rpms/kernel/devel linux-2.6-ext4-inode-alloc-race.patch, NONE, 1.1 kernel.spec, 1.1384, 1.1385

[Eric Sandeen]

Author: sandeen

Update of /cvs/pkgs/rpms/kernel/devel
In directory cvs1.fedora.phx.redhat.com:/tmp/cvs-serv17165

Modified Files:
	kernel.spec 
Added Files:
	linux-2.6-ext4-inode-alloc-race.patch 
Log Message:
* Thu Mar 05 2009 Eric Sandeen <sandeen at redhat.com>
- Fix ext4 race between inode bitmap set/clear


linux-2.6-ext4-inode-alloc-race.patch:

--- NEW FILE linux-2.6-ext4-inode-alloc-race.patch ---
Date: 	Tue, 03 Mar 2009 22:38:41 -0600
From: Eric Sandeen <sandeen at redhat.com>
To: ext4 development <linux-ext4 at vger.kernel.org>
CC: "Aneesh Kumar K.V" <aneesh.kumar at linux.vnet.ibm.com>
Subject: [PATCH] fix ext4_free_inode vs. ext4_claim_inode race

I was seeing fsck errors on inode bitmaps after a 4 thread
dbench run on a 4 cpu machine:

Inode bitmap differences: -50736 -(50752--50753) etc...

I believe that this is because ext4_free_inode() uses atomic
bitops, and although ext4_new_inode() *used* to also use atomic 
bitops for synchronization, commit 
393418676a7602e1d7d3f6e560159c65c8cbd50e changed this to use
the sb_bgl_lock, so that we could also synchronize against
read_inode_bitmap and initialization of uninit inode tables.

However, that change left ext4_free_inode using atomic bitops,
which I think leaves no synchronization between setting & 
unsetting bits in the inode table.

The below patch fixes it for me, although I wonder if we're 
getting at all heavy-handed with this spinlock...

Signed-off-by: Eric Sandeen <sandeen at redhat.com>
---

Index: linux-2.6/fs/ext4/ialloc.c
===================================================================
--- linux-2.6.orig/fs/ext4/ialloc.c
+++ linux-2.6/fs/ext4/ialloc.c
@@ -188,7 +188,7 @@ void ext4_free_inode(handle_t *handle, s
 	struct ext4_group_desc *gdp;
 	struct ext4_super_block *es;
 	struct ext4_sb_info *sbi;
-	int fatal = 0, err, count;
+	int fatal = 0, err, count, cleared;
 	ext4_group_t flex_group;
 
 	if (atomic_read(&inode->i_count) > 1) {
@@ -248,8 +248,10 @@ void ext4_free_inode(handle_t *handle, s
 		goto error_return;
 
 	/* Ok, now we can actually update the inode bitmaps.. */
-	if (!ext4_clear_bit_atomic(sb_bgl_lock(sbi, block_group),
-					bit, bitmap_bh->b_data))
+	spin_lock(sb_bgl_lock(sbi, block_group));
+	cleared = ext4_clear_bit(bit, bitmap_bh->b_data);
+	spin_unlock(sb_bgl_lock(sbi, block_group));
+	if (!cleared)
 		ext4_error(sb, "ext4_free_inode",
 			   "bit already cleared for inode %lu", ino);
 	else {




Index: kernel.spec
===================================================================
RCS file: /cvs/pkgs/rpms/kernel/devel/kernel.spec,v
retrieving revision 1.1384
retrieving revision 1.1385
diff -u -r1.1384 -r1.1385
--- kernel.spec	5 Mar 2009 00:27:56 -0000	1.1384
+++ kernel.spec	5 Mar 2009 05:23:27 -0000	1.1385
@@ -674,6 +674,9 @@
 Patch2899: linux-2.6-v4l-dvb-fixes.patch
 Patch2900: linux-2.6-v4l-dvb-update.patch
 
+# ext4 fixes
+Patch2910: linux-2.6-ext4-inode-alloc-race.patch
+
 Patch9001: revert-fix-modules_install-via-nfs.patch
 
 %endif
@@ -1086,6 +1089,7 @@
 #
 
 # ext4
+ApplyPatch linux-2.6-ext4-inode-alloc-race.patch
 
 # xfs
 
@@ -1792,6 +1796,9 @@
 # and build.
 
 %changelog
+* Thu Mar 05 2009 Eric Sandeen <sandeen at redhat.com>
+- Fix ext4 race between inode bitmap set/clear
+
 * Thu Mar 05 2009 Ben Skeggs <bskeggs at redhat.com>
 - drm-nouveau.patch: fix <nv50 chipsets, and ppc.