054: epoll_ctl(DEL), 6: Bad file descriptor lib/isc/unix/socket.c:3332: INSIST(!sock->pending_send) failed.
--- meta/recipes-connectivity/dhcp/dhcp.inc | 6 ++--- .../0014-Use-static-version-of-libbind.patch | 23 +++++++++++++++++++ meta/recipes-connectivity/dhcp/dhcp_4.4.2.bb | 1 + 3 files changed, 27 insertions(+), 3 deletions(-) create mode 100644 meta/recipes-connectivity/dhcp/dhcp/0014-Use-static-version-of-libbind.patch
meta/recipes-connectivity/dhcp/dhcp.inc +3 -3
@@ -10,7 +10,7 @@ HOMEPAGE = "http://www.isc.org/" LICENSE = "ISC" LIC_FILES_CHKSUM = "file://LICENSE;beginline=4;md5=004a4db50a1e20972e924a8618747c01" -DEPENDS = "openssl bind" +DEPENDS = "openssl zlib" SRC_URI = "http://ftp.isc.org/isc/dhcp/${PV}/dhcp-${PV}.tar.gz \ file://init-relay file://default-relay \
@@ -50,9 +50,9 @@ EXTRA_OECONF = "--with-srv-lease-file=${localstatedir}/lib/dhcp/dhcpd.leases \ --with-cli6-lease-file=${localstatedir}/lib/dhcp/dhclient6.leases \ --enable-paranoia --disable-static \ --with-randomdev=/dev/random \ - --with-libbind=${STAGING_DIR_HOST} \ - --enable-libtool \ + --enable-libtool \ " +PARALLEL_MAKE = "" #Enable shared libs per dhcp README do_configure_prepend () {
meta/recipes-connectivity/dhcp/dhcp/0014-Use-static-version-of-libbind.patch new file +23 -0
@@ -0,0 +1,23 @@ +From a96f65117d63bea4287f19e80595f7aab282410a Mon Sep 17 00:00:00 2001 +From: Vyacheslav Yurkov <[email protected]> +Date: Wed, 10 Jun 2020 07:56:20 +0200 +Subject: [PATCH] Use static version of libbind + +Signed-off-by: Vyacheslav Yurkov <[email protected]> +--- + bind/Makefile.in | 2 +- + 1 file changed, 1 insertion(+), 1 deletion(-) + +diff --git a/bind/Makefile.in b/bind/Makefile.in +index 8fe8883..902fc3d 100644 +--- a/bind/Makefile.in ++++ b/bind/Makefile.in +@@ -22,7 +22,7 @@ prefix = @prefix@ + exec_prefix = @exec_prefix@ + + bindconfig = --without-openssl --without-libxml2 --without-libjson \ +- --without-gssapi --disable-threads --without-lmdb \ ++ --without-gssapi --disable-threads --without-lmdb --disable-shared \ + --includedir=@includedir@ --libdir=@libdir@ --without-python\ + @BINDLT@ @BINDIOMUX@ @BINDCONFIG@ --enable-full-report +
meta/recipes-connectivity/dhcp/dhcp_4.4.2.bb +1 -0
@@ -10,6 +10,7 @@ SRC_URI += "file://0001-define-macro-_PATH_DHCPD_CONF-and-_PATH_DHCLIENT_CON.pat file://0012-dhcp-correct-the-intention-for-xml2-lib-search.patch \ file://0013-fixup_use_libbind.patch \ file://0001-workaround-busybox-limitation-in-linux-dhclient-script.patch \ + file://0014-Use-static-version-of-libbind.patch \ " SRC_URI[md5sum] = "2afdaf8498dc1edaf3012efdd589b3e1"
Signature · 1 line
2.25.1
✗ patchtest: failure for dhcp: use included bind version
== Series Details ==
Series: dhcp: use included bind version
Revision: 1
URL : https://patchwork.openembedded.org/series/24576/
State : failure
== Summary ==
Thank you for submitting this patch series to OpenEmbedded Core. This is an automated response. Several tests have been executed on the proposed series by patchtest resulting in the following failures:
* Issue Added patch file is missing Upstream-Status in the header [test_upstream_status_presence_format]
Suggested fix Add Upstream-Status: <Valid status> to the header of meta/recipes-connectivity/dhcp/dhcp/0014-Use-static-version-of-libbind.patch Standard format Upstream-Status: <Valid status> Valid status Pending, Accepted, Backport, Denied, Inappropriate [reason], Submitted [where]
If you believe any of these test results are incorrect, please reply to the mailing list ([email protected]) raising your concerns. Otherwise we would appreciate you correcting the issues and submitting a new version of the patchset if applicable. Please ensure you add/increment the version number when sending the new version (i.e. [PATCH] -> [PATCH v2] -> [PATCH v3] -> ...).
