Reduce object locking in zfs.
ClosedPublic
Actions

Authored by jeff on Feb 27 2020, 3:13 AM.

Details

Reviewers

kib
markj
mav
mmacy

Commits

rS358443: Eliminate object locking in zfs where possible with the new lockless grab

Summary

The VM no longer requires the object lock to lookup pages. If it need to create pages, it will acquire it depending on grab flags.

I have tested this only with the zfs tests in the stress2 suite. I have not performance tested it. In some workloads on larger machines the win is substantial for things like exec/tmpfs/md/buf cache/fault.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

jeff created this revision.Feb 27 2020, 3:13 AM

Herald added a subscriber: delphij. · View Herald TranscriptFeb 27 2020, 3:13 AM

Harbormaster completed remote builds in B29634: Diff 68868.Feb 27 2020, 3:13 AM

jeff edited the summary of this revision. (Show Details)Feb 27 2020, 3:14 AM

jeff added reviewers: kib, markj, mav, mmacy.

jeff set the repository for this revision to rS FreeBSD src repository - subversion.

Herald added a subscriber: imp. · View Herald TranscriptFeb 27 2020, 3:14 AM

markj accepted this revision.Feb 27 2020, 1:58 PM

This revision is now accepted and ready to land.Feb 27 2020, 1:58 PM

kib accepted this revision.Feb 28 2020, 1:23 PM

mmacy accepted this revision.Feb 28 2020, 7:05 PM

Closed by commit rS358443: Eliminate object locking in zfs where possible with the new lockless grab (authored by jeff). · Explain WhyFeb 28 2020, 8:30 PM

This revision was automatically updated to reflect the committed changes.

jeff added a commit: rS358443: Eliminate object locking in zfs where possible with the new lockless grab.

I think deadlock at https://lists.freebsd.org/pipermail/freebsd-current/2020-May/076100.html may be related to this but I don't understand a lot of it.
Edit: Nevermind, reverting did not help. Sorry for the noise.

In D23848#546300, @bdrewery wrote:

I think deadlock at https://lists.freebsd.org/pipermail/freebsd-current/2020-May/076100.html may be related to this but I don't understand a lot of it.

The basic problem is an LOR between the page busy lock and ZFS range locks. zfs_write() locks the range, performs the write, and then harmonizes the update with data in the page cache. This last operation is done in update_pages(), and busies the VM's copy of the data. In the other thread, vm_fault() busies the fault page during a read fault and then calls zfs_getpages(), which locks the range.

Revision Contents
Changeset List

Path

Size

head/

sys/

cddl/

contrib/

opensolaris/

uts/

common/

fs/

zfs/

dmu.c

6 lines

zfs_vnops.c

46 lines

Diff 68968

View Options

head/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c