On Mon 01-09-25 15:01:45, Sun Yongjian wrote:
> 在 2025/7/31 22:05, sunyongjian@xxxxxxxxxxxxxxx 写道:
> Gentle ping.
> > From: Yongjian Sun <sunyongjian1@xxxxxxxxxx>
> >
> > After running a stress test combined with fault injection,
> > we performed fsck -a followed by fsck -fn on the filesystem
> > image. During the second pass, fsck -fn reported:
> >
> > Inode 131512, end of extent exceeds allowed value
> > (logical block 405, physical block 1180540, len 2)
> >
> > This inode was not in the orphan list.
Thanks for report! Interesting... Which kernel were you using?
> > Analysis revealed the
> > following call chain that leads to the inconsistency:
> >
> > ext4_da_write_end()
> > //does not update i_disksize
Right, for any write beyond i_disksize to unallocated blocks we update
i_disksize only during page writeback.
> > ext4_punch_hole()
> > //truncate folio, keep size
So here offset + len passed to ext4_punch_hole() is important. Because
there's ext4_update_disksize_before_punch() call which updates i_disksize
to i_size if the punched hole reaches EOF. So did you punch hole in the
middle of the file?
> > ext4_page_mkwrite()
> > ext4_block_page_mkwrite()
> > ext4_block_write_begin()
> > ext4_get_block()
> > //insert written extent without update i_disksize
We should insert unwritten extent here, shouldn't we? We use
ext4_get_block_unwritten() when we are inside i_size. Ah, you mention below
you use nodioread_nolock. Nasty :)
> > journal commit
> > echo 1 > /sys/block/xxx/device/delete
> >
> > da-write path updates i_size but does not update i_disksize. Then
> > ext4_punch_hole truncates the da-folio yet still leaves i_disksize
> > unchanged. Then ext4_page_mkwrite sees ext4_nonda_switch return 1
> > and takes the nodioread_nolock path, the folio about to be written
> > has just been punched out, and it’s offset sits beyond the current
> > i_disksize. This may result in a written extent being inserted, but
> > again does not update i_disksize. If the journal gets committed and
> > then the block device is yanked, we might run into this.
> >
> > To fix this, we now check in ext4_block_page_mkwrite whether
> > i_disksize needs to be updated to cover the newly allocated blocks.
> >
> > Signed-off-by: Yongjian Sun <sunyongjian1@xxxxxxxxxx>
Hum, rather than complicating this niche code what if we just
unconditionally used ext4_get_block_unwritten() in
ext4_block_page_mkwrite() when delalloc gets disabled? It is far from any
performance critical path. What do people think? The code would actually
have to be something like:
if (ext4_should_journal_data(inode))
get_block = ext4_get_block;
else
get_block = ext4_get_block_unwritten;
to properly handle data journalling. I'm adding Ritesh to CC because I do
remember there used to be some issues with dioread_nolock with blocksize <
pagesize which he was able to trigger. But I think they were fixed.
Honza
> > ---
> > fs/ext4/inode.c | 10 ++++++++++
> > 1 file changed, 10 insertions(+)
> >
> > diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
> > index ed54c4d0f2f9..050270b265ae 100644
> > --- a/fs/ext4/inode.c
> > +++ b/fs/ext4/inode.c
> > @@ -6666,8 +6666,18 @@ static int ext4_block_page_mkwrite(struct inode *inode, struct folio *folio,
> > goto out_error;
> > if (!ext4_should_journal_data(inode)) {
> > + loff_t disksize = folio_pos(folio) + len;
> > block_commit_write(folio, 0, len);
> > folio_mark_dirty(folio);
> > + if (disksize > READ_ONCE(EXT4_I(inode)->i_disksize)) {
> > + down_write(&EXT4_I(inode)->i_data_sem);
> > + if (disksize > EXT4_I(inode)->i_disksize)
> > + EXT4_I(inode)->i_disksize = disksize;
> > + up_write(&EXT4_I(inode)->i_data_sem);
> > + ret = ext4_mark_inode_dirty(handle, inode);
> > + if (ret)
> > + goto out_error;
> > + }
> > } else {
> > ret = ext4_journal_folio_buffers(handle, folio, len);
> > if (ret)
>
--
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR
