bcache: fix kcrashes with fio in RAID5 backend dev
authorTang Junhui <tang.junhui@zte.com.cn>
Tue, 27 Feb 2018 17:49:30 +0000 (09:49 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 30 May 2018 05:50:32 +0000 (07:50 +0200)
[ Upstream commit 60eb34ec5526e264c2bbaea4f7512d714d791caf ]

Kernel crashed when run fio in a RAID5 backend bcache device, the call
trace is bellow:
[  440.012034] kernel BUG at block/blk-ioc.c:146!
[  440.012696] invalid opcode: 0000 [#1] SMP NOPTI
[  440.026537] CPU: 2 PID: 2205 Comm: md127_raid5 Not tainted 4.15.0 #8
[  440.027441] Hardware name: HP ProLiant MicroServer Gen8, BIOS J06 07/16
/2015
[  440.028615] RIP: 0010:put_io_context+0x8b/0x90
[  440.029246] RSP: 0018:ffffa8c882b43af8 EFLAGS: 00010246
[  440.029990] RAX: 0000000000000000 RBX: ffffa8c88294fca0 RCX: 0000000000
0f4240
[  440.031006] RDX: 0000000000000004 RSI: 0000000000000286 RDI: ffffa8c882
94fca0
[  440.032030] RBP: ffffa8c882b43b10 R08: 0000000000000003 R09: ffff949cb8
0c1700
[  440.033206] R10: 0000000000000104 R11: 000000000000b71c R12: 00000000000
01000
[  440.034222] R13: 0000000000000000 R14: ffff949cad84db70 R15: ffff949cb11
bd1e0
[  440.035239] FS:  0000000000000000(0000) GS:ffff949cba280000(0000) knlGS:
0000000000000000
[  440.060190] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  440.084967] CR2: 00007ff0493ef000 CR3: 00000002f1e0a002 CR4: 00000000001
606e0
[  440.110498] Call Trace:
[  440.135443]  bio_disassociate_task+0x1b/0x60
[  440.160355]  bio_free+0x1b/0x60
[  440.184666]  bio_put+0x23/0x30
[  440.208272]  search_free+0x23/0x40 [bcache]
[  440.231448]  cached_dev_write_complete+0x31/0x70 [bcache]
[  440.254468]  closure_put+0xb6/0xd0 [bcache]
[  440.277087]  request_endio+0x30/0x40 [bcache]
[  440.298703]  bio_endio+0xa1/0x120
[  440.319644]  handle_stripe+0x418/0x2270 [raid456]
[  440.340614]  ? load_balance+0x17b/0x9c0
[  440.360506]  handle_active_stripes.isra.58+0x387/0x5a0 [raid456]
[  440.380675]  ? __release_stripe+0x15/0x20 [raid456]
[  440.400132]  raid5d+0x3ed/0x5d0 [raid456]
[  440.419193]  ? schedule+0x36/0x80
[  440.437932]  ? schedule_timeout+0x1d2/0x2f0
[  440.456136]  md_thread+0x122/0x150
[  440.473687]  ? wait_woken+0x80/0x80
[  440.491411]  kthread+0x102/0x140
[  440.508636]  ? find_pers+0x70/0x70
[  440.524927]  ? kthread_associate_blkcg+0xa0/0xa0
[  440.541791]  ret_from_fork+0x35/0x40
[  440.558020] Code: c2 48 00 5b 41 5c 41 5d 5d c3 48 89 c6 4c 89 e7 e8 bb c2
48 00 48 8b 3d bc 36 4b 01 48 89 de e8 7c f7 e0 ff 5b 41 5c 41 5d 5d c3 <0f> 0b
0f 1f 00 0f 1f 44 00 00 55 48 8d 47 b8 48 89 e5 41 57 41
[  440.610020] RIP: put_io_context+0x8b/0x90 RSP: ffffa8c882b43af8
[  440.628575] ---[ end trace a1fd79d85643a73e ]--

All the crash issue happened when a bypass IO coming, in such scenario
s->iop.bio is pointed to the s->orig_bio. In search_free(), it finishes the
s->orig_bio by calling bio_complete(), and after that, s->iop.bio became
invalid, then kernel would crash when calling bio_put(). Maybe its upper
layer's faulty, since bio should not be freed before we calling bio_put(),
but we'd better calling bio_put() first before calling bio_complete() to
notify upper layer ending this bio.

This patch moves bio_complete() under bio_put() to avoid kernel crash.

[mlyle: fixed commit subject for character limits]

Reported-by: Matthias Ferdinand <bcache@mfedv.net>
Tested-by: Matthias Ferdinand <bcache@mfedv.net>
Signed-off-by: Tang Junhui <tang.junhui@zte.com.cn>
Reviewed-by: Michael Lyle <mlyle@lyle.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <alexander.levin@microsoft.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/md/bcache/request.c

index edb8d1a1a69f72e01130456ba1e963b319872541..bd6f6f4b4256e8c099f9df458d7613ee17f345e1 100644 (file)
@@ -633,11 +633,11 @@ static void do_bio_hook(struct search *s, struct bio *orig_bio)
 static void search_free(struct closure *cl)
 {
        struct search *s = container_of(cl, struct search, cl);
-       bio_complete(s);
 
        if (s->iop.bio)
                bio_put(s->iop.bio);
 
+       bio_complete(s);
        closure_debug_destroy(cl);
        mempool_free(s, s->d->c->search);
 }