case insensitivity and kernel release 4.7+

[RJVB]

I am experiencing an issue with a case-insensitive dataset running 0.6.5.8 built for Ubuntu 14.04 (see ppa:rjvbertin/misc) under mainstream kernels 4.7.2 and 4.8.15 built with the ConKolivas patches.

Any process trying to access a mounted case-insensitive dataset hangs, and that includes zfs.mount when mounting a case-sensitive child of the case-insensitive dataset. With hanging I mean that the process is completely unresponsive to signals and interrupts though it does not show up as a zombie:

For instance:

Dec 21 09:18:42 Bola kernel: [  734.989027] INFO: task updatedb.mlocat:8044 blocked for more than 120 seconds.
Dec 21 09:18:42 Bola kernel: [  734.989055]       Tainted: P     U     OE   4.8.15-ck8-mainline-core2-rjvb #1
Dec 21 09:18:42 Bola kernel: [  734.989067] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
Dec 21 09:18:42 Bola kernel: [  734.989080] updatedb.mlocat D ffff9b3798e07aa0     0  8044      1 0x00000004
Dec 21 09:18:42 Bola kernel: [  734.989107]  ffff9b3798e07aa0 ffff9b383fc990a0 0000000000000001 000000861b669c56
Dec 21 09:18:42 Bola kernel: [  734.989130]  ffff9b3812615348 0000000000000001 010000018020001a 00000000434464a3
Dec 21 09:18:42 Bola kernel: [  734.989148]  ffff9b3761593038 0000000000000000 0000000000000000 0000000000000001
Dec 21 09:18:42 Bola kernel: [  734.989169] Call Trace:
Dec 21 09:18:42 Bola kernel: [  734.989256]  [<ffffffff8ba93f7a>] schedule+0x3a/0xc0
Dec 21 09:18:42 Bola kernel: [  734.989273]  [<ffffffff8b2f15ba>] d_alloc_parallel+0x4da/0x5e0
Dec 21 09:18:42 Bola kernel: [  734.989290]  [<ffffffff8b0d9e50>] ? wake_up_process+0x20/0x20
Dec 21 09:18:42 Bola kernel: [  734.989323]  [<ffffffff8b2f1974>] d_add_ci+0x94/0x1e0
Dec 21 09:18:42 Bola kernel: [  734.989624]  [<ffffffffc06829d9>] zpl_lookup+0x1d9/0x210 [zfs]
Dec 21 09:18:42 Bola kernel: [  734.989661]  [<ffffffff8b2d8445>] lookup_slow+0x95/0x170
Dec 21 09:18:42 Bola kernel: [  734.989695]  [<ffffffff8b2dd47e>] path_lookupat+0x61e/0xac0
Dec 21 09:18:42 Bola kernel: [  734.989967]  [<ffffffffc065b287>] ? zfs_readdir+0x307/0x540 [zfs]
Dec 21 09:18:42 Bola kernel: [  734.989982]  [<ffffffff8b2ddaa0>] filename_lookup+0x180/0x2f0
Dec 21 09:18:42 Bola kernel: [  734.989996]  [<ffffffff8b0d627b>] ? preempt_count_add+0xbb/0xd0
Dec 21 09:18:42 Bola kernel: [  734.990009]  [<ffffffff8ba9c11a>] ? _raw_spin_unlock+0x1a/0x30
Dec 21 09:18:42 Bola kernel: [  734.990253]  [<ffffffffc05e2d9d>] ? rrw_exit+0x4d/0x130 [zfs]
Dec 21 09:18:42 Bola kernel: [  734.990313]  [<ffffffff8b2e0993>] ? getname_flags.part.39+0x53/0x200
Dec 21 09:18:42 Bola kernel: [  734.990327]  [<ffffffff8b2e0b9a>] user_path_at_empty+0x5a/0x80
Dec 21 09:18:42 Bola kernel: [  734.990343]  [<ffffffff8b2d0893>] vfs_fstatat.constprop.5+0x33/0x170
Dec 21 09:18:42 Bola kernel: [  734.990356]  [<ffffffff8b2d09f5>] SYSC_newlstat+0x15/0x30
Dec 21 09:18:42 Bola kernel: [  734.990370]  [<ffffffff8b2ea5ce>] ? SyS_poll+0x7e/0x160
Dec 21 09:18:42 Bola kernel: [  734.990381]  [<ffffffff8b2d1c6e>] SyS_newlstat+0xe/0x10
Dec 21 09:18:42 Bola kernel: [  734.990393]  [<ffffffff8ba9c876>] entry_SYSCALL_64_fastpath+0x1e/0xa8

The dataset properties:

