[PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero

[PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero

[Yangtao Li]

I recently found a case where de->name_len is 0 in f2fs_fill_dentries() easily reproduced,
and finally set the fsck flag.

Thread A					Thread B

f2fs_readdir
	f2fs_read_inline_dir
		ctx->pos = d.max
						f2fs_add_dentry
							f2fs_add_inline_entry
								do_convert_inline_dir
							f2fs_add_regular_entry
f2fs_readdir
	f2fs_fill_dentries
		set_sbi_flag(sbi, SBI_NEED_FSCK)

Process A opens the folder, and has been reading without closing it. During this period,
Process B created a file under the folder (occupying multiple f2fs_dir_entry, exceeding
the d.max of the inline dir). After creation, process A uses the d.max of inline dir to
read it again, and it will read that de->name_len is 0.

And Chao pointed out that w/o inline conversion, the race condition still can happen as below

dir_entry1: A
dir_entry2: B
dir_entry3: C
free slot: _
ctx->pos: ^

Before:
AAAABBBB___
	 ^
Thread B delete dir_entry2, and create dir_entry3.

After:
AAAACCCCC__
	 ^

In these scenarios, the file system is not damaged, and it's hard to avoid it. But we can bypass
tagging FSCK flag if:
a) bit_pos (:= ctx->pos % d->max) is non-zero & b) before bit_pos moves to first
valid dir_entry.

Signed-off-by: Yangtao Li <frank.li@vivo.com>
---
 fs/f2fs/dir.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)

diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
index 456651682daf..bfe942733b5e 100644
--- a/fs/f2fs/dir.c
+++ b/fs/f2fs/dir.c
@@ -1000,6 +1000,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
 	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
 	struct blk_plug plug;
 	bool readdir_ra = sbi->readdir_ra == 1;
+	bool found_valid_dirent  = false;
 	int err = 0;
 
 	bit_pos = ((unsigned long)ctx->pos % d->max);
@@ -1014,13 +1015,15 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
 
 		de = &d->dentry[bit_pos];
 		if (de->name_len == 0) {
+			if (found_valid_dirent || !bit_pos) {
+				printk_ratelimited(
+					"%sF2FS-fs (%s): invalid namelen(0), ino:%u, run fsck to fix.",
+					KERN_WARNING, sbi->sb->s_id,
+					le32_to_cpu(de->ino));
+				set_sbi_flag(sbi, SBI_NEED_FSCK);
+			}
 			bit_pos++;
 			ctx->pos = start_pos + bit_pos;
-			printk_ratelimited(
-				"%sF2FS-fs (%s): invalid namelen(0), ino:%u, run fsck to fix.",
-				KERN_WARNING, sbi->sb->s_id,
-				le32_to_cpu(de->ino));
-			set_sbi_flag(sbi, SBI_NEED_FSCK);
 			continue;
 		}
 
@@ -1063,6 +1066,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
 			f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
 
 		ctx->pos = start_pos + bit_pos;
+		found_valid_dirent = true;
 	}
 out:
 	if (readdir_ra)
-- 
2.32.0

Re: [PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero

[Chao Yu]

On 2021/7/20 15:06, Yangtao Li wrote:
> I recently found a case where de->name_len is 0 in f2fs_fill_dentries() easily reproduced,
> and finally set the fsck flag.
> 
> Thread A					Thread B
> 
> f2fs_readdir
> 	f2fs_read_inline_dir
> 		ctx->pos = d.max
> 						f2fs_add_dentry
> 							f2fs_add_inline_entry
> 								do_convert_inline_dir
> 							f2fs_add_regular_entry
> f2fs_readdir
> 	f2fs_fill_dentries
> 		set_sbi_flag(sbi, SBI_NEED_FSCK)
> 
> Process A opens the folder, and has been reading without closing it. During this period,
> Process B created a file under the folder (occupying multiple f2fs_dir_entry, exceeding
> the d.max of the inline dir). After creation, process A uses the d.max of inline dir to
> read it again, and it will read that de->name_len is 0.
> 
> And Chao pointed out that w/o inline conversion, the race condition still can happen as below
> 
> dir_entry1: A
> dir_entry2: B
> dir_entry3: C
> free slot: _
> ctx->pos: ^
> 
> Before:
> AAAABBBB___
> 	 ^

please use blank instead of tab before '^'

> Thread B delete dir_entry2, and create dir_entry3.
> 
> After:
> AAAACCCCC__
> 	 ^

