mbox series

[RFC,-next,v2,00/41] md/md-bitmap: introduce bitmap_operations and make structure internel

Message ID 20240814071113.346781-1-yukuai1@huaweicloud.com (mailing list archive)
Headers show
Series md/md-bitmap: introduce bitmap_operations and make structure internel | expand

Message

Yu Kuai Aug. 14, 2024, 7:10 a.m. UTC
From: Yu Kuai <yukuai3@huawei.com>

Changes in v2:
 - add patch 1-8 to prevent dereference bitmap directly, and the last
 patch to make bitmap structure internel.
 - use plain function alls "bitmap_ops->xxx()" directly;

The background is that currently bitmap is using a global spin_lock,
cauing lock contention and huge IO performance degration for all raid
levels.

However, it's impossible to implement a new lock free bitmap with
current situation that md-bitmap exposes the internal implementation
with lots of exported apis. Hence bitmap_operations is invented, to
describe bitmap core implementation, and a new bitmap can be introduced
with a new bitmap_operations, we only need to switch to the new one
during initialization.

And with this we can build bitmap as kernel module, but that's not
our concern for now.

Noted I just compile this patchset, not tested yet.

Yu Kuai (41):
  md/raid1: use md_bitmap_wait_behind_writes() in raid1_read_request()
  md/md-bitmap: replace md_bitmap_status() with a new helper
    md_bitmap_get_stats()
  md: use new helper md_bitmap_get_stats() in update_array_info()
  md/md-bitmap: add 'events_cleared' into struct md_bitmap_stats
  md/md-bitmap: add 'sync_size' into struct md_bitmap_stats
  md/md-bitmap: add 'file_pages' into struct md_bitmap_stats
  md/md-bitmap: add 'behind_writes' and 'behind_wait' into struct
    md_bitmap_stats
  md/md-cluster: use helper md_bitmap_get_stats() to get pages in
    resize_bitmaps()
  md/md-bitmap: add a new helper md_bitmap_set_pages()
  md/md-bitmap: introduce struct bitmap_operations
  md/md-bitmap: simplify md_bitmap_create() + md_bitmap_load()
  md/md-bitmap: merge md_bitmap_create() into bitmap_operations
  md/md-bitmap: merge md_bitmap_load() into bitmap_operations
  md/md-bitmap: merge md_bitmap_destroy() into bitmap_operations
  md/md-bitmap: merge md_bitmap_flush() into bitmap_operations
  md/md-bitmap: make md_bitmap_print_sb() internal
  md/md-bitmap: merge md_bitmap_update_sb() into bitmap_operations
  md/md-bitmap: merge md_bitmap_status() into bitmap_operations
  md/md-bitmap: remove md_bitmap_setallbits()
  md/md-bitmap: merge bitmap_write_all() into bitmap_operations
  md/md-bitmap: merge md_bitmap_dirty_bits() into bitmap_operations
  md/md-bitmap: merge md_bitmap_startwrite() into bitmap_operations
  md/md-bitmap: merge md_bitmap_endwrite() into bitmap_operations
  md/md-bitmap: merge md_bitmap_start_sync() into bitmap_operations
  md/md-bitmap: remove the parameter 'aborted' for md_bitmap_end_sync()
  md/md-bitmap: merge md_bitmap_end_sync() into bitmap_operations
  md/md-bitmap: merge md_bitmap_close_sync() into bitmap_operations
  md/md-bitmap: mrege md_bitmap_cond_end_sync() into bitmap_operations
  md/md-bitmap: merge md_bitmap_sync_with_cluster() into
    bitmap_operations
  md/md-bitmap: merge md_bitmap_unplug_async() into md_bitmap_unplug()
  md/md-bitmap: merge bitmap_unplug() into bitmap_operations
  md/md-bitmap: merge md_bitmap_daemon_work() into bitmap_operations
  md/md-bitmap: pass in mddev directly for md_bitmap_resize()
  md/md-bitmap: merge md_bitmap_resize() into bitmap_operations
  md/md-bitmap: merge get_bitmap_from_slot() into bitmap_operations
  md/md-bitmap: merge md_bitmap_copy_from_slot() into struct
    bitmap_operation.
  md/md-bitmap: merge md_bitmap_set_pages() into struct
    bitmap_operations
  md/md-bitmap: merge md_bitmap_free() into bitmap_operations
  md/md-bitmap: merge md_bitmap_wait_behind_writes() into
    bitmap_operations
  md/md-bitmap: merge md_bitmap_enabled() into bitmap_operations
  md/md-bitmap: make in memory structure internal

