From: Theodore Ts'o Subject: [PATCH 5/5] ext4: Replace lock/unlock_super() with an explicit lock for resizing Date: Sat, 25 Apr 2009 23:49:25 -0400 Message-ID: <1240717765-16572-6-git-send-email-tytso@mit.edu> References: <1240717765-16572-1-git-send-email-tytso@mit.edu> <1240717765-16572-2-git-send-email-tytso@mit.edu> <1240717765-16572-3-git-send-email-tytso@mit.edu> <1240717765-16572-4-git-send-email-tytso@mit.edu> <1240717765-16572-5-git-send-email-tytso@mit.edu> Cc: Linux Kernel Developers List , Christoph Hellwig , Al Viro , Theodore Ts'o To: Ext4 Developers List Return-path: Received: from THUNK.ORG ([69.25.196.29]:60012 "EHLO thunker.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753677AbZDZDtc (ORCPT ); Sat, 25 Apr 2009 23:49:32 -0400 In-Reply-To: <1240717765-16572-5-git-send-email-tytso@mit.edu> Sender: linux-ext4-owner@vger.kernel.org List-ID: Use a separate lock to protect s_groups_count and the other block group descriptors which get changed via an on-line resize operation, so we can stop overloading the use of lock_super(). Signed-off-by: "Theodore Ts'o" --- fs/ext4/ext4_sb.h | 1 + fs/ext4/resize.c | 35 ++++++++++++++++++----------------- fs/ext4/super.c | 1 + 3 files changed, 20 insertions(+), 17 deletions(-) diff --git a/fs/ext4/ext4_sb.h b/fs/ext4/ext4_sb.h index 4bda2f7..2d36223 100644 --- a/fs/ext4/ext4_sb.h +++ b/fs/ext4/ext4_sb.h @@ -72,6 +72,7 @@ struct ext4_sb_info { struct journal_s *s_journal; struct list_head s_orphan; struct mutex s_orphan_lock; + struct mutex s_resize_lock; unsigned long s_commit_interval; u32 s_max_batch_time; u32 s_min_batch_time; diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c index 546c7dd..e8ded13 100644 --- a/fs/ext4/resize.c +++ b/fs/ext4/resize.c @@ -193,7 +193,7 @@ static int setup_new_group_blocks(struct super_block *sb, if (IS_ERR(handle)) return PTR_ERR(handle); - lock_super(sb); + mutex_lock(&sbi->s_resize_lock); if (input->group != sbi->s_groups_count) { err = -EBUSY; goto exit_journal; @@ -302,7 +302,7 @@ exit_bh: brelse(bh); exit_journal: - unlock_super(sb); + mutex_unlock(&sbi->s_resize_lock); if ((err2 = ext4_journal_stop(handle)) && !err) err = err2; @@ -643,11 +643,12 @@ exit_free: * important part is that the new block and inode counts are in the backup * superblocks, and the location of the new group metadata in the GDT backups. * - * We do not need lock_super() for this, because these blocks are not - * otherwise touched by the filesystem code when it is mounted. We don't - * need to worry about last changing from sbi->s_groups_count, because the - * worst that can happen is that we do not copy the full number of backups - * at this time. The resize which changed s_groups_count will backup again. + * We do not need take the s_resize_lock for this, because these + * blocks are not otherwise touched by the filesystem code when it is + * mounted. We don't need to worry about last changing from + * sbi->s_groups_count, because the worst that can happen is that we + * do not copy the full number of backups at this time. The resize + * which changed s_groups_count will backup again. */ static void update_backups(struct super_block *sb, int blk_off, char *data, int size) @@ -809,7 +810,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) goto exit_put; } - lock_super(sb); + mutex_lock(&sbi->s_resize_lock); if (input->group != sbi->s_groups_count) { ext4_warning(sb, __func__, "multiple resizers run on filesystem!"); @@ -840,7 +841,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) /* * OK, now we've set up the new group. Time to make it active. * - * Current kernels don't lock all allocations via lock_super(), + * We do not lock all allocations via s_resize_lock * so we have to be safe wrt. concurrent accesses the group * data. So we need to be careful to set all of the relevant * group descriptor data etc. *before* we enable the group. @@ -900,12 +901,12 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) * * The precise rules we use are: * - * * Writers of s_groups_count *must* hold lock_super + * * Writers of s_groups_count *must* hold s_resize_lock * AND * * Writers must perform a smp_wmb() after updating all dependent * data and before modifying the groups count * - * * Readers must hold lock_super() over the access + * * Readers must hold s_resize_lock over the access * OR * * Readers must perform an smp_rmb() after reading the groups count * and before reading any dependent data. @@ -948,7 +949,7 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input) sb->s_dirt = 1; exit_journal: - unlock_super(sb); + mutex_unlock(&sbi->s_resize_lock); if ((err2 = ext4_journal_stop(handle)) && !err) err = err2; if (!err) { @@ -986,7 +987,7 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, /* We don't need to worry about locking wrt other resizers just * yet: we're going to revalidate es->s_blocks_count after - * taking lock_super() below. */ + * taking the s_resize_lock below. */ o_blocks_count = ext4_blocks_count(es); o_groups_count = EXT4_SB(sb)->s_groups_count; @@ -1056,11 +1057,11 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, goto exit_put; } - lock_super(sb); + mutex_lock(&EXT4_SB(sb)->s_resize_lock); if (o_blocks_count != ext4_blocks_count(es)) { ext4_warning(sb, __func__, "multiple resizers run on filesystem!"); - unlock_super(sb); + mutex_unlock(&EXT4_SB(sb)->s_resize_lock); ext4_journal_stop(handle); err = -EBUSY; goto exit_put; @@ -1070,14 +1071,14 @@ int ext4_group_extend(struct super_block *sb, struct ext4_super_block *es, EXT4_SB(sb)->s_sbh))) { ext4_warning(sb, __func__, "error %d on journal write access", err); - unlock_super(sb); + mutex_unlock(&EXT4_SB(sb)->s_resize_lock); ext4_journal_stop(handle); goto exit_put; } ext4_blocks_count_set(es, o_blocks_count + add); ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh); sb->s_dirt = 1; - unlock_super(sb); + mutex_unlock(&EXT4_SB(sb)->s_resize_lock); ext4_debug("freeing blocks %llu through %llu\n", o_blocks_count, o_blocks_count + add); /* We add the blocks to the bitmap and set the group need init bit */ diff --git a/fs/ext4/super.c b/fs/ext4/super.c index c23e82c..b4735e4 100644 --- a/fs/ext4/super.c +++ b/fs/ext4/super.c @@ -2624,6 +2624,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent) #endif INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */ mutex_init(&sbi->s_orphan_lock); + mutex_init(&sbi->s_resize_lock); sb->s_root = NULL; -- 1.5.6.3