Meeting notes for 2013

Meeting of Aug 29

Attending: Brian, Eric, Mark, Andrew, Troy, Ben, Rich, Phil, Geoffrey

3.12 content:

Remaining patches from last meeting (although we didn't discuss them in this meeting):

• From Chandra Seetharaman - [3/3] gfs2: Add support for the Q_XGETQSTATV (Waiting from Ack from Steven Whitehouse)
• From James Dingwall - Cleancache support in XFS (Although this is a one liner, this needs serious review)
• From Jan Kara - two direct-io patches (need review)

Patches we discussed during this meeting:

• From David Chinner - three patches that fix regressions in commit 00574da1 (to be pulled in)
• From Brian Foster - xfs: check correct status variable for xfs_inobt_get_rec() call (ready to be pulled in)
• From David Chinner - xfs: inode log reservations are too small (Mark noted that this fixes some asserts, it's ready to be pulled in)
• From David Chinner - xfs: finish removing IOP_* macros. (ready to be pulled in)
• From David Chinner - xfs: prevent transient corrupt states during log recovery (Ben was reluctant to pull this in for 3.12, but Eric wanted it, and Mark said it's ready, so it will be pulled in for 3.12)
• From David Chinner - XFS: Assertion failed: first <= last && last < BBTOB(bp->b_length), file: fs/xfs/xfs_trans_buf.c, line: 568

This is a directory bug found by Brian.  Mark thinks that there are actually two bugs here, this fixes one.  Pull this in for 3.12.

• From Eric Sandeen - Fix wrong flag ASSERT in xfs_attr_shortform_getvalue (Ben to pull this in)

Other items:

• Mark expressed some concerns about being able to turn crc calculation off in some situations, and there was some discussion of how it might be done.
• Ben noted that Chandra asked about his quotactl patch, there may be further discussion on what Christoph wants done, and this might be work that needs to go in past -rc1.
• Eric discussed the remaining issues related to the crc work. Mark noted that the kernel work is mostly pulled in. Ben noted that the biggest remaining piece is uerspace.

Discussion of xfstests:

• Eric and Rich discussed the merits of two methods of breaking up the xfstests development trees by filesystem:

1) centralized tree with multiple committers
2) distributed trees with pull requests

• Rich to discuss it further with Troy.

Meeting of Aug 22

Attending: Brian, Carlos, Troy, Rich, Geoffrey, Mark, Andrew, Ben

3.12 content:

Remaining patches:

• From Jan Kara - two direct-io patches (need review)
• From Jeff Liu - [v2] xfs: Simplify xfs_ail_min() with list_first_entry_or_null() (need review)
• From Richard Weinberger - xfs: Register hotcpu notifier after initialization (reviewed, applied, not pushed yet)
• From Dan Carpenter - xfs: check for underflow in xfs_iformat_fork() (BPM to ping Dan to see if he'd like to repost)
• From zwu.kernel@gmail.com - [v4] xfs: introduce object readahead to log recovery (BPM to finish review)
• From Chandra Seetharaman - [3/3] gfs2: Add support for the Q_XGETQSTATV (Waiting from Ack from Steven Whitehouse)
• From James Dingwall - Cleancache support in XFS (Although this is a one liner, this needs serious review)

Userspace:

• BPM to update his patch for the di_pad regression.
• Rich mentioned that eric submitted a fix for a bug that was reported, pulled these in:

xfs_repair: zero out unused parts of superblocks
xfs_repair: add prototype for alloc_ex_data()

• Mark to repost his SEEK_DATA/SEEK_HOLE patch
• Rich pulled in a couple of one liner fixes for bugs that were detected by coverity.
• Dimitry Monakov has posted a series on 7/9 that needs review.

Plumbers Conf:

• Carlos is working on his talk on thin provisioning for the plumbers conference, also doing some benchmarking related to snapshots. May have interesting numbers for the next meeting.

xfstests trees:

• Rich to make a proposal regarding the xfstest trees on the list. (This was prompted by a discussion on irc.)

Meeting of Aug 15

Attending: Troy, Geoffrey, Mark, Ben, Andrew, Rich, Phil, Carlos, Brian, Eric, Dave

3.12:

• Ben will pull in Dwight's kernel work.
• Rich will pull in Dwight's tests.
• Ben to pull in Chandra's Kernel work, will apply fixes for Rich's comments at commit time, as per Chandra's request.
• Rich pulled in Chandra's userspace work for quotas.
• Ben to pull in Zwu's log recovery readahead work.
• Ben to pull in Zwu's comment fixes.
• There was a short discussion of secondary superblocks and crcs, this is Eric's patch 'xfs: don't verify checksum on non-V5 superblocks'. Eric thinks this would wait for 3.13.
• Dave will repost the libxfs sync with the 3.12 kernel code 'maybe today'.
• Ben to commit 'xfs: use reference counts to free clean buffer items'.
• We discussed the dirent filetype feature 'xfs: Add read-only support for dirent filetype field' and 'xfs: Add write support for dirent filetype field' at length and settled on several actions for getting this in:

* Eric volunteered to do an xfstest
* xfs_repair, Dave said he was working on this
* v4 support in the kernel to be done by Mark
* v4 support for mkfs.xfs to be done by Mark
* Plan for 3.13

Meeting of Jul 19

Attending: Ric, Dave, Brian, Chandra, Troy, Ben, Mark, Andrew, Geoffrey, Eric

3.11:

