From: Akinobu Mita Subject: Re: [PATCH] ext4: fix ext4_init_block_bitmap() for metablock block group Date: Mon, 16 Jun 2008 23:10:17 +0900 Message-ID: <20080616141016.GA19080@localhost.localdomain> References: <20080615142455.GA3502@localhost.localdomain> <20080616051859.GK3726@webber.adilger.int> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-2022-jp Cc: linux-ext4@vger.kernel.org, Andrew Morton , Stephen Tweedie , adilger@clusterfs.com, Mingming Cao , Theodore Tso To: Andreas Dilger Return-path: Received: from rv-out-0506.google.com ([209.85.198.233]:11473 "EHLO rv-out-0506.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753656AbYFPOLl (ORCPT ); Mon, 16 Jun 2008 10:11:41 -0400 Received: by rv-out-0506.google.com with SMTP id k40so3602265rvb.1 for ; Mon, 16 Jun 2008 07:11:38 -0700 (PDT) Content-Disposition: inline In-Reply-To: <20080616051859.GK3726@webber.adilger.int> Sender: linux-ext4-owner@vger.kernel.org List-ID: On Sun, Jun 15, 2008 at 11:19:00PM -0600, Andreas Dilger wrote: > On Jun 15, 2008 23:24 +0900, Akinobu Mita wrote: > > When meta_bg feature is enabled and s_first_meta_bg != 0, > > ext4_init_block_bitmap() miscalculates the number of block used by > > the group descriptor table (0 or 1 for metablock block group) > > Can you please clarify why the calculation is incorrect? I admit that > I didn't test with META_BG enabled, so it could well be wrong, but looking > at the code I can't understand why it is incorrect. group_rel in the original code should be calculated by: group_rel = block_group % EXT4_DESC_PER_BLOCK(sb); No need to subtract s_first_meta_bg from brock_group. It will be equivalent what ext4_bg_num_gdb_meta() does > > @@ -121,12 +121,7 @@ unsigned ext4_init_block_bitmap(struct s > if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) || > block_group < le32_to_cpu(sbi->s_es->s_first_meta_bg) * > sbi->s_desc_per_block) { > if (bit_max) { > bit_max += ext4_bg_num_gdb(sb, block_group); > bit_max += > > le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); > > } > > } else { /* For META_BG_BLOCK_GROUPS */ > > - int group_rel = (block_group - > > - le32_to_cpu(sbi->s_es->s_first_meta_bg)) % > > - EXT4_DESC_PER_BLOCK(sb); > > - if (group_rel == 0 || group_rel == 1 || > > - (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1)) > > - bit_max += 1; > > + bit_max += ext4_bg_num_gdb(sb, block_group); > > } > > As you can see, the "if" checks if the block group is before s_first_meta_bg > to treat it as a "normal" group, and only uses the "else" once beyond the > start of the s_first_meta_bg limit. > > It definitely is less complex to use ext4_bg_num_gdb(), and this could > further be simplified by using ext4_bg_gdb_meta() in the "else" clause. > In fact, the whole if/else could be replaced with ext4_bg_num_gdb() if > it weren't for s_reserved_gdt_blocks. > > Maybe it makes sense (cleaner code, less chance for bugs) to change the > ext4_bg_num_gdb() function to take a parameter on whether it should > include the s_reserved_gdt_blocks or not: > > static unsigned long ext4_bg_num_gdb_nometa(struct super_block *sb, int group, > int reserved) > { > if (EXT4_HAS_RO_COMPAT_FEATURE(sb, > EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER) && > !ext4_group_sparse(group)) > return 0; > return EXT4_SB(sb)->s_gdb_count + > reserved ? EXT4_SB(sb)->s_reserved_gdt_blocks : 0; > } > > unsigned long ext4_bg_num_gdb(struct super_block *sb, int group, int reserved) > { > unsigned long first_meta_bg = > le32_to_cpu(EXT4_SB(sb)->s_es->s_first_meta_bg); > unsigned long metagroup = group / EXT4_DESC_PER_BLOCK(sb); > > if (!EXT4_HAS_INCOMPAT_FEATURE(sb,EXT4_FEATURE_INCOMPAT_META_BG) || > metagroup < first_meta_bg) > return ext4_bg_num_gdb_nometa(sb, group, reserved); > > return ext4_bg_num_gdb_meta(sb, group); > } > > The fewer places in the code that need to understand META_BG, the less > chance of having a bug. Now the code in ext4_init_block_bitmap() can be: > > bit_max += ext4_bg_num_gdb(sb, block_group, 1); Looks good to me. But I'd like to divide into the change into simple bugfix and cleanup patch in order to clarify the each change. Here is a bugfix only patch. From: Akinobu Mita Subject: ext4: fix ext4_init_block_bitmap() for metablock block group When meta_bg feature is enabled and s_first_meta_bg != 0, ext4_init_block_bitmap() miscalculates the number of block used by the group descriptor table (0 or 1 for metablock block group) group_rel in the original code should be calculated by: group_rel = block_group % EXT4_DESC_PER_BLOCK(sb); No need to subtract s_first_meta_bg from brock_group. It will be equivalent what ext4_bg_num_gdb_meta() does Signed-off-by: Akinobu Mita Cc: Andrew Morton Cc: Stephen Tweedie Cc: adilger@clusterfs.com Cc: Mingming Cao Cc: Theodore Tso --- fs/ext4/balloc.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) Index: 2.6-git/fs/ext4/balloc.c =================================================================== --- 2.6-git.orig/fs/ext4/balloc.c +++ 2.6-git/fs/ext4/balloc.c @@ -121,9 +121,8 @@ unsigned ext4_init_block_bitmap(struct s le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks); } } else { /* For META_BG_BLOCK_GROUPS */ - int group_rel = (block_group - - le32_to_cpu(sbi->s_es->s_first_meta_bg)) % - EXT4_DESC_PER_BLOCK(sb); + int group_rel = block_group % EXT4_DESC_PER_BLOCK(sb); + if (group_rel == 0 || group_rel == 1 || (group_rel == EXT4_DESC_PER_BLOCK(sb) - 1)) bit_max += 1;