---
Guidelines: https://www.openembedded.org/wiki/Commit_Patch_Message_Guidelines Test framework: http://git.yoctoproject.org/cgit/cgit.cgi/patchtest Test suite: http://git.yoctoproject.org/cgit/cgit.cgi/patchtest-oe
Quoted 16 lines · Vyacheslav Yurkov
From: Vyacheslav Yurkov <[email protected]>
ISC DHCP should link against BIND9 libraries, which are tailored specifically for DHCP. BIND9 package in Yocto core layer has different
configuraiton, in particular it has threads and epoll enabled.ISC DHCP isn't a multi-threaded application, running it with bind9 libraries compiled in with threading enabled is not something ISC can vouch for.
BIND9 libraries support a lot of options specifically geared towards optimizing DNS operations, many of which do not play nicely with ISC DHCP's "architecture". It isn't necessarily practical to build those libraries for both purposes.
Quoted 5 lines
If threading is enabled, then DHCP sporadically fails/crashes with messages like:
lib/isc/unix/socket.c:1054:epoll_ctl(DEL), 6: Bad file descriptorlib/isc/unix/socket.c:3332:INSIST(!sock->pending_send)failed.
You seem to disable parallel make yet no mention of that in the log?
Why doesn't dhcp build and link against a static libbind by default if these issues are present the the dhcp authors don't want to work with such libbinds?
Cheers,
Richard
Quoted 1 line · Richard Purdie
You seem to disable parallel make yet no mention of that in the log?
Right, I forgot to mention that in the commit message. Building bind at 'compile' stage breaks parallel make. I assume that can be addressed, but I'd need to take a look at how to set that dependency with autotools.
Quoted 3 lines
Why doesn't dhcp build and link against a static libbind by default if these issues are present the the dhcp authors don't want to work with such libbinds?
That's a good question. I assume the alternative is disable threads in bind recipes then a shared library can be used by both, dhcp and named (DNS). Hopefully Thomas from ISC (in CC) could clarify this better than me.
Quoted 5 lines
Cheers,
Richard
Regards,
Vyacheslav
Quoted 7 lines · PDT Vyacheslav Yurkov
Quoted 1 line · Richard Purdie
You seem to disable parallel make yet no mention of that in the log?
Right, I forgot to mention that in the commit message. Building bind at 'compile' stage breaks parallel make. I assume that can be addressed, but I'd need to take a look at how to set that dependency with autotools.
Quoted 8 lines
Quoted 3 lines
Why doesn't dhcp build and link against a static libbind by default if these issues are present the the dhcp authors don't want to work with such libbinds?
That's a good question. I assume the alternative is disable threads in bind recipes then a shared library can be used by both, dhcp and named (DNS). Hopefully Thomas from ISC (in CC) could clarify this better than me.
I think it would be good to share if we can. Shipping two versions of bind libraries is less than ideal, have you explored disabling threading in the bind recipe and see if named is happy with that ?
Quoted 6 lines
Quoted 3 lines
Cheers,
Richard
Regards,
Vyacheslav
Quoted 4 lines · Khem Raj
I think it would be good to share if we can. Shipping two versions of bind libraries is less than ideal, have you explored disabling threading in the bind recipe and see if named is happy with that ?
Yes, I have. It partially solves the issue. At least DHCP doesn't crash/fail anymore, but warnings from epoll enabled are still seen in the logs. The other question is do we have to strip down BIND functionality in order to make DHCP work?
I understand that maintenance effort is at stake and we are looking for the best possible solution here, but IMHO having one set of bind libraries for both BIND and DHCP servers is not one of them.
