BUG/MAJOR: server: prevent deadlock when using 'set maxconn server'
A deadlock is possible with 'set maxconn server' command, if there is
pending connection ready to be dequeued. This is caused by the locking
of server spinlock in both cli_parse_set_maxconn_server and
process_srv_queue.
Fix this by reducing the scope of the server lock into
server_parse_maxconn_change_request. If connection are dequeued, the
lock is taken a second time. This can be seen as suboptimal but as it
happens only during 'set maxconn server' it can be considered as
tolerable.
This issue was reported on the mailing list, for the 1.8.x branch.
It must be backported up to the 1.8.
(cherry picked from commit 79a88ba3d09f7e2b73ae27cb5d24cc087a548fa6)
Signed-off-by: Christopher Faulet <cfaulet@haproxy.com>
(cherry picked from commit 3cf814498e8196a9bc5dade38a5de77bd4ee6431)
Signed-off-by: Christopher Faulet <cfaulet@haproxy.com>
(cherry picked from commit ac458459c2db807bb5ff5b51c2e40a6633bf2101)
Signed-off-by: Christopher Faulet <cfaulet@haproxy.com>
(cherry picked from commit 2279c0a0364fde1c901004334df558e409eff848)
Signed-off-by: Christopher Faulet <cfaulet@haproxy.com>
1 file changed