ocfs2: fix the issue with discontiguous allocation in the global_bitmap
authorHeming Zhao <heming.zhao@suse.com>
Mon, 14 Apr 2025 06:01:23 +0000 (14:01 +0800)
committerAndrew Morton <akpm@linux-foundation.org>
Thu, 8 May 2025 06:39:37 +0000 (23:39 -0700)
commitbd1261b16d9131d79723d982d54295e7f309797a
treec6257740c2064b74b57c37ade934c13245fae95a
parentdf84d2fd35c6f2d6b2241e47f26e1829eab26186
ocfs2: fix the issue with discontiguous allocation in the global_bitmap

commit 4eb7b93e0310 ("ocfs2: improve write IO performance when
fragmentation is high") introduced another regression.

The following ocfs2-test case can trigger this issue:
> discontig_runner.sh => activate_discontig_bg.sh => resv_unwritten:
> ${RESV_UNWRITTEN_BIN} -f ${WORK_PLACE}/large_testfile -s 0 -l \
> $((${FILE_MAJOR_SIZE_M}*1024*1024))

In my env, test disk size (by "fdisk -l <dev>"):
53687091200 bytes, 104857600 sectors.

Above command is:
> /usr/local/ocfs2-test/bin/resv_unwritten -f \
> /mnt/ocfs2/ocfs2-activate-discontig-bg-dir/large_testfile -s 0 -l \
53187969024

Error log:
> [*] Reserve 50724M space for a LARGE file, reserve 200M space for future test.
> ioctl error 28: "No space left on device"
> resv allocation failed Unknown error -1
> reserve unwritten region from 0 to 53187969024.

Call flow:
__ocfs2_change_file_space //by ioctl OCFS2_IOC_RESVSP64
 ocfs2_allocate_unwritten_extents //start:0 len:53187969024
  while()
   + ocfs2_get_clusters //cpos:0, alloc_size:1623168 (cluster number)
   + ocfs2_extend_allocation
     + ocfs2_lock_allocators
     |  + choose OCFS2_AC_USE_MAIN & ocfs2_cluster_group_search
     |
     + ocfs2_add_inode_data
        ocfs2_add_clusters_in_btree
         __ocfs2_claim_clusters
          ocfs2_claim_suballoc_bits
          + During the allocation of the final part of the large file
    (after ~47GB), no chain had the required contiguous
            bits_wanted. Consequently, the allocation failed.

How to fix:
When OCFS2 is encountering fragmented allocation, the file system should
stop attempting bits_wanted contiguous allocation and instead provide the
largest available contiguous free bits from the cluster groups.

Link: https://lkml.kernel.org/r/20250414060125.19938-2-heming.zhao@suse.com
Fixes: 4eb7b93e0310 ("ocfs2: improve write IO performance when fragmentation is high")
Signed-off-by: Heming Zhao <heming.zhao@suse.com>
Reported-by: Gautham Ananthakrishna <gautham.ananthakrishna@oracle.com>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Changwei Ge <gechangwei@live.cn>
Cc: Jun Piao <piaojun@huawei.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
fs/ocfs2/suballoc.c
fs/ocfs2/suballoc.h