[PATCH v3 0/2] ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

[PATCH v3 0/2] ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

[Eric Biggers]

This series fixes a race between writepages and enabling EXT4_EXTENTS_FL
that could cause a WARN_ON() in ext4_add_complete_io() to be hit.  Patch
1 is a trivial renaming in preparation for patch 2 which is the actual
fix.  See patch 2 for the full details.

Changed in v3:
    Do the renaming in a separate patch.

Changed in v2:
    Instead of making ext4_writepages() read EXT4_EXTENTS_FL only once,
    make it so that EXT4_EXTENTS_FL can't be changed while
    ext4_writepages() is running.

Eric Biggers (2):
  ext4: rename s_journal_flag_rwsem to s_writepages_rwsem
  ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

 fs/ext4/ext4.h    |  7 +++++--
 fs/ext4/inode.c   | 14 +++++++-------
 fs/ext4/migrate.c | 27 +++++++++++++++++++--------
 fs/ext4/super.c   |  6 +++---
 4 files changed, 34 insertions(+), 20 deletions(-)


base-commit: c96dceeabf765d0b1b1f29c3bf50a5c01315b820
-- 
2.25.0

[PATCH v3 1/2] ext4: rename s_journal_flag_rwsem to s_writepages_rwsem

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

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 <ebiggers@google.com>
---
 fs/ext4/ext4.h  |  2 +-
 fs/ext4/inode.c | 14 +++++++-------
 fs/ext4/super.c |  6 +++---
 3 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 4441331d06cc4..487a7b430b9dd 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -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;
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index c04a15fc8b6ad..f49c48ea2f170 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -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);
diff --git a/fs/ext4/super.c b/fs/ext4/super.c
index b0b9150c97735..feb59c7ad395f 100644
--- a/fs/ext4/super.c
+++ b/fs/ext4/super.c
@@ -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
-- 
2.25.0

[PATCH v3 2/2] ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

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: syzbot+2202a584a00fffd19fbf@syzkaller.appspotmail.com
Fixes: 6b523df4fb5a ("ext4: use transaction reservation for extent conversion in ext4_end_io")
Cc: stable@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/ext4/ext4.h    |  5 ++++-
 fs/ext4/migrate.c | 27 +++++++++++++++++++--------
 2 files changed, 23 insertions(+), 9 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 487a7b430b9dd..0a59006c621a0 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -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
diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
index 89725fa425732..fb6520f371355 100644
--- a/fs/ext4/migrate.c
+++ b/fs/ext4/migrate.c
@@ -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;
 }
-- 
2.25.0

Re: [PATCH v3 1/2] ext4: rename s_journal_flag_rwsem to s_writepages_rwsem

[Jan Kara]

On Wed 19-02-20 10:30:46, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> 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 <ebiggers@google.com>

The patch looks good to me. You can add:

Reviewed-by: Jan Kara <jack@suse.cz>

								Honza

> ---
>  fs/ext4/ext4.h  |  2 +-
>  fs/ext4/inode.c | 14 +++++++-------
>  fs/ext4/super.c |  6 +++---
>  3 files changed, 11 insertions(+), 11 deletions(-)
> 
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 4441331d06cc4..487a7b430b9dd 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -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;
> diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> index c04a15fc8b6ad..f49c48ea2f170 100644
> --- a/fs/ext4/inode.c
> +++ b/fs/ext4/inode.c
> @@ -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);
> diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> index b0b9150c97735..feb59c7ad395f 100644
> --- a/fs/ext4/super.c
> +++ b/fs/ext4/super.c
> @@ -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
> -- 
> 2.25.0
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v3 2/2] ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

[Jan Kara]

On Wed 19-02-20 10:30:47, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> 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: syzbot+2202a584a00fffd19fbf@syzkaller.appspotmail.com
> Fixes: 6b523df4fb5a ("ext4: use transaction reservation for extent conversion in ext4_end_io")
> Cc: stable@kernel.org
> Signed-off-by: Eric Biggers <ebiggers@google.com>

The patch looks good to me. You can add:

Reviewed-by: Jan Kara <jack@suse.cz>

								Honza

> ---
>  fs/ext4/ext4.h    |  5 ++++-
>  fs/ext4/migrate.c | 27 +++++++++++++++++++--------
>  2 files changed, 23 insertions(+), 9 deletions(-)
> 
> diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
> index 487a7b430b9dd..0a59006c621a0 100644
> --- a/fs/ext4/ext4.h
> +++ b/fs/ext4/ext4.h
> @@ -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
> diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
> index 89725fa425732..fb6520f371355 100644
> --- a/fs/ext4/migrate.c
> +++ b/fs/ext4/migrate.c
> @@ -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;
>  }
> -- 
> 2.25.0
> 
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR

Re: [PATCH v3 1/2] ext4: rename s_journal_flag_rwsem to s_writepages_rwsem

[Theodore Y. Ts'o]

On Thu, Feb 20, 2020 at 10:14:58AM +0100, Jan Kara wrote:
> On Wed 19-02-20 10:30:46, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > 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 <ebiggers@google.com>
> 
> The patch looks good to me. You can add:
> 
> Reviewed-by: Jan Kara <jack@suse.cz>

Thanks, applied.

					- Ted

Re: [PATCH v3 2/2] ext4: fix race between writepages and enabling EXT4_EXTENTS_FL

[Theodore Y. Ts'o]

On Thu, Feb 20, 2020 at 10:15:48AM +0100, Jan Kara wrote:
> On Wed 19-02-20 10:30:47, Eric Biggers wrote:
> > From: Eric Biggers <ebiggers@google.com>
> > 
> > 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: syzbot+2202a584a00fffd19fbf@syzkaller.appspotmail.com
> > Fixes: 6b523df4fb5a ("ext4: use transaction reservation for extent conversion in ext4_end_io")
> > Cc: stable@kernel.org
> > Signed-off-by: Eric Biggers <ebiggers@google.com>
> 
> The patch looks good to me. You can add:
> 
> Reviewed-by: Jan Kara <jack@suse.cz>

Thanks, applied.

						- Ted