Andreas Dilger wrote:
On Sep 19, 2007 14:06 +0200, Valerie Clement wrote:
I ran some tests with the uninit_groups feature enabled and got error messages
when running e2fsck on my ext4 partition. e2fsck complains of an "invalid
unused inodes count" in some group descriptors.
These errors occur when checking groups which have only one inode in use. The
"free inodes" count has been decremented by one in these groups but not the
"unused inodes" count.
Index: linux-2.6.23-rc6/fs/ext4/ialloc.c
===================================================================
--- linux-2.6.23-rc6.orig/fs/ext4/ialloc.c 2007-09-19 11:31:01.000000000 +0200
+++ linux-2.6.23-rc6/fs/ext4/ialloc.c 2007-09-19 11:31:41.000000000 +0200
@@ -633,13 +633,10 @@ got:
/* If we didn't allocate from within the initialized part of the inode
* table then we need to initialize up to this inode. */
if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
- if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
+ if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))
gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
- free = EXT4_INODES_PER_GROUP(sb);
- } else {
- free = EXT4_INODES_PER_GROUP(sb) -
+ free = EXT4_INODES_PER_GROUP(sb) -
le16_to_cpu(gdp->bg_itable_unused);
- }
Hmm, this is indeed incorrect, but I'm not sure solution is the right one.
I guess in our testing we ran it for a long time and must have created
more than a single inode per group...
What about the following instead? I think the assumption in the original
code is that "it's a new group, all the inodes are free", but that is not
correct - we want to make NONE of the inodes free initially so that
bt_itable_unused is recalculated below:
if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) {
gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT);
- free = EXT4_INODES_PER_GROUP(sb);
+ free = 0;
} else {
free = EXT4_INODES_PER_GROUP(sb) -
le16_to_cpu(gdp->bg_itable_unused);
}
if (ino > free)
gdp->bg_itable_unused =
cpu_to_le16(EXT3_INODES_PER_GROUP(sb) - ino);
Still a bit uneasy about off-by-one errors here though. Is "ino" 0 or
1 for the first inode in the group. We might need to have a -1 in the
bg_itable_unused calculation still.
ino is incremented right before this code, so the first inode in the
group is represented by 1, not 0. So this fix looks good.
Aneesh has incorporated this fix and also removed the local crc16 code,
I will be reposting his new patch to lkml.
thanks
Avantika
-
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
