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 diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
  13 index 4880cc3..b1cdd8d 100644
  14 --- a/fs/ext4/ext4.h
  15 +++ b/fs/ext4/ext4.h
  16 @@ -540,6 +540,7 @@ do {                                                                               \
  17  #define EXT4_MOUNT_I_VERSION            0x2000000 /* i_version support */
  18  #define EXT4_MOUNT_DELALLOC            0x8000000 /* Delalloc support */
  19  #define EXT4_MOUNT_DATA_ERR_ABORT      0x10000000 /* Abort on file data write */
  20 +#define EXT4_MOUNT_AKPM_LOCK_HACK      0x20000000 /* akpm lock hack */
  21
  22  /* Compatibility, for having both ext2_fs.h and ext4_fs.h included at once */
  23  #ifndef _LINUX_EXT2_FS_H
  24 diff --git a/fs/ext4/super.c b/fs/ext4/super.c
  25 index 9b2b2bc..faebc2e 100644
  26 --- a/fs/ext4/super.c
  27 +++ b/fs/ext4/super.c
  28 @@ -693,6 +693,8 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs)
  29                 seq_puts(seq, ",journal_async_commit");
  30         if (test_opt(sb, NOBH))
  31                 seq_puts(seq, ",nobh");
  32 +       if (test_opt(sb, AKPM_LOCK_HACK))
  33 +               seq_puts(seq, ",akpm_lock_hack");
  34         if (!test_opt(sb, EXTENTS))
  35                 seq_puts(seq, ",noextents");
  36         if (test_opt(sb, I_VERSION))
  37 @@ -856,7 +858,7 @@ enum {
  38         Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota,
  39         Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version,
  40         Opt_stripe, Opt_delalloc, Opt_nodelalloc,
  41 -       Opt_inode_readahead_blks
  42 +       Opt_inode_readahead_blks, Opt_akpm_lock_hack,
  43  };
  44
  45  static const match_table_t tokens = {
  46 @@ -916,6 +918,7 @@ static const match_table_t tokens = {
  47         {Opt_delalloc, "delalloc"},
  48         {Opt_nodelalloc, "nodelalloc"},
  49         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
  50 +       {Opt_akpm_lock_hack, "akpm_lock_hack"},
  51         {Opt_err, NULL},
  52  };
  53
  54 @@ -1328,6 +1331,9 @@ set_qf_format:
  55                                 return 0;
  56                         sbi->s_inode_readahead_blks = option;
  57                         break;
  58 +               case Opt_akpm_lock_hack:
  59 +                       set_opt(sbi->s_mount_opt, AKPM_LOCK_HACK);
  60 +                       break;
  61                 default:
  62                         printk(KERN_ERR
  63                                "EXT4-fs: Unrecognized mount option \"%s\" "
  64 @@ -2489,6 +2495,10 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
  65                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
  66         else
  67                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
  68 +       if (test_opt(sb, AKPM_LOCK_HACK))
  69 +               journal->j_flags |= JBD2_LOCK_HACK;
  70 +       else
  71 +               journal->j_flags &= ~JBD2_LOCK_HACK;
  72         spin_unlock(&journal->j_state_lock);
  73  }
  74
  75 diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
  76 index 648624c..e48c225 100644
  77 --- a/fs/jbd2/transaction.c
  78 +++ b/fs/jbd2/transaction.c
  79 @@ -547,6 +547,7 @@ do_get_write_access(handle_t *handle, struct journal_head *jh,
  80         int error;
  81         char *frozen_buffer = NULL;
  82         int need_copy = 0;
  83 +       int locked = 0;
  84
  85         if (is_handle_aborted(handle))
  86                 return -EROFS;
  87 @@ -562,7 +563,13 @@ repeat:
  88
  89         /* @@@ Need to check for errors here at some point. */
  90
  91 -       lock_buffer(bh);
  92 +       if (journal->j_flags & JBD2_LOCK_HACK) {
  93 +               if (trylock_buffer(bh))
  94 +                       locked = 1;     /* lolz */
  95 +       } else {
  96 +               lock_buffer(bh);
  97 +               locked = 1;
  98 +       }
  99         jbd_lock_bh_state(bh);
 100
 101         /* We now hold the buffer lock so it is safe to query the buffer
 102 @@ -601,7 +608,8 @@ repeat:
 103                 jbd_unexpected_dirty_buffer(jh);
 104         }
 105
 106 -       unlock_buffer(bh);
 107 +       if (locked)
 108 +               unlock_buffer(bh);
 109
 110         error = -EROFS;
 111         if (is_handle_aborted(handle)) {
 112 diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h
 113 index 09458eb..6131c3c 100644
 114 --- a/include/linux/jbd2.h
 115 +++ b/include/linux/jbd2.h
 116 @@ -979,6 +979,7 @@ struct journal_s
 117  #define JBD2_ABORT_ON_SYNCDATA_ERR     0x040   /* Abort the journal on file
 118                                                  * data write error in ordered
 119                                                  * mode */
 120 +#define JBD2_LOCK_HACK 0x080   /* akpm's locking hack */
 121
 122  /*
 123   * Function declarations for the journaling transaction and buffer