Re: [f2fs-dev] [RFC PATCH v5] f2fs: support data compression

[Chao Yu <yuchao0@huawei.com>]

On 2020/1/15 6:48, Jaegeuk Kim wrote:
> On 01/14, Chao Yu wrote:
>> On 2020/1/14 0:11, Jaegeuk Kim wrote:
>>> On 01/13, Chao Yu wrote:
>>>> On 2020/1/12 2:02, Jaegeuk Kim wrote:
>>>>> On 01/11, Chao Yu wrote:
>>>>>> On 2020/1/11 7:52, Jaegeuk Kim wrote:
>>>>>>> On 01/06, Jaegeuk Kim wrote:
>>>>>>>> On 01/06, Chao Yu wrote:
>>>>>>>>> On 2020/1/3 14:50, Chao Yu wrote:
>>>>>>>>>> This works to me. Could you run fsstress tests on compressed root directory?
>>>>>>>>>> It seems still there are some bugs.
>>>>>>>>>
>>>>>>>>> Jaegeuk,
>>>>>>>>>
>>>>>>>>> Did you mean running por_fsstress testcase?
>>>>>>>>>
>>>>>>>>> Now, at least I didn't hit any problem for normal fsstress case.
>>>>>>>>
>>>>>>>> Yup. por_fsstress
>>>>>>>
>>>>>>> Please check https://github.com/jaegeuk/f2fs/commits/g-dev-test.
>>>>>>> I've fixed
>>>>>>> - truncation offset
>>>>>>> - i_compressed_blocks and its lock coverage
>>>>>>> - error handling
>>>>>>> - etc
>>>>>>
>>>>>> I changed as below, and por_fsstress stops panic the system.
>>>>>>
>>>>>> Could you merge all these fixes into original patch?
>>>>>
>>>>> Yup, let m roll up some early patches first once test results become good.
>>>>
>>>> I didn't encounter issue any more, how about por_fsstress test result in your
>>>> enviornment? If there is, please share the call stack with me.
>>>
>>> Sure, will do, once I hit an issue. BTW, I'm hitting another unreacheable nat
>>> entry issue during por_stress without compression. :(
>>
>> Did you enable any features during por_fsstress test?
>>
>> I only hit below warning during por_fsstress test on image w/o compression.
>>
>> ------------[ cut here ]------------
>> WARNING: CPU: 10 PID: 33483 at fs/fs-writeback.c:1448 __writeback_single_inode+0x28c/0x340
>> Call Trace:
>>  writeback_single_inode+0xad/0x120
>>  sync_inode_metadata+0x3d/0x60
>>  f2fs_sync_inode_meta+0x90/0xe0 [f2fs]
>>  block_operations+0x17c/0x360 [f2fs]
>>  f2fs_write_checkpoint+0x101/0xff0 [f2fs]
>>  f2fs_sync_fs+0xa8/0x130 [f2fs]
>>  f2fs_do_sync_file+0x19c/0x880 [f2fs]
>>  do_fsync+0x38/0x60
>>  __x64_sys_fsync+0x10/0x20
>>  do_syscall_64+0x5f/0x220
>>  entry_SYSCALL_64_after_hwframe+0x44/0xa9
> 
> Does gc_mutex patch fix this?

No, gc_mutex patch fixes another problem.

BTW, it looks like a bug of VFS.

