akpm-jbd2-locking-fix

   1 ext4: akpm's locking hack to fix locking delays
   2
   3 This is a port of the following patch from Andrew Morton to ext4:
   4
   5         http://lkml.org/lkml/2008/10/3/22
   6
   7 This fixes a major contention problem in do_get_write_access() when a
   8 buffer is modified in both the current and committing transaction.
   9
  10 Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
  11 Cc: akpm@linux-foundation.org
  12 ---
  13  fs/ext4/ext4.h        |  3 +++
  14  fs/ext4/super.c       | 11 +++++++++++
  15  fs/jbd2/transaction.c | 12 ++++++++++--
  16  include/linux/jbd2.h  |  1 +
  17  4 files changed, 25 insertions(+), 2 deletions(-)
  18
  19 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  20 index 4a17e88..d19ef59 100644
  21 --- a/fs/ext4/ext4.h
  22 +++ b/fs/ext4/ext4.h
  23 @@ -977,6 +977,9 @@ struct ext4_inode_info {
  24                                                       size of blocksize * 8
  25                                                       blocks */
  26
  27 +#define EXT4_MOUNT2_AKPM_LOCK_HACK     0x80000000 /* akpm lock hack */
  28 +
  29 +
  30  #define clear_opt(sb, opt)             EXT4_SB(sb)->s_mount_opt &= \
  31                                                 ~EXT4_MOUNT_##opt
  32  #define set_opt(sb, opt)               EXT4_SB(sb)->s_mount_opt |= \
  33 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
  34 index 79de0b1..86a80d2 100644
  35 --- a/fs/ext4/super.c
  36 +++ b/fs/ext4/super.c
  37 @@ -1151,6 +1151,7 @@ enum {
  38         Opt_dioread_nolock, Opt_dioread_lock,
  39         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
  40         Opt_max_dir_size_kb,
  41 +       Opt_akpm_lock_hack,
  42  };
  43
  44  static const match_table_t tokens = {
  45 @@ -1206,6 +1207,7 @@ static const match_table_t tokens = {
  46         {Opt_nobarrier, "nobarrier"},
  47         {Opt_i_version, "i_version"},
  48         {Opt_stripe, "stripe=%u"},
  49 +       {Opt_akpm_lock_hack, "akpm_lock_hack"},
  50         {Opt_delalloc, "delalloc"},
  51         {Opt_nodelalloc, "nodelalloc"},
  52         {Opt_removed, "mblk_io_submit"},
  53 @@ -1460,6 +1462,9 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
  54         case Opt_i_version:
  55                 sb->s_flags |= MS_I_VERSION;
  56                 return 1;
  57 +       case Opt_akpm_lock_hack:
  58 +               set_opt2(sb, AKPM_LOCK_HACK);
  59 +               return 1;
  60         }
  61
  62         for (m = ext4_mount_opts; m->token != Opt_err; m++)
  63 @@ -1779,6 +1784,8 @@ static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
  64                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
  65         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
  66                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
  67 +       if (test_opt2(sb, AKPM_LOCK_HACK))
  68 +               seq_puts(seq, ",akpm_lock_hack");
  69         if (sb->s_flags & MS_I_VERSION)
  70                 SEQ_OPTS_PUTS("i_version");
  71         if (nodefs || sbi->s_stripe)
  72 @@ -4299,6 +4306,10 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
  73                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
  74         else
  75                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
  76 +       if (test_opt2(sb, AKPM_LOCK_HACK))
  77 +               journal->j_flags |= JBD2_LOCK_HACK;
  78 +       else
  79 +               journal->j_flags &= ~JBD2_LOCK_HACK;
  80         write_unlock(&journal->j_state_lock);
  81  }
  82
  83 diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
  84 index e1c0b4a..6d6b6bc 100644
  85 --- a/fs/jbd2/transaction.c
  86 +++ b/fs/jbd2/transaction.c
  87 @@ -788,6 +788,7 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
  88         char *frozen_buffer = NULL;
  89         int need_copy = 0;
  90         unsigned long start_lock, time_lock;
  91 +       int locked = 0;
  92
  93         WARN_ON(!transaction);
  94         if (is_handle_aborted(handle))
  95 @@ -803,7 +804,13 @@ repeat:
  96         /* @@@ Need to check for errors here at some point. */
  97
  98         start_lock = jiffies;
  99 -       lock_buffer(bh);
 100 +       if (journal->j_flags & JBD2_LOCK_HACK) {
 101 +               if (trylock_buffer(bh))
 102 +                       locked = 1;     /* lolz */
 103 +       } else {
 104 +               lock_buffer(bh);
 105 +               locked = 1;
 106 +       }
 107         jbd_lock_bh_state(bh);
 108
 109         /* If it takes too long to lock the buffer, trace it */
 110 @@ -850,7 +857,8 @@ repeat:
 111                 set_buffer_jbddirty(bh);
 112         }
 113
 114 -       unlock_buffer(bh);
 115 +       if (locked)
 116 +               unlock_buffer(bh);
 117
 118         error = -EROFS;
 119         if (is_handle_aborted(handle)) {
 120 diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
 121 index d5b50a1..6f850b4 100644
 122 --- a/include/linux/jbd2.h
 123 +++ b/include/linux/jbd2.h
 124 @@ -996,6 +996,7 @@ struct journal_s
 125  #define JBD2_ABORT_ON_SYNCDATA_ERR     0x040   /* Abort the journal on file
 126                                                  * data write error in ordered
 127                                                  * mode */
 128 +#define JBD2_LOCK_HACK 0x080   /* akpm's locking hack */
 129
 130  /*
 131   * Function declarations for the journaling transaction and buffer