diff options
| author | Mikulas Patocka <mpatocka@redhat.com> | 2024-09-05 20:27:25 +0200 |
|---|---|---|
| committer | Mikulas Patocka <mpatocka@redhat.com> | 2024-09-06 12:38:16 +0200 |
| commit | f8e1ca92e35e9041cc0a1bc226ef07a853a22de4 (patch) | |
| tree | 2b95f92bb36523a38f786b473b924bb6b1141521 | |
| parent | faada2174c08662ae98b439c69efe3e79382c538 (diff) | |
dm-integrity: fix a race condition when accessing recalc_sector
There's a race condition when accessing the variable
ic->sb->recalc_sector. The function integrity_recalc writes to this
variable when it makes some progress and the function
dm_integrity_map_continue may read this variable concurrently.
One problem is that on 32-bit architectures the 64-bit variable is not
read and written atomically - it may be possible to read garbage if read
races with write.
Another problem is that memory accesses to this variable are not guarded
with memory barriers.
This commit fixes the race - it moves reading ic->sb->recalc_sector to an
earlier place where we hold &ic->endio_wait.lock.
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Cc: stable@vger.kernel.org
| -rw-r--r-- | drivers/md/dm-integrity.c | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/drivers/md/dm-integrity.c b/drivers/md/dm-integrity.c index 51e6964c1305..acff2f64f251 100644 --- a/drivers/md/dm-integrity.c +++ b/drivers/md/dm-integrity.c @@ -2174,6 +2174,7 @@ static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io)); unsigned int journal_section, journal_entry; unsigned int journal_read_pos; + sector_t recalc_sector; struct completion read_comp; bool discard_retried = false; bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ; @@ -2314,6 +2315,7 @@ offload_to_thread: goto lock_retry; } } + recalc_sector = le64_to_cpu(ic->sb->recalc_sector); spin_unlock_irq(&ic->endio_wait.lock); if (unlikely(journal_read_pos != NOT_FOUND)) { @@ -2368,7 +2370,7 @@ offload_to_thread: if (need_sync_io) { wait_for_completion_io(&read_comp); if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) && - dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector)) + dio->range.logical_sector + dio->range.n_sectors > recalc_sector) goto skip_check; if (ic->mode == 'B') { if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector, |