[PATCH v2] ext4: delete invalid ac_b_extent backup inside ext4_mb_use_best_found()

[PATCH v2] ext4: delete invalid ac_b_extent backup inside ext4_mb_use_best_found()

[brookxu]

Delete invalid ac_b_extent backup inside ext4_mb_use_best_found(),
we have done this operation in ext4_mb_new_group_pa() and
ext4_mb_new_inode_pa().

Signed-off-by: Chunguang Xu <brookxu@tencent.com>

diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 9b1c3ad..fb63e9f 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -1704,10 +1704,6 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
 	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
 	ret = mb_mark_used(e4b, &ac->ac_b_ex);
 
-	/* preallocation can change ac_b_ex, thus we store actually
-	 * allocated blocks for history */
-	ac->ac_f_ex = ac->ac_b_ex;
-
 	ac->ac_status = AC_STATUS_FOUND;
 	ac->ac_tail = ret & 0xffff;
 	ac->ac_buddy = ret >> 16;
@@ -1726,8 +1722,8 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
 	/* store last allocated for subsequent stream allocation */
 	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
 		spin_lock(&sbi->s_md_lock);
-		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
-		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
+		sbi->s_mb_last_group = ac->ac_b_ex.fe_group;
+		sbi->s_mb_last_start = ac->ac_b_ex.fe_start;
 		spin_unlock(&sbi->s_md_lock);
 	}
 	/*
-- 
1.8.3.1

Re: [PATCH v2] ext4: delete invalid ac_b_extent backup inside ext4_mb_use_best_found()

[Andreas Dilger]

[-- Attachment #1: Type: text/plain, Size: 2555 bytes --]

On Aug 7, 2020, at 5:32 AM, brookxu <brookxu.cn@gmail.com> wrote:
> 
> Delete invalid ac_b_extent backup inside ext4_mb_use_best_found(),
> we have done this operation in ext4_mb_new_group_pa() and
> ext4_mb_new_inode_pa().

I'm not sure I understand this patch completely.

The calls to ext4_mb_new_group_pa() and ext4_mb_new_inode_pa() are
done from ext4_mb_new_preallocation(), which is called at the *end*
of ext4_mb_use_best_found() (i.e. after the lines that are being
deleted).

Maybe I'm confused by the description "we *have done* this operation"
makes it seem like it was already done, but really it should be
"we *will do* this operation in ..."?

That said, it would make more sense to keep the one line here in
ext4_mb_use_best_found() and remove the two duplicate lines in
ext4_mb_new_group_pa() and ext4_mb_new_inode_pa()?  In that case,
the patch description would be more correct, like:

    Delete duplicate ac_b_extent backup in ext4_mb_new_group_pa()
    and ext4_mb_new_inode_pa(), since we have done this operation
    in ext4_mb_use_best_found() already.

Cheers, Andreas

PS: thank you for taking the time to look at this code and improve it.
I know it is complex and hard to understand, but going through it like
this and trimming off the bad bits makes it a bit easier to understand
and maintain with each small patch.

> Signed-off-by: Chunguang Xu <brookxu@tencent.com>
> 
> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
> index 9b1c3ad..fb63e9f 100644
> --- a/fs/ext4/mballoc.c
> +++ b/fs/ext4/mballoc.c
> @@ -1704,10 +1704,6 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
> 	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
> 	ret = mb_mark_used(e4b, &ac->ac_b_ex);
> 
> -	/* preallocation can change ac_b_ex, thus we store actually
> -	 * allocated blocks for history */
> -	ac->ac_f_ex = ac->ac_b_ex;
> -
> 	ac->ac_status = AC_STATUS_FOUND;
> 	ac->ac_tail = ret & 0xffff;
> 	ac->ac_buddy = ret >> 16;
> @@ -1726,8 +1722,8 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
> 	/* store last allocated for subsequent stream allocation */
> 	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
> 		spin_lock(&sbi->s_md_lock);
> -		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
> -		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
> +		sbi->s_mb_last_group = ac->ac_b_ex.fe_group;
> +		sbi->s_mb_last_start = ac->ac_b_ex.fe_start;
> 		spin_unlock(&sbi->s_md_lock);
> 	}
> 	/*
> --
> 1.8.3.1


Cheers, Andreas






[-- Attachment #2: Message signed with OpenPGP --]
[-- Type: application/pgp-signature, Size: 873 bytes --]

Re: [PATCH v2] ext4: delete invalid ac_b_extent backup inside ext4_mb_use_best_found()

[brookxu]

Thank you very much for taking the time to review this patch. Due to poor thinking, there are some
problems with this patch. I think this patch can be ignored. Thank you again for your time.

