Re: [PATCH V4] f2fs: Avoid double lock for cp_rwsem during checkpoint

[Chao Yu <yuchao0@huawei.com>]

On 2020/5/12 6:11, Jaegeuk Kim wrote:
> On 05/11, Chao Yu wrote:
>> On 2020/5/10 3:03, Jaegeuk Kim wrote:
>>> On 05/09, Chao Yu wrote:
>>>> On 2020/5/9 0:10, Jaegeuk Kim wrote:
>>>>> Hi Sayali,
>>>>>
>>>>> In order to address the perf regression, how about this?
>>>>>
>>>>> >From 48418af635884803ffb35972df7958a2e6649322 Mon Sep 17 00:00:00 2001
>>>>> From: Jaegeuk Kim <jaegeuk@kernel.org>
>>>>> Date: Fri, 8 May 2020 09:08:37 -0700
>>>>> Subject: [PATCH] f2fs: avoid double lock for cp_rwsem during checkpoint
>>>>>
>>>>> There could be a scenario where f2fs_sync_node_pages gets
>>>>> called during checkpoint, which in turn tries to flush
>>>>> inline data and calls iput(). This results in deadlock as
>>>>> iput() tries to hold cp_rwsem, which is already held at the
>>>>> beginning by checkpoint->block_operations().
>>>>>
>>>>> Call stack :
>>>>>
>>>>> Thread A		Thread B
>>>>> f2fs_write_checkpoint()
>>>>> - block_operations(sbi)
>>>>>  - f2fs_lock_all(sbi);
>>>>>   - down_write(&sbi->cp_rwsem);
>>>>>
>>>>>                         - open()
>>>>>                          - igrab()
>>>>>                         - write() write inline data
>>>>>                         - unlink()
>>>>> - f2fs_sync_node_pages()
>>>>>  - if (is_inline_node(page))
>>>>>   - flush_inline_data()
>>>>>    - ilookup()
>>>>>      page = f2fs_pagecache_get_page()
>>>>>      if (!page)
>>>>>       goto iput_out;
>>>>>      iput_out:
>>>>> 			-close()
>>>>> 			-iput()
>>>>>        iput(inode);
>>>>>        - f2fs_evict_inode()
>>>>>         - f2fs_truncate_blocks()
>>>>>          - f2fs_lock_op()
>>>>>            - down_read(&sbi->cp_rwsem);
>>>>>
>>>>> Fixes: 2049d4fcb057 ("f2fs: avoid multiple node page writes due to inline_data")
>>>>> Signed-off-by: Sayali Lokhande <sayalil@codeaurora.org>
>>>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>>>>> ---
>>>>>  fs/f2fs/node.c | 4 ++--
>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
>>>>> index 1db8cabf727ef..626d7daca09de 100644
>>>>> --- a/fs/f2fs/node.c
>>>>> +++ b/fs/f2fs/node.c
>>>>> @@ -1870,8 +1870,8 @@ int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
>>>>>  				goto continue_unlock;
>>>>>  			}
>>>>>  
>>>>> -			/* flush inline_data */
>>>>> -			if (is_inline_node(page)) {
>>>>> +			/* flush inline_data, if it's not sync path. */
>>>>> +			if (do_balance && is_inline_node(page)) {
>>>>
>>>> IIRC, this flow was designed to avoid running out of free space issue
>>>> during checkpoint:
>>>>
>>>> 2049d4fcb057 ("f2fs: avoid multiple node page writes due to inline_data")
>>>>
>>>> The sceanrio is:
>>>> 1. create fully node blocks
>>>> 2. flush node blocks
>>>> 3. write inline_data for all the node blocks again
>>>> 4. flush node blocks redundantly
>>>>
>>>> I guess this may cause failing one case of fstest.
>>>
>>> Yeah, actually I was hitting 204 failure, and thus, revised like this.
>>> Now, I don't see any regression in fstest.
>>>
>>> >From 8f1882acfb0a5fc43e5a2bbd576a8f3c681a7d2c Mon Sep 17 00:00:00 2001
>>> From: Sayali Lokhande <sayalil@codeaurora.org>
>>> Date: Thu, 30 Apr 2020 16:28:29 +0530
>>> Subject: [PATCH] f2fs: Avoid double lock for cp_rwsem during checkpoint
>>>
>>> There could be a scenario where f2fs_sync_node_pages gets
>>> called during checkpoint, which in turn tries to flush
>>> inline data and calls iput(). This results in deadlock as
>>> iput() tries to hold cp_rwsem, which is already held at the
>>> beginning by checkpoint->block_operations().
>>>
>>> Call stack :
>>>
>>> Thread A		Thread B
>>> f2fs_write_checkpoint()
>>> - block_operations(sbi)
>>>  - f2fs_lock_all(sbi);
>>>   - down_write(&sbi->cp_rwsem);
>>>
>>>                         - open()
>>>                          - igrab()
>>>                         - write() write inline data
