[PATCH 0/2] ext4: Fix two bugs in journalled writepages rework

[PATCH 0/2] ext4: Fix two bugs in journalled writepages rework

[Jan Kara]

Hello!

These two patches fix two problems introduced by the rewrite of journalled
writeback path.

								Honza

[PATCH 1/2] ext4: Fix data=journal writeback of DMA pinned page

[Jan Kara]

The condition in ext4_writepages() checking whether the filesystem is
frozen had a brown paper bag bug resulting in us never writing back
DMA pinned pages. Fix the condition.

Reported-by: Eric Biggers <ebiggers@kernel.org>
Link: https://lore.kernel.org/all/20230319183617.GA896@sol.localdomain
Fixes: 18b7f4107219 ("ext4: Fix warnings when freezing filesystem with journaled data")
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/ext4/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 3e311c85df08..15bac8181798 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2436,7 +2436,7 @@ static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
 	 * loop clears them.
 	 */
 	if (ext4_should_journal_data(mpd->inode) &&
-	    sb->s_writers.frozen >= SB_FREEZE_FS) {
+	    sb->s_writers.frozen < SB_FREEZE_FS) {
 		handle = ext4_journal_start(mpd->inode, EXT4_HT_WRITE_PAGE,
 					    bpp);
 		if (IS_ERR(handle))
-- 
2.35.3

[PATCH 2/2] ext4: Fix crash on shutdown filesystem

[Jan Kara]

Test generic/388 triggered a crash in mpage_release_unused_pages()
because a page in mpd->first_page..mpd->next_page range was not locked.
This can happen in data=journal mode when we exit from
mpage_prepare_extent_to_map() before actually initializing
mpd->next_page. Move the initialization to a place before we can exit
with error from mpage_prepare_extent_to_map().

Fixes: f7233fb54d18 ("ext4: Convert data=journal writeback to use ext4_writepages()")
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/ext4/inode.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index 15bac8181798..dbcc8b48c7ba 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -2428,6 +2428,8 @@ static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
 	else
 		tag = PAGECACHE_TAG_DIRTY;
 
+	mpd->map.m_len = 0;
+	mpd->next_page = index;
 	/*
 	 * Start a transaction for writeback of journalled data. We don't start
 	 * the transaction if the filesystem is frozen. In that case we
@@ -2443,8 +2445,6 @@ static int mpage_prepare_extent_to_map(struct mpage_da_data *mpd)
 			return PTR_ERR(handle);
 	}
 	folio_batch_init(&fbatch);
-	mpd->map.m_len = 0;
-	mpd->next_page = index;
 	while (index <= end) {
 		nr_folios = filemap_get_folios_tag(mapping, &index, end,
 				tag, &fbatch);
-- 
2.35.3

Re: [PATCH 0/2] ext4: Fix two bugs in journalled writepages rework

[Theodore Ts'o]

On Thu, Mar 23, 2023 at 03:53:57PM +0100, Jan Kara wrote:
> Hello!
> 
> These two patches fix two problems introduced by the rewrite of journalled
> writeback path.

I've folded these patches into the "ext4: Covert data=journal
writeback to use ext4_writepages()" commit in my tree.  Thanks,

	     	 		    	      - Ted