kernel/linux.git/fs/ext4/resize.c, branch v6.12.80

ext4: fix race in buffer_head read fault injection

2024-12-05T13:01:15+00:00

[ Upstream commit 2f3d93e210b9c2866c8b3662adae427d5bf511ec ] When I enabled ext4 debug for fault injection testing, I encountered the following warning: EXT4-fs error (device sda): ext4_read_inode_bitmap:201: comm fsstress: Cannot read inode bitmap - block_group = 8, inode_bitmap = 1051 WARNING: CPU: 0 PID: 511 at fs/buffer.c:1181 mark_buffer_dirty+0x1b3/0x1d0 The root cause of the issue lies in the improper implementation of ext4's buffer_head read fault injection. The actual completion of buffer_head read and the buffer_head fault injection are not atomic, which can lead to the uptodate flag being cleared on normally used buffer_heads in race conditions. [CPU0] [CPU1] [CPU2] ext4_read_inode_bitmap ext4_read_bh() ext4_read_inode_bitmap if (buffer_uptodate(bh)) return bh jbd2_journal_commit_transaction __jbd2_journal_refile_buffer __jbd2_journal_unfile_buffer __jbd2_journal_temp_unlink_buffer ext4_simulate_fail_bh() clear_buffer_uptodate mark_buffer_dirty WARN_ON_ONCE(!buffer_uptodate(bh)) The best approach would be to perform fault injection in the IO completion callback function, rather than after IO completion. However, the IO completion callback function cannot get the fault injection code in sb. Fix it by passing the result of fault injection into the bh read function, we simulate faults within the bh read function itself. This requires adding an extra parameter to the bh read functions that need fault injection. Fixes: 46f870d690fe ("ext4: simulate various I/O and checksum errors when reading metadata") Signed-off-by: Long Li Link: https://patch.msgid.link/20240906091746.510163-1-leo.lilong@huawei.com Signed-off-by: Theodore Ts'o Signed-off-by: Sasha Levin

ext4: fix off by one issue in alloc_flex_gd()

2024-10-04T21:36:28+00:00

Wesley reported an issue: ================================================================== EXT4-fs (dm-5): resizing filesystem from 7168 to 786432 blocks ------------[ cut here ]------------ kernel BUG at fs/ext4/resize.c:324! CPU: 9 UID: 0 PID: 3576 Comm: resize2fs Not tainted 6.11.0+ #27 RIP: 0010:ext4_resize_fs+0x1212/0x12d0 Call Trace: __ext4_ioctl+0x4e0/0x1800 ext4_ioctl+0x12/0x20 __x64_sys_ioctl+0x99/0xd0 x64_sys_call+0x1206/0x20d0 do_syscall_64+0x72/0x110 entry_SYSCALL_64_after_hwframe+0x76/0x7e ================================================================== While reviewing the patch, Honza found that when adjusting resize_bg in alloc_flex_gd(), it was possible for flex_gd->resize_bg to be bigger than flexbg_size. The reproduction of the problem requires the following: o_group = flexbg_size * 2 * n; o_size = (o_group + 1) * group_size; n_group: [o_group + flexbg_size, o_group + flexbg_size * 2) o_size = (n_group + 1) * group_size; Take n=0,flexbg_size=16 as an example: last:15 |o---------------|--------------n-| o_group:0 resize to n_group:30 The corresponding reproducer is: img=test.img rm -f $img truncate -s 600M $img mkfs.ext4 -F $img -b 1024 -G 16 8M dev=`losetup -f --show $img` mkdir -p /tmp/test mount $dev /tmp/test resize2fs $dev 248M Delete the problematic plus 1 to fix the issue, and add a WARN_ON_ONCE() to prevent the issue from happening again. [ Note: another reproucer which this commit fixes is: img=test.img rm -f $img truncate -s 25MiB $img mkfs.ext4 -b 4096 -E nodiscard,lazy_itable_init=0,lazy_journal_init=0 $img truncate -s 3GiB $img dev=`losetup -f --show $img` mkdir -p /tmp/test mount $dev /tmp/test resize2fs $dev 3G umount $dev losetup -d $dev -- TYT ] Reported-by: Wesley Hershberger Closes: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2081231 Reported-by: Stéphane Graber Closes: https://lore.kernel.org/all/20240925143325.518508-1-aleksandr.mikhalitsyn@canonical.com/ Tested-by: Alexander Mikhalitsyn Tested-by: Eric Sandeen Fixes: 665d3e0af4d3 ("ext4: reduce unnecessary memory allocation in alloc_flex_gd()") Cc: stable@vger.kernel.org Signed-off-by: Baokun Li Reviewed-by: Jan Kara Link: https://patch.msgid.link/20240927133329.1015041-1-libaokun@huaweicloud.com Signed-off-by: Theodore Ts'o

ext4: move checksum length calculation of inode bitmap into ext4_inode_bitmap_csum_[verify/set]() functions

2024-09-04T02:12:15+00:00

There are some little improve: 1. remove repeat code to calculate checksum length of inode bitmap 2. remove unnecessary checksum length calculation if checksum is not enabled. 3. use more efficient bit shift operation instead of div opreation. Signed-off-by: Kemeng Shi Link: https://patch.msgid.link/20240820132234.2759926-6-shikemeng@huaweicloud.com Signed-off-by: Theodore Ts'o

ext4: enable meta_bg only when new desc blocks are needed

2024-03-07T18:32:54+00:00

This patch addresses an issue observed when resize_inode is disabled and an online extension of a filesysyem is performed. When a filesystem is expanded to a size that does not require a addition of a new descriptor block, the meta_bg feature is being enabled even though no part of the filesystem uses this layout. This patch ensures that the meta_bg feature is only enabled if any of the added block groups utilize meta_bg layout. Signed-off-by: Srivathsa Dara Link: https://lore.kernel.org/r/20240227131329.2608466-1-srivathsa.d.dara@oracle.com Signed-off-by: Theodore Ts'o

ext4: fix corruption during on-line resize

2024-03-07T18:32:54+00:00

We observed a corruption during on-line resize of a file system that is larger than 16 TiB with 4k block size. With having more then 2^32 blocks resize_inode is turned off by default by mke2fs. The issue can be reproduced on a smaller file system for convenience by explicitly turning off resize_inode. An on-line resize across an 8 GiB boundary (the size of a meta block group in this setup) then leads to a corruption: dev=/dev/ # should be >= 16 GiB mkdir -p /corruption /sbin/mke2fs -t ext4 -b 4096 -O ^resize_inode $dev $((2 * 2**21 - 2**15)) mount -t ext4 $dev /corruption dd if=/dev/zero bs=4096 of=/corruption/test count=$((2*2**21 - 4*2**15)) sha1sum /corruption/test # 79d2658b39dcfd77274e435b0934028adafaab11 /corruption/test /sbin/resize2fs $dev $((2*2**21)) # drop page cache to force reload the block from disk echo 1 > /proc/sys/vm/drop_caches sha1sum /corruption/test # 3c2abc63cbf1a94c9e6977e0fbd72cd832c4d5c3 /corruption/test 2^21 = 2^15*2^6 equals 8 GiB whereof 2^15 is the number of blocks per block group and 2^6 are the number of block groups that make a meta block group. The last checksum might be different depending on how the file is laid out across the physical blocks. The actual corruption occurs at physical block 63*2^15 = 2064384 which would be the location of the backup of the meta block group's block descriptor. During the on-line resize the file system will be converted to meta_bg starting at s_first_meta_bg which is 2 in the example - meaning all block groups after 16 GiB. However, in ext4_flex_group_add we might add block groups that are not part of the first meta block group yet. In the reproducer we achieved this by substracting the size of a whole block group from the point where the meta block group would start. This must be considered when updating the backup block group descriptors to follow the non-meta_bg layout. The fix is to add a test whether the group to add is already part of the meta block group or not. Fixes: 01f795f9e0d67 ("ext4: add online resizing support for meta_bg and 64-bit file systems") Cc: Signed-off-by: Maximilian Heyne Tested-by: Srivathsa Dara Reviewed-by: Srivathsa Dara Link: https://lore.kernel.org/r/20240215155009.94493-1-mheyne@amazon.de Signed-off-by: Theodore Ts'o

ext4: reduce unnecessary memory allocation in alloc_flex_gd()

