Commit 07a8109d authored by Zhihao Cheng's avatar Zhihao Cheng
Browse files

ext4: Stop trying writing pages if no free blocks generated

hulk inclusion
category: bugfix
bugzilla: https://gitee.com/openeuler/kernel/issues/I7CBCS



--------------------------------

Folllowing steps could make ext4_wripages trap into a dead loop:

1. Consume free_clusters until free_clusters > 2 * sbi->s_resv_clusters,
   and free_clusters > EXT4_FREECLUSTERS_WATERMARK.
   // eg. free_clusters = 1422, sbi->s_resv_clusters = 512
   // nr_cpus = 4, EXT4_FREECLUSTERS_WATERMARK = 512
2. umount && mount.  // dirty_clusters = 0
3. Run free_clusters tasks concurrently to write different files, many
   tasks write(appendant) 4K data by da_write method. And each inode will
   consume one data block and one extent block in map_block.
   // There are (free_clusters - EXT4_FREECLUSTERS_WATERMARK = 910)
   // tasks choosing da_write method, left 512 tasks choose write_begin
   // method. If tasks which chooses da_write path run first.
   // dirty_clusters = 910, free_clusters = 1422
   // Tasks which choose write_begin path will get ENOSPC:
   //  free_clusters < (nclusters + dirty_clusters + resv_clusters)
   //  1422 < (1 + 910 + 512)
4. After certain number of map_block iterations in ext4_writepages.
   // free_clusters = 0,
   // dirty_clusters = 910 - (1422 / 2) = 199
5. Delete one 4K file.  // free_clusters = 1
6. ext4_writepages traps into dead loop:
    mpage_map_and_submit_extent
     mpage_map_one_extent // ret = ENOSPC
       ext4_map_blocks -> ext4_ext_map_blocks -> ext4_mb_new_blocks ->
       ext4_claim_free_clusters:
         if (free_clusters >= (nclusters + dirty_clusters)) // false
     if (err == -ENOSPC && ext4_count_free_clusters(sb)) // true
       return err
     *give_up_on_write = true // won't be executed

Fix it by terminating ext4_writepages if no free blocks generated.

Signed-off-by: default avatarZhihao Cheng <chengzhihao1@huawei.com>
parent fc143182
Loading
Loading
Loading
Loading
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Please to comment