From: Andreas Dilger Subject: [PATCH] (unfinished) don't alloc/free in system zones Date: Fri, 3 Aug 2007 17:50:13 -0600 Message-ID: <20070803235013.GA6142@schatzie.adilger.int> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii To: linux-ext4@vger.kernel.org Return-path: Received: from mail.clusterfs.com ([74.0.229.162]:42721 "EHLO mail.clusterfs.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1758807AbXHCXuQ (ORCPT ); Fri, 3 Aug 2007 19:50:16 -0400 Received: from localhost.adilger.int (S01060004e23cfc51.ek.shawcable.net [24.66.76.119]) (using TLSv1 with cipher DHE-RSA-AES256-SHA (256/256 bits)) (No client certificate requested) by mail.clusterfs.com (Postfix) with ESMTP id C35284E4567 for ; Fri, 3 Aug 2007 17:50:14 -0600 (MDT) Content-Disposition: inline Sender: linux-ext4-owner@vger.kernel.org List-Id: linux-ext4.vger.kernel.org It seems there is a long-standing defect in ext2/ext3/ext4 in the 2.6 kernels that hasn't been fixed, even though I thought it was. There are checks in ext[234]_{new,free}_blocks() to see if the block(s) being allocated or freed overlap with the bitmaps or inode tables, but if the filesystem is mounted with "errors=continue" (the default), then even though an error is printed the allocation or free will succeed and this will lead to filesystem corruption. There was a patch posted in http://marc.info/?l=linux-ext4&m=116360109620982&w=2 that should be added to the ext[234] patch queue that fixes this problem. We might also consider making "errors=remount-ro" the default for ext4, this has long been the default for Debian users of ext[23]. Signed-Off-By: Andreas Dilger Signed-Off-By: Andre Noll --- fs/ext3/balloc.c | 102 +++++++++++++++++++++++++++++++++++++++-------------- 1 files changed, 75 insertions(+), 27 deletions(-) diff --git a/fs/ext3/balloc.c b/fs/ext3/balloc.c index 063d994..763b7a0 100644 --- a/fs/ext3/balloc.c +++ b/fs/ext3/balloc.c @@ -359,17 +359,6 @@ do_more: if (!desc) goto error_return; - if (in_range (le32_to_cpu(desc->bg_block_bitmap), block, count) || - in_range (le32_to_cpu(desc->bg_inode_bitmap), block, count) || - in_range (block, le32_to_cpu(desc->bg_inode_table), - sbi->s_itb_per_group) || - in_range (block + count - 1, le32_to_cpu(desc->bg_inode_table), - sbi->s_itb_per_group)) - ext3_error (sb, "ext3_free_blocks", - "Freeing blocks in system zones - " - "Block = "E3FSBLK", count = %lu", - block, count); - /* * We are about to start releasing blocks in the bitmap, * so we need undo access. @@ -392,7 +381,17 @@ do_more: jbd_lock_bh_state(bitmap_bh); - for (i = 0, group_freed = 0; i < count; i++) { + for (i = 0, group_freed = 0; i < count; i++, block++) { + struct ext3_group_desc *gdp = ext3_get_group_desc(sb, i, NULL); + if (block == le32_to_cpu(gdp->bg_block_bitmap) || + block == le32_to_cpu(gdp->bg_inode_bitmap) || + in_range(block, le32_to_cpu(gdp->bg_inode_table), + EXT3_SB(sb)->s_itb_per_group)) { + ext3_error(sb, __FUNCTION__, + "Freeing block in system zone - block = %lu", + block); + continue; + } /* * An HJ special. This is expensive... */ @@ -400,7 +399,7 @@ do_more: jbd_unlock_bh_state(bitmap_bh); { struct buffer_head *debug_bh; - debug_bh = sb_find_get_block(sb, block + i); + debug_bh = sb_find_get_block(sb, block); if (debug_bh) { BUFFER_TRACE(debug_bh, "Deleted!"); if (!bh2jh(bitmap_bh)->b_committed_data) @@ -452,7 +451,7 @@ do_more: jbd_unlock_bh_state(bitmap_bh); ext3_error(sb, __FUNCTION__, "bit already cleared