Ditto

> 
> In these scenarios, the file system is not damaged, and it's hard to avoid it. But we can bypass
> tagging FSCK flag if:
> a) bit_pos (:= ctx->pos % d->max) is non-zero & b) before bit_pos moves to first
> valid dir_entry.
> 
> Signed-off-by: Yangtao Li <frank.li@vivo.com>
> ---
>   fs/f2fs/dir.c | 14 +++++++++-----
>   1 file changed, 9 insertions(+), 5 deletions(-)
> 
> diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
> index 456651682daf..bfe942733b5e 100644
> --- a/fs/f2fs/dir.c
> +++ b/fs/f2fs/dir.c
> @@ -1000,6 +1000,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
>   	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
>   	struct blk_plug plug;
>   	bool readdir_ra = sbi->readdir_ra == 1;
> +	bool found_valid_dirent  = false;

One more blank before '='.

bool found_valid_dirent = false;

Thanks,

>   	int err = 0;
>   
>   	bit_pos = ((unsigned long)ctx->pos % d->max);
> @@ -1014,13 +1015,15 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
>   
>   		de = &d->dentry[bit_pos];
>   		if (de->name_len == 0) {
> +			if (found_valid_dirent || !bit_pos) {
> +				printk_ratelimited(
> +					"%sF2FS-fs (%s): invalid namelen(0), ino:%u, run fsck to fix.",
> +					KERN_WARNING, sbi->sb->s_id,
> +					le32_to_cpu(de->ino));
> +				set_sbi_flag(sbi, SBI_NEED_FSCK);
> +			}
>   			bit_pos++;
>   			ctx->pos = start_pos + bit_pos;
> -			printk_ratelimited(
> -				"%sF2FS-fs (%s): invalid namelen(0), ino:%u, run fsck to fix.",
> -				KERN_WARNING, sbi->sb->s_id,
> -				le32_to_cpu(de->ino));
> -			set_sbi_flag(sbi, SBI_NEED_FSCK);
>   			continue;
>   		}
>   
> @@ -1063,6 +1066,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
>   			f2fs_ra_node_page(sbi, le32_to_cpu(de->ino));
>   
>   		ctx->pos = start_pos + bit_pos;
> +		found_valid_dirent = true;
>   	}
>   out:
>   	if (readdir_ra)
> 

Re:Re: [PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero

[李扬韬]

HI Chao，

From: Chao Yu <chao@kernel.org>
Date: 2021-07-20 18:23:15
To:  Yangtao Li <frank.li@vivo.com>,jaegeuk@kernel.org
Cc:  linux-f2fs-devel@lists.sourceforge.net,linux-kernel@vger.kernel.org
Subject: Re: [PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero>On 2021/7/20 15:06, Yangtao Li wrote:
>> I recently found a case where de->name_len is 0 in f2fs_fill_dentries() easily reproduced,
>> and finally set the fsck flag.
>> 
>> Thread A					Thread B
>> 
>> f2fs_readdir
>> 	f2fs_read_inline_dir
>> 		ctx->pos = d.max
>> 						f2fs_add_dentry
>> 							f2fs_add_inline_entry
>> 								do_convert_inline_dir
>> 							f2fs_add_regular_entry
>> f2fs_readdir
>> 	f2fs_fill_dentries
>> 		set_sbi_flag(sbi, SBI_NEED_FSCK)
>> 
>> Process A opens the folder, and has been reading without closing it. During this period,
>> Process B created a file under the folder (occupying multiple f2fs_dir_entry, exceeding
>> the d.max of the inline dir). After creation, process A uses the d.max of inline dir to
>> read it again, and it will read that de->name_len is 0.
>> 
>> And Chao pointed out that w/o inline conversion, the race condition still can happen as below
>> 
>> dir_entry1: A
>> dir_entry2: B
>> dir_entry3: C
>> free slot: _
>> ctx->pos: ^
>> 
>> Before:
>> AAAABBBB___
>> 	 ^
>
>please use blank instead of tab before '^'

I don't know exactly what happened. In fact, in v2, spaces were used. Then it was changed to tab in v3.