 drivers/md/dm-raid.c     |   7 +-
 drivers/md/md-bitmap.c   | 561 +++++++++++++++++++++++++++++----------
 drivers/md/md-bitmap.h   | 272 ++++---------------
 drivers/md/md-cluster.c  |  79 +++---
 drivers/md/md.c          | 133 ++++++----
 drivers/md/md.h          |   3 +-
 drivers/md/raid1-10.c    |   9 +-
 drivers/md/raid1.c       |  78 +++---
 drivers/md/raid10.c      |  73 ++---
 drivers/md/raid5-cache.c |   8 +-
 drivers/md/raid5.c       |  62 ++---
 11 files changed, 731 insertions(+), 554 deletions(-)

Comments

Su Yue Aug. 19, 2024, 8:18 a.m. UTC | #1
On Wed 14 Aug 2024 at 15:10, Yu Kuai <yukuai1@huaweicloud.com> 
wrote:

> From: Yu Kuai <yukuai3@huawei.com>
>
> Changes in v2:
>  - add patch 1-8 to prevent dereference bitmap directly, and the 
>  last
>  patch to make bitmap structure internel.
>  - use plain function alls "bitmap_ops->xxx()" directly;
>
> The background is that currently bitmap is using a global 
> spin_lock,
> cauing lock contention and huge IO performance degration for all 
> raid
> levels.
>
> However, it's impossible to implement a new lock free bitmap 
> with
> current situation that md-bitmap exposes the internal 
> implementation
> with lots of exported apis. Hence bitmap_operations is invented, 
> to
> describe bitmap core implementation, and a new bitmap can be 
> introduced
> with a new bitmap_operations, we only need to switch to the new 
> one
> during initialization.
>
Is the new bitmap in plan?
>
> And with this we can build bitmap as kernel module, but that's 
> not
> our concern for now.
>
> Noted I just compile this patchset, not tested yet.
>

I've looked through the patchset almostly. The changes are quite
straightforward. IMO, it's good timing to  test it and drop RFC in 
next
version.

