Skip to contentSkip to conversation

one writeback regression fix for .36

GIT PULL Jens Axboe 4 participants 5 messages Oct 6, 2010 VFS

lore
Share & tools
lore

Jens Axboe

author
#0 lore raw

Hi Linus,

The recent patch series from Jan introduced a WARN_ON() regression. The real fix for this is adding a super_operations->get_bdi(), which we'll do for 2.6.37.

Please pull.

are available in the git repository at:
git://git.kernel.dk/linux-2.6-block.git for-linus

Christoph Hellwig (1):

      writeback: always use sb->s_bdi for writeback purposes
 fs/fs-writeback.c |   19 ++++---------------
 1 files changed, 4 insertions(+), 15 deletions(-)
fs/fs-writeback.c +4 -15
index 5581122..ab38fef 100644
@@ -72,22 +72,11 @@ int writeback_in_progress(struct backing_dev_info *bdi)
 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
 {
 	struct super_block *sb = inode->i_sb;
-	struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info;
 
-	/*
-	 * For inodes on standard filesystems, we use superblock's bdi. For
-	 * inodes on virtual filesystems, we want to use inode mapping's bdi
-	 * because they can possibly point to something useful (think about
-	 * block_dev filesystem).
-	 */
-	if (sb->s_bdi && sb->s_bdi != &noop_backing_dev_info) {
-		/* Some device inodes could play dirty tricks. Catch them... */
-		WARN(bdi != sb->s_bdi && bdi_cap_writeback_dirty(bdi),
-			"Dirtiable inode bdi %s != sb bdi %s\n",
-			bdi->name, sb->s_bdi->name);
-		return sb->s_bdi;
-	}
-	return bdi;
+	if (strcmp(sb->s_type->name, "bdev") == 0)
+		return inode->i_mapping->backing_dev_info;
+
+	return sb->s_bdi;
 }
 
 static void bdi_queue_work(struct backing_dev_info *bdi,
Signature · 1 line
Jens Axboe
#1 lore raw
Quoted 5 lines · Jens Axboe

The recent patch series from Jan introduced a WARN_ON() regression. The real fix for this is adding a super_operations->get_bdi(), which we'll do for 2.6.37.

Quoted 7 lines

Please pull.

are available in the git repository at:
 git://git.kernel.dk/linux-2.6-block.git for-linus

Christoph Hellwig (1):
     writeback: always use sb->s_bdi for writeback purposes

This is a f*cking disgrace. It's now the second patch I see during this release window that works by parsing random strings in the block device layer.

This needs to stop. There's something seriously wrong in the whole subsystem. This kind of hackery is a disease, and it seems to be endemic.

So let's just make sure that 2.6.37 really does clean these things up. And dammit, I don't want to see more random crap added. There has been too much crazyness going on, with too little taste in the whole writeback and IO scheduler area. Jens, you need to put the brakes on. No more crap. Really. Make 2.6.37 a stabilization and cleanup release, without new crazy features or clever tweaking. Ok? Because writeback was a disaster in 2.6.35, and it's been this kind of crazy ugly in 2.6.36.

So I'm not taking any more writeback changes unless I feel that you are actively working on cleaning up the crazy tasteless crap without adding new code.

Linus

maintainer
#2 lore raw
Quoted 10 lines · Linus Torvalds
Quoted 2 lines

Christoph Hellwig (1):
? ? ?writeback: always use sb->s_bdi for writeback purposes

This is a f*cking disgrace. It's now the second patch I see during this release window that works by parsing random strings in the block device layer.

This needs to stop. There's something seriously wrong in the whole subsystem. This kind of hackery is a disease, and it seems to be endemic.

Eh... I'm no fonder of that than you are (and strcmp is fucking stupid), but... that thing is really sb_is_blkdev_sb() trying to make a comeback in fs/fs-writeback.c. IOW, it's sick, but not for the reasons you are mentioning; strcmp() use is trivially removable.

author
#4 lore raw
In reply to Linus Torvalds
Quoted 14 lines · Linus Torvalds
Quoted 5 lines · Jens Axboe

The recent patch series from Jan introduced a WARN_ON() regression. The real fix for this is adding a super_operations->get_bdi(), which we'll do for 2.6.37.

Quoted 7 lines

Please pull.

are available in the git repository at:
git://git.kernel.dk/linux-2.6-block.git for-linus

Christoph Hellwig (1):

     writeback: always use sb->s_bdi for writeback purposes
Quoted 7 lines

This is a f*cking disgrace. It's now the second patch I see during this release window that works by parsing random strings in the block device layer.

This needs to stop. There's something seriously wrong in the whole subsystem. This kind of hackery is a disease, and it seems to be endemic.

I'm not enjoying this patch anymore than you are, trust me. As Al mentioned, it's not as bad as it appears (which is good, because it appears like a bastard). We could work around this ugliness, but at this point it'll just be a prettied up turd anyway. So may as well go for the simple variant of that at least.

Quoted 9 lines

So let's just make sure that 2.6.37 really does clean these things up. And dammit, I don't want to see more random crap added. There has been too much crazyness going on, with too little taste in the whole writeback and IO scheduler area. Jens, you need to put the brakes on. No more crap. Really. Make 2.6.37 a stabilization and cleanup release, without new crazy features or clever tweaking. Ok? Because writeback was a disaster in 2.6.35, and it's been this kind of crazy ugly in 2.6.36.

Quoted 3 lines

So I'm not taking any more writeback changes unless I feel that you are actively working on cleaning up the crazy tasteless crap without adding new code.

Agree, there's no complex stuff queued up for .37. Brakes are on, we do need a stability release on that side.

Signature · 1 line
Jens Axboe
#3 lore raw
In reply to Al Viro
Quoted 4 lines · Al Viro

Eh... I'm no fonder of that than you are (and strcmp is fucking stupid), but... that thing is really sb_is_blkdev_sb() trying to make a comeback in fs/fs-writeback.c. IOW, it's sick, but not for the reasons you are mentioning; strcmp() use is trivially removable.

The alternative basically is exporting bd_type so that it can be used here and comparing ->s_type. The real problem is that right now the internal block device filesystem is the only one having more than one backing device per filesystem. I have patches to generalize that, which is something we'll need for writeback performance on large filesystem sooner or later. I was planning to get this ready for 2.6.37, but if you prefer no big chance in this area I can calm down and move other things up on my TODO list.