ipvs: fix rtnl_lock lockups caused by start_sync_thread
authorJulian Anastasov <ja@ssi.bg>
Sat, 7 Apr 2018 12:50:47 +0000 (15:50 +0300)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 16 May 2018 08:10:22 +0000 (10:10 +0200)
commit88c43b469db5d4c39fba20c293b44cbe78b66e3a
tree6d4d3a4a8f4fa74ca80c01109d3aa0d6c46def8b
parentfc72a4171174dd6b7ddefe5eeaa12cec9a162704
ipvs: fix rtnl_lock lockups caused by start_sync_thread

commit 5c64576a77894a50be80be0024bed27171b55989 upstream.

syzkaller reports for wrong rtnl_lock usage in sync code [1] and [2]

We have 2 problems in start_sync_thread if error path is
taken, eg. on memory allocation error or failure to configure
sockets for mcast group or addr/port binding:

1. recursive locking: holding rtnl_lock while calling sock_release
which in turn calls again rtnl_lock in ip_mc_drop_socket to leave
the mcast group, as noticed by Florian Westphal. Additionally,
sock_release can not be called while holding sync_mutex (ABBA
deadlock).

2. task hung: holding rtnl_lock while calling kthread_stop to
stop the running kthreads. As the kthreads do the same to leave
the mcast group (sock_release -> ip_mc_drop_socket -> rtnl_lock)
they hang.

Fix the problems by calling rtnl_unlock early in the error path,
now sock_release is called after unlocking both mutexes.

Problem 3 (task hung reported by syzkaller [2]) is variant of
problem 2: use _trylock to prevent one user to call rtnl_lock and
then while waiting for sync_mutex to block kthreads that execute
sock_release when they are stopped by stop_sync_thread.

[1]
IPVS: stopping backup sync thread 4500 ...
WARNING: possible recursive locking detected
4.16.0-rc7+ #3 Not tainted
--------------------------------------------
syzkaller688027/4497 is trying to acquire lock:
  (rtnl_mutex){+.+.}, at: [<00000000bb14d7fb>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74

but task is already holding lock:
IPVS: stopping backup sync thread 4495 ...
  (rtnl_mutex){+.+.}, at: [<00000000bb14d7fb>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74

other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(rtnl_mutex);
   lock(rtnl_mutex);

  *** DEADLOCK ***

  May be due to missing lock nesting notation

2 locks held by syzkaller688027/4497:
  #0:  (rtnl_mutex){+.+.}, at: [<00000000bb14d7fb>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74
  #1:  (ipvs->sync_mutex){+.+.}, at: [<00000000703f78e3>]
do_ip_vs_set_ctl+0x10f8/0x1cc0 net/netfilter/ipvs/ip_vs_ctl.c:2388

