1. 17 Jul, 2018 2 commits
  2. 16 Jul, 2018 2 commits
    • Phoebe Buckheister's avatar
      client: set posix acl to uncached on read · 573d194b
      Phoebe Buckheister authored
      returning the acl without calling forget_cached_acl will cause the
      currently cached acl (default: empty) to be user on all further acl
      accesses for the inode, until the inode is flushed. flushed happen only
      under memory pressure and during unmount, thus the current behaviour is
      far from what users expect.
      573d194b
    • Phoebe Buckheister's avatar
      ctl: improve error message for meta buddy group add failures · a9b44fdc
      Phoebe Buckheister authored
      meta buddy groups that contain the root inode must be created such that
      owner of the non-mirrored root inode becomes the primary of the newly
      created group. adding a group that does not satisfy this condition
      should give the user some info about why the process failed, not just
      *that* it failed and an error message no user can be expected to
      understand.
      
      see #941
      a9b44fdc
  3. 29 Jun, 2018 2 commits
  4. 28 Jun, 2018 2 commits
  5. 22 Jun, 2018 1 commit
  6. 21 Jun, 2018 1 commit
  7. 20 Jun, 2018 3 commits
  8. 19 Jun, 2018 2 commits
  9. 18 Jun, 2018 3 commits
  10. 06 Jun, 2018 4 commits
    • Bernd Lietzow's avatar
      Merge branch '948-owner-calculation-v6' into 'v6' · 2790927d
      Bernd Lietzow authored
      Resolve "Remote IO error when creating the same file from two different mount points simultaneously, with metadata mirroring on"
      
      See merge request beegfs/projects0!788
      2790927d
    • Phoebe Buckheister's avatar
      meta: expect correct owner node ID for lookup-intent stat · d5ec1748
      Phoebe Buckheister authored
      when two clients race to create the same file in a meta-mirrored setup
      only one of them can succeed. the other will have to either fail
      entirely (for exclusive creates) or use the file created by the winning
      client (for non-exclusive creates).
      
      creates are usually accompanied by stat to retrieve important file
      metadata as well. the losing client will see that the file has already
      been created and only attempt to stat the file it has just found.
      during stat it expects the file to be on the wrong node: owner checks
      are done expecting primary-of-group(localNodeID) instead of expecting
      localGroupID. for groups created with the id of their primary server
      this usually succeeds (until failover happens), on others it will fail.
      
      fixes #948
      
      (cherry picked from commit ed46b0cc5e7916632b83cd67e8506573c5b20f0f)
      d5ec1748
    • Bernd Lietzow's avatar
      Merge branch '954-int-overflow-v6' into 'v6' · f25bfe9b
      Bernd Lietzow authored
      Resolve "integer overflow with certain chunk sizes and number of targets"
      
      See merge request beegfs/projects0!785
      f25bfe9b
    • Oliver Steffen's avatar
      common: fix uint overflow in getStripeTargetIndex() · 91ff4e41
      Oliver Steffen authored
      expressed formula differently:
      
         (pos % stripeSetSize) / chunkSize
       = (pos % (numTargets * chunkSize)) / chunkSize
       = (pos / chunkSize) % numTargets
      
       (pos / chunkSize): number of chunks from beginning of file
      
      removes the previously overflowing stripeSetSize variable.
      
      (cherry picked from da4e5efa2b931a10b0d460458316c41f4a639563)
      91ff4e41
  11. 05 Jun, 2018 2 commits
    • Bernd Lietzow's avatar
      Merge branch '947-ofed-4.3-v6' into 'v6' · 3deca9fa
      Bernd Lietzow authored
      Resolve "beegfs-client (v6) on OFED 4.3 and CentOS 7.5 doesn't build (RT #8015)"
      
      See merge request beegfs/projects0!781
      3deca9fa
    • Oliver Steffen's avatar
      client: fix build with OFED4.3 · d435ab39
      Oliver Steffen authored
      OFED 4.3 changed rdma_create_id to a macro, which is not found
      by the feature detection. Using a check for that in the code.
      
      (cherry-picked from 823497768594d0e204be983b01c26b1a511eb053)
      d435ab39
  12. 26 Apr, 2018 3 commits
    • Bernd Lietzow's avatar
      Merge branch 'modsync-v6' into 'v6' · 0d8b1b1c
      Bernd Lietzow authored
      Modsync/locking fixes for v6
      
      See merge request beegfs/projects0!715
      0d8b1b1c
    • Phoebe Buckheister's avatar
      meta: lock moved directory during rename · 15039642
      Phoebe Buckheister authored
      the ctime of the directory is changed (iff the directory lives on the
      local node!), so we must lock it to prevent races with modsync and
      rmdir.
      15039642
    • Phoebe Buckheister's avatar
      meta: lock disposed files during unlink-after-client-removal · ee95848d
      Phoebe Buckheister authored
      removing a client from the system causes the meta server to close file
      handles held by that client. if any file handle refers to an unlinked
      file, that close may remove the file from its disposal directory. if
      unlink and close happen simultaneously they can then race with each
      other, with unlink adding the file to disposal and adding a "changed
      file" modsync event, and close removing the file before the "changed"
      event was synced. this causes the resync to abort with modsync errors
      afterwars.
      ee95848d
  13. 18 Apr, 2018 2 commits
  14. 17 Apr, 2018 2 commits
    • Patrick Reh's avatar
      client: add an option to limit concurrent connection attempts · cd053b9e
      Patrick Reh authored
      This is made especially to counter problems with RDMA over omni-path. It turned out that
      rdma_route_resolve() fails sometimes when there are massive amounts of connections opened
      at the same time (in our case the problem starts when opening several hundred connections at
      once, for example when using a benchmarking tool). This is kind of nondeterministic, so it doesn't
      happen all the time.
      
      The introduced option limits the amount of parallel connection attempts so this doesn't happen
      anymore. There is no recommended value  since it's hard to find out where it stops working.
      A value of 10 to 100 seems to be pretty good though, the connection process even seems to be faster.
      
      (cherry picked from commit 7ab3302c4447d1110ab30dc1dbf178e5855212f4)
      cd053b9e
    • Phoebe Buckheister's avatar
      add more symlinks for toplevel folders · 8fbfca53
      Phoebe Buckheister authored
      v7 and master have renamed their toplevel directories to drop prefixes
      and old symlinks. v6 should just add more symlinks instead for
      consistency: some packages depend on the names of the toplevel
      directories to create names for folder *inside* the packages. those
      should not be disturbed if at all possible.
      8fbfca53
  15. 16 Apr, 2018 2 commits
    • Bernd Lietzow's avatar
      Merge branch 'rhel75-readdir-fix-v6' into 'v6' · b4191cae
      Bernd Lietzow authored
      client: fix readdir in rhel 7.5
      
      See merge request beegfs/projects0!742
      b4191cae
    • Phoebe Buckheister's avatar
      client: fix readdir in rhel 7.5 · 39887b10
      Phoebe Buckheister authored
      backports backported changes to the readdir interface, but reverted
      part of it. other parts were left behind and used in special
      circumstances, ie when the user has previously indicated that the new
      interface is available.
      
      we used the new interface (ever since 3.11 introduced it), but did not
      indicate that we support it (because that's a redhat thing).
      39887b10
  16. 09 Apr, 2018 4 commits
  17. 04 Apr, 2018 3 commits