ext4: akpm's locking hack to fix locking delays This is a port of the following patch from Andrew Morton to ext4: http://lkml.org/lkml/2008/10/3/22 This fixes a major contention problem in do_get_write_access() when a buffer is modified in both the current and committing transaction. Signed-off-by: "Theodore Ts'o" Cc: akpm@linux-foundation.org diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h index f46a513..23822fb 100644 --- a/fs/ext4/ext4.h +++ b/fs/ext4/ext4.h @@ -540,6 +540,7 @@ do { \ #define EXT4_MOUNT_JOURNAL_ASYNC_COMMIT 0x1000000 /* Journal Async Commit */ #define EXT4_MOUNT_I_VERSION 0x2000000 /* i_version support */ #define EXT4_MOUNT_DELALLOC 0x8000000 /* Delalloc support */ +#define EXT4_MOUNT_AKPM_LOCK_HACK 0x10000000 /* akpm lock hack */ /* Compatibility, for having both ext2_fs.h and ext4_fs.h included at once */ #ifndef _LINUX_EXT2_FS_H #define clear_opt(o, opt) o &= ~EXT4_MOUNT_##opt diff --git a/fs/ext4/super.c b/fs/ext4/super.c index 67ebefb..f4e7157 100644 --- a/fs/ext4/super.c +++ b/fs/ext4/super.c @@ -752,6 +752,8 @@ static int ext4_show_options(struct seq_file *seq, struct vfsmount *vfs) seq_puts(seq, ",journal_async_commit"); if (test_opt(sb, NOBH)) seq_puts(seq, ",nobh"); + if (test_opt(sb, AKPM_LOCK_HACK)) + seq_puts(seq, ",akpm_lock_hack"); if (!test_opt(sb, EXTENTS)) seq_puts(seq, ",noextents"); if (test_opt(sb, I_VERSION)) @@ -911,7 +913,7 @@ enum { Opt_ignore, Opt_barrier, Opt_err, Opt_resize, Opt_usrquota, Opt_grpquota, Opt_extents, Opt_noextents, Opt_i_version, Opt_mballoc, Opt_nomballoc, Opt_stripe, Opt_delalloc, Opt_nodelalloc, - Opt_inode_readahead_blks + Opt_inode_readahead_blks, Opt_akpm_lock_hack, }; static match_table_t tokens = { @@ -973,6 +975,7 @@ static match_table_t tokens = { {Opt_delalloc, "delalloc"}, {Opt_nodelalloc, "nodelalloc"}, {Opt_inode_readahead_blks, "inode_readahead_blks=%u"}, + {Opt_akpm_lock_hack, "akpm_lock_hack"}, {Opt_err, NULL}, }; @@ -1382,6 +1385,9 @@ set_qf_format: return 0; sbi->s_inode_readahead_blks = option; break; + case Opt_akpm_lock_hack: + set_opt(sbi->s_mount_opt, AKPM_LOCK_HACK); + break; default: printk(KERN_ERR "EXT4-fs: Unrecognized mount option \"%s\" " @@ -2534,6 +2540,10 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal) journal->j_flags |= JBD2_BARRIER; else journal->j_flags &= ~JBD2_BARRIER; + if (test_opt(sb, AKPM_LOCK_HACK)) + journal->j_flags |= JBD2_LOCK_HACK; + else + journal->j_flags &= ~JBD2_LOCK_HACK; spin_unlock(&journal->j_state_lock); } diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c index e5d5405..32c288a 100644 --- a/fs/jbd2/transaction.c +++ b/fs/jbd2/transaction.c @@ -546,6 +546,7 @@ do_get_write_access(handle_t *handle, struct journal_head *jh, int error; char *frozen_buffer = NULL; int need_copy = 0; + int locked = 0; if (is_handle_aborted(handle)) return -EROFS; @@ -561,7 +562,13 @@ repeat: /* @@@ Need to check for errors here at some point. */ - lock_buffer(bh); + if (journal->j_flags & JBD2_LOCK_HACK) { + if (trylock_buffer(bh)) + locked = 1; /* lolz */ + } else { + lock_buffer(bh); + locked = 1; + } jbd_lock_bh_state(bh); /* We now hold the buffer lock so it is safe to query the buffer @@ -600,7 +607,8 @@ repeat: jbd_unexpected_dirty_buffer(jh); } - unlock_buffer(bh); + if (locked) + unlock_buffer(bh); error = -EROFS; if (is_handle_aborted(handle)) { diff --git a/include/linux/jbd2.h b/include/linux/jbd2.h index 66c3499..c614dae 100644 --- a/include/linux/jbd2.h +++ b/include/linux/jbd2.h @@ -967,6 +967,7 @@ struct journal_s #define JBD2_FLUSHED 0x008 /* The journal superblock has been flushed */ #define JBD2_LOADED 0x010 /* The journal superblock has been loaded */ #define JBD2_BARRIER 0x020 /* Use IDE barriers */ +#define JBD2_LOCK_HACK 0x040 /* akpm's locking hack */ /* * Function declarations for the journaling transaction and buffer