[LU-14186] fallocate does not extend uninitialized PFL components

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.14.0
• Affects Version/s: Lustre 2.14.0
• Severity: 3
• Assignee: Wang Shilong (Inactive)
• Reporter: Andreas Dilger

Description

I was testing "fallocate" as a fast way to allocate space in the file, but I found that it only allocated space on the first (initialized) component of a PFL file. It does not allocate space for any of the later components if those components are uninitialized.

# lfs setstripe -E1M -c1 -E16M -c3 -Eeof -c 4 /mnt/testfs/pflfile
# ls -l /mnt/testfs/pflfile
0 -rw-r--r--. 1 root 0 Dec  5 13:24 /mnt/testfs/pflfile
# fallocate -l $((1048576 * 1024)) /mnt/testfs/pflfile
# ls -l /mnt/testfs/pflfile
1024 -rw-r--r--. 1 root 1048576 Dec  5 13:33 /mnt/testfs/pflfile
# lfs getstripe /mnt/testfs/pflfile
/mnt/testfs/pflfile
  lcm_layout_gen:    3
  lcm_mirror_count:  1
  lcm_entry_count:   3
    lcme_id:             1
    lcme_mirror_id:      0
    lcme_flags:          init
    lcme_extent.e_start: 0
    lcme_extent.e_end:   1048576
      lmm_stripe_count:  1
      lmm_stripe_size:   1048576
      lmm_pattern:       raid0
      lmm_layout_gen:    0
      lmm_stripe_offset: 3
      lmm_objects:
      - 0: { l_ost_idx: 3, l_fid: [0x100030000:0x60a:0x0] }

    lcme_id:             2
    lcme_mirror_id:      0
    lcme_flags:          0
    lcme_extent.e_start: 1048576
    lcme_extent.e_end:   16777216
      lmm_stripe_count:  3
      lmm_stripe_size:   1048576
      lmm_pattern:       raid0
      lmm_layout_gen:    0
      lmm_stripe_offset: -1

    lcme_id:             3
    lcme_mirror_id:      0
    lcme_flags:          0
    lcme_extent.e_start: 16777216
    lcme_extent.e_end:   1073741824
      lmm_stripe_count:  4
      lmm_stripe_size:   2097152
      lmm_pattern:       raid0
      lmm_layout_gen:    0
      lmm_stripe_offset: -1

If the file is truncated, the later components are initialized:

# truncate -s 1023M /mnt/testfs/pflfile
# lfs getstripe /mnt/testfs/pflfile
/mnt/testfs/pflfile
  lcm_layout_gen:    4
  lcm_mirror_count:  1
  lcm_entry_count:   3
    lcme_id:             1
    lcme_mirror_id:      0
    lcme_flags:          init
    lcme_extent.e_start: 0
    lcme_extent.e_end:   1048576
      lmm_stripe_count:  1
      lmm_stripe_size:   1048576
      lmm_pattern:       raid0
      lmm_layout_gen:    0
      lmm_stripe_offset: 3
      lmm_objects:
      - 0: { l_ost_idx: 3, l_fid: [0x100030000:0x60a:0x0] }

    lcme_id:             2
    lcme_mirror_id:      0
    lcme_flags:          init
    lcme_extent.e_start: 1048576
    lcme_extent.e_end:   16777216
      lmm_stripe_count:  3
      lmm_stripe_size:   1048576
      lmm_pattern:       raid0
      lmm_layout_gen:    65535
      lmm_stripe_offset: 0
      lmm_objects:
      - 0: { l_ost_idx: 0, l_fid: [0x100000000:0x1ca:0x0] }
      - 1: { l_ost_idx: 1, l_fid: [0x100010000:0x8aa:0x0] }
      - 2: { l_ost_idx: 2, l_fid: [0x100020000:0x6aa:0x0] }

    lcme_id:             3
    lcme_mirror_id:      0
    lcme_flags:          init
    lcme_extent.e_start: 16777216
    lcme_extent.e_end:   1073741824
      lmm_stripe_count:  4
      lmm_stripe_size:   2097152
      lmm_pattern:       raid0
      lmm_layout_gen:    65535
      lmm_stripe_offset: 0
      lmm_objects:
      - 0: { l_ost_idx: 0, l_fid: [0x100000000:0x1cb:0x0] }
      - 1: { l_ost_idx: 1, l_fid: [0x100010000:0x8ab:0x0] }
      - 2: { l_ost_idx: 2, l_fid: [0x100020000:0x6ab:0x0] }
      - 3: { l_ost_idx: 3, l_fid: [0x100030000:0x60b:0x0] }

# lfs df
UUID                   1K-blocks        Used   Available Use% Mounted on
testfs-MDT0000_UUID       125368        2596      111536   3% /mnt/testfs[MDT:0]
testfs-MDT0001_UUID       125368        3112      111020   3% /mnt/testfs[MDT:1]
testfs-OST0000_UUID       325368       13796      284412   5% /mnt/testfs[OST:0]
testfs-OST0001_UUID       325368       14304      283904   5% /mnt/testfs[OST:1]
testfs-OST0002_UUID       325368       13664      284544   5% /mnt/testfs[OST:2]
testfs-OST0003_UUID       325368       14692      283516   5% /mnt/testfs[OST:3]

filesystem_summary:      1301472       56456     1136376   5% /mnt/testfs

If the fallocate is run again, this time the space is allocated:

# fallocate -l $((1048576 * 1024)) /mnt/testfs/pflfile
# ls -l /mnt/testfs/pflfile
1048576 -rw-r--r--. 1 root 1073741824 Dec  5 13:37 /mnt/testfs/pflfile
# lfs df
UUID                   1K-blocks        Used   Available Use% Mounted on
testfs-MDT0000_UUID       125368        2596      111536   3% /mnt/testfs[MDT:0]
testfs-MDT0001_UUID       125368        3112      111020   3% /mnt/testfs[MDT:1]
testfs-OST0000_UUID       325368      276964       21244  93% /mnt/testfs[OST:0]
testfs-OST0001_UUID       325368      277472       20736  94% /mnt/testfs[OST:1]
testfs-OST0002_UUID       325368      276832       21376  93% /mnt/testfs[OST:2]
testfs-OST0003_UUID       325368      272740       25468  92% /mnt/testfs[OST:3]

filesystem_summary:      1301472     1104008       88824  93% /mnt/testfs

Comments

[Peter Jones]

Landed for 2.14

[Gerrit Updater]

Oleg Drokin ([e-mail]) merged in patch https://review.whamcloud.com/40885/
Subject: LU-14186 lov: instantiate components layout for fallocate
Project: fs/lustre-release
Branch: master
Current Patch Set:
Commit: 7e25e6c7d0a710ea25aa0ca277e61fbd18ab3dd3

[Gerrit Updater]

Wang Shilong ([e-mail]) uploaded a new patch: https://review.whamcloud.com/40885
Subject: LU-14186 lov: instantiate components layout for fallocate
Project: fs/lustre-release
Branch: master
Current Patch Set: 1
Commit: f7714e0cc88a0aa427cebb2677be2a1ec5209bf2

[Andreas Dilger]

I suspect what is missing is that fallocate() needs to internally get the intent lock with the MDS for the full range of data, like a write() or truncate() call, which will cause it to initialize those components. Then, the later RPCs to the OSTs will "just work".

The initialization of later components is probably not needed for PUNCH_HOLE|KEEP_SIZE, but is needed if PUNCH_HOLE by itself acts like truncate() to extend the size, since the size of the file is determined by the size of the OST objects themselves.