stack backtrace:
CPU: 1 PID: 4497 Comm: syzkaller688027 Not tainted 4.16.0-rc7+ #3
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
Google 01/01/2011
Call Trace:
  __dump_stack lib/dump_stack.c:17 [inline]
  dump_stack+0x194/0x24d lib/dump_stack.c:53
  print_deadlock_bug kernel/locking/lockdep.c:1761 [inline]
  check_deadlock kernel/locking/lockdep.c:1805 [inline]
  validate_chain kernel/locking/lockdep.c:2401 [inline]
  __lock_acquire+0xe8f/0x3e00 kernel/locking/lockdep.c:3431
  lock_acquire+0x1d5/0x580 kernel/locking/lockdep.c:3920
  __mutex_lock_common kernel/locking/mutex.c:756 [inline]
  __mutex_lock+0x16f/0x1a80 kernel/locking/mutex.c:893
  mutex_lock_nested+0x16/0x20 kernel/locking/mutex.c:908
  rtnl_lock+0x17/0x20 net/core/rtnetlink.c:74
  ip_mc_drop_socket+0x88/0x230 net/ipv4/igmp.c:2643
  inet_release+0x4e/0x1c0 net/ipv4/af_inet.c:413
  sock_release+0x8d/0x1e0 net/socket.c:595
  start_sync_thread+0x2213/0x2b70 net/netfilter/ipvs/ip_vs_sync.c:1924
  do_ip_vs_set_ctl+0x1139/0x1cc0 net/netfilter/ipvs/ip_vs_ctl.c:2389
  nf_sockopt net/netfilter/nf_sockopt.c:106 [inline]
  nf_setsockopt+0x67/0xc0 net/netfilter/nf_sockopt.c:115
  ip_setsockopt+0x97/0xa0 net/ipv4/ip_sockglue.c:1261
  udp_setsockopt+0x45/0x80 net/ipv4/udp.c:2406
  sock_common_setsockopt+0x95/0xd0 net/core/sock.c:2975
  SYSC_setsockopt net/socket.c:1849 [inline]
  SyS_setsockopt+0x189/0x360 net/socket.c:1828
  do_syscall_64+0x281/0x940 arch/x86/entry/common.c:287
  entry_SYSCALL_64_after_hwframe+0x42/0xb7
RIP: 0033:0x446a69
RSP: 002b:00007fa1c3a64da8 EFLAGS: 00000246 ORIG_RAX: 0000000000000036
RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000446a69
RDX: 000000000000048b RSI: 0000000000000000 RDI: 0000000000000003
RBP: 00000000006e29fc R08: 0000000000000018 R09: 0000000000000000
R10: 00000000200000c0 R11: 0000000000000246 R12: 00000000006e29f8
R13: 00676e697279656b R14: 00007fa1c3a659c0 R15: 00000000006e2b60

[2]
IPVS: sync thread started: state = BACKUP, mcast_ifn = syz_tun, syncid = 4,
id = 0
IPVS: stopping backup sync thread 25415 ...
INFO: task syz-executor7:25421 blocked for more than 120 seconds.
       Not tainted 4.16.0-rc6+ #284
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
syz-executor7   D23688 25421   4408 0x00000004
Call Trace:
  context_switch kernel/sched/core.c:2862 [inline]
  __schedule+0x8fb/0x1ec0 kernel/sched/core.c:3440
  schedule+0xf5/0x430 kernel/sched/core.c:3499
  schedule_timeout+0x1a3/0x230 kernel/time/timer.c:1777
  do_wait_for_common kernel/sched/completion.c:86 [inline]
  __wait_for_common kernel/sched/completion.c:107 [inline]
  wait_for_common kernel/sched/completion.c:118 [inline]
  wait_for_completion+0x415/0x770 kernel/sched/completion.c:139
  kthread_stop+0x14a/0x7a0 kernel/kthread.c:530
  stop_sync_thread+0x3d9/0x740 net/netfilter/ipvs/ip_vs_sync.c:1996
  do_ip_vs_set_ctl+0x2b1/0x1cc0 net/netfilter/ipvs/ip_vs_ctl.c:2394
  nf_sockopt net/netfilter/nf_sockopt.c:106 [inline]
  nf_setsockopt+0x67/0xc0 net/netfilter/nf_sockopt.c:115
  ip_setsockopt+0x97/0xa0 net/ipv4/ip_sockglue.c:1253
  sctp_setsockopt+0x2ca/0x63e0 net/sctp/socket.c:4154
  sock_common_setsockopt+0x95/0xd0 net/core/sock.c:3039
  SYSC_setsockopt net/socket.c:1850 [inline]
  SyS_setsockopt+0x189/0x360 net/socket.c:1829
  do_syscall_64+0x281/0x940 arch/x86/entry/common.c:287
  entry_SYSCALL_64_after_hwframe+0x42/0xb7
RIP: 0033:0x454889
RSP: 002b:00007fc927626c68 EFLAGS: 00000246 ORIG_RAX: 0000000000000036
RAX: ffffffffffffffda RBX: 00007fc9276276d4 RCX: 0000000000454889
RDX: 000000000000048c RSI: 0000000000000000 RDI: 0000000000000017
RBP: 000000000072bf58 R08: 0000000000000018 R09: 0000000000000000
R10: 0000000020000000 R11: 0000000000000246 R12: 00000000ffffffff
R13: 000000000000051c R14: 00000000006f9b40 R15: 0000000000000001

