Skip to contentSkip to conversation

Btrfs changes for 2.6.30-rc

GIT PULL Chris Mason 1 participant 2 messages Apr 21, 2009

lore
Share & tools
lore
#0 lore raw

Hello everyone,

The master branch of:

git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable.git

Has some fixes for oopsen and a cluster of bugs inside of one of the btree readahead functions.

I also jumped on the fsync latency train for btrfs and switched over to WRITE_SYNC for the writes I was going to wait on.

This is a bigger change in btrfs because I had to add a high priority list to the async helper threads, but it had a big impact on performance overall. Before, the writes we were waiting on would get stuck behind the less critical writes in the async helper threads, and now they pop out much faster.

Average fsync latencies concurrent with a streaming writer went down from ~5 seconds (with longer peaks) to < 1 second.

Chris Mason (4):
Btrfs: use WRITE_SYNC for synchronous writes
Btrfs: add a priority queue to the async thread helpers
Btrfs: fix oops on page->mapping->host during writepage
Btrfs: use the right node in reada_for_balance

 fs/btrfs/async-thread.c |   60 ++++++++++++++++++-----
 fs/btrfs/async-thread.h |    2
 fs/btrfs/ctree.c        |   17 ++++--
 fs/btrfs/disk-io.c      |    9 ++-
 fs/btrfs/extent_io.c    |   86 ++++++++++++++++++++++++---------
 fs/btrfs/file.c         |    2
 fs/btrfs/ordered-data.c |    2
 fs/btrfs/volumes.c      |  124 +++++++++++++++++++++++++++++++++++++-----------
 fs/btrfs/volumes.h      |   13 ++++-

-chris

#1 lore raw
Quoted 12 lines · Chris Mason

Hello everyone,

The master branch of:

git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable.git

Has some fixes for oopsen and a cluster of bugs inside of one of the btree readahead functions.

I also jumped on the fsync latency train for btrfs and switched over to WRITE_SYNC for the writes I was going to wait on.

Quoted 9 lines

This is a bigger change in btrfs because I had to add a high priority list to the async helper threads, but it had a big impact on performance overall. Before, the writes we were waiting on would get stuck behind the less critical writes in the async helper threads, and now they pop out much faster.

Average fsync latencies concurrent with a streaming writer went down from ~5 seconds (with longer peaks) to < 1 second.

Jens Axboe hit a bug in fallocate, I've pushed that fix out today as well. Updated shortlog/diffstat:

Chris Mason (5):
Btrfs: use WRITE_SYNC for synchronous writes
Btrfs: add a priority queue to the async thread helpers
Btrfs: fix oops on page->mapping->host during writepage
Btrfs: use the right node in reada_for_balance
Btrfs: fix btrfs fallocate oops and deadlock

 fs/btrfs/async-thread.c |   60 ++++++++++++++++++-----
 fs/btrfs/async-thread.h |    2
 fs/btrfs/ctree.c        |   17 ++++--
 fs/btrfs/disk-io.c      |    9 ++-
 fs/btrfs/extent_io.c    |   86 ++++++++++++++++++++++++---------
 fs/btrfs/file.c         |    6 +-
 fs/btrfs/inode.c        |   36 ++++++++++---
 fs/btrfs/ordered-data.c |    2
 fs/btrfs/volumes.c      |  124 +++++++++++++++++++++++++++++++++++++-----------
 fs/btrfs/volumes.h      |   13 ++++-

-chris