Thanks,

> 
>>
>> Thanks,
>>
>>>
>>> Thanks,
>>>
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>>>
>>>>>> >From bb17d7d77fe0b8a3e3632a7026550800ab9609e9 Mon Sep 17 00:00:00 2001
>>>>>> From: Chao Yu <yuchao0@huawei.com>
>>>>>> Date: Sat, 11 Jan 2020 16:58:20 +0800
>>>>>> Subject: [PATCH] f2fs: compress: fix f2fs_put_rpages_mapping()
>>>>>>
>>>>>> Signed-off-by: Chao Yu <yuchao0@huawei.com>
>>>>>> ---
>>>>>>  fs/f2fs/compress.c | 30 +++++++++++++++---------------
>>>>>>  1 file changed, 15 insertions(+), 15 deletions(-)
>>>>>>
>>>>>> diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
>>>>>> index 502cd0ddc2a7..5c6a31d84ce4 100644
>>>>>> --- a/fs/f2fs/compress.c
>>>>>> +++ b/fs/f2fs/compress.c
>>>>>> @@ -74,18 +74,10 @@ static void f2fs_put_compressed_page(struct page *page)
>>>>>>  }
>>>>>>
>>>>>>  static void f2fs_drop_rpages(struct compress_ctx *cc,
>>>>>> -		struct address_space *mapping, int len, bool unlock)
>>>>>> +					int len, bool unlock)
>>>>>>  {
>>>>>>  	unsigned int i;
>>>>>>  	for (i = 0; i < len; i++) {
>>>>>> -		if (mapping) {
>>>>>> -			pgoff_t start = start_idx_of_cluster(cc);
>>>>>> -			struct page *page = find_get_page(mapping, start + i);
>>>>>> -
>>>>>> -			put_page(page);
>>>>>> -			put_page(page);
>>>>>> -			cc->rpages[i] = NULL;
>>>>>> -		}
>>>>>>  		if (!cc->rpages[i])
>>>>>>  			continue;
>>>>>>  		if (unlock)
>>>>>> @@ -97,18 +89,25 @@ static void f2fs_drop_rpages(struct compress_ctx *cc,
>>>>>>
>>>>>>  static void f2fs_put_rpages(struct compress_ctx *cc)
>>>>>>  {
>>>>>> -	f2fs_drop_rpages(cc, NULL, cc->cluster_size, false);
>>>>>> +	f2fs_drop_rpages(cc, cc->cluster_size, false);
>>>>>>  }
>>>>>>
>>>>>>  static void f2fs_unlock_rpages(struct compress_ctx *cc, int len)
>>>>>>  {
>>>>>> -	f2fs_drop_rpages(cc, NULL, len, true);
>>>>>> +	f2fs_drop_rpages(cc, len, true);
>>>>>>  }
>>>>>>
>>>>>>  static void f2fs_put_rpages_mapping(struct compress_ctx *cc,
>>>>>> -				struct address_space *mapping, int len)
>>>>>> +				struct address_space *mapping,
>>>>>> +				pgoff_t start, int len)
>>>>>>  {
>>>>>> -	f2fs_drop_rpages(cc, mapping, len, false);
>>>>>> +	int i;
>>>>>> +	for (i = 0; i < len; i++) {
>>>>>> +		struct page *page = find_get_page(mapping, start + i);
>>>>>> +
>>>>>> +		put_page(page);
>>>>>> +		put_page(page);
>>>>>> +	}
>>>>>>  }
>>>>>>
>>>>>>  static void f2fs_put_rpages_wbc(struct compress_ctx *cc,
>>>>>> @@ -680,7 +679,8 @@ static int prepare_compress_overwrite(struct compress_ctx *cc,
>>>>>>
>>>>>>  		if (!PageUptodate(page)) {
>>>>>>  			f2fs_unlock_rpages(cc, i + 1);
>>>>>> -			f2fs_put_rpages_mapping(cc, mapping, cc->cluster_size);
>>>>>> +			f2fs_put_rpages_mapping(cc, mapping, start_idx,
>>>>>> +					cc->cluster_size);
>>>>>>  			f2fs_destroy_compress_ctx(cc);
>>>>>>  			goto retry;
>>>>>>  		}
>>>>>> @@ -714,7 +714,7 @@ static int prepare_compress_overwrite(struct compress_ctx *cc,
>>>>>>  unlock_pages:
>>>>>>  	f2fs_unlock_rpages(cc, i);
>>>>>>  release_pages:
>>>>>> -	f2fs_put_rpages_mapping(cc, mapping, i);
>>>>>> +	f2fs_put_rpages_mapping(cc, mapping, start_idx, i);
>>>>>>  	f2fs_destroy_compress_ctx(cc);
>>>>>>  	return ret;
>>>>>>  }
>>>>>> -- 
>>>>>> 2.18.0.rc1
>>>>>>
>>>>>>
>>>>>>
>>>>>>>
>>>>>>> One another fix in f2fs-tools as well.
>>>>>>> https://github.com/jaegeuk/f2fs-tools
>>>>>>>
>>>>>>>>
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>> .
>>>>>>>
>>>>> .
>>>>>
>>> .
>>>
> .
> 


_______________________________________________
Linux-f2fs-devel mailing list
Linux-f2fs-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel