ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

From: Eric Biggers <[email protected]>

In preparation for making s_journal_flag_rwsem synchronize ext4_writepages() with changes to both the EXTENTS and JOURNAL_DATA flags (rather than just JOURNAL_DATA as it does currently), rename it to s_writepages_rwsem.

Signed-off-by Eric Biggers <[email protected]>

---
 fs/ext4/ext4.h  |  2 +-
 fs/ext4/inode.c | 14 +++++++-------
 fs/ext4/super.c |  6 +++---
 3 files changed, 11 insertions(+), 11 deletions(-)

fs/ext4/ext4.h +1 -1

index 4441331d06cc4..487a7b430b9dd 100644
@@ -1553,7 +1553,7 @@ struct ext4_sb_info {
 	struct ratelimit_state s_msg_ratelimit_state;
 
 	/* Barrier between changing inodes' journal flags and writepages ops. */
-	struct percpu_rw_semaphore s_journal_flag_rwsem;
+	struct percpu_rw_semaphore s_writepages_rwsem;
 	struct dax_device *s_daxdev;
 #ifdef CONFIG_EXT4_DEBUG
 	unsigned long s_simulate_fail;

fs/ext4/inode.c +7 -7

index c04a15fc8b6ad..f49c48ea2f170 100644
@@ -2628,7 +2628,7 @@ static int ext4_writepages(struct address_space *mapping,
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
 
-	percpu_down_read(&sbi->s_journal_flag_rwsem);
+	percpu_down_read(&sbi->s_writepages_rwsem);
 	trace_ext4_writepages(inode, wbc);
 
 	/*

@@ -2849,7 +2849,7 @@ static int ext4_writepages(struct address_space *mapping,
 out_writepages:
 	trace_ext4_writepages_result(inode, wbc, ret,
 				     nr_to_write - wbc->nr_to_write);
-	percpu_up_read(&sbi->s_journal_flag_rwsem);
+	percpu_up_read(&sbi->s_writepages_rwsem);
 	return ret;
 }
 

@@ -2864,13 +2864,13 @@ static int ext4_dax_writepages(struct address_space *mapping,
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
 
-	percpu_down_read(&sbi->s_journal_flag_rwsem);
+	percpu_down_read(&sbi->s_writepages_rwsem);
 	trace_ext4_writepages(inode, wbc);
 
 	ret = dax_writeback_mapping_range(mapping, inode->i_sb->s_bdev, wbc);
 	trace_ext4_writepages_result(inode, wbc, ret,
 				     nr_to_write - wbc->nr_to_write);
-	percpu_up_read(&sbi->s_journal_flag_rwsem);
+	percpu_up_read(&sbi->s_writepages_rwsem);
 	return ret;
 }
 

@@ -5861,7 +5861,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 		}
 	}
 
-	percpu_down_write(&sbi->s_journal_flag_rwsem);
+	percpu_down_write(&sbi->s_writepages_rwsem);
 	jbd2_journal_lock_updates(journal);
 
 	/*

@@ -5878,7 +5878,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 		err = jbd2_journal_flush(journal);
 		if (err < 0) {
 			jbd2_journal_unlock_updates(journal);
-			percpu_up_write(&sbi->s_journal_flag_rwsem);
+			percpu_up_write(&sbi->s_writepages_rwsem);
 			return err;
 		}
 		ext4_clear_inode_flag(inode, EXT4_INODE_JOURNAL_DATA);

@@ -5886,7 +5886,7 @@ int ext4_change_inode_journal_flag(struct inode *inode, int val)
 	ext4_set_aops(inode);
 
 	jbd2_journal_unlock_updates(journal);
-	percpu_up_write(&sbi->s_journal_flag_rwsem);
+	percpu_up_write(&sbi->s_writepages_rwsem);
 
 	if (val)
 		up_write(&EXT4_I(inode)->i_mmap_sem);

fs/ext4/super.c +3 -3

index b0b9150c97735..feb59c7ad395f 100644
@@ -1054,7 +1054,7 @@ static void ext4_put_super(struct super_block *sb)
 	percpu_counter_destroy(&sbi->s_freeinodes_counter);
 	percpu_counter_destroy(&sbi->s_dirs_counter);
 	percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
-	percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
+	percpu_free_rwsem(&sbi->s_writepages_rwsem);
 #ifdef CONFIG_QUOTA
 	for (i = 0; i < EXT4_MAXQUOTAS; i++)
 		kfree(get_qf_name(sb, sbi, i));

@@ -4600,7 +4600,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 		err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
 					  GFP_KERNEL);
 	if (!err)
-		err = percpu_init_rwsem(&sbi->s_journal_flag_rwsem);
+		err = percpu_init_rwsem(&sbi->s_writepages_rwsem);
 
 	if (err) {
 		ext4_msg(sb, KERN_ERR, "insufficient memory");