We could also go Ubuntu way here, by having two bind-9.11.x for DHCP and upgrading BIND server to the latest stable (9.16.x if I'm not mistaken). Then security support could be only done for 9.11.x version, the latest stable should have most of security patches included already.
Vyacheslav
This brings me to a question. Should we replace dhcp with kea?
Alex
Quoted 24 lines · Vyacheslav Yurkov
Quoted 1 line · Khem Raj
I think it would be good to share if we can. Shipping two versions of
bind
Quoted 1 line
libraries is less than ideal, have you explored disabling threading in
the
Quoted 2 lines
bind recipe and see if named is happy with that ?
Yes, I have. It partially solves the issue. At least DHCP doesn't crash/fail anymore, but warnings from epoll enabled are still seen in the logs. The other question is do we have to strip down BIND functionality in order to make DHCP work?
I understand that maintenance effort is at stake and we are looking for the best possible solution here, but IMHO having one set of bind libraries for both BIND and DHCP servers is not one of them.
We could also go Ubuntu way here, by having two bind-9.11.x for DHCP and upgrading BIND server to the latest stable (9.16.x if I'm not mistaken). Then security support could be only done for 9.11.x version, the latest stable should have most of security patches included already.
Vyacheslav
Quoted 6 lines · Vyacheslav Yurkov
...
BIND9 libraries support a lot of options specifically geared towards optimizing DNS operations, many of which do not play nicely with ISC DHCP's "architecture". It isn't necessarily practical to build those libraries for both purposes.
...
Providing security support for several different copies of the BIND code would be a nightmare.
cu
Adrian
Quoted 7 lines · Adrian Bunk
Quoted 6 lines · Vyacheslav Yurkov
...
BIND9 libraries support a lot of options specifically geared towards optimizing DNS operations, many of which do not play nicely with ISC DHCP's "architecture". It isn't necessarily practical to build those libraries for both purposes.
...
Quoted 5 lines
Providing security support for several different copies of the BIND code would be a nightmare.
cu
Adrian
Right, but I don't think there's a way to build two configurations out of one recipe, isn't it?
Vyacheslav
Quoted 8 lines · Vyacheslav Yurkov
Quoted 7 lines · Adrian Bunk
Quoted 6 lines · Vyacheslav Yurkov
...
BIND9 libraries support a lot of options specifically geared towards optimizing DNS operations, many of which do not play nicely with ISC DHCP's "architecture". It isn't necessarily practical to build those libraries for both purposes.
...
Quoted 8 lines
Quoted 5 lines
Providing security support for several different copies of the BIND code would be a nightmare.
cu
AdrianRight, but I don't think there's a way to build two configurations out of one recipe, isn't it?
It is possible.
More common is sharing the sources between two (or more) recipes with an .inc file.
Unrelated to that, I am a bit suprised why the current Yocto setup is suddenly a problem. Other distributions like Debian/Ubuntu are also sharing one build of the BIND libraries between BIND and dhcpd.
Quoted 1 line
Vyacheslav
cu
Adrian
Quoted 5 lines · Adrian Bunk
Quoted 2 lines
Right, but I don't think there's a way to build two configurations out of one recipe, isn't it?
It is possible.
More common is sharing the sources between two (or more) recipes with an .inc file.
That's not really what I had in mind. That would result in two set of recipes anyway, perhaps with the same SRC_URI.
Quoted 5 lines
Unrelated to that, I am a bit suprised why the current Yocto setup is suddenly a problem. Other distributions like Debian/Ubuntu are also sharing one build of the BIND libraries between BIND and dhcpd.
The thing is, they don't. This is the 'depends' output from my Ubuntu 20.04 installation:
% sudo apt-cache depends isc-dhcp-server
...
Depends: libdns-export1109
Depends: libirs-export161
Depends: libisc-export1105
% sudo apt-cache depends bind9
bind9
...
Depends: bind9-libs
bind9-libs is bind-9.16.1, but export libraries are bind-9.11.16.
So answering Richard's question, having shared linking is also possible, but that would result in two sets of bind libraries.
Quoted 2 lines
cu
Adrian
Regards,
Vyacheslav