2023-12-14T15:54:47+00:00

When a large flex_bg file system is resized, the number of groups to be added may be small, and a large amount of memory that will not be used will be allocated. Therefore, resize_bg can be set to the size after the number of new_group_data to be used is aligned upwards to the power of 2. This does not affect the disk layout after online resize and saves some memory. Signed-off-by: Baokun Li Reviewed-by: Jan Kara Link: https://lore.kernel.org/r/20231023013057.2117948-5-libaokun1@huawei.com Signed-off-by: Theodore Ts'o

ext4: avoid online resizing failures due to oversized flex bg

2023-12-14T15:54:47+00:00

When we online resize an ext4 filesystem with a oversized flexbg_size, mkfs.ext4 -F -G 67108864 $dev -b 4096 100M mount $dev $dir resize2fs $dev 16G the following WARN_ON is triggered: ================================================================== WARNING: CPU: 0 PID: 427 at mm/page_alloc.c:4402 __alloc_pages+0x411/0x550 Modules linked in: sg(E) CPU: 0 PID: 427 Comm: resize2fs Tainted: G E 6.6.0-rc5+ #314 RIP: 0010:__alloc_pages+0x411/0x550 Call Trace: __kmalloc_large_node+0xa2/0x200 __kmalloc+0x16e/0x290 ext4_resize_fs+0x481/0xd80 __ext4_ioctl+0x1616/0x1d90 ext4_ioctl+0x12/0x20 __x64_sys_ioctl+0xf0/0x150 do_syscall_64+0x3b/0x90 ================================================================== This is because flexbg_size is too large and the size of the new_group_data array to be allocated exceeds MAX_ORDER. Currently, the minimum value of MAX_ORDER is 8, the minimum value of PAGE_SIZE is 4096, the corresponding maximum number of groups that can be allocated is: (PAGE_SIZE << MAX_ORDER) / sizeof(struct ext4_new_group_data) ≈ 21845 And the value that is down-aligned to the power of 2 is 16384. Therefore, this value is defined as MAX_RESIZE_BG, and the number of groups added each time does not exceed this value during resizing, and is added multiple times to complete the online resizing. The difference is that the metadata in a flex_bg may be more dispersed. Signed-off-by: Baokun Li Reviewed-by: Jan Kara Link: https://lore.kernel.org/r/20231023013057.2117948-4-libaokun1@huawei.com Signed-off-by: Theodore Ts'o

ext4: remove unnecessary check from alloc_flex_gd()

2023-12-14T15:54:46+00:00

In commit 967ac8af4475 ("ext4: fix potential integer overflow in alloc_flex_gd()"), an overflow check is added to alloc_flex_gd() to prevent the allocated memory from being smaller than expected due to the overflow. However, after kmalloc() is replaced with kmalloc_array() in commit 6da2ec56059c ("treewide: kmalloc() -> kmalloc_array()"), the kmalloc_array() function has an overflow check, so the above problem will not occur. Therefore, the extra check is removed. Signed-off-by: Baokun Li Reviewed-by: Jan Kara Link: https://lore.kernel.org/r/20231023013057.2117948-3-libaokun1@huawei.com Signed-off-by: Theodore Ts'o

ext4: unify the type of flexbg_size to unsigned int

2023-12-14T15:54:46+00:00

The maximum value of flexbg_size is 2^31, but the maximum value of int is (2^31 - 1), so overflow may occur when the type of flexbg_size is declared as int. For example, when uninit_mask is initialized in ext4_alloc_group_tables(), if flexbg_size == 2^31, the initialized uninit_mask is incorrect, and this may causes set_flexbg_block_bitmap() to trigger a BUG_ON(). Therefore, the flexbg_size type is declared as unsigned int to avoid overflow and memory waste. Signed-off-by: Baokun Li Reviewed-by: Jan Kara Link: https://lore.kernel.org/r/20231023013057.2117948-2-libaokun1@huawei.com Signed-off-by: Theodore Ts'o

ext4: remove unnecessary initialization of count2 in set_flexbg_block_bitmap

2023-10-06T02:32:15+00:00

We always overwrite count2 to "EXT4_CLUSTERS_PER_GROUP(sb) - (first_cluster - start)" after its initialization in for loop initialization statement . Just remove unnecessary initialization of count2. Signed-off-by: Kemeng Shi Reviewed-by: Theodore Ts'o Link: https://lore.kernel.org/r/20230826174712.4059355-14-shikemeng@huaweicloud.com Signed-off-by: Theodore Ts'o