Re: [PATCH v2 06/17] btrfs-progs: lowmem check: introduce force_cow_in_new_chunk()

[Su Yue <suy.fnst@cn.fujitsu.com>]

On 12/21/2017 03:12 PM, Qu Wenruo wrote:
> 
> 
> On 2017年12月21日 15:09, Su Yue wrote:
>>
>>
>> On 12/21/2017 02:51 PM, Qu Wenruo wrote:
>>>
>>>
>>> On 2017年12月20日 16:37, Qu Wenruo wrote:
>>>>
>>>>
>>>> On 2017年12月20日 16:21, Su Yue wrote:
>>>>>
>>>>>
>>>>> On 12/20/2017 01:41 PM, Qu Wenruo wrote:
>>>>>>
>>>>>>
>>>>>> On 2017年12月20日 12:57, Su Yue wrote:
>>>>>>> Introduce create_chunk_and_block_block_group() to allocate new chunk
>>>>>>> and corresponding block group.
>>>>>>>
>>>>>>> The new function force_cow_in_new_chunk() first allocates new chunk
>>>>>>> and records its start.
>>>>>>> Then it modifies all metadata block groups cached and full.
>>>>>>> Finally it marks the new block group uncached and unfree.
>>>>>>> In the next CoW, extents states will be updated automatically by
>>>>>>> cache_block_group().
>>>>>>>
>>>>>>> Suggested-by: Qu Wenruo <wqu@suse.com>
>>>>>>> Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
>>>>>>> ---
>>>>>>>     cmds-check.c | 80
>>>>>>> ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>>>     1 file changed, 80 insertions(+)
>>>>>>>
>>>>>>> diff --git a/cmds-check.c b/cmds-check.c
>>>>>>> index d98d4bda7357..311c8a9f45e8 100644
>>>>>>> --- a/cmds-check.c
>>>>>>> +++ b/cmds-check.c
>>>>>>> @@ -10911,6 +10911,86 @@ out:
>>>>>>>         return ret;
>>>>>>>     }
>>>>>>>     +static int create_chunk_and_block_group(struct btrfs_fs_info
>>>>>>> *fs_info,
>>>>>>> +                    u64 flags, u64 *start, u64 *nbytes)
>>>>>>> +{
>>>>>>> +    struct btrfs_trans_handle *trans;
>>>>>>> +    struct btrfs_root *root = fs_info->extent_root;
>>>>>>> +    int ret;
>>>>>>> +
>>>>>>> +    if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)
>>>>>>> +        return -EINVAL;
>>>>>>> +
>>>>>>> +    trans = btrfs_start_transaction(root, 1);
>>>>>>> +    if (IS_ERR(trans)) {
>>>>>>> +        ret = PTR_ERR(trans);
>>>>>>> +        error("error starting transaction %s", strerror(-ret));
>>>>>>> +        return ret;
>>>>>>> +    }
>>>>>>> +    ret = btrfs_alloc_chunk(trans, fs_info, start, nbytes, flags);
>>>>>>> +    if (ret) {
>>>>>>> +        error("fail to allocate new chunk %s", strerror(-ret));
>>>>>>> +        goto out;
>>>>>>> +    }
>>>>>>> +    ret = btrfs_make_block_group(trans, fs_info, 0, flags,
>>>>>>> +                 BTRFS_FIRST_CHUNK_TREE_OBJECTID, *start, *nbytes);
>>>>>>> +    if (ret) {
>>>>>>> +        error("fail to make block group for chunk %llu %llu %s",
>>>>>>> +              *start, *nbytes, strerror(-ret));
>>>>>>> +        goto out;
>>>>>>> +    }
>>>>>>> +out:
>>>>>>> +    btrfs_commit_transaction(trans, root);
>>>>>>> +    return ret;
>>>>>>> +}
>>>>>>> +
>>>>>>> +static int force_cow_in_new_chunk(struct btrfs_fs_info *fs_info)
>>>>>>> +{
>>>>>>> +    struct btrfs_block_group_cache *bg;
>>>>>>> +    u64 start;
>>>>>>> +    u64 nbytes;
>>>>>>> +    u64 alloc_profile;
>>>>>>> +    u64 flags;
>>>>>>> +    int ret;
>>>>>>> +
>>>>>>> +    alloc_profile = (fs_info->avail_metadata_alloc_bits &
>>>>>>> +             fs_info->metadata_alloc_profile);
>>>>>>> +    flags = BTRFS_BLOCK_GROUP_METADATA | alloc_profile;
>>>>>>> +    if (btrfs_fs_incompat(fs_info, MIXED_GROUPS))
>>>>>>> +        flags |= BTRFS_BLOCK_GROUP_DATA;
>>>>>>> +
>>>>>>> +    ret = create_chunk_and_block_group(fs_info, flags, &start,
>>>>>>> &nbytes);
>>>>>>
>>>>>> Why bother allocating a new chunk by yourself?
>>>>>>
>>>>>> Just mark all block groups full and that's all.
>>>>>>
>>>>>> Any later tree block allocation like btrfs_search_slot() with @cow = 1
>>>>>> will trigger chunk allocation automatically.
>>>>>>
>>>>>
>>>>> Tried to let it happen but BUG_ON with -ENOSPC.
>>>>
>>>> Then fix it.
>>>>
>>>> It's not a normal behavior in this case.
>>>>
>>>> Thanks,
>>>> Qu
>>>
>>> And I think the fix to allow btrfs_reserve_extent() to allocate new
>>> chunk will solve a lot of strange BUG_ON().
>>>
>>> it just occurred to me that, a lot of use cases relies on the assumption
>>> that btrfs_reserve_extent() will try to allocate new chunks.
>>>
>>> Especially for case like convert, certain btrfs check --repair, some
>>> rescue tools.
>>>
>> Sorry for the previous wrong format mail.
>>
>> Yes, it has many dependency so I considered to do chunk allocation
>> manually in the patchset. If fix is not good enough, many funtions
>> of btrfs-progs will behave abnormal.
>> Since you ask it, I will go to fix it.
> 
> Manually allocation in advance has its advantage, like we can determine
> if there is enough space for new chunk instead of checking every return
> value with ENOSPC.
> 
> 
> However in current case, your metadata usage is limited to the new chunk
> only.
> If there extent tree has quite a lot of problem, and the chunk allocated
> is small (if using single profile and small fs), it can easily hit
> ENOSPC again, since btrfs doesn't allocate new chunk for later metadata
> write.
> 
SAD. After I tried to implement above nice idea, infinite recursive
brings me back to the reality.

