linux-kernel - Re: [PATCH v3] f2fs: fix missing discard candidates in fstrim

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <e0d3d08a-8809-49b7-8144-eec405464bde@vivo.com>
Date: Wed, 21 May 2025 13:51:30 +0000
From: Chunhai Guo <guochunhai@...o.com>
To: Chunhai Guo <guochunhai@...o.com>, Jaegeuk Kim <jaegeuk@...nel.org>
CC: "chao@...nel.org" <chao@...nel.org>,
	"linux-f2fs-devel@...ts.sourceforge.net"
	<linux-f2fs-devel@...ts.sourceforge.net>, "linux-kernel@...r.kernel.org"
	<linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v3] f2fs: fix missing discard candidates in fstrim

在 3/14/2025 6:38 PM, Chunhai Guo 写道:
> 在 3/12/2025 10:27 PM, Jaegeuk Kim 写道:
>> On 03/12, Chunhai Guo wrote:
>>> fstrim may miss candidates that need to be discarded, as shown in the
>>> examples below.
>>>
>>> The root cause is that when cpc->reason is set with CP_DISCARD,
>>> add_discard_addrs() expects that ckpt_valid_map and cur_valid_map have
>>> been synced by seg_info_to_raw_sit() [1], and it tries to find the
>>> candidates based on ckpt_valid_map and discard_map. However,
>>> seg_info_to_raw_sit() does not actually run before
>>> f2fs_exist_trim_candidates(), resulting in the failure.
>> I think we need to fix the above logic.

Hi Jaegeuk,

Could you please provide some advice on how to fix the above logic when you have a moment?

Thanks,


>
> Hi Jaegeuk,
>
> I initially attempted to address this issue by checking whether
> seg_info_to_raw_sit() runs before add_discard_addrs(). The
> implementation can be found in the V1 patch here:
> https://lore.kernel.org/linux-f2fs-devel/20250102101310.580277-1-guochunhai@vivo.com/
>
> However, after discussing with Chao, I realized that perhaps a better
> way is to find all the discard blocks based only on discard_map. This
> approach is much simpler and makes add_discard_addrs() independent of
> the pre-run of seg_info_to_raw_sit().
>
> Could you please detail your opinion?
>
> Thanks,
>
>
>>> The code logic can be simplified for all cases by finding all the
>>> discard blocks based only on discard_map. This might result in more
>>> discard blocks being sent for the segment during the first checkpoint
>>> after mounting, which were originally expected to be sent only in
>>> fstrim. Regardless, these discard blocks should eventually be sent, and
>>> the simplified code makes sense in this context.
>>>
>>> root# cp testfile /f2fs_mountpoint
>>>
>>> root# f2fs_io fiemap 0 1 /f2fs_mountpoint/testfile
>>> Fiemap: offset = 0 len = 1
>>>           logical addr.    physical addr.   length           flags
>>> 0       0000000000000000 0000000406a00000 000000003d800000 00001000
>>>
>>> root# rm /f2fs_mountpoint/testfile
>>>
>>> root# fstrim -v -o 0x406a00000 -l 1024M /f2fs_mountpoint -- no candidate is found
>>> /f2fs_mountpoint: 0 B (0 bytes) trimmed
>>>
>>> Relevant code process of the root cause:
>>> f2fs_trim_fs()
>>>       f2fs_write_checkpoint()
>>>           ...
>>>           if (cpc->reason & CP_DISCARD) {
>>>                   if (!f2fs_exist_trim_candidates(sbi, cpc)) {
>>>                       unblock_operations(sbi);
>>>                       goto out; // No candidates are found here, and it exits.
>>>                   }
>>>               ...
>>>           }
>>>
>>> [1] Please refer to commit d7bc2484b8d4 ("f2fs: fix small discards not
>>> to issue redundantly") for the relationship between
>>> seg_info_to_raw_sit() and add_discard_addrs().
>>>
>>> Fixes: 25290fa5591d ("f2fs: return fs_trim if there is no candidate")
>>> Signed-off-by: Chunhai Guo <guochunhai@...o.com>
>>> ---
>>> v2->v3: Add f2fs_bug_on() to make sure it never issues discard to valid data's block address.
>>> v1->v2: Find all the discard blocks based only on discard_map in add_discard_addrs().
>>> v1: https://lore.kernel.org/linux-f2fs-devel/20250102101310.580277-1-guochunhai@vivo.com/
>>> ---
>>>    fs/f2fs/segment.c | 8 ++++----
>>>    1 file changed, 4 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>> index 86e547f008f9..c8ad8e3bfebb 100644
>>> --- a/fs/f2fs/segment.c
>>> +++ b/fs/f2fs/segment.c
>>> @@ -2075,7 +2075,6 @@ static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
>>>    	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
>>>    	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
>>>    	unsigned long *cur_map = (unsigned long *)se->cur_valid_map;
>>> -	unsigned long *ckpt_map = (unsigned long *)se->ckpt_valid_map;
>>>    	unsigned long *discard_map = (unsigned long *)se->discard_map;
>>>    	unsigned long *dmap = SIT_I(sbi)->tmp_map;
>>>    	unsigned int start = 0, end = -1;
>>> @@ -2097,9 +2096,10 @@ static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
>>>    	}
>>>    
>>>    	/* SIT_VBLOCK_MAP_SIZE should be multiple of sizeof(unsigned long) */
>>> -	for (i = 0; i < entries; i++)
>>> -		dmap[i] = force ? ~ckpt_map[i] & ~discard_map[i] :
>>> -				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
>>> +	for (i = 0; i < entries; i++) {
>>> +		dmap[i] = ~discard_map[i];
>>> +		f2fs_bug_on(sbi, (cur_map[i] ^ discard_map[i]) & cur_map[i]);
>>> +	}
>>>    
>>>    	while (force || SM_I(sbi)->dcc_info->nr_discards <=
>>>    				SM_I(sbi)->dcc_info->max_discards) {
>>> -- 
>>> 2.34.1
>