Re: [PATCH] btrfs: adjust subpage bit start based on sectorsize

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

在 2025/4/16 02:53, Josef Bacik 写道:
> On Tue, Apr 15, 2025 at 07:38:29AM +0930, Qu Wenruo wrote:
>>
>>
[...]
>> The problem is, we can not ensure all extent buffers are nodesize aligned.
>>
>> If we have an eb whose bytenr is only block aligned but not node size
>> aligned, this will lead to the same problem.
>>
>> We need an extra check to reject tree blocks which are not node size
>> aligned, which is another big change and not suitable for a quick fix.
> 
> We already have this.

The checks inside check_eb_alignment() only ensure the subpage eb will 
not cross page boundary, not strong nodesize alignment:

         if (fs_info->nodesize < PAGE_SIZE &&
             offset_in_page(start) + fs_info->nodesize > PAGE_SIZE) {
                 btrfs_err(fs_info,
                 "tree block crosses page boundary, start %llu nodesize %u",
                           start, fs_info->nodesize);
                 return true;
         }

In fact, it even allows such unaligned tree blocks, but with some warnings:

         if (!IS_ALIGNED(start, fs_info->nodesize) &&
             !test_and_set_bit(BTRFS_FS_UNALIGNED_TREE_BLOCK, 
&fs_info->flags)) {
                 btrfs_warn(fs_info,
"tree block not nodesize aligned, start %llu nodesize %u, can be 
resolved by a full metadata balance",
                               start, fs_info->nodesize);
         }

Thanks,
Qu

> 
>>
>>
>> Can we do a gang radix tree lookup for the involved ranges that can cover
>> the block, then increase bit_start to the end of the found eb instead?
> 
> That's a followup patch that I'm testing now.  I've started with the simplest
> fix so that they can be pulled back to all the affected kernels, and then I'm
> following up with much more invasive changes to make these problems go away in
> general.  Thanks,
> 
> Josef
>