NAME                                       PROPERTY              VALUE                                        SOURCE
bolaLNX/bertinsDevel/Scratch/caseInsensed  type                  filesystem                                   -
bolaLNX/bertinsDevel/Scratch/caseInsensed  creation              Thu Sep 29 10:04 2016                        -
bolaLNX/bertinsDevel/Scratch/caseInsensed  used                  672K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  available             131G                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  referenced            576K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  compressratio         1.35x                                        -
bolaLNX/bertinsDevel/Scratch/caseInsensed  mounted               yes                                          -
bolaLNX/bertinsDevel/Scratch/caseInsensed  quota                 none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  reservation           none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  recordsize            128K                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  mountpoint            /home/bertin/cworks/Scratch/caseInsenseTest  local
bolaLNX/bertinsDevel/Scratch/caseInsensed  sharenfs              off                                          inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  checksum              off                                          inherited from bolaLNX/bertinsDevel/Scratch
bolaLNX/bertinsDevel/Scratch/caseInsensed  compression           lz4                                          inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  atime                 off                                          inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  devices               on                                           default
bolaLNX/bertinsDevel/Scratch/caseInsensed  exec                  on                                           default
bolaLNX/bertinsDevel/Scratch/caseInsensed  setuid                on                                           default
bolaLNX/bertinsDevel/Scratch/caseInsensed  readonly              off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  zoned                 off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  snapdir               visible                                      inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  aclinherit            restricted                                   default
bolaLNX/bertinsDevel/Scratch/caseInsensed  canmount              on                                           default
bolaLNX/bertinsDevel/Scratch/caseInsensed  xattr                 sa                                           inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  copies                1                                            inherited from bolaLNX/bertinsDevel/Scratch
bolaLNX/bertinsDevel/Scratch/caseInsensed  version               5                                            -
bolaLNX/bertinsDevel/Scratch/caseInsensed  utf8only              on                                           -
bolaLNX/bertinsDevel/Scratch/caseInsensed  normalization         formD                                        -
bolaLNX/bertinsDevel/Scratch/caseInsensed  casesensitivity       insensitive                                  -
bolaLNX/bertinsDevel/Scratch/caseInsensed  vscan                 off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  nbmand                off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  sharesmb              on                                           inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  refquota              none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  refreservation        none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  primarycache          all                                          inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  secondarycache        all                                          inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  usedbysnapshots       0                                            -
bolaLNX/bertinsDevel/Scratch/caseInsensed  usedbydataset         576K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  usedbychildren        96K                                          -
bolaLNX/bertinsDevel/Scratch/caseInsensed  usedbyrefreservation  0                                            -
bolaLNX/bertinsDevel/Scratch/caseInsensed  logbias               throughput                                   inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  dedup                 off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  mlslabel              none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  sync                  standard                                     inherited from bolaLNX/bertinsDevel/Scratch
bolaLNX/bertinsDevel/Scratch/caseInsensed  refcompressratio      1.38x                                        -
bolaLNX/bertinsDevel/Scratch/caseInsensed  written               576K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  logicalused           596K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  logicalreferenced     556K                                         -
bolaLNX/bertinsDevel/Scratch/caseInsensed  filesystem_limit      none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  snapshot_limit        none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  filesystem_count      none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  snapshot_count        none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  snapdev               hidden                                       default
bolaLNX/bertinsDevel/Scratch/caseInsensed  acltype               off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  context               none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  fscontext             none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  defcontext            none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  rootcontext           none                                         default
bolaLNX/bertinsDevel/Scratch/caseInsensed  relatime              off                                          default
bolaLNX/bertinsDevel/Scratch/caseInsensed  redundant_metadata    most                                         inherited from bolaLNX
bolaLNX/bertinsDevel/Scratch/caseInsensed  overlay               on                                           inherited from bolaLNX

Running kernel 4.5.7 with the same patches I do not have this issue, suggestion that it is related to the modifications made to enable ZFS on 4.7+ kernels.

[kernelOfTruth]

Can you test whether this also occurs with kernels that use CFS (stock) cpu scheduler ?

[loli10K]

running 0.6.5.8
under mainstream kernels 4.7.2 and 4.8.15

@RJVB this should be fixed by #5141 (Linux 4.7: fix deadlock during lookup on case-insensitive). You can find a patch for 0.6.5.8 in the comments.

[RJVB]

Great, I'll test that straightaway. If that fixes the issue it is highly likely that the CK patches are not to blame.

[RJVB]

On Wednesday December 21 2016 03:53:04 LOLi wrote: @RJVB this should be fixed by #5141 (_Linux 4.7: fix deadlock during lookup on case-insensitive_). You can find a patch for 0.6.5.8 in the comments.

I can confirm now: that patch fixed my issue on 4.8.15 , with the ck8 patch still in place. FWIW: Con Kolivas has been working on a new scheduler ("muqss") that should be a bit more appropriate for normal desktop use, esp. if you often run applications at background priority levels (http://www.phoronix.com/scan.php?page=news_item&px=MuQSS-First-Major-Release)

[gmelikov]

@RJVB as i see the issue is already solved. Feel free to reopen if it's not.