>>>                         - unlink()
>>> - f2fs_sync_node_pages()
>>>  - if (is_inline_node(page))
>>>   - flush_inline_data()
>>>    - ilookup()
>>>      page = f2fs_pagecache_get_page()
>>>      if (!page)
>>>       goto iput_out;
>>>      iput_out:
>>> 			-close()
>>> 			-iput()
>>>        iput(inode);
>>>        - f2fs_evict_inode()
>>>         - f2fs_truncate_blocks()
>>>          - f2fs_lock_op()
>>>            - down_read(&sbi->cp_rwsem);
>>>
>>> Fixes: 2049d4fcb057 ("f2fs: avoid multiple node page writes due to inline_data")
>>> Signed-off-by: Sayali Lokhande <sayalil@codeaurora.org>
>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>>> ---
>>>  fs/f2fs/checkpoint.c |  9 ++++++++-
>>>  fs/f2fs/f2fs.h       |  4 ++--
>>>  fs/f2fs/node.c       | 10 +++++-----
>>>  3 files changed, 15 insertions(+), 8 deletions(-)
>>>
>>> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
>>> index d49f7a01d8a26..928aea4ff663d 100644
>>> --- a/fs/f2fs/checkpoint.c
>>> +++ b/fs/f2fs/checkpoint.c
>>> @@ -1168,6 +1168,12 @@ static int block_operations(struct f2fs_sb_info *sbi)
>>>  	};
>>>  	int err = 0, cnt = 0;
>>>  
>>> +	/*
>>> +	 * Let's flush node pages first to flush inline_data.
>>> +	 * We'll actually guarantee everything below under f2fs_lock_all.
>>> +	 */
>>> +	f2fs_sync_node_pages(sbi, &wbc, false, false, FS_CP_NODE_IO);
>>
>> It is possible that user write a large number of inline data in between
>> f2fs_sync_node_pages() and f2fs_lock_all(), it will cause the no-space issue in
>> race condition.
>>
>> Also, if there is huge number of F2FS_DIRTY_IMETA, after this change, we will
>> flush inode page twice which is unneeded.
>>
>> f2fs_sync_node_pages() --- flush dirty inode page
>> f2fs_lock_all()
>> ...
>> f2fs_sync_inode_meta() --- update dirty inode page
>> f2fs_sync_node_pages() --- flush dirty inode page again.
>>
> 
> Another version:
> 
>>From 6b430b72af57c65c20dea7b87f7ba7e9df36be98 Mon Sep 17 00:00:00 2001
> From: Sayali Lokhande <sayalil@codeaurora.org>
> Date: Thu, 30 Apr 2020 16:28:29 +0530
> Subject: [PATCH] f2fs: Avoid double lock for cp_rwsem during checkpoint
> 
> There could be a scenario where f2fs_sync_node_pages gets
> called during checkpoint, which in turn tries to flush
> inline data and calls iput(). This results in deadlock as
> iput() tries to hold cp_rwsem, which is already held at the
> beginning by checkpoint->block_operations().
> 
> Call stack :
> 
> Thread A		Thread B
> f2fs_write_checkpoint()
> - block_operations(sbi)
>  - f2fs_lock_all(sbi);
>   - down_write(&sbi->cp_rwsem);
> 
>                         - open()
>                          - igrab()
>                         - write() write inline data
>                         - unlink()
> - f2fs_sync_node_pages()
>  - if (is_inline_node(page))
>   - flush_inline_data()
>    - ilookup()
>      page = f2fs_pagecache_get_page()
>      if (!page)
>       goto iput_out;
>      iput_out:
> 			-close()
> 			-iput()
>        iput(inode);
>        - f2fs_evict_inode()
>         - f2fs_truncate_blocks()
>          - f2fs_lock_op()
>            - down_read(&sbi->cp_rwsem);
> 
> Fixes: 2049d4fcb057 ("f2fs: avoid multiple node page writes due to inline_data")
> Signed-off-by: Sayali Lokhande <sayalil@codeaurora.org>
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>  fs/f2fs/checkpoint.c |  5 +++++
>  fs/f2fs/f2fs.h       |  1 +
>  fs/f2fs/node.c       | 51 ++++++++++++++++++++++++++++++++++++++++++--
>  3 files changed, 55 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> index d49f7a01d8a26..79e605f38f4fa 100644
> --- a/fs/f2fs/checkpoint.c
> +++ b/fs/f2fs/checkpoint.c
> @@ -1168,6 +1168,11 @@ static int block_operations(struct f2fs_sb_info *sbi)
>  	};
>  	int err = 0, cnt = 0;
>  
> +	/*
> +	 * Let's flush inline_data in dirty node pages.
> +	 */
> +	f2fs_flush_inline_data(sbi);

