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       | 13 ++++++++++++-
  15  fs/jbd2/transaction.c | 12 ++++++++++--
  16  include/linux/jbd2.h  |  1 +
  17  4 files changed, 26 insertions(+), 3 deletions(-)
  18
  19 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  20 index ef9bf04..5c64ba6 100644
  21 --- a/fs/ext4/ext4.h
  22 +++ b/fs/ext4/ext4.h
  23 @@ -1010,6 +1010,9 @@ struct ext4_inode_info {
  24                                                       file systems */
  25  #define EXT4_MOUNT2_DUMMY_ENCRYPTION   0x80000000 /* Use dummy encryption */
  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 54ac1e5..d179efc 100644
  35 --- a/fs/ext4/super.c
  36 +++ b/fs/ext4/super.c
  37 @@ -1149,7 +1149,8 @@ enum {
  38         Opt_inode_readahead_blks, Opt_journal_ioprio,
  39         Opt_dioread_nolock, Opt_dioread_lock,
  40         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
  41 -       Opt_max_dir_size_kb, Opt_encrypt_key_sig, Opt_dummy_encryption
  42 +       Opt_max_dir_size_kb, Opt_encrypt_key_sig, Opt_dummy_encryption,
  43 +       Opt_akpm_lock_hack,
  44  };
  45
  46  static const match_table_t tokens = {
  47 @@ -1206,6 +1207,7 @@ static const match_table_t tokens = {
  48         {Opt_nobarrier, "nobarrier"},
  49         {Opt_i_version, "i_version"},
  50         {Opt_stripe, "stripe=%u"},
  51 +       {Opt_akpm_lock_hack, "akpm_lock_hack"},
  52         {Opt_delalloc, "delalloc"},
  53         {Opt_nodelalloc, "nodelalloc"},
  54         {Opt_removed, "mblk_io_submit"},
  55 @@ -1466,6 +1468,9 @@ static int handle_mount_opt(struct super_block *sb, char *opt, int token,
  56         case Opt_i_version:
  57                 sb->s_flags |= MS_I_VERSION;
  58                 return 1;
  59 +       case Opt_akpm_lock_hack:
  60 +               set_opt2(sb, AKPM_LOCK_HACK);
  61 +               return 1;
  62         }
  63
  64         for (m = ext4_mount_opts; m->token != Opt_err; m++)
  65 @@ -1834,6 +1839,8 @@ static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
  66                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
  67         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
  68                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
  69 +       if (test_opt2(sb, AKPM_LOCK_HACK))
  70 +               seq_puts(seq, ",akpm_lock_hack");
  71         if (sb->s_flags & MS_I_VERSION)
  72                 SEQ_OPTS_PUTS("i_version");
  73         if (nodefs || sbi->s_stripe)
  74 @@ -4431,6 +4438,10 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
  75                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
  76         else
  77                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
  78 +       if (test_opt2(sb, AKPM_LOCK_HACK))
  79 +               journal->j_flags |= JBD2_LOCK_HACK;
  80 +       else
  81 +               journal->j_flags &= ~JBD2_LOCK_HACK;
  82         write_unlock(&journal->j_state_lock);
  83  }
  84
  85 diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
  86 index 93ffee2..4ba2b76 100644
  87 --- a/fs/jbd2/transaction.c
  88 +++ b/fs/jbd2/transaction.c
  89 @@ -784,6 +784,7 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
  90         char *frozen_buffer = NULL;
  91         int need_copy = 0;
  92         unsigned long start_lock, time_lock;
  93 +       int locked = 0;
  94
  95         WARN_ON(!transaction);
  96         if (is_handle_aborted(handle))
  97 @@ -799,7 +800,13 @@ repeat:
  98         /* @@@ Need to check for errors here at some point. */
  99
 100         start_lock = jiffies;
 101 -       lock_buffer(bh);
 102 +       if (journal->j_flags & JBD2_LOCK_HACK) {
 103 +               if (trylock_buffer(bh))
 104 +                       locked = 1;     /* lolz */
 105 +       } else {
 106 +               lock_buffer(bh);
 107 +               locked = 1;
 108 +       }
 109         jbd_lock_bh_state(bh);
 110
 111         /* If it takes too long to lock the buffer, trace it */
 112 @@ -846,7 +853,8 @@ repeat:
 113                 set_buffer_jbddirty(bh);
 114         }
 115
 116 -       unlock_buffer(bh);
 117 +       if (locked)
 118 +               unlock_buffer(bh);
 119
 120         error = -EROFS;
 121         if (is_handle_aborted(handle)) {
 122 diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
 123 index 704b9a5..907a108 100644
 124 --- a/include/linux/jbd2.h
 125 +++ b/include/linux/jbd2.h
 126 @@ -1007,6 +1007,7 @@ struct journal_s
 127  #define JBD2_ABORT_ON_SYNCDATA_ERR     0x040   /* Abort the journal on file
 128                                                  * data write error in ordered
 129                                                  * mode */
 130 +#define JBD2_LOCK_HACK 0x080   /* akpm's locking hack */
 131
 132  /*
 133   * Function declarations for the journaling transaction and buffer