Do not listen if rtnl_send() fails in ip link iplink_have_newlink() test
Subsystems (2)
PATCH 1/1 Do not listen if rtnl_send() fails in ip link iplink_have_newlink() test
If rtnl_send() fails in iplink_have_newlink() test, listening for response will result in indefinite hang. This can be demonstrated by "ip link show" while SELinux preventing from sending the RTM_NEWLINK over netlink.
This patch checks for the return value as is done at all other rtnl_send() calls.
--- ip/iplink.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
ip/iplink.c +2 -2
@@ -178,8 +178,8 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ; req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); -rtnl_listen (&rth, accept_msg,NULL ); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) >=0 ) +rtnl_listen (&rth, accept_msg,NULL ); }return have_rtnl_newlink; }
Signature · 1 line
1.8.3.1
Quoted 18 lines · On Wed, 11 Dec 2013 14:01:29 +0100 Petr Písař <[email protected]> wrote:
If
rtnl_send()fails iniplink_have_newlink()test, listening for response will result in indefinite hang. This can be demonstrated by "ip link show" while SELinux preventing from sending theRTM_NEWLINKover netlink.This patch checks for the return value as is done at all other
rtnl_send()calls.Signed-off-by Petr Písař <[email protected]>--- ip/iplink.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)ip/iplink.c +0
-0
ip/iplink.c static int iplink_have_newlink(void) · line 185
Quoted 10 lines
ip/iplink.c +2
-2@@ -178,8 +178,8 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ; req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); -rtnl_listen (&rth, accept_msg,NULL ); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) >=0 ) +rtnl_listen (&rth, accept_msg,NULL ); }return have_rtnl_newlink; }
I think it should print an error messag, not silently ignore the send failure.
PATCH 1/1 Do not listen if rtnl_send() fails in ip link iplink_have_newlink() test
If rtnl_send() fails in iplink_have_newlink() test, listening for response will result in indefinite hang. This can be demonstrated by "ip link show" while SELinux preventing from sending the RTM_NEWLINK over netlink.
This patch checks for the return value as is done at all other rtnl_send() calls. It falls back to IOCTL in case of failure.
--- ip/iplink.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
ip/iplink.c +7 -2
@@ -178,8 +178,13 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ; req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); -rtnl_listen (&rth, accept_msg,NULL ); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) <0 ) { +perror ("Could not check for " +"link configuration over netlink support" ); + have_rtnl_newlink =0 ; + }else { +rtnl_listen (&rth, accept_msg,NULL ); + } }return have_rtnl_newlink; }
Signature · 1 line
1.8.3.1
I took your idea and enhanced it to all of iproute2 by doing the following:
From c4b6330a3a033bd9c9b0664c5f844493137ae599 Mon Sep 17 00:00:00 2001
From: Stephen Hemminger <[email protected]>
Date: Fri, 20 Dec 2013 08:15:02 -0800
Subject: [PATCH] check return value of rtnl_send and related functions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Use warn_unused_result to enforce checking return value of rtnl_send, and fix where the errors are.
Suggested by initial patch from Petr Písař <[email protected]>
--- include/libnetlink.h | 28 ++++++++++++++++++++-------- ip/iplink.c | 5 ++++- ip/ipnetconf.c | 5 ++++- misc/arpd.c | 6 +++++- misc/ss.c | 4 +++- 5 files changed, 36 insertions(+), 12 deletions(-)
include/libnetlink.h +20 -8
@@ -22,13 +22,22 @@ struct rtnl_handleextern int rcvbuf; -extern int rtnl_open (struct rtnl_handle *rth,unsigned subscriptions); -extern int rtnl_open_byproto (struct rtnl_handle *rth,unsigned subscriptions,int protocol); +extern int rtnl_open (struct rtnl_handle *rth,unsigned subscriptions) + __attribute__((warn_unused_result)); + +extern int rtnl_open_byproto (struct rtnl_handle *rth,unsigned subscriptions, +int protocol) + __attribute__((warn_unused_result)); +extern void rtnl_close (struct rtnl_handle *rth); -extern int rtnl_wilddump_request (struct rtnl_handle *rth,int fam,int type); +extern int rtnl_wilddump_request (struct rtnl_handle *rth,int fam,int type) + __attribute__((warn_unused_result));extern int rtnl_wilddump_req_filter (struct rtnl_handle *rth,int fam,int type, -__u32 filt_mask); -extern int rtnl_dump_request (struct rtnl_handle *rth,int type,void *req,int len); +__u32 filt_mask) + __attribute__((warn_unused_result)); +extern int rtnl_dump_request (struct rtnl_handle *rth,int type,void *req, +int len) + __attribute__((warn_unused_result));typedef int (*rtnl_filter_t )(const struct sockaddr_nl *,struct nlmsghdr *n,void *);
