mm,vmscan: Make unregister_shrinker() no-op if register_shrinker() failed.
authorTetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Mon, 18 Dec 2017 11:31:41 +0000 (20:31 +0900)
committerAl Viro <viro@zeniv.linux.org.uk>
Mon, 18 Dec 2017 20:03:09 +0000 (15:03 -0500)
commitbb422a738f6566f7439cd347d54e321e4fe92a9f
tree6749680cf7e571d7a60ca6bdfa60baba83ab492d
parentd7ee946942bdd12394809305e3df05aa4c8b7b8f
mm,vmscan: Make unregister_shrinker() no-op if register_shrinker() failed.

Syzbot caught an oops at unregister_shrinker() because combination of
commit 1d3d4437eae1bb29 ("vmscan: per-node deferred work") and fault
injection made register_shrinker() fail and the caller of
register_shrinker() did not check for failure.

----------
[  554.881422] FAULT_INJECTION: forcing a failure.
[  554.881422] name failslab, interval 1, probability 0, space 0, times 0
[  554.881438] CPU: 1 PID: 13231 Comm: syz-executor1 Not tainted 4.14.0-rc8+ #82
[  554.881443] Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
[  554.881445] Call Trace:
[  554.881459]  dump_stack+0x194/0x257
[  554.881474]  ? arch_local_irq_restore+0x53/0x53
[  554.881486]  ? find_held_lock+0x35/0x1d0
[  554.881507]  should_fail+0x8c0/0xa40
[  554.881522]  ? fault_create_debugfs_attr+0x1f0/0x1f0
[  554.881537]  ? check_noncircular+0x20/0x20
[  554.881546]  ? find_next_zero_bit+0x2c/0x40
[  554.881560]  ? ida_get_new_above+0x421/0x9d0
[  554.881577]  ? find_held_lock+0x35/0x1d0
[  554.881594]  ? __lock_is_held+0xb6/0x140
[  554.881628]  ? check_same_owner+0x320/0x320
[  554.881634]  ? lock_downgrade+0x990/0x990
[  554.881649]  ? find_held_lock+0x35/0x1d0
[  554.881672]  should_failslab+0xec/0x120
[  554.881684]  __kmalloc+0x63/0x760
[  554.881692]  ? lock_downgrade+0x990/0x990
[  554.881712]  ? register_shrinker+0x10e/0x2d0
[  554.881721]  ? trace_event_raw_event_module_request+0x320/0x320
[  554.881737]  register_shrinker+0x10e/0x2d0
[  554.881747]  ? prepare_kswapd_sleep+0x1f0/0x1f0
[  554.881755]  ? _down_write_nest_lock+0x120/0x120
[  554.881765]  ? memcpy+0x45/0x50
[  554.881785]  sget_userns+0xbcd/0xe20
(...snipped...)
[  554.898693] kasan: CONFIG_KASAN_INLINE enabled
[  554.898724] kasan: GPF could be caused by NULL-ptr deref or user memory access
[  554.898732] general protection fault: 0000 [#1] SMP KASAN
[  554.898737] Dumping ftrace buffer:
[  554.898741]    (ftrace buffer empty)
[  554.898743] Modules linked in:
[  554.898752] CPU: 1 PID: 13231 Comm: syz-executor1 Not tainted 4.14.0-rc8+ #82
[  554.898755] Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
[  554.898760] task: ffff8801d1dbe5c0 task.stack: ffff8801c9e38000
[  554.898772] RIP: 0010:__list_del_entry_valid+0x7e/0x150
[  554.898775] RSP: 0018:ffff8801c9e3f108 EFLAGS: 00010246
[  554.898780] RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000000
[  554.898784] RDX: 0000000000000000 RSI: ffff8801c53c6f98 RDI: ffff8801c53c6fa0
[  554.898788] RBP: ffff8801c9e3f120 R08: 1ffff100393c7d55 R09: 0000000000000004
[  554.898791] R10: ffff8801c9e3ef70 R11: 0000000000000000 R12: 0000000000000000
[  554.898795] R13: dffffc0000000000 R14: 1ffff100393c7e45 R15: ffff8801c53c6f98
[  554.898800] FS:  0000000000000000(0000) GS:ffff8801db300000(0000) knlGS:0000000000000000
[  554.898804] CS:  0010 DS: 002b ES: 002b CR0: 0000000080050033
[  554.898807] CR2: 00000000dbc23000 CR3: 00000001c7269000 CR4: 00000000001406e0
[  554.898813] DR0: 0000000020000000 DR1: 0000000020000000 DR2: 0000000000000000
[  554.898816] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000600
[  554.898818] Call Trace:
[  554.898828]  unregister_shrinker+0x79/0x300
[  554.898837]  ? perf_trace_mm_vmscan_writepage+0x750/0x750
[  554.898844]  ? down_write+0x87/0x120
[  554.898851]  ? deactivate_super+0x139/0x1b0
[  554.898857]  ? down_read+0x150/0x150
[  554.898864]  ? check_same_owner+0x320/0x320
[  554.898875]  deactivate_locked_super+0x64/0xd0
[  554.898883]  deactivate_super+0x141/0x1b0
----------

Since allowing register_shrinker() callers to call unregister_shrinker()
when register_shrinker() failed can simplify error recovery path, this
patch makes unregister_shrinker() no-op when register_shrinker() failed.
Also, reset shrinker->nr_deferred in case unregister_shrinker() was
by error called twice.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Aliaksei Karaliou <akaraliou.dev@gmail.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Cc: Glauber Costa <glauber@scylladb.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
mm/vmscan.c