Re: [f2fs-dev] [PATCH] f2fs: support resizable tail section and unify pinned allocation

[Chao Yu <chao@kernel.org>]

On 8/26/26 00:14, Daeho Jeong wrote:
> On Mon, Aug 24, 2026 at 8:16 PM Chao Yu <chao@kernel.org> wrote:
>>
>> On 8/25/26 03:14, Daeho Jeong wrote:
>>> From: Daeho Jeong <daehojeong@google.com>
>>>
>>> Currently, zoned block devices restrict pinned file allocations to
>>> conventional zones at the beginning of the storage (before
>>> first_seq_zone_segno), triggering range GC when conventional space is
>>> exhausted.
>>>
>>> On regular block devices, when preparing for future online filesystem
>>> resizing (e.g. partition shrinking), pinned files must not be allocated
>>> in the tail area that will be truncated, as pinned files cannot be
>>> relocated by GC. Specifying the resizable tail area size (in sections)
>>> allows uniform mount configuration across devices of different storage
>>> capacities.
>>>
>>> To support this, introduce a unified `pinned_area_max_secno` boundary
>>> abstraction in `f2fs_sb_info`:
>>> 1. Add `-o resizable_tail_secno=%u` mount option to specify the number
>>>     of sections at the tail of the filesystem reserved for resizing.
>>> 2. In `f2fs_fill_super()`, initialize `sbi->pinned_area_max_secno` as:
>>>     min(MAIN_SECS(sbi) - resizable_tail_sec, zoned_max_sec).
>>> 3. In `get_new_segment()`, restrict segment allocation for pinned files
>>>     (`pinning == true`) to `0 .. sbi->pinned_area_max_secno - 1`. If no
>>>     free section is available in the pinned area, return -EAGAIN.
>>> 4. In `f2fs_allocate_pinning_section()`, unify the range GC trigger to
>>>     run `f2fs_gc_range()` up to `sbi->pinned_area_max_secno` whenever
>>>     `sbi->pinned_area_max_secno < MAIN_SECS(sbi)` and allocation
>>>     returns -EAGAIN.
>>> 5. Expose `/sys/fs/f2fs/<dev>/pinned_area_max_secno` as a read-only
>>>     sysfs node.
>>>
>>> Signed-off-by: Daeho Jeong <daehojeong@google.com>
>>> Signed-off-by: Sunmin Jeong <s_min.jeong@samsung.com>
>>> ---
>>>   Documentation/ABI/testing/sysfs-fs-f2fs |  7 +++++
>>>   Documentation/filesystems/f2fs.rst      |  8 +++++-
>>>   fs/f2fs/f2fs.h                          |  2 ++
>>>   fs/f2fs/segment.c                       | 33 ++++++++++++-----------
>>>   fs/f2fs/segment.h                       |  1 +
>>>   fs/f2fs/super.c                         | 36 +++++++++++++++++++++++++
>>>   fs/f2fs/sysfs.c                         |  2 ++
>>>   7 files changed, 72 insertions(+), 17 deletions(-)
>>>
>>> diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
>>> index 85194e4c7f01..0cebc89799dd 100644
>>> --- a/Documentation/ABI/testing/sysfs-fs-f2fs
>>> +++ b/Documentation/ABI/testing/sysfs-fs-f2fs
>>> @@ -1013,3 +1013,10 @@ Description:   Every time a write operation completes f2fs_write_end_io() is
>>>               the maximum size of a write bio that is completed in atomic
>>>               (atc) context. The default value for this attribute is UINT_MAX
>>>               which means that this functionality is disabled by default.
>>> +
>>> +What:                /sys/fs/f2fs/<disk>/pinned_area_max_secno
>>> +Date:                August 2026
>>> +Contact:     "Daeho Jeong" <daehojeong@google.com>
>>> +Description: This is a read-only entry to show the upper bound section number
>>> +             for pinned files. Pinned files will only be allocated within
>>> +             sections 0 to pinned_area_max_secno - 1.
