Re: [PATCH] ext4: cache es->s_journal_inum in ext4_sb_info

[Baokun Li]

On 2025/3/26 1:57, Ojaswin Mujoo wrote:
> On Tue, Mar 18, 2025 at 10:31:29PM -0400, Theodore Ts'o wrote:
> > On Tue, Mar 18, 2025 at 01:42:31PM +0530, Ojaswin Mujoo wrote:
> > > > So this is something we need to do if the journal is actived, and if
> > > > it's active, then sbi->s_journal will be non-NULL, and so we can just
> > > > check to see if inode == sbi->s_journal instead.  This will simplify
> > > I believe you mean inode == sbi->s_journal->j_inode here right?
> > Yes, that's what I meant; sorry for the not catching this before I
> > sent my reply.
> >
> > Cheers,
> >
> > 					- Ted
> Hi Ted, Baokun,
>
> I got some time to revisit this. Seems like checking against
> s_journal->j_inode is not enough. This is because both
> ext4_check_blockref() and check_block_validity() can be called even
> before journal->j_inode is set:
>
> ext4_open_inode_journal
>    ext4_get_journal_inode
> 	  __ext4_iget
> 		  ext4_ind_check_inode
> 			  ext4_check_blockref  /* j_inode not set */
>
>    journal = jbd2_journal_init_inode
> 	  bmap
> 		  ext4_bmap
> 			 iomap_bmap
> 			   ext4_iomap_begin
> 				   ext4_map_blocks
> 					   check_block_validity
>
>    journal->j_inode = inode
>
>
> Now, I think in this case the best solution might be to use the extra
> field like we do in this patch but set  EXT4_SB(sb)->s_journal_ino
> sufficiently early.
>
> Thoughts?

Because system zone setup happens after the journal are loaded, I think we
can skip the check if the journal haven't been loaded yet, like this:

diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index d04d8a7f12e7..38dc72ff7e78 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -383,9 +383,10 @@ static int __check_block_validity(struct inode 
*inode, const char *func,
                                unsigned int line,
                                struct ext4_map_blocks *map)
 {
+       journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
+
        if (ext4_has_feature_journal(inode->i_sb) &&
-           (inode->i_ino ==
- le32_to_cpu(EXT4_SB(inode->i_sb)->s_es->s_journal_inum)))
+           (!journal || inode == journal->j_inode))
                return 0;
        if (!ext4_inode_block_valid(inode, map->m_pblk, map->m_len)) {
                ext4_error_inode(inode, func, line, map->m_pblk,

If any part of the journal area overlaps with the system zone, we'll catch
it when we add the journal area to the system zone later.


Cheers,
Baokun