ext4: fast commit may miss file actions
authorXin Yin <yinxin.x@bytedance.com>
Mon, 17 Jan 2022 09:36:55 +0000 (17:36 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 8 Mar 2022 18:12:32 +0000 (19:12 +0100)
[ Upstream commit bdc8a53a6f2f0b1cb5f991440f2100732299eb93 ]

in the follow scenario:
1. jbd start transaction n
2. task A get new handle for transaction n+1
3. task A do some actions and add inode to FC_Q_MAIN fc_q
4. jbd complete transaction n and clear FC_Q_MAIN fc_q
5. task A call fsync

Fast commit will lost the file actions during a full commit.

we should also add updates to staging queue during a full commit.
and in ext4_fc_cleanup(), when reset a inode's fc track range, check
it's i_sync_tid, if it bigger than current transaction tid, do not
rest it, or we will lost the track range.

And EXT4_MF_FC_COMMITTING is not needed anymore, so drop it.

Signed-off-by: Xin Yin <yinxin.x@bytedance.com>
Link: https://lore.kernel.org/r/20220117093655.35160-3-yinxin.x@bytedance.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/ext4/ext4.h
fs/ext4/fast_commit.c
fs/ext4/super.c

index 2b67c7d1677bbe9e9eac2a319d97364208be57c1..db981619f6c8ef0faecff8fa713d3039003944ab 100644 (file)
@@ -1792,10 +1792,7 @@ static inline int ext4_valid_inum(struct super_block *sb, unsigned long ino)
 enum {
        EXT4_MF_MNTDIR_SAMPLED,
        EXT4_MF_FS_ABORTED,     /* Fatal error detected */
-       EXT4_MF_FC_INELIGIBLE,  /* Fast commit ineligible */
-       EXT4_MF_FC_COMMITTING   /* File system underoing a fast
-                                * commit.
-                                */
+       EXT4_MF_FC_INELIGIBLE   /* Fast commit ineligible */
 };
 
 static inline void ext4_set_mount_flag(struct super_block *sb, int bit)
index 96fac6ac18f468b907ec4ad42fa229d2406c19b9..285c91b0166c0f48d00e381d3c31b8362ed5feaa 100644 (file)
@@ -377,7 +377,8 @@ static int ext4_fc_track_template(
        spin_lock(&sbi->s_fc_lock);
        if (list_empty(&EXT4_I(inode)->i_fc_list))
                list_add_tail(&EXT4_I(inode)->i_fc_list,
-                               (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_COMMITTING)) ?
+                               (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
+                                sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING) ?
                                &sbi->s_fc_q[FC_Q_STAGING] :
                                &sbi->s_fc_q[FC_Q_MAIN]);
        spin_unlock(&sbi->s_fc_lock);
@@ -430,7 +431,8 @@ static int __track_dentry_update(struct inode *inode, void *arg, bool update)
        node->fcd_name.len = dentry->d_name.len;
 
        spin_lock(&sbi->s_fc_lock);
-       if (ext4_test_mount_flag(inode->i_sb, EXT4_MF_FC_COMMITTING))
+       if (sbi->s_journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
+               sbi->s_journal->j_flags & JBD2_FAST_COMMIT_ONGOING)
                list_add_tail(&node->fcd_list,
                                &sbi->s_fc_dentry_q[FC_Q_STAGING]);
        else
@@ -894,7 +896,6 @@ static int ext4_fc_submit_inode_data_all(journal_t *journal)
        int ret = 0;
 
        spin_lock(&sbi->s_fc_lock);
-       ext4_set_mount_flag(sb, EXT4_MF_FC_COMMITTING);
        list_for_each_entry(ei, &sbi->s_fc_q[FC_Q_MAIN], i_fc_list) {
                ext4_set_inode_state(&ei->vfs_inode, EXT4_STATE_FC_COMMITTING);
                while (atomic_read(&ei->i_fc_updates)) {
@@ -1212,7 +1213,8 @@ static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
                list_del_init(&iter->i_fc_list);
                ext4_clear_inode_state(&iter->vfs_inode,
                                       EXT4_STATE_FC_COMMITTING);
-               ext4_fc_reset_inode(&iter->vfs_inode);
+               if (iter->i_sync_tid <= tid)
+                       ext4_fc_reset_inode(&iter->vfs_inode);
                /* Make sure EXT4_STATE_FC_COMMITTING bit is clear */
                smp_mb();
 #if (BITS_PER_LONG < 64)
@@ -1241,7 +1243,6 @@ static void ext4_fc_cleanup(journal_t *journal, int full, tid_t tid)
        list_splice_init(&sbi->s_fc_q[FC_Q_STAGING],
                                &sbi->s_fc_q[FC_Q_MAIN]);
 
-       ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
        if (tid >= sbi->s_fc_ineligible_tid) {
                sbi->s_fc_ineligible_tid = 0;
                ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
index eb5d55cf277ce230f4eb727a4674664dd7954c82..fd4d34deb9fc6dec98a1114c067b401ce40bbfe3 100644 (file)
@@ -4619,7 +4619,6 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
        INIT_LIST_HEAD(&sbi->s_fc_dentry_q[FC_Q_STAGING]);
        sbi->s_fc_bytes = 0;
        ext4_clear_mount_flag(sb, EXT4_MF_FC_INELIGIBLE);
-       ext4_clear_mount_flag(sb, EXT4_MF_FC_COMMITTING);
        sbi->s_fc_ineligible_tid = 0;
        spin_lock_init(&sbi->s_fc_lock);
        memset(&sbi->s_fc_stats, 0, sizeof(sbi->s_fc_stats));