Re: [PATCH v5 00/18] btrfs: add read-only support for subpage sector size

[Qu Wenruo <quwenruo.btrfs@gmx.com>]

On 2021/1/28 上午12:17, Josef Bacik wrote:
> On 1/26/21 3:33 AM, Qu Wenruo wrote:
>> Patches can be fetched from github:
>> https://github.com/adam900710/linux/tree/subpage
>> Currently the branch also contains partial RW data support (still some
>> ordered extent and data csum mismatch problems)
>>
>> Great thanks to David/Nikolay/Josef for their effort reviewing and
>> merging the preparation patches into misc-next.
>>
>> === What works ===
>> Just from the patchset:
>> - Data read
>>    Both regular and compressed data, with csum check.
>>
>> - Metadata read
>>
>> This means, with these patchset, 64K page systems can at least mount
>> btrfs with 4K sector size read-only.
>> This should provide the ability to migrate data at least.
>>
>> While on the github branch, there are already experimental RW supports,
>> there are still ordered extent related bugs for me to fix.
>> Thus only the RO part is sent for review and testing.
>>
>> === Patchset structure ===
>> Patch 01~02:    Preparation patches which don't have functional change
>> Patch 03~12:    Subpage metadata allocation and freeing
>> Patch 13~15:    Subpage metadata read path
>> Patch 16~17:    Subpage data read path
>> Patch 18:    Enable subpage RO support
>>
>> === Changelog ===
>> v1:
>> - Separate the main implementation from previous huge patchset
>>    Huge patchset doesn't make much sense.
>>
>> - Use bitmap implementation
>>    Now page::private will be a pointer to btrfs_subpage structure, which
>>    contains bitmaps for various page status.
>>
>> v2:
>> - Use page::private as btrfs_subpage for extra info
>>    This replace old extent io tree based solution, which reduces latency
>>    and don't require memory allocation for its operations.
>>
>> - Cherry-pick new preparation patches from RW development
>>    Those new preparation patches improves the readability by their own.
>>
>> v3:
>> - Make dummy extent buffer to follow the same subpage accessors
>>    Fsstress exposed several ASSERT() for dummy extent buffers.
>>    It turns out we need to make dummy extent buffer to own the same
>>    btrfs_subpage structure to make eb accessors to work properly
>>
>> - Two new small __process_pages_contig() related preparation patches
>>    One to make __process_pages_contig() to enhance the error handling
>>    path for locked_page, one to merge one macro.
>>
>> - Extent buffers refs count update
>>    Except try_release_extent_buffer(), all other eb uses will try to
>>    increase the ref count of the eb.
>>    For try_release_extent_buffer(), the eb refs check will happen inside
>>    the rcu critical section to avoid eb being freed.
>>
>> - Comment updates
>>    Addressing the comments from the mail list.
>>
>> v4:
>> - Get rid of btrfs_subpage::tree_block_bitmap
>>    This is to reduce lock complexity (no need to bother extra subpage
>>    lock for metadata, all locks are existing locks)
>>    Now eb looking up mostly depends on radix tree, with small help from
>>    btrfs_subpage::under_alloc.
>>    Now I haven't experieneced metadata related problems any more during
>>    my local fsstress tests.
>>
>> - Fix a race where metadata page dirty bit can race
>>    Fixed in the metadata RW patchset though.
>>
>> - Rebased to latest misc-next branch
>>    With 4 patches removed, as they are already in misc-next.
>>
>> v5:
>> - Use the updated version from David as base
>>    Most comment/commit message update should be kept as is.
>>
>> - A new separate patch to move UNMAPPED bit set timing
>>
>> - New comment on why we need to prealloc subpage inside a loop
>>    Mostly for further 16K page size support, where we can have
>>    eb across multiple pages.
>>
>> - Remove one patch which is too RW specific
>>    Since it introduces functional change which only makes sense for RW
>>    support, it's not a good idea to include it in RO support.
>>
>> - Error handling fixes
>>    Great thanks to Josef.
>>
>> - Refactor btrfs_subpage allocation/freeing
>>    Now we have btrfs_alloc_subpage() and btrfs_free_subpage() helpers to
>>    do all the allocation/freeing.
>>    It's pretty easy to convert to kmem_cache using above helpers.
>>    (already internally tested using kmem_cache without problem, in fact
>>     it's all the problems found in kmem_cache test leads to the new
>>     interface)
>>
>> - Use btrfs_subpage::eb_refs to replace old under_alloc
>>    This makes checking whether the page has any eb left much easier.
>>
>> Qu Wenruo (18):
>>    btrfs: merge PAGE_CLEAR_DIRTY and PAGE_SET_WRITEBACK to
>>      PAGE_START_WRITEBACK
>>    btrfs: set UNMAPPED bit early in btrfs_clone_extent_buffer() for
>>      subpage support
>>    btrfs: introduce the skeleton of btrfs_subpage structure
>>    btrfs: make attach_extent_buffer_page() handle subpage case
>>    btrfs: make grab_extent_buffer_from_page() handle subpage case
>>    btrfs: support subpage for extent buffer page release
>
> I don't have this patch in my inbox so I can't reply to it directly, but
> you include refcount.h, but then use normal atomics.  Please used the
> actual refcount_t, as it gets us all the debugging stuff that makes
> finding problems much easier.  Thanks,

My bad, my initial plan is to use refcount, but the use case has valid 0
refcount usage, thus refcount is not good here.

I'll remove the remaining including line.

Thanks,
Qu
>
> Josef