Re: [PATCH v4 4/4] f2fs: stop checkpoint when get a out-of-bounds segment

[Chao Yu]

On 2024/2/20 14:11, Zhiguo Niu wrote:
> There is low probability that an out-of-bounds segment will be got
> on a small-capacity device. In order to prevent subsequent write requests
> allocating block address from this invalid segment, which may cause
> unexpected issue, stop checkpoint should be performed.
>
> Also introduce a new stop cp reason: STOP_CP_REASON_NO_SEGMENT.
>
> Signed-off-by: Zhiguo Niu <zhiguo.niu@xxxxxxxxxx>
> ---
> changes of v4: use more suitable MACRO name according to Chao's suggestions
> changes of v3: correct MACRO spelling and update based on the lastes code
> ---
> ---
>   fs/f2fs/segment.c       | 7 ++++++-
>   include/linux/f2fs_fs.h | 1 +
>   2 files changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> index c25aaec..e42e34c 100644
> --- a/fs/f2fs/segment.c
> +++ b/fs/f2fs/segment.c
> @@ -2665,7 +2665,12 @@ static void get_new_segment(struct f2fs_sb_info *sbi,
>   	if (secno >= MAIN_SECS(sbi)) {
>   		secno = find_first_zero_bit(free_i->free_secmap,
>   							MAIN_SECS(sbi));
> -		f2fs_bug_on(sbi, secno >= MAIN_SECS(sbi));
> +		if (secno >= MAIN_SECS(sbi)) {
> +			f2fs_stop_checkpoint(sbi, false,
> +				STOP_CP_REASON_NO_SEGMENT);

We should relocate stop_checkpoint(sbi, false, STOP_CP_REASON_NO_SEGMENT) outside
segmap_lock spinlock, due to it may sleep in f2fs_flush_merged_writes().

Thanks,

> +			f2fs_bug_on(sbi, 1);
> +		}
> +
>   	}
>   	segno = GET_SEG_FROM_SEC(sbi, secno);
>   	zoneno = GET_ZONE_FROM_SEC(sbi, secno);
> diff --git a/include/linux/f2fs_fs.h b/include/linux/f2fs_fs.h
> index 9b69c50..755e9a4 100644
> --- a/include/linux/f2fs_fs.h
> +++ b/include/linux/f2fs_fs.h
> @@ -75,6 +75,7 @@ enum stop_cp_reason {
>   	STOP_CP_REASON_CORRUPTED_SUMMARY,
>   	STOP_CP_REASON_UPDATE_INODE,
>   	STOP_CP_REASON_FLUSH_FAIL,
> +	STOP_CP_REASON_NO_SEGMENT,
>   	STOP_CP_REASON_MAX,
>   };
>