Re: [PATCH v2] f2fs: fix to keep isolation of atomic write

[Chao Yu <yuchao0@huawei.com>]

On 2021/1/29 0:21, Jaegeuk Kim wrote:
> On 01/20, Chao Yu wrote:
>> On 2021/1/20 3:06, Jaegeuk Kim wrote:
>>> On 01/15, Chao Yu wrote:
>>>> On 2021/1/15 5:53, Jaegeuk Kim wrote:
>>>>> On 12/30, Chao Yu wrote:
>>>>>> ThreadA					ThreadB
>>>>>> - f2fs_ioc_start_atomic_write
>>>>>> - write
>>>>>> - f2fs_ioc_commit_atomic_write
>>>>>>     - f2fs_commit_inmem_pages
>>>>>>     - f2fs_drop_inmem_pages
>>>>>>     - f2fs_drop_inmem_pages
>>>>>>      - __revoke_inmem_pages
>>>>>> 					- f2fs_vm_page_mkwrite
>>>>>> 					 - set_page_dirty
>>>>>> 					  - tag ATOMIC_WRITTEN_PAGE and add page
>>>>>> 					    to inmem_pages list
>>>>>>      - clear_inode_flag(FI_ATOMIC_FILE)
>>>>>> 					- f2fs_vm_page_mkwrite
>>>>>> 					  - set_page_dirty
>>>>>> 					   - f2fs_update_dirty_page
>>>>>> 					    - f2fs_trace_pid
>>>>>> 					     - tag inmem page private to pid
>>>>>
>>>>> Hmm, how about removing fs/f2fs/trace.c to make private more complicated
>>>>> like this? I think we can get IO traces from tracepoints.
>>>>
>>>> Hmm, actually, there is are issues, one is the trace IO, the other is the
>>>> race issue (atomic_start,commit,drop vs mkwrite) which can make isolation
>>>> semantics of transaction be broken.
>>>>
>>>> Or can we avoid atomic file racing with file mmap?
>>
>> Otherwise I think we should add i_mmap_sem to avoid the race.
>>
>>>
>>> No, we can't. We may need to find other way to check the race. :)
>>
>> Well, any thoughts about this issue?
>>
>> Thanks,
>>
>>>
>>>>
>>>> - atomic_start			- file_mmap
>>>> 				 - inode_lock
>>>> 				 - if (FI_ATOMIC_FILE) return
>>>>    - inode_lock
>>>>    - if (FI_MMAP_FILE) return
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>>> 					- truncate
>>>>>> 					 - f2fs_invalidate_page
>>>>>> 					 - set page->mapping to NULL
>>>>>> 					  then it will cause panic once we
>>>>>> 					  access page->mapping
> 
> Are we hitting this, since page was referenced by in-mem list?

Yes, we hit this NULL point dereferencing issue when running fuzz tool,
but the condition is not the same, because in our product,
CONFIG_F2FS_IO_TRACE was not set.

Thanks,

> 
>>>>>>
>>>>>> The root cause is we missed to keep isolation of atomic write in the case
>>>>>> of commit_atomic_write vs mkwrite, let commit_atomic_write helds i_mmap_sem
>>>>>> lock to avoid this issue.
>>>>>>
>>>>>> Signed-off-by: Chao Yu <yuchao0@huawei.com>
>>>>>> ---
>>>>>> v2:
>>>>>> - use i_mmap_sem to avoid mkwrite racing with below flows:
>>>>>>     * f2fs_ioc_start_atomic_write
>>>>>>     * f2fs_drop_inmem_pages
>>>>>>     * f2fs_commit_inmem_pages
>>>>>>
>>>>>>     fs/f2fs/file.c    | 3 +++
>>>>>>     fs/f2fs/segment.c | 7 +++++++
>>>>>>     2 files changed, 10 insertions(+)
>>>>>>
>>>>>> diff --git a/fs/f2fs/file.c b/fs/f2fs/file.c
>>>>>> index 4e6d4b9120a8..a48ec650d691 100644
>>>>>> --- a/fs/f2fs/file.c
>>>>>> +++ b/fs/f2fs/file.c
>>>>>> @@ -2050,6 +2050,7 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>>>>>>     		goto out;
>>>>>>     	down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>>>>> +	down_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     	/*
>>>>>>     	 * Should wait end_io to count F2FS_WB_CP_DATA correctly by
>>>>>> @@ -2060,6 +2061,7 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>>>>>>     			  inode->i_ino, get_dirty_pages(inode));
>>>>>>     	ret = filemap_write_and_wait_range(inode->i_mapping, 0, LLONG_MAX);
>>>>>>     	if (ret) {
>>>>>> +		up_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     		up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>>>>>     		goto out;
>>>>>>     	}
>>>>>> @@ -2073,6 +2075,7 @@ static int f2fs_ioc_start_atomic_write(struct file *filp)
>>>>>>     	/* add inode in inmem_list first and set atomic_file */
>>>>>>     	set_inode_flag(inode, FI_ATOMIC_FILE);
>>>>>>     	clear_inode_flag(inode, FI_ATOMIC_REVOKE_REQUEST);
>>>>>> +	up_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     	up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
>>>>>>     	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
>>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>>> index d8570b0359f5..dab870d9faf6 100644
>>>>>> --- a/fs/f2fs/segment.c
>>>>>> +++ b/fs/f2fs/segment.c
>>>>>> @@ -327,6 +327,8 @@ void f2fs_drop_inmem_pages(struct inode *inode)
>>>>>>     	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
>>>>>>     	struct f2fs_inode_info *fi = F2FS_I(inode);
>>>>>> +	down_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>> +
>>>>>>     	while (!list_empty(&fi->inmem_pages)) {
>>>>>>     		mutex_lock(&fi->inmem_lock);
>>>>>>     		__revoke_inmem_pages(inode, &fi->inmem_pages,
>>>>>> @@ -344,6 +346,8 @@ void f2fs_drop_inmem_pages(struct inode *inode)
>>>>>>     		sbi->atomic_files--;
>>>>>>     	}
>>>>>>     	spin_unlock(&sbi->inode_lock[ATOMIC_FILE]);
>>>>>> +
>>>>>> +	up_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     }
>>>>>>     void f2fs_drop_inmem_page(struct inode *inode, struct page *page)
>>>>>> @@ -467,6 +471,7 @@ int f2fs_commit_inmem_pages(struct inode *inode)
>>>>>>     	f2fs_balance_fs(sbi, true);
>>>>>>     	down_write(&fi->i_gc_rwsem[WRITE]);
>>>>>> +	down_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     	f2fs_lock_op(sbi);
>>>>>>     	set_inode_flag(inode, FI_ATOMIC_COMMIT);
>>>>>> @@ -478,6 +483,8 @@ int f2fs_commit_inmem_pages(struct inode *inode)
>>>>>>     	clear_inode_flag(inode, FI_ATOMIC_COMMIT);
>>>>>>     	f2fs_unlock_op(sbi);
>>>>>> +
>>>>>> +	up_write(&F2FS_I(inode)->i_mmap_sem);
>>>>>>     	up_write(&fi->i_gc_rwsem[WRITE]);
>>>>>>     	return err;
>>>>>> -- 
>>>>>> 2.29.2
>>>>> .
>>>>>
>>> .
>>>
> .
>