@@ -4694,7 +4694,7 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 	percpu_counter_destroy(&sbi->s_freeinodes_counter);
 	percpu_counter_destroy(&sbi->s_dirs_counter);
 	percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
-	percpu_free_rwsem(&sbi->s_journal_flag_rwsem);
+	percpu_free_rwsem(&sbi->s_writepages_rwsem);
 failed_mount5:
 	ext4_ext_release(sb);
 	ext4_release_system_zone(sb);

base-commit: c96dceeabf765d0b1b1f29c3bf50a5c01315b820

Signature · 1 line

2.25.0

Quoted 7 lines · Eric Biggers

From: Eric Biggers <[email protected]>

In preparation for making s_journal_flag_rwsem synchronize ext4_writepages() with changes to both the EXTENTS and JOURNAL_DATA flags (rather than just JOURNAL_DATA as it does currently), rename it to s_writepages_rwsem.

Quoted 1 line

Signed-off-by Eric Biggers <[email protected]>

The patch looks good to me. You can add:

Reviewed-by Jan Kara <[email protected]>

From: Eric Biggers <[email protected]>

If EXT4_EXTENTS_FL is set on an inode while ext4_writepages() is running on it, the following warning in ext4_add_complete_io() can be hit:

WARNING: CPU: 1 PID: 0 at fs/ext4/page-io.c:234 ext4_put_io_end_defer+0xf0/0x120

Here's a minimal reproducer (not 100% reliable) (root isn't required):

while true; do
sync
done &
while true; do
rm -f file
touch file
chattr -e file
echo X >> file
chattr +e file
done

The problem is that in ext4_writepages(), ext4_should_dioread_nolock() (which only returns true on extent-based files) is checked once to set the number of reserved journal credits, and also again later to select the flags for ext4_map_blocks() and copy the reserved journal handle to ext4_io_end::handle. But if EXT4_EXTENTS_FL is being concurrently set, the first check can see dioread_nolock disabled while the later one can see it enabled, causing the reserved handle to unexpectedly be NULL.

Since changing EXT4_EXTENTS_FL is uncommon, and there may be other races related to doing so as well, fix this by synchronizing changing EXT4_EXTENTS_FL with ext4_writepages() via the existing s_writepages_rwsem (previously called s_journal_flag_rwsem).

This was originally reported by syzbot without a reproducer at
https://syzkaller.appspot.com/bug?extid=2202a584a00fffd19fbf,
but now that dioread_nolock is the default I also started seeing this when running syzkaller locally.

Reported-by [email protected]

Fixes 6b523df4fb5a ("ext4: use transaction reservation for extent conversion in ext4_end_io")

Cc [email protected]

Signed-off-by Eric Biggers <[email protected]>

---
 fs/ext4/ext4.h    |  5 ++++-
 fs/ext4/migrate.c | 27 +++++++++++++++++++--------
 2 files changed, 23 insertions(+), 9 deletions(-)

fs/ext4/ext4.h +4 -1

index 487a7b430b9dd..0a59006c621a0 100644
@@ -1552,7 +1552,10 @@ struct ext4_sb_info {
 	struct ratelimit_state s_warning_ratelimit_state;
 	struct ratelimit_state s_msg_ratelimit_state;
 
-	/* Barrier between changing inodes' journal flags and writepages ops. */
+	/*
+	 * Barrier between writepages ops and changing any inode's JOURNAL_DATA
+	 * or EXTENTS flag.
+	 */
 	struct percpu_rw_semaphore s_writepages_rwsem;
 	struct dax_device *s_daxdev;
 #ifdef CONFIG_EXT4_DEBUG

fs/ext4/migrate.c +19 -8

index 89725fa425732..fb6520f371355 100644
@@ -407,6 +407,7 @@ static int free_ext_block(handle_t *handle, struct inode *inode)
 
 int ext4_ext_migrate(struct inode *inode)
 {
+	struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
 	handle_t *handle;
 	int retval = 0, i;
 	__le32 *i_data;

@@ -431,6 +432,8 @@ int ext4_ext_migrate(struct inode *inode)
 		 */
 		return retval;
 
+	percpu_down_write(&sbi->s_writepages_rwsem);
+
 	/*
 	 * Worst case we can touch the allocation bitmaps, a bgd
 	 * block, and a block to link in the orphan list.  We do need

@@ -441,7 +444,7 @@ int ext4_ext_migrate(struct inode *inode)
 
 	if (IS_ERR(handle)) {
 		retval = PTR_ERR(handle);
-		return retval;
+		goto out_unlock;
 	}
 	goal = (((inode->i_ino - 1) / EXT4_INODES_PER_GROUP(inode->i_sb)) *
 		EXT4_INODES_PER_GROUP(inode->i_sb)) + 1;

