linux-btrfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v2 00/12] btrfs: raid56: use submit-and-wait method to handle raid56 worload
@ 2022-11-01 11:16 Qu Wenruo
  2022-11-01 11:16 ` [PATCH v2 01/12] btrfs: raid56: extract the vertical stripe recovery code into recover_vertical() Qu Wenruo
                   ` (12 more replies)
  0 siblings, 13 replies; 16+ messages in thread
From: Qu Wenruo @ 2022-11-01 11:16 UTC (permalink / raw)
  To: linux-btrfs

[CHANGELOG]
v2:
- Add coverage for raid56 recover and scrub paths
  In fact, with full coverage we can do better cleanups, which gets
  reflected to the changed lines.

- Better naming schemes
  We now have 3 main entrances:
  * recover_rbio()
  * rmw_rbio()
  * scrub_rbio()

  And their work related functions will be called:
  * {recover|rmw|scrub}_rbio_work()
  * {recover|rmw|scrub}_rbio_work_locked()

  The later is for unlock_stripe() to call, where we hold the full
  stripe lock.

- More extracted helpers
  Now we have the following helpers:
  * {recover|rmw|scrub}_assemble_{read|write}_bios()
  * submit_read_bios()
  * submit_write_bios()

Currently btrfs raid56 has very chaotic jumps using endio functions:

E.g. for raid_parity_write(), if we go sub-stripe, the function jumps
would be:

  raid_parity_write()
   |
   v
  __raid56_parity_write()
   |
   v
  partial_stripe_write()
   |
   v
  start_async_work(rmw_work) <<< Delayed work
   |
   v
  raid56_rmw_stripe()
   |
   v
  raid56_rmw_end_io_work() <<< End io jump
   |
   v
  validate_rbio_for_rmw()
   |
   v
  finish_rmw()
   |
   v
  raid_write_end_io() <<< End io jump
   |
   v
  rbio_orig_end_io()

During a simple sub-stripe write, we have to go through over 10
functions, two end_io jump, at least one delayed work.

With this patchset, we only go like this:

  raid_parity_write()
   |
   v
  rmw_rbio_work() <<< Delayed work
   |
   v
  rbio_work()
   |
   v
  rbio_orig_end_io()

And inside rbio_work(), there is no more end io jumps or extra delayed
work.
Everything except IO is single threaded, and the IO is just
submit-and-wait.

This patchset will rework the raid56 write path (recover and scrub path
is not touched yet) by:

- Introduce a single entrance for recover/rmw/scrub.
  The main function will be called {recover|rmw|scrub}_rbio(),
  executed in rmw_worker workqueue.

- Unified handling for all writes (full/sub-stripe, cached/non-cached,
  degraded or not), recover (dedicated, and in writes/scrub path) and
  scrub.

- No special handling for cases we can skip some workload
  E.g. for sub-stripe write, if we have a cached rbio, we can skip the
  read.

  Now we just assemble the read bio list, submit all bios (none in
  this case) and wait for the IO to finish.

  Since we submitted zero bios, the rbio::stripes_pending is 0, the
  wait immediately returned, needing no extra handling.

- No more need for end_io_work or raid56_endio_workers
  Since we don't rely on endio functions to jump to the next step.

By this, we have unified entrance for all raid56 writes, and no extra
jumping/workqueue mess to interrupt the workflow.

This would make later destructive RMW fix much easier to add, as the
timing of each step in RMW cycle should be very easy to grasp.

Thus I hope this series can be merged before the previous RFC series of
destructive RMW fix.

Qu Wenruo (12):
  btrfs: raid56: extract the vertical stripe recovery code into
    recover_vertical()
  btrfs: raid56: extract the pq generation code into a helper
  btrfs: raid56: extract the recovery bio list build code into a helper
  btrfs: raid56: extract sector recovery code into a helper
  btrfs: raid56: switch recovery path to a single function
  btrfs: raid56: extract the rmw bio list build code into a helper
  btrfs: raid56: extract rwm write bios assembly into a helper
  btrfs: raid56: introduce the a main entrance for rmw path
  btrfs: raid56: switch write path to rmw_rbio()
  btrfs: raid56: extract scrub read bio list assembly code into a helper
  btrfs: raid56: switch scrub path to use a single function
  btrfs: remove the unused btrfs_fs_info::endio_raid56_workers and
    btrfs_raid_bio::end_io_work

 fs/btrfs/disk-io.c |    6 +-
 fs/btrfs/fs.h      |    1 -
 fs/btrfs/raid56.c  | 1422 ++++++++++++++++++++------------------------
 fs/btrfs/raid56.h  |    2 +-
 4 files changed, 631 insertions(+), 800 deletions(-)

-- 
2.38.1


^ permalink raw reply	[flat|nested] 16+ messages in thread
end of thread, other threads:[~2022-11-02 23:40 UTC | newest]

Thread overview: 16+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2022-11-01 11:16 [PATCH v2 00/12] btrfs: raid56: use submit-and-wait method to handle raid56 worload Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 01/12] btrfs: raid56: extract the vertical stripe recovery code into recover_vertical() Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 02/12] btrfs: raid56: extract the pq generation code into a helper Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 03/12] btrfs: raid56: extract the recovery bio list build " Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 04/12] btrfs: raid56: extract sector recovery " Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 05/12] btrfs: raid56: switch recovery path to a single function Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 06/12] btrfs: raid56: extract the rmw bio list build code into a helper Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 07/12] btrfs: raid56: extract rwm write bios assembly " Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 08/12] btrfs: raid56: introduce the a main entrance for rmw path Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 09/12] btrfs: raid56: switch write path to rmw_rbio() Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 10/12] btrfs: raid56: extract scrub read bio list assembly code into a helper Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 11/12] btrfs: raid56: switch scrub path to use a single function Qu Wenruo
2022-11-01 11:16 ` [PATCH v2 12/12] btrfs: remove the unused btrfs_fs_info::endio_raid56_workers and btrfs_raid_bio::end_io_work Qu Wenruo
2022-11-02 14:59 ` [PATCH v2 00/12] btrfs: raid56: use submit-and-wait method to handle raid56 worload David Sterba
2022-11-02 22:32   ` Qu Wenruo
2022-11-02 23:34     ` David Sterba

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).