> Yu Kuai (41):
>   md/raid1: use md_bitmap_wait_behind_writes() in 
>   raid1_read_request()
>   md/md-bitmap: replace md_bitmap_status() with a new helper
>     md_bitmap_get_stats()
>   md: use new helper md_bitmap_get_stats() in 
>   update_array_info()
>   md/md-bitmap: add 'events_cleared' into struct md_bitmap_stats
>   md/md-bitmap: add 'sync_size' into struct md_bitmap_stats
>   md/md-bitmap: add 'file_pages' into struct md_bitmap_stats
>   md/md-bitmap: add 'behind_writes' and 'behind_wait' into 
>   struct
>     md_bitmap_stats
>   md/md-cluster: use helper md_bitmap_get_stats() to get pages 
>   in
>     resize_bitmaps()
>   md/md-bitmap: add a new helper md_bitmap_set_pages()
>   md/md-bitmap: introduce struct bitmap_operations
>   md/md-bitmap: simplify md_bitmap_create() + md_bitmap_load()
>   md/md-bitmap: merge md_bitmap_create() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_load() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_destroy() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_flush() into bitmap_operations
>   md/md-bitmap: make md_bitmap_print_sb() internal
>   md/md-bitmap: merge md_bitmap_update_sb() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_status() into bitmap_operations
>   md/md-bitmap: remove md_bitmap_setallbits()
>   md/md-bitmap: merge bitmap_write_all() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_dirty_bits() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_startwrite() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_endwrite() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_start_sync() into 
>   bitmap_operations
>   md/md-bitmap: remove the parameter 'aborted' for 
>   md_bitmap_end_sync()
>   md/md-bitmap: merge md_bitmap_end_sync() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_close_sync() into 
>   bitmap_operations
>   md/md-bitmap: mrege md_bitmap_cond_end_sync() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_sync_with_cluster() into
>     bitmap_operations
>   md/md-bitmap: merge md_bitmap_unplug_async() into 
>   md_bitmap_unplug()
>   md/md-bitmap: merge bitmap_unplug() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_daemon_work() into 
>   bitmap_operations
>   md/md-bitmap: pass in mddev directly for md_bitmap_resize()
>   md/md-bitmap: merge md_bitmap_resize() into bitmap_operations
>   md/md-bitmap: merge get_bitmap_from_slot() into 
>   bitmap_operations
>   md/md-bitmap: merge md_bitmap_copy_from_slot() into struct
>     bitmap_operation.
>   md/md-bitmap: merge md_bitmap_set_pages() into struct
>     bitmap_operations
>   md/md-bitmap: merge md_bitmap_free() into bitmap_operations
>   md/md-bitmap: merge md_bitmap_wait_behind_writes() into
>     bitmap_operations
>   md/md-bitmap: merge md_bitmap_enabled() into bitmap_operations
>   md/md-bitmap: make in memory structure internal
>
>  drivers/md/dm-raid.c     |   7 +-
>  drivers/md/md-bitmap.c   | 561 
>  +++++++++++++++++++++++++++++----------
>  drivers/md/md-bitmap.h   | 272 ++++---------------
>  drivers/md/md-cluster.c  |  79 +++---
>  drivers/md/md.c          | 133 ++++++----
>  drivers/md/md.h          |   3 +-
>  drivers/md/raid1-10.c    |   9 +-
>  drivers/md/raid1.c       |  78 +++---
>  drivers/md/raid10.c      |  73 ++---
>  drivers/md/raid5-cache.c |   8 +-
>  drivers/md/raid5.c       |  62 ++---
>  11 files changed, 731 insertions(+), 554 deletions(-)
Yu Kuai Aug. 19, 2024, 11:13 a.m. UTC | #2
Hi,

在 2024/08/19 16:18, Su Yue 写道:
> 
> On Wed 14 Aug 2024 at 15:10, Yu Kuai <yukuai1@huaweicloud.com> wrote:
> 
>> From: Yu Kuai <yukuai3@huawei.com>
>>
>> Changes in v2:
>>  - add patch 1-8 to prevent dereference bitmap directly, and the  last
>>  patch to make bitmap structure internel.
>>  - use plain function alls "bitmap_ops->xxx()" directly;
>>
>> The background is that currently bitmap is using a global spin_lock,
>> cauing lock contention and huge IO performance degration for all raid
>> levels.
>>
>> However, it's impossible to implement a new lock free bitmap with
>> current situation that md-bitmap exposes the internal implementation
>> with lots of exported apis. Hence bitmap_operations is invented, to
>> describe bitmap core implementation, and a new bitmap can be introduced
>> with a new bitmap_operations, we only need to switch to the new one
>> during initialization.
>>
> Is the new bitmap in plan?

Yes, I'm working on this now in my spare time, based on this set.
>>
>> And with this we can build bitmap as kernel module, but that's not
>> our concern for now.
>>
>> Noted I just compile this patchset, not tested yet.
>>
> 
> I've looked through the patchset almostly. The changes are quite
> straightforward. IMO, it's good timing to  test it and drop RFC in next
> version.