for block "E3FSBLK, - block + i); + block); jbd_lock_bh_state(bitmap_bh); BUFFER_TRACE(bitmap_bh, "bit already cleared"); } else { @@ -479,7 +478,6 @@ do_more: *pdquot_freed_blocks += group_freed; if (overflow && !err) { - block += count; count = overflow; goto do_more; } @@ -1192,6 +1190,63 @@ int ext3_should_retry_alloc(struct super } /* + * Check if given blocks are metadata blocks. + * + * We should also check the backup group descriptors and the superblock, + * but it is too expensive to do so for each allocated/freed block. So, + * let's defer that check until we have per-group data structs. + */ +static inline int ext3_blocks_are_metadata(ext3_fsblk_t block, + unsigned long num, + struct ext3_group_desc *gdp, + struct super_block *sb) +{ + if (in_range(le32_to_cpu(gdp->bg_block_bitmap), block, num)) + return 1; + if (in_range(le32_to_cpu(gdp->bg_inode_bitmap), block, num)) + return 1; + if (in_range(block, le32_to_cpu(gdp->bg_inode_table), + EXT3_SB(sb)->s_itb_per_group)) + return 1; + if (in_range(block + num - 1, le32_to_cpu(gdp->bg_inode_table), + EXT3_SB(sb)->s_itb_per_group)) + return 1; + return 0; +} + +/* + * + * set the bits for all of the metadata blocks in the group + * + * Note: This will potentially use up some of the handle's buffer credits. + * Normally we have way too many credits, so that is OK. In _very_ rare cases it + * might not be OK. We will trigger an assertion if we run out of credits, and we + * will have to do a full fsck of the filesystem - better than randomly corrupting + * filesystem metadata. + */ +static void fix_group(int group, struct super_block *sb) +{ + int i; + ext3_fsblk_t bit; + unsigned long gdblocks; + struct buffer_head *gdp_bh; + struct ext3_group_desc *gdp = ext3_get_group_desc(sb, group, &gdp_bh); + + if (ext3_bg_has_super(sb, group)) + ext3_set_bit(0, gdp_bh->b_data); + gdblocks = ext3_bg_num_gdb(sb, group); + for (i = 0, bit = 1; i < gdblocks; i++, bit++) + ext3_set_bit(i, gdp_bh->b_data); + ext3_set_bit(gdp->bg_inode_bitmap % EXT3_BLOCKS_PER_GROUP(sb), + gdp_bh->b_data); + ext3_set_bit(gdp->bg_block_bitmap % EXT3_BLOCKS_PER_GROUP(sb), + gdp_bh->b_data); + for (i = 0, bit = gdp->bg_inode_table % EXT3_BLOCKS_PER_GROUP(sb); + i < EXT3_SB(sb)->s_itb_per_group; i++, bit++) + ext3_set_bit(i, gdp_bh->b_data); +} + +/* * ext3_new_block uses a goal block to assist allocation. If the goal is * free, or there is a free block within 32 blocks of the goal, that block * is allocated. Otherwise a forward search is made for a free block; within @@ -1260,7 +1315,7 @@ ext3_fsblk_t ext3_new_blocks(handle_t *h *errp = -ENOSPC; goto out; } - +repeat: /* * First, test whether the goal block is free. */ @@ -1367,17 +1422,10 @@ allocated: ret_block = grp_alloc_blk + ext3_group_first_block_no(sb, group_no); - if (in_range(le32_to_cpu(gdp->bg_block_bitmap), ret_block, num) || - in_range(le32_to_cpu(gdp->bg_inode_bitmap), ret_block, num) || - in_range(ret_block, le32_to_cpu(gdp->bg_inode_table), - EXT3_SB(sb)->s_itb_per_group) || - in_range(ret_block + num - 1, le32_to_cpu(gdp->bg_inode_table), - EXT3_SB(sb)->s_itb_per_group)) - ext3_error(sb, "ext3_new_block", - "Allocating block in system zone - " - "blocks from "E3FSBLK", length %lu", - ret_block, num); - + if (ext3_blocks_are_metadata(ret_block, num, gdp, sb)) { + fix_group(group_no, sb); + goto repeat; + } performed_allocation = 1; #ifdef CONFIG_JBD_DEBUG Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.