linux-ext4 - [RFC][PATCH] set_page_buffer_dirty should skip unmapped buffers

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-Id: <1157125829.30578.6.camel@dyn9047017100.beaverton.ibm.com>
Date:	Fri, 01 Sep 2006 08:50:29 -0700
From:	Badari Pulavarty <pbadari@...ibm.com>
To:	sct@...hat.com, akpm@...l.org
Cc:	linux-fsdevel <linux-fsdevel@...r.kernel.org>,
	lkml <linux-kernel@...r.kernel.org>,
	ext4 <linux-ext4@...r.kernel.org>
Subject: [RFC][PATCH] set_page_buffer_dirty should skip unmapped buffers

Hi Andrew,

I have been running into following bug while running fsx
tests on 1k (ext3) filesystem all the time. 

----------- [cut here ] --------- [please bite here ] ---------
Kernel BUG at fs/buffer.c:2791
invalid opcode: 0000 [1] SMP

Its complaining about BUG_ON(!buffer_mapped(bh)).

It was hard to track it down, needed lots of debug - but here 
is the problem & fix.  Since the fix is in __set_page_buffer_dirty()
code - I am wondering how it would effect others :(

With this fix fsx tests ran for more than 16 hours (and still
running).

Please let me know, what you think.

Thanks,
Badari 

Patch to fix: Kernel BUG at fs/buffer.c:2791
on 1k (2k) filesystems while running fsx.

journal_commit_transaction collects lots of dirty buffer from
and does a single ll_rw_block() to write them out. ll_rw_block()
locks the buffer and checks to see if they are dirty and submits
them for IO.

In the mean while, journal_unmap_buffers() as part of
truncate can unmap the buffer and throw it away. Since its
a 1k (2k) filesystem - each page (4k) will have more than
one buffer_head attached to the page and and we can't free 
up buffer_heads attached to the page (if we are not
invalidating the whole page).

Now, any call to set_page_dirty() (like msync_interval)
could end up setting all the buffer heads attached to
this page again dirty, including the ones those got
cleaned up :(

If ll_rw_block() runs now and sees the dirty bit it does
submit_bh() on those buffer_heads and triggers the assert.

Fix is to check if the buffer is mapped before setting its
dirty bit in __set_page_dirty_buffers().

Signed-off-by: Badari Pulavarty <pbadari@...ibm.com>
---
 fs/buffer.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

Index: linux-2.6.18-rc5/fs/buffer.c
===================================================================
--- linux-2.6.18-rc5.orig/fs/buffer.c	2006-09-01 08:20:51.000000000 -0700
+++ linux-2.6.18-rc5/fs/buffer.c	2006-09-01 08:41:01.000000000 -0700
@@ -846,7 +846,13 @@ int __set_page_dirty_buffers(struct page
 		struct buffer_head *bh = head;

 		do {
-			set_buffer_dirty(bh);
+			/*
+			 * Its possible that, not all buffers attached to
+			 * this page are mapped (cleaned up by truncate).
+			 * If so, skip them.
+			 */
+			if (buffer_mapped(bh))
+				set_buffer_dirty(bh);
 			bh = bh->b_this_page;
 		} while (bh != head);
 	}

-
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html