dm thin: do not queue freed thin mapping for next stage processing
authorVallish Vaidyeshwara <vallish@amazon.com>
Fri, 23 Jun 2017 18:53:06 +0000 (18:53 +0000)
committerMike Snitzer <snitzer@redhat.com>
Tue, 27 Jun 2017 19:14:34 +0000 (15:14 -0400)
commit00a0ea33b495ee6149bf5a77ac5807ce87323abb
treeb33f6834021fb03250a7fd78e9d2f2d836dac4f7
parentc4d097d13052d1e6f29b8798264aed6135d99568
dm thin: do not queue freed thin mapping for next stage processing

process_prepared_discard_passdown_pt1() should cleanup
dm_thin_new_mapping in cases of error.

dm_pool_inc_data_range() can fail trying to get a block reference:

metadata operation 'dm_pool_inc_data_range' failed: error = -61

When dm_pool_inc_data_range() fails, dm thin aborts current metadata
transaction and marks pool as PM_READ_ONLY. Memory for thin mapping
is released as well. However, current thin mapping will be queued
onto next stage as part of queue_passdown_pt2() or passdown_endio().
This dangling thin mapping memory when processed and accessed in
next stage will lead to device mapper crashing.

Code flow without fix:
-> process_prepared_discard_passdown_pt1(m)
   -> dm_thin_remove_range()
   -> discard passdown
      --> passdown_endio(m) queues m onto next stage
   -> dm_pool_inc_data_range() fails, frees memory m
            but does not remove it from next stage queue

-> process_prepared_discard_passdown_pt2(m)
   -> processes freed memory m and crashes

One such stack:

Call Trace:
[<ffffffffa037a46f>] dm_cell_release_no_holder+0x2f/0x70 [dm_bio_prison]
[<ffffffffa039b6dc>] cell_defer_no_holder+0x3c/0x80 [dm_thin_pool]
[<ffffffffa039b88b>] process_prepared_discard_passdown_pt2+0x4b/0x90 [dm_thin_pool]
[<ffffffffa0399611>] process_prepared+0x81/0xa0 [dm_thin_pool]
[<ffffffffa039e735>] do_worker+0xc5/0x820 [dm_thin_pool]
[<ffffffff8152bf54>] ? __schedule+0x244/0x680
[<ffffffff81087e72>] ? pwq_activate_delayed_work+0x42/0xb0
[<ffffffff81089f53>] process_one_work+0x153/0x3f0
[<ffffffff8108a71b>] worker_thread+0x12b/0x4b0
[<ffffffff8108a5f0>] ? rescuer_thread+0x350/0x350
[<ffffffff8108fd6a>] kthread+0xca/0xe0
[<ffffffff8108fca0>] ? kthread_park+0x60/0x60
[<ffffffff81530b45>] ret_from_fork+0x25/0x30

The fix is to first take the block ref count for discarded block and
then do a passdown discard of this block. If block ref count fails,
then bail out aborting current metadata transaction, mark pool as
PM_READ_ONLY and also free current thin mapping memory (existing error
handling code) without queueing this thin mapping onto next stage of
processing. If block ref count succeeds, then passdown discard of this
block. Discard callback of passdown_endio() will queue this thin mapping
onto next stage of processing.

Code flow with fix:
-> process_prepared_discard_passdown_pt1(m)
   -> dm_thin_remove_range()
   -> dm_pool_inc_data_range()
      --> if fails, free memory m and bail out
   -> discard passdown
      --> passdown_endio(m) queues m onto next stage

Cc: stable <stable@vger.kernel.org> # v4.9+
Reviewed-by: Eduardo Valentin <eduval@amazon.com>
Reviewed-by: Cristian Gafton <gafton@amazon.com>
Reviewed-by: Anchal Agarwal <anchalag@amazon.com>
Signed-off-by: Vallish Vaidyeshwara <vallish@amazon.com>
Reviewed-by: Joe Thornber <ejt@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
drivers/md/dm-thin.c