swap: avoid read_swap_cache_async() race to deadlock while waiting on discard I/O...
authorRafael Aquini <aquini@redhat.com>
Wed, 12 Jun 2013 21:04:49 +0000 (14:04 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 12 Jun 2013 23:29:45 +0000 (16:29 -0700)
commitcbab0e4eec299e9059199ebe6daf48730be46d2b
treed1bf3596a81a521b25814c3693cf4a3769109a08
parent24b8256a1fb28d357bc6fa09184ba29b4255ba5c
swap: avoid read_swap_cache_async() race to deadlock while waiting on discard I/O completion

read_swap_cache_async() can race against get_swap_page(), and stumble
across a SWAP_HAS_CACHE entry in the swap map whose page wasn't brought
into the swapcache yet.

This transient swap_map state is expected to be transitory, but the
actual placement of discard at scan_swap_map() inserts a wait for I/O
completion thus making the thread at read_swap_cache_async() to loop
around its -EEXIST case, while the other end at get_swap_page() is
scheduled away at scan_swap_map().  This can leave the system deadlocked
if the I/O completion happens to be waiting on the CPU waitqueue where
read_swap_cache_async() is busy looping and !CONFIG_PREEMPT.

This patch introduces a cond_resched() call to make the aforementioned
read_swap_cache_async() busy loop condition to bail out when necessary,
thus avoiding the subtle race window.

Signed-off-by: Rafael Aquini <aquini@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Shaohua Li <shli@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/swap_state.c