src/patches/suse-2.6.27.31/patches.fixes/do_anonymous_page-race

   1 From: Andrea Arcangeli <andrea@suse.de>
   2 Subject: Race condition in userspace testcase
   3 References: 46948, LTC11574
   4
   5
   6 Additional Comment #103 From Andrea Arcangeli 2004-10-15 19:41
   7 the last patch I attached is the safest I believe.
   8
   9 I'm not sure if a lock_unlock or lock_unlock is always guaranteed to happen
  10 after the I/O, and this makes sure no race can happen anymore.
  11
  12 ---
  13  fs/bio.c |   10 ++++++++++
  14  1 file changed, 10 insertions(+)
  15
  16 --- a/fs/bio.c
  17 +++ b/fs/bio.c
  18 @@ -1192,6 +1192,16 @@ void bio_endio(struct bio *bio, int erro
  19         else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
  20                 error = -EIO;
  21
  22 +       if (bio_data_dir(bio) == READ)
  23 +               /*
  24 +                * If the current cpu has written to the page by hand
  25 +                * without dma, we must enforce ordering to be sure
  26 +                * this written data will be visible before we expose
  27 +                * the page contents to other cpus (for example with
  28 +                * a set_pte).
  29 +                */
  30 +               smp_wmb();
  31 +
  32         if (bio->bi_end_io)
  33                 bio->bi_end_io(bio, error);
  34  }