]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
mlock: fix mlock count can not decrease in race condition
authorYisheng Xie <xieyisheng1@huawei.com>
Fri, 2 Jun 2017 21:46:43 +0000 (14:46 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 2 Jun 2017 22:07:38 +0000 (15:07 -0700)
Kefeng reported that when running the follow test, the mlock count in
meminfo will increase permanently:

 [1] testcase
 linux:~ # cat test_mlockal
 grep Mlocked /proc/meminfo
  for j in `seq 0 10`
  do
  for i in `seq 4 15`
  do
  ./p_mlockall >> log &
  done
  sleep 0.2
 done
 # wait some time to let mlock counter decrease and 5s may not enough
 sleep 5
 grep Mlocked /proc/meminfo

 linux:~ # cat p_mlockall.c
 #include <sys/mman.h>
 #include <stdlib.h>
 #include <stdio.h>

 #define SPACE_LEN 4096

 int main(int argc, char ** argv)
 {
  int ret;
  void *adr = malloc(SPACE_LEN);
  if (!adr)
  return -1;

  ret = mlockall(MCL_CURRENT | MCL_FUTURE);
  printf("mlcokall ret = %d\n", ret);

  ret = munlockall();
  printf("munlcokall ret = %d\n", ret);

  free(adr);
  return 0;
 }

In __munlock_pagevec() we should decrement NR_MLOCK for each page where
we clear the PageMlocked flag.  Commit 1ebb7cc6a583 ("mm: munlock: batch
NR_MLOCK zone state updates") has introduced a bug where we don't
decrement NR_MLOCK for pages where we clear the flag, but fail to
isolate them from the lru list (e.g.  when the pages are on some other
cpu's percpu pagevec).  Since PageMlocked stays cleared, the NR_MLOCK
accounting gets permanently disrupted by this.

Fix it by counting the number of page whose PageMlock flag is cleared.

Fixes: 1ebb7cc6a583 (" mm: munlock: batch NR_MLOCK zone state updates")
Link: http://lkml.kernel.org/r/1495678405-54569-1-git-send-email-xieyisheng1@huawei.com
Signed-off-by: Yisheng Xie <xieyisheng1@huawei.com>
Reported-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Tested-by: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Joern Engel <joern@logfs.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Michel Lespinasse <walken@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Xishi Qiu <qiuxishi@huawei.com>
Cc: zhongjiang <zhongjiang@huawei.com>
Cc: Hanjun Guo <guohanjun@huawei.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/mlock.c

index c483c5c20b4bd12bcca50972c9f74a0dbd3a713e..b562b5523a6544e6c0ae6e4f792943441f6217a3 100644 (file)
@@ -284,7 +284,7 @@ static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
 {
        int i;
        int nr = pagevec_count(pvec);
-       int delta_munlocked;
+       int delta_munlocked = -nr;
        struct pagevec pvec_putback;
        int pgrescued = 0;
 
@@ -304,6 +304,8 @@ static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
                                continue;
                        else
                                __munlock_isolation_failed(page);
+               } else {
+                       delta_munlocked++;
                }
 
                /*
@@ -315,7 +317,6 @@ static void __munlock_pagevec(struct pagevec *pvec, struct zone *zone)
                pagevec_add(&pvec_putback, pvec->pages[i]);
                pvec->pages[i] = NULL;
        }
-       delta_munlocked = -nr + pagevec_count(&pvec_putback);
        __mod_zone_page_state(zone, NR_MLOCK, delta_munlocked);
        spin_unlock_irq(zone_lru_lock(zone));