| Message ID | cover.1720448663.git.fdmanana@suse.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | btrfs: some fixes/updates for the extent map shrinker | expand |
On Mon, Jul 08, 2024 at 03:42:42PM +0100, fdmanana@kernel.org wrote: > From: Filipe Manana <fdmanana@suse.com> > > A few fixes for the extent map shrinker that fix some reports from users > where their desktops became very unresponsive. > > Several notes here: > > 1) The first patch was already sent before to the list and it's in > for-next already. It's unrelated to the reports from those users but > it's related to a report from syzbot for a deadlock in case the > shrinker does an iput on an inode with 0 links that needs to be > deleted, and when the inode still has links but it's dirty, it can > make the iput wait for writeback to complete, slowing down the > shrinker. I've included it here just to group things in a logical > way; > > 2) These patches apply to 6.10-rc; > > 3) At least the first 2 patches should go to 6.10 final IMO; > > 4) In case they land in 6.10-rc, then for-next needs to be rebased since > there are minor and trivial conflicts to solve due to the last patch in > for-next that reduces the size of struct btrfs_inode down to 1024 bytes. > > Also the following patch which landed on for-next should be dropped > because it's made obsolete by the second patch in this patchset: > > https://lore.kernel.org/linux-btrfs/cb12212b9c599817507f3978c9102767267625b2.1719825714.git.fdmanana@suse.com/ > > 5) There will be further improvements to the shrinker, namely to > reduce the latency of finding open inodes in a root, but those > will come later and may be too much for 6.10 final. It would also > require different patch versions for 6.10 and 6.11 (for-next) since > in the former we use a rbtree while in the later we have a xarray > now. > > Thanks. > > Filipe Manana (3): > btrfs: use delayed iput during extent map shrinking > btrfs: stop extent map shrinker if reschedule is needed > btrfs: avoid races when tracking progress for extent map shrinking Reviewed-by: Josef Bacik <josef@toxicpanda.com> Thanks, Josef
On Mon, Jul 08, 2024 at 03:42:42PM +0100, fdmanana@kernel.org wrote: > From: Filipe Manana <fdmanana@suse.com> > > A few fixes for the extent map shrinker that fix some reports from users > where their desktops became very unresponsive. > > Several notes here: > > 1) The first patch was already sent before to the list and it's in > for-next already. It's unrelated to the reports from those users but > it's related to a report from syzbot for a deadlock in case the > shrinker does an iput on an inode with 0 links that needs to be > deleted, and when the inode still has links but it's dirty, it can > make the iput wait for writeback to complete, slowing down the > shrinker. I've included it here just to group things in a logical > way; > > 2) These patches apply to 6.10-rc; > > 3) At least the first 2 patches should go to 6.10 final IMO; > > 4) In case they land in 6.10-rc, then for-next needs to be rebased since > there are minor and trivial conflicts to solve due to the last patch in > for-next that reduces the size of struct btrfs_inode down to 1024 bytes. > > Also the following patch which landed on for-next should be dropped > because it's made obsolete by the second patch in this patchset: > > https://lore.kernel.org/linux-btrfs/cb12212b9c599817507f3978c9102767267625b2.1719825714.git.fdmanana@suse.com/ > > 5) There will be further improvements to the shrinker, namely to > reduce the latency of finding open inodes in a root, but those > will come later and may be too much for 6.10 final. It would also > require different patch versions for 6.10 and 6.11 (for-next) since > in the former we use a rbtree while in the later we have a xarray > now. > > Thanks. > > Filipe Manana (3): > btrfs: use delayed iput during extent map shrinking > btrfs: stop extent map shrinker if reschedule is needed > btrfs: avoid races when tracking progress for extent map shrinking Thanks for the fixes, this seems to be urgent, I'd rather not delay getting the fixes via stable with a 2 week delay. I'm going to send a pull request for 6.10, the problems have been reported by early testing users and will be most likely seen by more after release. I hope this gives it enough justification despite the patch sizes and ETA of 6.10 release.
From: Filipe Manana <fdmanana@suse.com> A few fixes for the extent map shrinker that fix some reports from users where their desktops became very unresponsive. Several notes here: 1) The first patch was already sent before to the list and it's in for-next already. It's unrelated to the reports from those users but it's related to a report from syzbot for a deadlock in case the shrinker does an iput on an inode with 0 links that needs to be deleted, and when the inode still has links but it's dirty, it can make the iput wait for writeback to complete, slowing down the shrinker. I've included it here just to group things in a logical way; 2) These patches apply to 6.10-rc; 3) At least the first 2 patches should go to 6.10 final IMO; 4) In case they land in 6.10-rc, then for-next needs to be rebased since there are minor and trivial conflicts to solve due to the last patch in for-next that reduces the size of struct btrfs_inode down to 1024 bytes. Also the following patch which landed on for-next should be dropped because it's made obsolete by the second patch in this patchset: https://lore.kernel.org/linux-btrfs/cb12212b9c599817507f3978c9102767267625b2.1719825714.git.fdmanana@suse.com/ 5) There will be further improvements to the shrinker, namely to reduce the latency of finding open inodes in a root, but those will come later and may be too much for 6.10 final. It would also require different patch versions for 6.10 and 6.11 (for-next) since in the former we use a rbtree while in the later we have a xarray now. Thanks. Filipe Manana (3): btrfs: use delayed iput during extent map shrinking btrfs: stop extent map shrinker if reschedule is needed btrfs: avoid races when tracking progress for extent map shrinking fs/btrfs/disk-io.c | 2 + fs/btrfs/extent_map.c | 123 ++++++++++++++++++++++++++--------- fs/btrfs/fs.h | 1 + include/trace/events/btrfs.h | 18 ++--- 4 files changed, 107 insertions(+), 37 deletions(-)