Re: linux-2.6.20-rc4-mm1 Reiser4 filesystem freeze and corruption

[Edward Shishkin <edward@namesys.com>]

Zan Lynx wrote:

> On Sat, 2007-01-20 at 03:34 +0300, Vladimir V. Saveliev wrote:
>  
>
> > Hello
> >
> > On Friday 19 January 2007 20:58, Zan Lynx wrote:
> >    
> >
> > > I have been running 2.6.20-rc2-mm1 without problems, but both rc3-mm1
> > > and rc4-mm1 have been giving me these freezes.  They were happening
> > > inside X and without external console it was impossible to get anything,
> > > plus I was reluctant to test it since the freeze sometimes requires a
> > > full fsck.reiser4 --build-fs to recover the filesystem.
> > >
> > > But I finally got some output in a console session.  I wasn't able to
> > > get it all, I made some notes of what I think the problem is.  I may try
> > > again later once I get netconsole working (netconsole fails as a
> > > built-in, I'll try it as a module next).
> > >      
> [snip]
>  
>
> > yes, please provide more information. Full kernel output at time of freeze is very desirable.
> >    
>
> Here comes a full sized bug report, as best as I can do it.  This is
> kernel 2.6.20-rc6-mm3 instead of rc4-mm1.  Still has the problem.
>  

Thanks for the dump.

> [ 3138.456588]  [<ffffffff8033f5de>] current_atom_finish_all_fq+0x12e/0x280
> [ 3138.456661]  [<ffffffff80296510>] autoremove_wake_function+0x0/0x30
> [ 3138.456674]  [<ffffffff803350ac>] submit_wb_list+0x11c/0x130
> [ 3138.456690]  [<ffffffff80335409>] reiser4_txn_end+0x349/0x530
> [ 3138.456710]  [<ffffffff803355f9>] reiser4_txn_restart+0x9/0x20
> [ 3138.456781]  [<ffffffff80335680>] force_commit_atom+0x50/0x60
> [ 3138.456793]  [<ffffffff8034cfb1>] writepages_unix_file+0x671/0x780
> [ 3138.456824]  [<ffffffff802590b3>] do_writepages+0x43/0x80
> [ 3138.456838]  [<ffffffff8024dbf8>] __filemap_fdatawrite_range+0x58/0x70
> [ 3138.456914]  [<ffffffff8024e19d>] do_fsync+0x3d/0xe0
> [ 3138.456930]  [<ffffffff802c2473>] sys_msync+0x143/0x1f0
> [ 3138.456945]  [<ffffffff8025c11e>] system_call+0x7e/0x83
>  

This is waiting for IO completion, and no success because of new plugging
policy introduced by block layer folks. The attached patch should help.
Andrew, please apply.

Thanks,
Edward.

Signed-off-by: Edward Shishkin <edward@namesys.com>
---
 linux-2.6.20-rc6-mm3/fs/reiser4/status_flags.c |    2 ++
 linux-2.6.20-rc6-mm3/fs/reiser4/wander.c       |   18 +++++++++++-------
 2 files changed, 13 insertions(+), 7 deletions(-)

--- linux-2.6.20-rc6-mm3/fs/reiser4/status_flags.c.orig
+++ linux-2.6.20-rc6-mm3/fs/reiser4/status_flags.c
@@ -63,6 +63,7 @@
 	}
 	lock_page(page);
 	submit_bio(READ, bio);
+	blk_replug_current_nested();
 	wait_on_page_locked(page);
 	if (!PageUptodate(page)) {
 		warning("green-2007",
@@ -157,6 +158,7 @@
 	lock_page(get_super_private(sb)->status_page);	// Safe as nobody should touch our page.
 	/* We can block now, but we have no other choice anyway */
 	submit_bio(WRITE, bio);
+	blk_replug_current_nested();
 	return 0;		// We do not wait for io to finish.
 }
 
--- linux-2.6.20-rc6-mm3/fs/reiser4/wander.c.orig
+++ linux-2.6.20-rc6-mm3/fs/reiser4/wander.c
@@ -718,6 +718,7 @@
 	jnode *first, int nr, const reiser4_block_nr *block_p,
 	flush_queue_t *fq, int flags)
 {
+	int ret = 0;
 	struct super_block *super = reiser4_get_current_sb();
 	int write_op = ( flags & WRITEOUT_BARRIER ) ? WRITE_BARRIER : WRITE;
 	int max_blocks;
@@ -738,9 +739,10 @@
 		int nr_used;
 
 		bio = bio_alloc(GFP_NOIO, nr_blocks);
-		if (!bio)
-			return RETERR(-ENOMEM);
-
+		if (!bio) {
+			ret = RETERR(-ENOMEM);
+			break;
+		}
 		bio->bi_bdev = super->s_bdev;
 		bio->bi_sector = block * (super->s_blocksize >> 9);
 		for (nr_used = 0, i = 0; i < nr_blocks; i++) {
@@ -843,8 +845,10 @@
 				reiser4_submit_bio(write_op, bio);
 				not_supported = bio_flagged(bio, BIO_EOPNOTSUPP);
 				bio_put(bio);
-				if (not_supported)
-					return -EOPNOTSUPP;
+				if (not_supported) {
+					ret = -EOPNOTSUPP;
+					break;
+				}
 			}
 
 			block += nr_used - 1;
@@ -855,8 +859,8 @@
 		}
 		nr -= nr_used;
 	}
-
-	return 0;
+	blk_replug_current_nested();
+	return ret;
 }
 
 /* This is a procedure which recovers a contiguous sequences of disk block