>>> diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
>>> index 1a5fd4afe609..a3c3b6948734 100644
>>> --- a/Documentation/filesystems/f2fs.rst
>>> +++ b/Documentation/filesystems/f2fs.rst
>>> @@ -417,7 +417,13 @@ lookup_mode=%s            Control the directory lookup behavior for casefolded
>>>                            auto               F2FS determines the mode based on the
>>>                                               on-disk `SB_ENC_NO_COMPAT_FALLBACK_FL`
>>>                                               flag.
>>> -                          ================== ========================================
>>> +resizable_tail_secno=%u  Control the number of sections at the tail of the
>>> +                      filesystem reserved for online resizing. Pinned files
>>> +                      will only be allocated within sections 0 to
>>> +                      (MAIN_SECS - resizable_tail_secno) - 1. If set to 0
>>> +                      (default), there is no tail restriction unless running
>>> +                      on a zoned block device where conventional zones are
>>> +                      used.
>>>   ======================== ============================================================
>>>
>>>   Debugfs Entries
>>> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
>>> index b0a9c14de595..16720f1f0a9c 100644
>>> --- a/fs/f2fs/f2fs.h
>>> +++ b/fs/f2fs/f2fs.h
>>> @@ -255,6 +255,7 @@ struct f2fs_mount_info {
>>>       block_t unusable_cap;           /* Amount of space allowed to be
>>>                                        * unusable when disabling checkpoint
>>>                                        */
>>> +     unsigned int resizable_tail_secno;      /* number of resizable tail sections */
>>>
>>>       /* For compression */
>>>       unsigned char compress_algorithm;       /* algorithm type */
>>> @@ -2005,6 +2006,7 @@ struct f2fs_sb_info {
>>>       spinlock_t dev_lock;                    /* protect dirty_device */
>>>       bool aligned_blksize;                   /* all devices has the same logical blksize */
>>>       unsigned int first_seq_zone_segno;      /* first segno in sequential zone */
>>> +     unsigned int pinned_area_max_secno;     /* upper bound section for pinned files */
>>>       unsigned int bggc_io_aware;             /* For adjust the BG_GC priority when pending IO */
>>>       unsigned int allocate_section_hint;     /* the boundary position between devices */
>>>       unsigned int allocate_section_policy;   /* determine the section writing priority */
>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>> index 56decf9c691c..ed6f2947210b 100644
>>> --- a/fs/f2fs/segment.c
>>> +++ b/fs/f2fs/segment.c
>>> @@ -2877,6 +2877,7 @@ static int get_new_segment(struct f2fs_sb_info *sbi,
>>>       unsigned int old_zoneno = GET_ZONE_FROM_SEG(sbi, *newseg);
>>>       unsigned int alloc_policy = sbi->allocate_section_policy;
>>>       unsigned int alloc_hint = sbi->allocate_section_hint;
>>> +     unsigned int max_secno = MAIN_SECS(sbi);
>>>       bool init = true;
>>>       bool looped = false;
>>>       int i, devi;
>>> @@ -2908,7 +2909,7 @@ static int get_new_segment(struct f2fs_sb_info *sbi,
>>>        */
>>>       if (f2fs_sb_has_blkzoned(sbi)) {
>>>               /* Prioritize writing to conventional zones */
>>> -             if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV || pinning)