thanks

Andreas Dilger wrote on 2020/8/13 16:44:
> On Aug 7, 2020, at 5:32 AM, brookxu <brookxu.cn@gmail.com> wrote:
>>
>> Delete invalid ac_b_extent backup inside ext4_mb_use_best_found(),
>> we have done this operation in ext4_mb_new_group_pa() and
>> ext4_mb_new_inode_pa().
> 
> I'm not sure I understand this patch completely.
> 
> The calls to ext4_mb_new_group_pa() and ext4_mb_new_inode_pa() are
> done from ext4_mb_new_preallocation(), which is called at the *end*
> of ext4_mb_use_best_found() (i.e. after the lines that are being
> deleted).
> 
> Maybe I'm confused by the description "we *have done* this operation"
> makes it seem like it was already done, but really it should be
> "we *will do* this operation in ..."?
> 
> That said, it would make more sense to keep the one line here in
> ext4_mb_use_best_found() and remove the two duplicate lines in
> ext4_mb_new_group_pa() and ext4_mb_new_inode_pa()?  In that case,
> the patch description would be more correct, like:
> 
>     Delete duplicate ac_b_extent backup in ext4_mb_new_group_pa()
>     and ext4_mb_new_inode_pa(), since we have done this operation
>     in ext4_mb_use_best_found() already.
> 
> Cheers, Andreas
> 
> PS: thank you for taking the time to look at this code and improve it.
> I know it is complex and hard to understand, but going through it like
> this and trimming off the bad bits makes it a bit easier to understand
> and maintain with each small patch.
> 
>> Signed-off-by: Chunguang Xu <brookxu@tencent.com>
>>
>> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
>> index 9b1c3ad..fb63e9f 100644
>> --- a/fs/ext4/mballoc.c
>> +++ b/fs/ext4/mballoc.c
>> @@ -1704,10 +1704,6 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
>> 	ac->ac_b_ex.fe_logical = ac->ac_g_ex.fe_logical;
>> 	ret = mb_mark_used(e4b, &ac->ac_b_ex);
>>
>> -	/* preallocation can change ac_b_ex, thus we store actually
>> -	 * allocated blocks for history */
>> -	ac->ac_f_ex = ac->ac_b_ex;
>> -
>> 	ac->ac_status = AC_STATUS_FOUND;
>> 	ac->ac_tail = ret & 0xffff;
>> 	ac->ac_buddy = ret >> 16;
>> @@ -1726,8 +1722,8 @@ static void ext4_mb_use_best_found(struct ext4_allocation_context *ac,
>> 	/* store last allocated for subsequent stream allocation */
>> 	if (ac->ac_flags & EXT4_MB_STREAM_ALLOC) {
>> 		spin_lock(&sbi->s_md_lock);
>> -		sbi->s_mb_last_group = ac->ac_f_ex.fe_group;
>> -		sbi->s_mb_last_start = ac->ac_f_ex.fe_start;
>> +		sbi->s_mb_last_group = ac->ac_b_ex.fe_group;
>> +		sbi->s_mb_last_start = ac->ac_b_ex.fe_start;
>> 		spin_unlock(&sbi->s_md_lock);
>> 	}
>> 	/*
>> --
>> 1.8.3.1
> 
> 
> Cheers, Andreas
> 
> 
> 
> 
> 

Re: [PATCH v2] ext4: delete invalid ac_b_extent backup inside ext4_mb_use_best_found()

[Ritesh Harjani]

On 8/13/20 2:14 PM, Andreas Dilger wrote:
> On Aug 7, 2020, at 5:32 AM, brookxu <brookxu.cn@gmail.com> wrote:
>>
>> Delete invalid ac_b_extent backup inside ext4_mb_use_best_found(),
>> we have done this operation in ext4_mb_new_group_pa() and
>> ext4_mb_new_inode_pa().
> 
> I'm not sure I understand this patch completely.
> 
> The calls to ext4_mb_new_group_pa() and ext4_mb_new_inode_pa() are
> done from ext4_mb_new_preallocation(), which is called at the *end*
> of ext4_mb_use_best_found() (i.e. after the lines that are being
> deleted).
> 
> Maybe I'm confused by the description "we *have done* this operation"
> makes it seem like it was already done, but really it should be
> "we *will do* this operation in ..."?
> 
> That said, it would make more sense to keep the one line here in
> ext4_mb_use_best_found() and remove the two duplicate lines in
> ext4_mb_new_group_pa() and ext4_mb_new_inode_pa()?  In that case,
> the patch description would be more correct, like:
> 
>      Delete duplicate ac_b_extent backup in ext4_mb_new_group_pa()
>      and ext4_mb_new_inode_pa(), since we have done this operation
>      in ext4_mb_use_best_found() already.
> 

Looked into the mballoc code and I agree with Andreas points here.

-ritesh