Mailing List Archive: [PATCH 6/8] ext4: add EXT4_MB_HINT_GOAL_ONLY test in ext4_mb_use

ext4_mb_use_preallocated will ignore the demand to alloc goal blocks,
although the EXT4_MB_HINT_GOAL_ONLY is requested.
For group pa, ext4_mb_group_or_file will not set EXT4_MB_HINT_GROUP_ALLOC
if EXT4_MB_HINT_GOAL_ONLY is set. So we will not alloc goal blocks from
group pa if EXT4_MB_HINT_GOAL_ONLY is set.
For inode pa, ext4_mb_pa_goal_check is added to check if free extent in
found inode pa meets goal blocks when EXT4_MB_HINT_GOAL_ONLY is set.

Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
Suggested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
---
fs/ext4/mballoc.c | 34 +++++++++++++++++++++++++++++++++-
1 file changed, 33 insertions(+), 1 deletion(-)

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index b963111eeec6..5b837918c624 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -4351,6 +4351,37 @@ ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
return pa;
}

+/*
+ * check if found pa meets EXT4_MB_HINT_GOAL_ONLY
+ */
+static bool
+ext4_mb_pa_goal_check(struct ext4_allocation_context *ac,
+ struct ext4_prealloc_space *pa)
+{
+ struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
+ ext4_fsblk_t start;
+
+ if (likely(!(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY)))
+ return true;
+
+ /*
+ * If EXT4_MB_HINT_GOAL_ONLY is set, ac_g_ex will not be adjusted
+ * in ext4_mb_normalize_request and will keep same with ac_o_ex
+ * from ext4_mb_initialize_context. Choose ac_g_ex here to keep
+ * consistent with ext4_mb_find_by_goal.
+ */
+ start = pa->pa_pstart +
+ (ac->ac_g_ex.fe_logical - pa->pa_lstart);
+ if (ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex) != start)
+ return false;
+
+ if (ac->ac_g_ex.fe_len > pa->pa_len -
+ EXT4_B2C(sbi, ac->ac_g_ex.fe_logical - pa->pa_lstart))
+ return false;
+
+ return true;
+}
+
/*
* search goal blocks in preallocated space
*/
@@ -4387,7 +4418,8 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac)

/* found preallocated blocks, use them */
spin_lock(&pa->pa_lock);
- if (pa->pa_deleted == 0 && pa->pa_free) {
+ if (pa->pa_deleted == 0 && pa->pa_free &&
+ likely(ext4_mb_pa_goal_check(ac, pa))) {
atomic_inc(&pa->pa_count);
ext4_mb_use_inode_pa(ac, pa);
spin_unlock(&pa->pa_lock);
--
2.30.0

On Wed, Mar 22, 2023 at 12:12:18AM +0800, Kemeng Shi wrote:
> ext4_mb_use_preallocated will ignore the demand to alloc goal blocks,
> although the EXT4_MB_HINT_GOAL_ONLY is requested.
> For group pa, ext4_mb_group_or_file will not set EXT4_MB_HINT_GROUP_ALLOC
> if EXT4_MB_HINT_GOAL_ONLY is set. So we will not alloc goal blocks from
> group pa if EXT4_MB_HINT_GOAL_ONLY is set.
> For inode pa, ext4_mb_pa_goal_check is added to check if free extent in
> found inode pa meets goal blocks when EXT4_MB_HINT_GOAL_ONLY is set.
>
> Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com>
> Suggested-by: Ojaswin Mujoo <ojaswin@linux.ibm.com>
> ---
> fs/ext4/mballoc.c | 34 +++++++++++++++++++++++++++++++++-
> 1 file changed, 33 insertions(+), 1 deletion(-)
>
> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
> index b963111eeec6..5b837918c624 100644
> --- a/fs/ext4/mballoc.c
> +++ b/fs/ext4/mballoc.c
> @@ -4351,6 +4351,37 @@ ext4_mb_check_group_pa(ext4_fsblk_t goal_block,
> return pa;
> }
>
> +/*
> + * check if found pa meets EXT4_MB_HINT_GOAL_ONLY
> + */
> +static bool
> +ext4_mb_pa_goal_check(struct ext4_allocation_context *ac,
> + struct ext4_prealloc_space *pa)
> +{
> + struct ext4_sb_info *sbi = EXT4_SB(ac->ac_sb);
> + ext4_fsblk_t start;
> +
> + if (likely(!(ac->ac_flags & EXT4_MB_HINT_GOAL_ONLY)))
> + return true;
> +
> + /*
> + * If EXT4_MB_HINT_GOAL_ONLY is set, ac_g_ex will not be adjusted
> + * in ext4_mb_normalize_request and will keep same with ac_o_ex
> + * from ext4_mb_initialize_context. Choose ac_g_ex here to keep
> + * consistent with ext4_mb_find_by_goal.
> + */
> + start = pa->pa_pstart +
> + (ac->ac_g_ex.fe_logical - pa->pa_lstart);
> + if (ext4_grp_offs_to_block(ac->ac_sb, &ac->ac_g_ex) != start)
> + return false;
> +
> + if (ac->ac_g_ex.fe_len > pa->pa_len -
> + EXT4_B2C(sbi, ac->ac_g_ex.fe_logical - pa->pa_lstart))
> + return false;
> +
> + return true;
> +}
> +
> /*
> * search goal blocks in preallocated space
> */
> @@ -4387,7 +4418,8 @@ ext4_mb_use_preallocated(struct ext4_allocation_context *ac)
>
> /* found preallocated blocks, use them */
> spin_lock(&pa->pa_lock);
> - if (pa->pa_deleted == 0 && pa->pa_free) {
> + if (pa->pa_deleted == 0 && pa->pa_free &&
> + likely(ext4_mb_pa_goal_check(ac, pa))) {
Alright so this looks good to me, I just have one concern.

Consider the following scenario:

1. We find a PA that cover our logical blocks but the physical blocks in
the PA don't match our goal and we return false.

2. We then proceed in ext4_mb_regular_allocator and actually manage to
allocate the goal blocks.

3. Now the logical blocks of the file are associated to the blocks
allocated in Step 2 and the logical blocks of the PA would actually never get
utilized, and hence they are wasted space.

That being said, I think fixing this would be a bit tricky and I'm not sure if
it's even worth the trouble as GOAL_ONLY is not a very common case anyways. Would like
to hear other people's thoughts on this.

Regards,
ojaswin
> atomic_inc(&pa->pa_count);
> ext4_mb_use_inode_pa(ac, pa);
> spin_unlock(&pa->pa_lock);
> --
> 2.30.0
>