>>> +             if (sbi->blkzone_alloc_policy == BLKZONE_ALLOC_PRIOR_CONV)
>>>                       segno = 0;
>>>               else
>>>                       segno = max(sbi->first_seq_zone_segno, *newseg);
>>> @@ -2924,19 +2925,24 @@ static int get_new_segment(struct f2fs_sb_info *sbi,
>>>               alloc_hint > MAIN_SECS(sbi))
>>>               alloc_hint = MAIN_SECS(sbi);
>>>
>>> -     if (alloc_policy == ALLOCATE_FORWARD_FROM_HINT &&
>>> -             hint < alloc_hint)
>>> +     if (pinning) {
>>> +             max_secno = sbi->pinned_area_max_secno;
>>> +             hint = 0;
>>> +     } else if (alloc_policy == ALLOCATE_FORWARD_FROM_HINT &&
>>> +             hint < alloc_hint) {
>>>               hint = alloc_hint;
>>> -     else if (alloc_policy == ALLOCATE_FORWARD_WITHIN_HINT &&
>>> -                     hint >= alloc_hint)
>>> +     } else if (alloc_policy == ALLOCATE_FORWARD_WITHIN_HINT &&
>>> +                     hint >= alloc_hint) {
>>>               hint = 0;
>>> +     }
>>>
>>>   find_other_zone:
>>> -     secno = find_next_zero_bit(free_i->free_secmap, MAIN_SECS(sbi), hint);
>>> +     secno = find_next_zero_bit(free_i->free_secmap, max_secno, hint);
>>>
>>> -     if (secno >= MAIN_SECS(sbi)) {
>>> +     if (secno >= max_secno) {
>>>               if (looped) {
>>> -                     ret = -ENOSPC;
>>> +                     ret = (pinning && has_pinned_area(sbi)) ?
>>> +                                     -EAGAIN : -ENOSPC;
>>>                       f2fs_bug_on(sbi, !pinning);
>>>                       goto out_unlock;
>>>               }
>>> @@ -3001,12 +3007,6 @@ static int get_new_segment(struct f2fs_sb_info *sbi,
>>>               goto out_unlock;
>>>       }
>>>
>>> -     /* no free section in conventional device or conventional zone */
>>> -     if (new_sec && pinning &&
>>> -             f2fs_is_sequential_zone_area(sbi, START_BLOCK(sbi, segno))) {
>>> -             ret = -EAGAIN;
>>> -             goto out_unlock;
>>> -     }
>>>       __set_inuse(sbi, segno);
>>>       *newseg = segno;
>>>   out_unlock:
>>> @@ -3472,8 +3472,9 @@ int f2fs_allocate_pinning_section(struct f2fs_sb_info *sbi)
>>>       err = f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
>>>       f2fs_unlock_op(sbi, &lc);
>>>
>>> -     if (f2fs_sb_has_blkzoned(sbi) && err == -EAGAIN && gc_required) {
>>> -             err = f2fs_gc_range(sbi, 0, sbi->first_seq_zone_segno - 1,
>>> +     if (has_pinned_area(sbi) && err == -EAGAIN && gc_required) {
>>> +             err = f2fs_gc_range(sbi, 0,
>>> +                             sbi->pinned_area_max_secno * SEGS_PER_SEC(sbi) - 1,
>>>                               true, ZONED_PIN_SEC_REQUIRED_COUNT, true);
>>>               if (err)
>>>                       return err;
>>> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h
>>> index db1079169a23..1dd8a2fa929d 100644
>>> --- a/fs/f2fs/segment.h
>>> +++ b/fs/f2fs/segment.h
>>> @@ -43,6 +43,7 @@ static inline void sanity_check_seg_type(struct f2fs_sb_info *sbi,
>>>
>>>   #define MAIN_SEGS(sbi)      (SM_I(sbi)->main_segments)
>>>   #define MAIN_SECS(sbi)      ((sbi)->total_sections)
>>> +#define has_pinned_area(sbi) ((sbi)->pinned_area_max_secno < MAIN_SECS(sbi))
>>
>> Looks more like has_unpinned_area(sbi), since we assign pinned_area_max_secno
>> w/ MAIN_SECS(sbi) by default, and will decrease it only if there is conv sections
>> or valid resizable_tail_secno.

