]> git.karo-electronics.de Git - linux-beck.git/commit
inet: fix possible panic in reqsk_queue_unlink()
authorEric Dumazet <edumazet@google.com>
Fri, 24 Apr 2015 01:03:44 +0000 (18:03 -0700)
committerDavid S. Miller <davem@davemloft.net>
Fri, 24 Apr 2015 15:39:15 +0000 (11:39 -0400)
commitb357a364c57c940ddb932224542494363df37378
tree84657c956c5c64a2ae058b8285271f778b1aea67
parent1d8dc3d3c8f1d8ee1da9d54c5d7c8694419ade42
inet: fix possible panic in reqsk_queue_unlink()

[ 3897.923145] BUG: unable to handle kernel NULL pointer dereference at
 0000000000000080
[ 3897.931025] IP: [<ffffffffa9f27686>] reqsk_timer_handler+0x1a6/0x243

There is a race when reqsk_timer_handler() and tcp_check_req() call
inet_csk_reqsk_queue_unlink() on the same req at the same time.

Before commit fa76ce7328b2 ("inet: get rid of central tcp/dccp listener
timer"), listener spinlock was held and race could not happen.

To solve this bug, we change reqsk_queue_unlink() to not assume req
must be found, and we return a status, to conditionally release a
refcount on the request sock.

This also means tcp_check_req() in non fastopen case might or not
consume req refcount, so tcp_v6_hnd_req() & tcp_v4_hnd_req() have
to properly handle this.

(Same remark for dccp_check_req() and its callers)

inet_csk_reqsk_queue_drop() is now too big to be inlined, as it is
called 4 times in tcp and 3 times in dccp.

Fixes: fa76ce7328b2 ("inet: get rid of central tcp/dccp listener timer")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/inet_connection_sock.h
include/net/request_sock.h
net/dccp/ipv4.c
net/dccp/ipv6.c
net/dccp/minisocks.c
net/ipv4/inet_connection_sock.c
net/ipv4/tcp_ipv4.c
net/ipv4/tcp_minisocks.c
net/ipv6/tcp_ipv6.c