@@ -44,9 +53,12 @@ extern int rtnl_dump_filter_l(struct rtnl_handle *rth,extern int rtnl_dump_filter (struct rtnl_handle *rth,rtnl_filter_t filter,void *arg);extern int rtnl_talk (struct rtnl_handle *rtnl,struct nlmsghdr *n,pid_t peer, -unsigned groups,struct nlmsghdr *answer); -extern int rtnl_send (struct rtnl_handle *rth,const void *buf,int ); -extern int rtnl_send_check (struct rtnl_handle *rth,const void *buf,int ); +unsigned groups,struct nlmsghdr *answer) + __attribute__((warn_unused_result)); +extern int rtnl_send (struct rtnl_handle *rth,const void *buf,int ) + __attribute__((warn_unused_result)); +extern int rtnl_send_check (struct rtnl_handle *rth,const void *buf,int ) + __attribute__((warn_unused_result));extern int addattr (struct nlmsghdr *n,int maxlen,int type);extern int addattr8 (struct nlmsghdr *n,int maxlen,int type,__u8 data);
ip/iplink.c +4 -1
@@ -178,7 +178,10 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ; req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) <0 ) { +perror ("request send failed" ); +exit (1 ); + }rtnl_listen (&rth, accept_msg,NULL ); }return have_rtnl_newlink;
ip/ipnetconf.c +4 -1
@@ -161,7 +161,10 @@ static int do_show(int argc, char **argv)addattr_l (&req.n,sizeof (req),NETCONFA_IFINDEX , &filter.ifindex,sizeof (filter.ifindex)); -rtnl_send (&rth, &req.n, req.n.nlmsg_len); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) <0 ) { +perror ("Can not send request" ); +exit (1 ); + }rtnl_listen (&rth, print_netconf, stdout); }else { dump:
misc/arpd.c +5 -1
@@ -428,7 +428,11 @@ static int do_one_request(struct nlmsghdr *n)static void load_initial_table (void ) { -rtnl_wilddump_request (&rth,AF_INET ,RTM_GETNEIGH ); +if (rtnl_wilddump_request (&rth,AF_INET ,RTM_GETNEIGH ) <0 ) { +perrror ("dump request failed" ); +exit (1 ); + } + }static void get_kern_msg (void )
misc/ss.c +3 -1
@@ -996,7 +996,9 @@ static int xll_initted = 0;static void xll_init (void ) {struct rtnl_handle rth; -rtnl_open (&rth,0 ); +if (rtnl_open (&rth,0 ) <0 ) +exit (1 ); +ll_init_map (&rth);rtnl_close (&rth); xll_initted =1 ;
Signature · 1 line
1.7.10.4
Quoted 13 lines · Stephen Hemminger
I took your idea and enhanced it to all of iproute2 by doing the following:
From
c4b6330a3a033bd9c9b0664c5f844493137ae599Mon Sep 17 00:00:00 2001
From: Stephen Hemminger <[email protected]>
Date: Fri, 20 Dec 2013 08:15:02 -0800
Subject: [PATCH] check return value ofrtnl_sendand related functions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bitUse
warn_unused_resultto enforce checking return value ofrtnl_send, and fix where the errors are.
[...]
Quoted 6 lines
ip/iplink.c +0
-0@@ -178,7 +178,10 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ;
ip/iplink.c static int iplink_have_newlink(void) · line 187
Quoted 10 lines
ip/iplink.c +4
-1req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) <0 ) { +perror ("request send failed" ); +exit (1 ); + }rtnl_listen (&rth, accept_msg,NULL ); }return have_rtnl_newlink;
This one exits instead of falling back to IOCTL. iplink_have_newlink() is called even from iplink_usage(). This even prevents from printing usage.
I think failure in this very place should be treated as I suggested.
-- Petr
Quoted 35 lines · On Thu, 2 Jan 2014 08:42:57 +0100 Petr Pisar <[email protected]> wrote:
Quoted 13 lines · Stephen Hemminger
I took your idea and enhanced it to all of iproute2 by doing the following:
From
c4b6330a3a033bd9c9b0664c5f844493137ae599Mon Sep 17 00:00:00 2001
From: Stephen Hemminger <[email protected]>
Date: Fri, 20 Dec 2013 08:15:02 -0800
Subject: [PATCH] check return value ofrtnl_sendand related functions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bitUse
warn_unused_resultto enforce checking return value ofrtnl_send, and fix where the errors are.[...]
Quoted 6 lines
ip/iplink.c +0
-0@@ -178,7 +178,10 @@ static int iplink_have_newlink(void) req.n.nlmsg_type =RTM_NEWLINK ;ip/iplink.c static int iplink_have_newlink(void) · line 187
Quoted 10 lines
ip/iplink.c +4
-1req.i.ifi_family =AF_UNSPEC ; -rtnl_send (&rth, &req.n, req.n.nlmsg_len); +if (rtnl_send (&rth, &req.n, req.n.nlmsg_len) <0 ) { +perror ("request send failed" ); +exit (1 ); + }rtnl_listen (&rth, accept_msg,NULL ); }return have_rtnl_newlink;This one exits instead of falling back to IOCTL.
iplink_have_newlink()is called even fromiplink_usage(). This even prevents from printing usage.
Quoted 3 lines
I think failure in this very place should be treated as I suggested.
-- Petr
The code gets used in multiple paths, some should fail and others have fallback. I will fix the cases where fallback is possible.