iplink: emit the error message if open_fds_add failed
Subsystems (2)
PATCH 1/1 iplink: emit the error message if open_fds_add failed
open_fds_add may fail since it adds an open fd
to the fixed array. Print the error message in
the such case.
--- ip/iplink.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
ip/iplink.c +3 -1
@@ -666,7 +666,9 @@ int iplink_parse(int argc, char **argv, struct iplink_req *req, char **type)if (netns <0 )invarg ("Invalid \"netns\" value\n" , *argv); -open_fds_add (netns); +if (open_fds_add (netns)) +invarg ("No descriptors left\n" , *argv); +addattr_l (&req->n,sizeof (*req),IFLA_NET_NS_FD , &netns,4 ); move_netns = true;
Signature · 1 line
2.43.0
Quoted 14 lines · On Tue, 21 Jan 2025 17:16:42 +0300 Denis Kirjanov <[email protected]> wrote:
open_fds_addmay fail since it adds an open fd
to the fixed array. Print the error message in
the such case.Signed-off-by Denis Kirjanov <[email protected]>--- ip/iplink.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)ip/iplink.c +0
-0
ip/iplink.c int iplink_parse(int argc, char **argv, struct iplink_req *req, char **type) · line 674
Quoted 10 lines
ip/iplink.c +3
-1@@ -666,7 +666,9 @@ int iplink_parse(int argc, char **argv, struct iplink_req *req, char **type)if (netns <0 )invarg ("Invalid \"netns\" value\n" , *argv); -open_fds_add (netns); +if (open_fds_add (netns)) +invarg ("No descriptors left\n" , *argv); +addattr_l (&req->n,sizeof (*req),IFLA_NET_NS_FD , &netns,4 ); move_netns = true;
Not an acceptable fix.
It is not an invalid argument, that would just be confusing. How to reproduce this?
The original patch that added this made arbitrary choice of 5 fd's. If this can be reproduced, that should be increased or switch to a dynamic list.
commit 57daf8ff8c6c357a5a083657e5b03d2883cbc4f9
Author: Nicolas Dichtel <[email protected]> Date: Wed Sep 18 18:49:41 2024 +0200
iplink: fix fd leak when playing with netns
The command 'ip link set foo netns mynetns' opens a file descriptor to fill the netlink attribute IFLA_NET_NS_FD. This file descriptor is never closed. When batch mode is used, the number of file descriptor may grow greatly and reach the maximum file descriptor number that can be opened.
This fd can be closed only after the netlink answer. Moreover, a second fd could be opened because some (struct link_util)->parse_opt() handlers call iplink_parse().
Let's add a helper to manage these fds:
- open_fds_add() stores a fd, up to 5 (arbitrary choice, it seems enough);
- open_fds_close() closes all stored fds.