Still there is a gap, user can write a large number of inline data here...

Would that be enough? I doubt we can suffer this issue in below pathes
as well:

- block_operations
 - f2fs_sync_dirty_inodes
  - iput
 - f2fs_sync_inode_meta
  - iput

Thanks,

> +
>  retry_flush_quotas:
>  	f2fs_lock_all(sbi);
>  	if (__need_flush_quota(sbi)) {
> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
> index 2a8ea81c52a15..7f3d259e7e376 100644
> --- a/fs/f2fs/f2fs.h
> +++ b/fs/f2fs/f2fs.h
> @@ -3282,6 +3282,7 @@ void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid);
>  struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid);
>  struct page *f2fs_get_node_page_ra(struct page *parent, int start);
>  int f2fs_move_node_page(struct page *node_page, int gc_type);
> +int f2fs_flush_inline_data(struct f2fs_sb_info *sbi);
>  int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
>  			struct writeback_control *wbc, bool atomic,
>  			unsigned int *seq_id);
> diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> index 1db8cabf727ef..e632de10aedab 100644
> --- a/fs/f2fs/node.c
> +++ b/fs/f2fs/node.c
> @@ -1807,6 +1807,53 @@ static bool flush_dirty_inode(struct page *page)
>  	return true;
>  }
>  
> +int f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
> +{
> +	pgoff_t index = 0;
> +	struct pagevec pvec;
> +	int nr_pages;
> +	int ret = 0;
> +
> +	pagevec_init(&pvec);
> +
> +	while ((nr_pages = pagevec_lookup_tag(&pvec,
> +			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
> +		int i;
> +
> +		for (i = 0; i < nr_pages; i++) {
> +			struct page *page = pvec.pages[i];
> +
> +			if (!IS_DNODE(page))
> +				continue;
> +
> +			lock_page(page);
> +
> +			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
> +continue_unlock:
> +				unlock_page(page);
> +				continue;
> +			}
> +
> +			if (!PageDirty(page)) {
> +				/* someone wrote it for us */
> +				goto continue_unlock;
> +			}
> +
> +			/* flush inline_data, if it's async context. */
> +			if (is_inline_node(page)) {
> +				clear_inline_node(page);
> +				unlock_page(page);
> +				flush_inline_data(sbi, ino_of_node(page));
> +				continue;
> +			}
> +			unlock_page(page);
> +		}
> +		pagevec_release(&pvec);
> +		cond_resched();
> +	}
> +	return ret;
> +}
> +
>  int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
>  				struct writeback_control *wbc,
>  				bool do_balance, enum iostat_type io_type)
> @@ -1870,8 +1917,8 @@ int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
>  				goto continue_unlock;
>  			}
>  
> -			/* flush inline_data */
> -			if (is_inline_node(page)) {
> +			/* flush inline_data, if it's async context. */
> +			if (do_balance && is_inline_node(page)) {
>  				clear_inline_node(page);
>  				unlock_page(page);
>  				flush_inline_data(sbi, ino_of_node(page));
>