Daeho,

Any comments about this macro naming?

>>
>>>
>>>   #define TOTAL_SEGS(sbi)                                                     \
>>>       (SM_I(sbi) ? SM_I(sbi)->segment_count :                                 \
>>> diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
>>> index 3bdb0f891c35..c1e315282ec6 100644
>>> --- a/fs/f2fs/super.c
>>> +++ b/fs/f2fs/super.c
>>> @@ -235,6 +235,7 @@ enum {
>>>       Opt_jqfmt,
>>>       Opt_checkpoint,
>>>       Opt_lookup_mode,
>>> +     Opt_resizable_tail_secno,
>>>       Opt_err,
>>>   };
>>>
>>> @@ -366,6 +367,7 @@ static const struct fs_parameter_spec f2fs_param_specs[] = {
>>>       fsparam_flag("age_extent_cache", Opt_age_extent_cache),
>>>       fsparam_enum("errors", Opt_errors, f2fs_param_errors),
>>>       fsparam_enum("lookup_mode", Opt_lookup_mode, f2fs_param_lookup_mode),
>>> +     fsparam_u32("resizable_tail_secno", Opt_resizable_tail_secno),
>>>       {}
>>>   };
>>>
>>> @@ -551,6 +553,17 @@ static inline void adjust_unusable_cap_perc(struct f2fs_sb_info *sbi)
>>>                       F2FS_OPTION(sbi).unusable_cap_perc);
>>>   }
>>>
>>> +static inline void adjust_pinned_area_boundary(struct f2fs_sb_info *sbi)
>>> +{
>>> +     sbi->pinned_area_max_secno = MAIN_SECS(sbi);
>>> +     if (f2fs_sb_has_blkzoned(sbi) && sbi->first_seq_zone_segno != NULL_SEGNO)
>>> +             sbi->pinned_area_max_secno = min(sbi->pinned_area_max_secno,
>>> +                             GET_SEC_FROM_SEG(sbi, sbi->first_seq_zone_segno));
>>> +     if (F2FS_OPTION(sbi).resizable_tail_secno)
>>> +             sbi->pinned_area_max_secno = min(sbi->pinned_area_max_secno,
>>> +                             MAIN_SECS(sbi) - F2FS_OPTION(sbi).resizable_tail_secno);
>>> +}
>>> +
>>>   static void init_once(void *foo)
>>>   {
>>>       struct f2fs_inode_info *fi = (struct f2fs_inode_info *) foo;
>>> @@ -1235,6 +1248,9 @@ static int f2fs_parse_param(struct fs_context *fc, struct fs_parameter *param)
>>>               F2FS_CTX_INFO(ctx).lookup_mode = result.uint_32;
>>>               ctx->spec_mask |= F2FS_SPEC_lookup_mode;
>>>               break;
>>> +     case Opt_resizable_tail_secno:
>>> +             F2FS_CTX_INFO(ctx).resizable_tail_secno = result.uint_32;
>>> +             break;
>>>       }
>>>       return 0;
>>>   }
>>> @@ -1771,6 +1787,12 @@ static void f2fs_apply_options(struct fs_context *fc, struct super_block *sb)
>>>
>>>   static int f2fs_sanity_check_options(struct f2fs_sb_info *sbi, bool remount)
>>>   {
>>> +     if (remount &&
>>> +         F2FS_OPTION(sbi).resizable_tail_secno >= MAIN_SECS(sbi)) {
>>> +             f2fs_err(sbi, "Option resizable_tail_secno is larger than or equal to total sections (%u >= %u)",
>>> +                             F2FS_OPTION(sbi).resizable_tail_secno, MAIN_SECS(sbi));
>>> +             return -EINVAL;
>>> +     }
>>>       if (f2fs_sb_has_device_alias(sbi) &&
>>>           !test_opt(sbi, READ_EXTENT_CACHE)) {
>>>               f2fs_err(sbi, "device aliasing requires extent cache");
>>> @@ -2544,6 +2566,10 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)
>>>       else if (F2FS_OPTION(sbi).lookup_mode == LOOKUP_AUTO)
>>>               seq_show_option(seq, "lookup_mode", "auto");
>>>
>>> +     if (F2FS_OPTION(sbi).resizable_tail_secno)
>>> +             seq_printf(seq, ",resizable_tail_secno=%u",
>>> +                             F2FS_OPTION(sbi).resizable_tail_secno);
>>> +
>>>       return 0;
>>>   }
>>>
>>> @@ -2586,6 +2612,7 @@ static void default_options(struct f2fs_sb_info *sbi, bool remount)
>>>       F2FS_OPTION(sbi).bggc_mode = BGGC_MODE_ON;
>>>       F2FS_OPTION(sbi).memory_mode = MEMORY_MODE_NORMAL;
>>>       F2FS_OPTION(sbi).errors = MOUNT_ERRORS_CONTINUE;
>>> +     F2FS_OPTION(sbi).resizable_tail_secno = 0;
>>>
>>>       set_opt(sbi, INLINE_XATTR);
>>>       set_opt(sbi, INLINE_DATA);
>>> @@ -3045,6 +3072,7 @@ static int __f2fs_remount(struct fs_context *fc, struct super_block *sb)
>>>       sb->s_flags = (sb->s_flags & ~SB_POSIXACL) |
>>>               (test_opt(sbi, POSIX_ACL) ? SB_POSIXACL : 0);
>>>
>>> +     adjust_pinned_area_boundary(sbi);
>>>       limit_reserve_root(sbi);
>>>       fc->sb_flags = (flags & ~SB_LAZYTIME) | (sb->s_flags & SB_LAZYTIME);
>>>
>>> @@ -5287,6 +5315,14 @@ static int f2fs_fill_super(struct super_block *sb, struct fs_context *fc)
>>>       /* get segno of first zoned block device */
>>>       sbi->first_seq_zone_segno = get_first_seq_zone_segno(sbi);
>>>
>>> +     if (F2FS_OPTION(sbi).resizable_tail_secno >= MAIN_SECS(sbi)) {
>>> +             f2fs_err(sbi, "Option resizable_tail_secno is larger than or equal to total sections (%u >= %u)",
>>> +                             F2FS_OPTION(sbi).resizable_tail_secno, MAIN_SECS(sbi));
>>> +             err = -EINVAL;
>>> +             goto free_nm;
>>> +     }
>>
>> How about merging this check into the one in f2fs_sanity_check_options()?
> 
> The reason it was kept separate is that during the initial mount,
> f2fs_sanity_check_options(sbi, false) is called before init_sb_info(sbi).
> At that point, sbi->total_sections has not been initialized yet, so
> MAIN_SECS(sbi) is still 0.