I'm testing now, however, I met some problem that's likely related to
the mdadm tests, not this set. I'll send the formal version soon.

Thansk for the review!
Kuai

> 
>> Yu Kuai (41):
>>   md/raid1: use md_bitmap_wait_behind_writes() in   raid1_read_request()
>>   md/md-bitmap: replace md_bitmap_status() with a new helper
>>     md_bitmap_get_stats()
>>   md: use new helper md_bitmap_get_stats() in   update_array_info()
>>   md/md-bitmap: add 'events_cleared' into struct md_bitmap_stats
>>   md/md-bitmap: add 'sync_size' into struct md_bitmap_stats
>>   md/md-bitmap: add 'file_pages' into struct md_bitmap_stats
>>   md/md-bitmap: add 'behind_writes' and 'behind_wait' into   struct
>>     md_bitmap_stats
>>   md/md-cluster: use helper md_bitmap_get_stats() to get pages   in
>>     resize_bitmaps()
>>   md/md-bitmap: add a new helper md_bitmap_set_pages()
>>   md/md-bitmap: introduce struct bitmap_operations
>>   md/md-bitmap: simplify md_bitmap_create() + md_bitmap_load()
>>   md/md-bitmap: merge md_bitmap_create() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_load() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_destroy() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_flush() into bitmap_operations
>>   md/md-bitmap: make md_bitmap_print_sb() internal
>>   md/md-bitmap: merge md_bitmap_update_sb() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_status() into bitmap_operations
>>   md/md-bitmap: remove md_bitmap_setallbits()
>>   md/md-bitmap: merge bitmap_write_all() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_dirty_bits() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_startwrite() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_endwrite() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_start_sync() into   bitmap_operations
>>   md/md-bitmap: remove the parameter 'aborted' for   md_bitmap_end_sync()
>>   md/md-bitmap: merge md_bitmap_end_sync() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_close_sync() into   bitmap_operations
>>   md/md-bitmap: mrege md_bitmap_cond_end_sync() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_sync_with_cluster() into
>>     bitmap_operations
>>   md/md-bitmap: merge md_bitmap_unplug_async() into   md_bitmap_unplug()
>>   md/md-bitmap: merge bitmap_unplug() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_daemon_work() into   bitmap_operations
>>   md/md-bitmap: pass in mddev directly for md_bitmap_resize()
>>   md/md-bitmap: merge md_bitmap_resize() into bitmap_operations
>>   md/md-bitmap: merge get_bitmap_from_slot() into   bitmap_operations
>>   md/md-bitmap: merge md_bitmap_copy_from_slot() into struct
>>     bitmap_operation.
>>   md/md-bitmap: merge md_bitmap_set_pages() into struct
>>     bitmap_operations
>>   md/md-bitmap: merge md_bitmap_free() into bitmap_operations
>>   md/md-bitmap: merge md_bitmap_wait_behind_writes() into
>>     bitmap_operations
>>   md/md-bitmap: merge md_bitmap_enabled() into bitmap_operations
>>   md/md-bitmap: make in memory structure internal
>>
>>  drivers/md/dm-raid.c     |   7 +-
>>  drivers/md/md-bitmap.c   | 561  +++++++++++++++++++++++++++++----------
>>  drivers/md/md-bitmap.h   | 272 ++++---------------
>>  drivers/md/md-cluster.c  |  79 +++---
>>  drivers/md/md.c          | 133 ++++++----
>>  drivers/md/md.h          |   3 +-
>>  drivers/md/raid1-10.c    |   9 +-
>>  drivers/md/raid1.c       |  78 +++---
>>  drivers/md/raid10.c      |  73 ++---
>>  drivers/md/raid5-cache.c |   8 +-
>>  drivers/md/raid5.c       |  62 ++---
>>  11 files changed, 731 insertions(+), 554 deletions(-)
> .
>