Re: [PATCH 1/2] f2fs: fix missing discard candidates in fstrim

[Chao Yu]

On 2025/1/3 16:07, Chunhai Guo wrote:
> 在 1/3/2025 11:26 AM, Chao Yu 写道:
> > On 2025/1/2 18:13, Chunhai Guo wrote:
> > > fstrim may miss candidates that need to be discarded in fstrim, 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 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(), which results in failure.
> > Chunhai,
> >
> > Can you please use nodiscard option due to fstrim stopped to return
> > trimmed length after below commit:
> >
> > 5a6154920faf ("f2fs: don't issue discard commands in online discard is on")
>
> Thank you for your explanation, but I guess this issue is not relevant
> to this commit, and I understand that '0 B (0 bytes) trimmed' is fine.
>
> The real problem is that there are actually many candidates that should
> be handled in fstrim, but it cannot find any of them.
>
> f2fs_trim_fs()
>       f2fs_write_checkpoint()
>           ...
>           if (cpc->reason & CP_DISCARD) {
>                   if (!f2fs_exist_trim_candidates(sbi, cpc)) {
>                       unblock_operations(sbi);
>                       goto out; // Not candidate is found here and exit.
>                   }
>               ...
>           }
>
> > > 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
> > >
> > > [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@xxxxxxxx>
> > > ---
> > >     fs/f2fs/segment.c | 10 +++++-----
> > >     1 file changed, 5 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > index eade36c5ef13..8fe9f794b581 100644
> > > --- a/fs/f2fs/segment.c
> > > +++ b/fs/f2fs/segment.c
> > > @@ -2070,7 +2070,7 @@ static int f2fs_issue_discard(struct f2fs_sb_info *sbi,
> > >     }
> > >     
> > >     static bool add_discard_addrs(struct f2fs_sb_info *sbi, struct cp_control *cpc,
> > > -							bool check_only)
> > > +					bool synced, bool check_only)
> > >     {
> > >     	int entries = SIT_VBLOCK_MAP_SIZE / sizeof(unsigned long);
> > >     	struct seg_entry *se = get_seg_entry(sbi, cpc->trim_start);
> > > @@ -2098,7 +2098,7 @@ 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] :
> > > +		dmap[i] = synced ? ~ckpt_map[i] & ~discard_map[i] :
> > I guess this condition "force ? ~ckpt_map[i] & ~discard_map[i]" didn't cover
> > all below cases, thoughts?
> > - ckpt_map[i] == 0 // write data, and then remove data before checkpoint
> > - ckpt_map[i] != 0 // remove data existed in previous checkpoint
>
>   From the handling of ckpt_valid_map in update_sit_entry(), I guess the
> condition can cover both cases.
> For example, when the checkpoint is enabled, all the set bits in the
> ckpt_valid_map remain set before the checkpoint (even when the blocks
> are deleted), which makes it find all the right bits in both cases.

My point is for fstrim case, we only need to check discard_map bitmap? 
once bit(s) in discard_map bitmap is zero, no matter the status of 
bit(s) in ckpt_map bitmap, we need to trigger a checkpoit for following 
discard submission?

Thanks,

> Thanks,
>
> > Thanks,
> >
> > >     				(cur_map[i] ^ ckpt_map[i]) & ckpt_map[i];
> > >     
> > >     	while (force || SM_I(sbi)->dcc_info->nr_discards <=
> > > @@ -3275,7 +3275,7 @@ bool f2fs_exist_trim_candidates(struct f2fs_sb_info *sbi,
> > >     
> > >     	down_write(&SIT_I(sbi)->sentry_lock);
> > >     	for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++) {
> > > -		if (add_discard_addrs(sbi, cpc, true)) {
> > > +		if (add_discard_addrs(sbi, cpc, false, true)) {
> > >     			has_candidate = true;
> > >     			break;
> > >     		}
> > > @@ -4611,7 +4611,7 @@ void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
> > >     			/* add discard candidates */
> > >     			if (!(cpc->reason & CP_DISCARD)) {
> > >     				cpc->trim_start = segno;
> > > -				add_discard_addrs(sbi, cpc, false);
> > > +				add_discard_addrs(sbi, cpc, false, false);
> > >     			}
> > >     
> > >     			if (to_journal) {
> > > @@ -4653,7 +4653,7 @@ void f2fs_flush_sit_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
> > >     		__u64 trim_start = cpc->trim_start;
> > >     
> > >     		for (; cpc->trim_start <= cpc->trim_end; cpc->trim_start++)
> > > -			add_discard_addrs(sbi, cpc, false);
> > > +			add_discard_addrs(sbi, cpc, true, false);
> > >     
> > >     		cpc->trim_start = trim_start;
> > >     	}