Ah, I see, thanks for the explanation.

> 
> If we simply check `F2FS_OPTION(sbi).resizable_tail_secno >= MAIN_SECS(sbi)`
> in f2fs_sanity_check_options(), it evaluates against 0 and causes the initial
> mount to unconditionally fail whenever resizable_tail_secno is specified.
> 
> This is similar to how `max_open_zones` is checked only on remount in
> f2fs_sanity_check_options() and checked separately during the initial mount.
> 
> However, if you'd prefer to merge them into f2fs_sanity_check_options(), we
> can use le32_to_cpu(sbi->raw_super->section_count),
> since sbi->raw_super is already loaded in memory by then.

I think latter one is better, more clean and better to maintain, please merge
into f2fs_sanity_check_options().

Thanks,

> 
> Which one do you prefer?
> 
> Thanks,
> 
>>
>> Thanks,
>>
>>> +     adjust_pinned_area_boundary(sbi);
>>> +
>>>       sbi->reserved_pin_section = f2fs_sb_has_blkzoned(sbi) ?
>>>                       ZONED_PIN_SEC_REQUIRED_COUNT :
>>>                       GET_SEC_FROM_SEG(sbi, overprovision_segments(sbi));
>>> diff --git a/fs/f2fs/sysfs.c b/fs/f2fs/sysfs.c
>>> index 3201e2185fea..811e350a1430 100644
>>> --- a/fs/f2fs/sysfs.c
>>> +++ b/fs/f2fs/sysfs.c
>>> @@ -1313,6 +1313,7 @@ F2FS_SBI_GENERAL_RW_ATTR(blkzone_alloc_policy);
>>>   #endif
>>>   F2FS_SBI_GENERAL_RW_ATTR(carve_out);
>>>   F2FS_SBI_GENERAL_RW_ATTR(reserved_pin_section);
>>> +F2FS_SBI_GENERAL_RO_ATTR(pinned_area_max_secno);
>>>   F2FS_SBI_GENERAL_RW_ATTR(bggc_io_aware);
>>>   F2FS_SBI_GENERAL_RW_ATTR(max_lock_elapsed_time);
>>>   F2FS_SBI_GENERAL_RW_ATTR(lock_duration_priority);
>>> @@ -1525,6 +1526,7 @@ static struct attribute *f2fs_attrs[] = {
>>>       ATTR_LIST(max_read_extent_count),
>>>       ATTR_LIST(carve_out),
>>>       ATTR_LIST(reserved_pin_section),
>>> +     ATTR_LIST(pinned_area_max_secno),
>>>       ATTR_LIST(allocate_section_hint),
>>>       ATTR_LIST(allocate_section_policy),
>>>       ATTR_LIST(max_lock_elapsed_time),
>>