• Ben notes the remaining content is limited to two patches from Chandra's quota work
• Chandra, Dave, and Ben discussed the remaining issues with the patches, and Ben requested that when user/group/prj quotas are active the old interface continue to work for compatability
• Chandra agreed to repost the patch without the -EINVAL return code
• A bugfix from Chandra for a CIL deadlock (will be appropriate for 3.11) is posted to the list
• The quota related hang at unmount in xfs/50 is reproduced by Dave, and Jeff. Chandra has not hit it
• Dave notes a log space hang with generic/204 with a 512 bytes block size
• Dave notes a log reservation overflow in generic/070

3.12:

• Dave notes that there are changes to shrinkers and LRU lists coming through the MM tree
• Dave is going to post the rearrangement for libxfs later this morning
• Ben notes there is work from Dwight, Jeff, and Chandra on the list for 3.12
• Log reservation size verification from Jeff
• Dave split his CIL optimisation patch into 5 separate patches
• Dave has been working on dtype support to optimise lookups in directories to avoid having to stat each inode
• Dave is working on swapextents with CRCS.
• Dave requested an xfsprogs release to go with 3.12
• Mark reported a bug with xfs/227, which Dave things is likely related to the log reservation overflow in generic/070

Meeting of Jul 12

Attending: Dave, Chandra, Andrew, Brian, Mark, Ben, Troy

• Chandra posted his quotas series again yesterday, will separate out the QSTAT work, and post a new rev soon
• Dave suggested a complete separation of the new QSTAT interface from the old QSTAT interface
• Discussed the proper behavior for the old QSTAT interface when user/group/proj quotas are all active, and came to the conclusion that it can return -EINVAL
• Agreed we'd like to get the remaining patch into 3.11, but are aware it is very close to the merge window
• Dave mentioned that lately we've been rushing to get stuff in at the last minute, and it would be better to get things in earlier in the development window. Ben agrees.
• Dave is going to repost the rearrangement of code for libxfs syncs for 3.12
• Dave mentioned that there are changes coming up which are going through the MM trees.

Meeting of Jun 6

Attending: Carlos, Brian, Rich, Ben, Troy, Mark, Dave, Geoffrey, Andrew

• Ben sent pull request with remaining 3.10 content
• Dave has been focusing on a log recovery problem reported by Dave Jones, and a corruption issue reported by Sage
• Ben is planning on pulling in crc userspace work into a 'crc-dev' branch once Dave reposts
• Jeff had a talk at Linux Con

• Around the table:
  • Ben: Working an NFS bug, sent pull request, runnign xfstests in the background
  • Rich: Working on a SuSE bug
  • Geoffrey: Nothing XFS related at this time
  • Mark: Parent pointers, userspace patches for CRCs, a SuSE bug
  • Andrew: Working an NFS bug
  • Dave: Log recovery issue with Dave Jones, stuff for 3.11: Inode create transaction, minor optimizations for inode creates, Doco updates for mount options in xfs.5 manpage which is shipped with xfsprogs. Slowly fixing things in CRC code
  • Brian: Helping out with reviews, working some RH internal xfs bugs
  • Carlos: Working on an issue with ACLs and set group id

Meeting of 30 May 2013

Held 3:00pm CDT

Attending: Phil White, Brian Foster, Mark Tinguely, Ben Myers, Geoffrey Wehrman, Andrew Dahl, Troy McCorkell, Rich Johnston, Dave Chinner, Eric Sandeen

• Mark noted that he hit the BMV_DELALLOC when testing patches for 3.10-rc4. Dave mentioned that this test hangs on 32 bit machines.

• Ben has a list of patches he's comfortable with for 3.10-rc4

Fixes:

• f648167f xfs: avoid nesting transactions in xfs_qm_scall_setqlim()
• xfs: kill suid/sgid through the truncate path.

Brian said this is low risk Dave said this should go to -stable kernels since it has potential security implications

Wanted for CRCS:

• xfs: fix split buffer vector log recovery support

Mark says this one isn't CRC specific Dave says this has problem has been there since 2002 Ben asked does this need to go back to -stable Dave said probably not since we don't have bug reports

• xfs: fix dir3 freespace block corruption

Dave: fix isn't for the padding but for the corruption related to the initialization. Addname_int removes initialization of a copule of fields using a dirv2 header format. Ben: So I missed the point in my review.

• 9a5b6dee xfs: fix incorrect remote symlink block count

Ben and brian reviewed this... fairly straightforward.

• xfs: disable swap extents ioctl on CRC enabled filesystems

Ben: this prevents crc users from corrupting their filesystems with xfs_fsr.

After discussing the patches on Ben's list, we discussed whether the rest of the queue should be pulled in for 3.10. 1) attribute related fixes, 2) fixes for quotas, 3) fixes for inode unlinked list, 4) max acl count...

Ben was resistant to pulling this work in due to the risk it poses to non-crc users. Dave and Eric wanted it pulled in. Geoffrey mentioned that the ACL update need not be tied to the CRC feature bit, and that not all users will want to use CRCs due to the performance impact.

<discussion ensued>

Actions: >>> We're going to do a pull request with the stuff BPM outlined plus the attr stuff for 3.10-rc4. >>> The inode unlinked list work and quota stuff still needs review for a pull request in 3.10-rc5.