f2fs: fix to do sanity check on valid node/block count
authorChao Yu <yuchao0@huawei.com>
Mon, 15 Apr 2019 07:30:50 +0000 (15:30 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Thu, 9 May 2019 04:23:09 +0000 (21:23 -0700)
As Jungyeon reported in bugzilla:

https://bugzilla.kernel.org/show_bug.cgi?id=203229

- Overview
When mounting the attached crafted image, following errors are reported.
Additionally, it hangs on sync after trying to mount it.

The image is intentionally fuzzed from a normal f2fs image for testing.
Compile options for F2FS are as follows.
CONFIG_F2FS_FS=y
CONFIG_F2FS_STAT_FS=y
CONFIG_F2FS_FS_XATTR=y
CONFIG_F2FS_FS_POSIX_ACL=y
CONFIG_F2FS_CHECK_FS=y

- Reproduces
mkdir test
mount -t f2fs tmp.img test
sync

- Kernel message
 kernel BUG at fs/f2fs/recovery.c:591!
 RIP: 0010:recover_data+0x12d8/0x1780
 Call Trace:
  f2fs_recover_fsync_data+0x613/0x710
  f2fs_fill_super+0x1043/0x1aa0
  mount_bdev+0x16d/0x1a0
  mount_fs+0x4a/0x170
  vfs_kern_mount+0x5d/0x100
  do_mount+0x200/0xcf0
  ksys_mount+0x79/0xc0
  __x64_sys_mount+0x1c/0x20
  do_syscall_64+0x43/0xf0
  entry_SYSCALL_64_after_hwframe+0x44/0xa9

With corrupted image wihch has out-of-range valid node/block count, during
recovery, once we failed due to no free space, it will trigger kernel
panic.

Adding sanity check on valid node/block count in f2fs_sanity_check_ckpt()
to detect such condition, so that potential panic can be avoided.

Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/super.c

index 2cd7858..cbbb1e3 100644 (file)
@@ -2592,7 +2592,8 @@ int f2fs_sanity_check_ckpt(struct f2fs_sb_info *sbi)
        unsigned int log_blocks_per_seg;
        unsigned int segment_count_main;
        unsigned int cp_pack_start_sum, cp_payload;
-       block_t user_block_count;
+       block_t user_block_count, valid_user_blocks;
+       block_t avail_node_count, valid_node_count;
        int i, j;
 
        total = le32_to_cpu(raw_super->segment_count);
@@ -2627,6 +2628,24 @@ int f2fs_sanity_check_ckpt(struct f2fs_sb_info *sbi)
                return 1;
        }
 
+       valid_user_blocks = le64_to_cpu(ckpt->valid_block_count);
+       if (valid_user_blocks > user_block_count) {
+               f2fs_msg(sbi->sb, KERN_ERR,
+                       "Wrong valid_user_blocks: %u, user_block_count: %u",
+                       valid_user_blocks, user_block_count);
+               return 1;
+       }
+
+       valid_node_count = le32_to_cpu(ckpt->valid_node_count);
+       avail_node_count = sbi->total_node_count - sbi->nquota_files -
+                                               F2FS_RESERVED_NODE_NUM;
+       if (valid_node_count > avail_node_count) {
+               f2fs_msg(sbi->sb, KERN_ERR,
+                       "Wrong valid_node_count: %u, avail_node_count: %u",
+                       valid_node_count, avail_node_count);
+               return 1;
+       }
+
        main_segs = le32_to_cpu(raw_super->segment_count_main);
        blocks_per_seg = sbi->blocks_per_seg;