Here is the reason why btrfs_reserve_extent can not allocate chunk
by itself if ENOSPC hints:

btrfs_cow_block
...
   btrfs_reserve_extent
     btrfs_alloc_chunk
       btrfs_alloc_dev_extent
         btrfs_insert_empty_item
           ...
            btrfs_cow_block

Thanks,
Su

> So here, we still need to allow btrfs allocate new meta chunk, even we
> pre-allocate one meta chunk in advance.
> 
> Thanks,
> Qu
>>
>> Thanks,
>> Su
>>
>>> So this would be a quite nice start point for such fix.
>>>
>>> Thanks,
>>> Qu
>>>
>>>>
>>>>> Do you mean do_chunk_alloc() in btrfs_reserve_extent() which is called
>>>>> while doing CoW?> In progs, allocation of new chunk during CoW
>>>>> depends @root->ref_cows.
>>>>> However, @root->ref_cows should be set only if @root is root of a fs
>>>>> trees.
>>>>>
>>>>> Thanks,
>>>>> Su
>>>>>
>>>>>> Thanks,
>>>>>> Qu
>>>>>>
>>>>>>> +    if (!ret)
>>>>>>> +        printf("Create new chunk %llu %llu\n", start, nbytes);
>>>>>>> +    else
>>>>>>> +        goto err;
>>>>>>> +
>>>>>>> +    flags = BTRFS_BLOCK_GROUP_METADATA;
>>>>>>> +    /* Mark all metadata block groups cached and full in free
>>>>>>> space*/
>>>>>>> +    ret = modify_block_groups_cache(fs_info, flags, 1);
>>>>>>> +    if (ret)
>>>>>>> +        goto clear_bg_cache;
>>>>>>> +
>>>>>>> +    bg = btrfs_lookup_block_group(fs_info, start);
>>>>>>> +    if (!bg) {
>>>>>>> +        ret = -ENOENT;
>>>>>>> +        error("fail to look up block group %llu %llu", start,
>>>>>>> nbytes);
>>>>>>> +        goto clear_bg_cache;
>>>>>>> +    }
>>>>>>> +
>>>>>>> +    /* Clear block group cache just allocated */
>>>>>>> +    ret = modify_block_group_cache(fs_info, bg, 0);
>>>>>>> +    if (ret)
>>>>>>> +        goto clear_bg_cache;
>>>>>>> +
>>>>>>> +    return 0;
>>>>>>> +
>>>>>>> +clear_bg_cache:
>>>>>>> +    modify_block_groups_cache(fs_info, flags, 0);
>>>>>>> +err:
>>>>>>> +    return ret;
>>>>>>> +}
>>>>>>> +
>>>>>>>     static int check_extent_refs(struct btrfs_root *root,
>>>>>>>                      struct cache_tree *extent_cache)
>>>>>>>     {
>>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> -- 
>>>>> To unsubscribe from this list: send the line "unsubscribe
>>>>> linux-btrfs" in
>>>>> the body of a message to majordomo@vger.kernel.org
>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>
>>>
>>
>>
>