block: fix q->flush_rq NULL pointer crash on dm-mpath flush
authorMike Snitzer <snitzer@redhat.com>
Sun, 9 Mar 2014 00:20:01 +0000 (17:20 -0700)
committerJens Axboe <axboe@fb.com>
Sun, 9 Mar 2014 00:20:01 +0000 (17:20 -0700)
Commit 1874198 ("blk-mq: rework flush sequencing logic") switched
->flush_rq from being an embedded member of the request_queue structure
to being dynamically allocated in blk_init_queue_node().

Request-based DM multipath doesn't use blk_init_queue_node(), instead it
uses blk_alloc_queue_node() + blk_init_allocated_queue().  Because
commit 1874198 placed the dynamic allocation of ->flush_rq in
blk_init_queue_node() any flush issued to a dm-mpath device would crash
with a NULL pointer, e.g.:

BUG: unable to handle kernel NULL pointer dereference at           (null)
IP: [<ffffffff8125037e>] blk_rq_init+0x1e/0xb0
PGD bb3c7067 PUD bb01d067 PMD 0
Oops: 0002 [#1] SMP
...
CPU: 5 PID: 5028 Comm: dt Tainted: G        W  O 3.14.0-rc3.snitm+ #10
...
task: ffff88032fb270e0 ti: ffff880079564000 task.ti: ffff880079564000
RIP: 0010:[<ffffffff8125037e>]  [<ffffffff8125037e>] blk_rq_init+0x1e/0xb0
RSP: 0018:ffff880079565c98  EFLAGS: 00010046
RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000030
RDX: ffff880260c74048 RSI: 0000000000000000 RDI: 0000000000000000
RBP: ffff880079565ca8 R08: ffff880260aa1e98 R09: 0000000000000001
R10: ffff88032fa78500 R11: 0000000000000246 R12: 0000000000000000
R13: ffff880260aa1de8 R14: 0000000000000650 R15: 0000000000000000
FS:  00007f8d36a2a700(0000) GS:ffff88033fca0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000079b36000 CR4: 00000000000007e0
Stack:
 0000000000000000 ffff880260c74048 ffff880079565cd8 ffffffff81257a47
 ffff880260aa1de8 ffff880260c74048 0000000000000001 0000000000000000
 ffff880079565d08 ffffffff81257c2d 0000000000000000 ffff880260aa1de8
Call Trace:
 [<ffffffff81257a47>] blk_flush_complete_seq+0x2d7/0x2e0
 [<ffffffff81257c2d>] blk_insert_flush+0x1dd/0x210
 [<ffffffff8124ec59>] __elv_add_request+0x1f9/0x320
 [<ffffffff81250681>] ? blk_account_io_start+0x111/0x190
 [<ffffffff81253a4b>] blk_queue_bio+0x25b/0x330
 [<ffffffffa0020bf5>] dm_request+0x35/0x40 [dm_mod]
 [<ffffffff812530c0>] generic_make_request+0xc0/0x100
 [<ffffffff81253173>] submit_bio+0x73/0x140
 [<ffffffff811becdd>] submit_bio_wait+0x5d/0x80
 [<ffffffff81257528>] blkdev_issue_flush+0x78/0xa0
 [<ffffffff811c1f6f>] blkdev_fsync+0x3f/0x60
 [<ffffffff811b7fde>] vfs_fsync_range+0x1e/0x20
 [<ffffffff811b7ffc>] vfs_fsync+0x1c/0x20
 [<ffffffff811b81f1>] do_fsync+0x41/0x80
 [<ffffffff8118874e>] ? SyS_lseek+0x7e/0x80
 [<ffffffff811b8260>] SyS_fsync+0x10/0x20
 [<ffffffff8154c2d2>] system_call_fastpath+0x16/0x1b

Fix this by moving the ->flush_rq allocation from blk_init_queue_node()
to blk_init_allocated_queue().  blk_init_queue_node() also calls
blk_init_allocated_queue() so this change is functionality equivalent
for all blk_init_queue_node() callers.

Reported-by: Hannes Reinecke <hare@suse.de>
Reported-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
block/blk-core.c

index 853f92749202cbfe5b252d557f667b7f970ac872..4cd5ffc184427cc808808cdd9cb6acbc86b381d9 100644 (file)
@@ -693,20 +693,11 @@ blk_init_queue_node(request_fn_proc *rfn, spinlock_t *lock, int node_id)
        if (!uninit_q)
                return NULL;
 
-       uninit_q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
-       if (!uninit_q->flush_rq)
-               goto out_cleanup_queue;
-
        q = blk_init_allocated_queue(uninit_q, rfn, lock);
        if (!q)
-               goto out_free_flush_rq;
-       return q;
+               blk_cleanup_queue(uninit_q);
 
-out_free_flush_rq:
-       kfree(uninit_q->flush_rq);
-out_cleanup_queue:
-       blk_cleanup_queue(uninit_q);
-       return NULL;
+       return q;
 }
 EXPORT_SYMBOL(blk_init_queue_node);
 
@@ -717,6 +708,10 @@ blk_init_allocated_queue(struct request_queue *q, request_fn_proc *rfn,
        if (!q)
                return NULL;
 
+       q->flush_rq = kzalloc(sizeof(struct request), GFP_KERNEL);
+       if (!q->flush_rq)
+               return NULL;
+
        if (blk_init_rl(&q->root_rl, q, GFP_KERNEL))
                return NULL;