>
>> Thread B delete dir_entry2, and create dir_entry3.
>> 
>> After:
>> AAAACCCCC__
>> 	 ^
>
>Ditto
>
>> 
>> In these scenarios, the file system is not damaged, and it's hard to avoid it. But we can bypass
>> tagging FSCK flag if:
>> a) bit_pos (:= ctx->pos % d->max) is non-zero & b) before bit_pos moves to first
>> valid dir_entry.
>> 
>> Signed-off-by: Yangtao Li <frank.li@vivo.com>
>> ---
>>   fs/f2fs/dir.c | 14 +++++++++-----
>>   1 file changed, 9 insertions(+), 5 deletions(-)
>> 
>> diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
>> index 456651682daf..bfe942733b5e 100644
>> --- a/fs/f2fs/dir.c
>> +++ b/fs/f2fs/dir.c
>> @@ -1000,6 +1000,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
>>   	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
>>   	struct blk_plug plug;
>>   	bool readdir_ra = sbi->readdir_ra == 1;
>> +	bool found_valid_dirent  = false;
>
>One more blank before '='.
>
>bool found_valid_dirent = false;
>

OK.

Thx，

Re: [PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero

[Chao Yu]

Hello, Yangtao,

I did some cleanup in your patch, and resend for you, please
note that. :)

Thanks,

On 2021/7/20 20:31, 李扬韬 wrote:
> HI Chao，
> 
> From: Chao Yu <chao@kernel.org>
> Date: 2021-07-20 18:23:15
> To:  Yangtao Li <frank.li@vivo.com>,jaegeuk@kernel.org
> Cc:  linux-f2fs-devel@lists.sourceforge.net,linux-kernel@vger.kernel.org
> Subject: Re: [PATCH v3] f2fs: Reduce the scope of setting fsck tag when de->name_len is zero>On 2021/7/20 15:06, Yangtao Li wrote:
>>> I recently found a case where de->name_len is 0 in f2fs_fill_dentries() easily reproduced,
>>> and finally set the fsck flag.
>>>
>>> Thread A					Thread B
>>>
>>> f2fs_readdir
>>> 	f2fs_read_inline_dir
>>> 		ctx->pos = d.max
>>> 						f2fs_add_dentry
>>> 							f2fs_add_inline_entry
>>> 								do_convert_inline_dir
>>> 							f2fs_add_regular_entry
>>> f2fs_readdir
>>> 	f2fs_fill_dentries
>>> 		set_sbi_flag(sbi, SBI_NEED_FSCK)
>>>
>>> Process A opens the folder, and has been reading without closing it. During this period,
>>> Process B created a file under the folder (occupying multiple f2fs_dir_entry, exceeding
>>> the d.max of the inline dir). After creation, process A uses the d.max of inline dir to
>>> read it again, and it will read that de->name_len is 0.
>>>
>>> And Chao pointed out that w/o inline conversion, the race condition still can happen as below
>>>
>>> dir_entry1: A
>>> dir_entry2: B
>>> dir_entry3: C
>>> free slot: _
>>> ctx->pos: ^
>>>
>>> Before:
>>> AAAABBBB___
>>> 	 ^
>>
>> please use blank instead of tab before '^'
> 
> I don't know exactly what happened. In fact, in v2, spaces were used. Then it was changed to tab in v3.
> 
>>
>>> Thread B delete dir_entry2, and create dir_entry3.
>>>
>>> After:
>>> AAAACCCCC__
>>> 	 ^
>>
>> Ditto
>>
>>>
>>> In these scenarios, the file system is not damaged, and it's hard to avoid it. But we can bypass
>>> tagging FSCK flag if:
>>> a) bit_pos (:= ctx->pos % d->max) is non-zero & b) before bit_pos moves to first
>>> valid dir_entry.
>>>
>>> Signed-off-by: Yangtao Li <frank.li@vivo.com>
>>> ---
>>>    fs/f2fs/dir.c | 14 +++++++++-----
>>>    1 file changed, 9 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/fs/f2fs/dir.c b/fs/f2fs/dir.c
>>> index 456651682daf..bfe942733b5e 100644
>>> --- a/fs/f2fs/dir.c
>>> +++ b/fs/f2fs/dir.c
>>> @@ -1000,6 +1000,7 @@ int f2fs_fill_dentries(struct dir_context *ctx, struct f2fs_dentry_ptr *d,
>>>    	struct f2fs_sb_info *sbi = F2FS_I_SB(d->inode);
>>>    	struct blk_plug plug;
>>>    	bool readdir_ra = sbi->readdir_ra == 1;
>>> +	bool found_valid_dirent  = false;
>>
>> One more blank before '='.
>>
>> bool found_valid_dirent = false;
>>
> 
> OK.
> 
> Thx，
> 
> 
>