@@ -452,7 +455,7 @@ int ext4_ext_migrate(struct inode *inode)
 	if (IS_ERR(tmp_inode)) {
 		retval = PTR_ERR(tmp_inode);
 		ext4_journal_stop(handle);
-		return retval;
+		goto out_unlock;
 	}
 	i_size_write(tmp_inode, i_size_read(inode));
 	/*

@@ -494,7 +497,7 @@ int ext4_ext_migrate(struct inode *inode)
 		 */
 		ext4_orphan_del(NULL, tmp_inode);
 		retval = PTR_ERR(handle);
-		goto out;
+		goto out_tmp_inode;
 	}
 
 	ei = EXT4_I(inode);

@@ -576,10 +579,11 @@ int ext4_ext_migrate(struct inode *inode)
 	ext4_ext_tree_init(handle, tmp_inode);
 out_stop:
 	ext4_journal_stop(handle);
-out:
+out_tmp_inode:
 	unlock_new_inode(tmp_inode);
 	iput(tmp_inode);
-
+out_unlock:
+	percpu_up_write(&sbi->s_writepages_rwsem);
 	return retval;
 }
 

@@ -589,7 +593,8 @@ int ext4_ext_migrate(struct inode *inode)
 int ext4_ind_migrate(struct inode *inode)
 {
 	struct ext4_extent_header	*eh;
-	struct ext4_super_block		*es = EXT4_SB(inode->i_sb)->s_es;
+	struct ext4_sb_info		*sbi = EXT4_SB(inode->i_sb);
+	struct ext4_super_block		*es = sbi->s_es;
 	struct ext4_inode_info		*ei = EXT4_I(inode);
 	struct ext4_extent		*ex;
 	unsigned int			i, len;

@@ -613,9 +618,13 @@ int ext4_ind_migrate(struct inode *inode)
 	if (test_opt(inode->i_sb, DELALLOC))
 		ext4_alloc_da_blocks(inode);
 
+	percpu_down_write(&sbi->s_writepages_rwsem);
+
 	handle = ext4_journal_start(inode, EXT4_HT_MIGRATE, 1);
-	if (IS_ERR(handle))
-		return PTR_ERR(handle);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		goto out_unlock;
+	}
 
 	down_write(&EXT4_I(inode)->i_data_sem);
 	ret = ext4_ext_check_inode(inode);

@@ -650,5 +659,7 @@ int ext4_ind_migrate(struct inode *inode)
 errout:
 	ext4_journal_stop(handle);
 	up_write(&EXT4_I(inode)->i_data_sem);
+out_unlock:
+	percpu_up_write(&sbi->s_writepages_rwsem);
 	return ret;
 }

Signature · 1 line

2.25.0

Quoted 38 lines · Eric Biggers

From: Eric Biggers <[email protected]>

If EXT4_EXTENTS_FL is set on an inode while ext4_writepages() is running on it, the following warning in ext4_add_complete_io() can be hit:

WARNING: CPU: 1 PID: 0 at fs/ext4/page-io.c:234 ext4_put_io_end_defer+0xf0/0x120

Here's a minimal reproducer (not 100% reliable) (root isn't required):

while true; do
sync
done &
while true; do
rm -f file
touch file
chattr -e file
echo X >> file
chattr +e file
done

The problem is that in ext4_writepages(), ext4_should_dioread_nolock() (which only returns true on extent-based files) is checked once to set the number of reserved journal credits, and also again later to select the flags for ext4_map_blocks() and copy the reserved journal handle to ext4_io_end::handle. But if EXT4_EXTENTS_FL is being concurrently set, the first check can see dioread_nolock disabled while the later one can see it enabled, causing the reserved handle to unexpectedly be NULL.

Since changing EXT4_EXTENTS_FL is uncommon, and there may be other races related to doing so as well, fix this by synchronizing changing EXT4_EXTENTS_FL with ext4_writepages() via the existing s_writepages_rwsem (previously called s_journal_flag_rwsem).

This was originally reported by syzbot without a reproducer at
https://syzkaller.appspot.com/bug?extid=2202a584a00fffd19fbf,
but now that dioread_nolock is the default I also started seeing this when running syzkaller locally.

Quoted 4 lines

Reported-by [email protected]

Fixes 6b523df4fb5a ("ext4: use transaction reservation for extent conversion in ext4_end_io")

Cc [email protected]

Signed-off-by Eric Biggers <[email protected]>

The patch looks good to me. You can add:

Reviewed-by Jan Kara <[email protected]>