git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From cbab0e4eec299e9059199ebe6daf48730be46d2b Mon Sep 17 00:00:00 2001
   2 From: Rafael Aquini <aquini@redhat.com>
   3 Date: Wed, 12 Jun 2013 14:04:49 -0700
   4 Subject: swap: avoid read_swap_cache_async() race to deadlock while waiting on discard I/O completion
   5
   6 From: Rafael Aquini <aquini@redhat.com>
   7
   8 commit cbab0e4eec299e9059199ebe6daf48730be46d2b upstream.
   9
  10 read_swap_cache_async() can race against get_swap_page(), and stumble
  11 across a SWAP_HAS_CACHE entry in the swap map whose page wasn't brought
  12 into the swapcache yet.
  13
  14 This transient swap_map state is expected to be transitory, but the
  15 actual placement of discard at scan_swap_map() inserts a wait for I/O
  16 completion thus making the thread at read_swap_cache_async() to loop
  17 around its -EEXIST case, while the other end at get_swap_page() is
  18 scheduled away at scan_swap_map().  This can leave the system deadlocked
  19 if the I/O completion happens to be waiting on the CPU waitqueue where
  20 read_swap_cache_async() is busy looping and !CONFIG_PREEMPT.
  21
  22 This patch introduces a cond_resched() call to make the aforementioned
  23 read_swap_cache_async() busy loop condition to bail out when necessary,
  24 thus avoiding the subtle race window.
  25
  26 Signed-off-by: Rafael Aquini <aquini@redhat.com>
  27 Acked-by: Johannes Weiner <hannes@cmpxchg.org>
  28 Acked-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
  29 Acked-by: Hugh Dickins <hughd@google.com>
  30 Cc: Shaohua Li <shli@kernel.org>
  31 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  32 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  33 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  34
  35 ---
  36  mm/swap_state.c |   18 +++++++++++++++++-
  37  1 file changed, 17 insertions(+), 1 deletion(-)
  38
  39 --- a/mm/swap_state.c
  40 +++ b/mm/swap_state.c
  41 @@ -315,8 +315,24 @@ struct page *read_swap_cache_async(swp_e
  42                  * Swap entry may have been freed since our caller observed it.
  43                  */
  44                 err = swapcache_prepare(entry);
  45 -               if (err == -EEXIST) {   /* seems racy */
  46 +               if (err == -EEXIST) {
  47                         radix_tree_preload_end();
  48 +                       /*
  49 +                        * We might race against get_swap_page() and stumble
  50 +                        * across a SWAP_HAS_CACHE swap_map entry whose page
  51 +                        * has not been brought into the swapcache yet, while
  52 +                        * the other end is scheduled away waiting on discard
  53 +                        * I/O completion at scan_swap_map().
  54 +                        *
  55 +                        * In order to avoid turning this transitory state
  56 +                        * into a permanent loop around this -EEXIST case
  57 +                        * if !CONFIG_PREEMPT and the I/O completion happens
  58 +                        * to be waiting on the CPU waitqueue where we are now
  59 +                        * busy looping, we just conditionally invoke the
  60 +                        * scheduler here, if there are some more important
  61 +                        * tasks to run.
  62 +                        */
  63 +                       cond_resched();
  64                         continue;
  65                 }
  66                 if (err) {              /* swp entry is obsolete ? */