Showing all locks held in the system:
2 locks held by khungtaskd/868:
  #0:  (rcu_read_lock){....}, at: [<00000000a1a8f002>]
check_hung_uninterruptible_tasks kernel/hung_task.c:175 [inline]
  #0:  (rcu_read_lock){....}, at: [<00000000a1a8f002>] watchdog+0x1c5/0xd60
kernel/hung_task.c:249
  #1:  (tasklist_lock){.+.+}, at: [<0000000037c2f8f9>]
debug_show_all_locks+0xd3/0x3d0 kernel/locking/lockdep.c:4470
1 lock held by rsyslogd/4247:
  #0:  (&f->f_pos_lock){+.+.}, at: [<000000000d8d6983>]
__fdget_pos+0x12b/0x190 fs/file.c:765
2 locks held by getty/4338:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4339:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4340:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4341:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4342:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4343:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
2 locks held by getty/4344:
  #0:  (&tty->ldisc_sem){++++}, at: [<00000000bee98654>]
ldsem_down_read+0x37/0x40 drivers/tty/tty_ldsem.c:365
  #1:  (&ldata->atomic_read_lock){+.+.}, at: [<00000000c1d180aa>]
n_tty_read+0x2ef/0x1a40 drivers/tty/n_tty.c:2131
3 locks held by kworker/0:5/6494:
  #0:  ((wq_completion)"%s"("ipv6_addrconf")){+.+.}, at:
[<00000000a062b18e>] work_static include/linux/workqueue.h:198 [inline]
  #0:  ((wq_completion)"%s"("ipv6_addrconf")){+.+.}, at:
[<00000000a062b18e>] set_work_data kernel/workqueue.c:619 [inline]
  #0:  ((wq_completion)"%s"("ipv6_addrconf")){+.+.}, at:
[<00000000a062b18e>] set_work_pool_and_clear_pending kernel/workqueue.c:646
[inline]
  #0:  ((wq_completion)"%s"("ipv6_addrconf")){+.+.}, at:
[<00000000a062b18e>] process_one_work+0xb12/0x1bb0 kernel/workqueue.c:2084
  #1:  ((addr_chk_work).work){+.+.}, at: [<00000000278427d5>]
process_one_work+0xb89/0x1bb0 kernel/workqueue.c:2088
  #2:  (rtnl_mutex){+.+.}, at: [<00000000066e35ac>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74
1 lock held by syz-executor7/25421:
  #0:  (ipvs->sync_mutex){+.+.}, at: [<00000000d414a689>]
do_ip_vs_set_ctl+0x277/0x1cc0 net/netfilter/ipvs/ip_vs_ctl.c:2393
2 locks held by syz-executor7/25427:
  #0:  (rtnl_mutex){+.+.}, at: [<00000000066e35ac>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74
  #1:  (ipvs->sync_mutex){+.+.}, at: [<00000000e6d48489>]
do_ip_vs_set_ctl+0x10f8/0x1cc0 net/netfilter/ipvs/ip_vs_ctl.c:2388
1 lock held by syz-executor7/25435:
  #0:  (rtnl_mutex){+.+.}, at: [<00000000066e35ac>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74
1 lock held by ipvs-b:2:0/25415:
  #0:  (rtnl_mutex){+.+.}, at: [<00000000066e35ac>] rtnl_lock+0x17/0x20
net/core/rtnetlink.c:74

Reported-and-tested-by: syzbot+a46d6abf9d56b1365a72@syzkaller.appspotmail.com
Reported-and-tested-by: syzbot+5fe074c01b2032ce9618@syzkaller.appspotmail.com
Fixes: e0b26cc997d5 ("ipvs: call rtnl_lock early")
Signed-off-by: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Simon Horman <horms@verge.net.au>
Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Zubin Mithra <zsm@chromium.org>
Cc: Guenter Roeck <groeck@chromium.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
net/netfilter/ipvs/ip_vs_ctl.c
net/netfilter/ipvs/ip_vs_sync.c