diff mbox series

[RFC,v3,for-6.8/block,11/17] erofs: use bdev api

Message ID 20231221085826.1768395-1-yukuai1@huaweicloud.com
State New
Headers show
Series block: don't access bd_inode directly from other modules | expand

Commit Message

Yu Kuai Dec. 21, 2023, 8:58 a.m. UTC 
From: Yu Kuai <yukuai3@huawei.com>

Avoid to access bd_inode directly, prepare to remove bd_inode from
block_device.

Signed-off-by: Yu Kuai <yukuai3@huawei.com>
---
 fs/erofs/data.c     | 18 ++++++++++++------
 fs/erofs/internal.h |  2 ++
 2 files changed, 14 insertions(+), 6 deletions(-)

Comments

Jan Kara Jan. 4, 2024, 12:02 p.m. UTC | #1 
On Thu 21-12-23 16:58:26, Yu Kuai wrote:
> From: Yu Kuai <yukuai3@huawei.com>
> 
> Avoid to access bd_inode directly, prepare to remove bd_inode from
> block_device.
> 
> Signed-off-by: Yu Kuai <yukuai3@huawei.com>

I'm not erofs maintainer but IMO this is quite ugly and grows erofs_buf
unnecessarily. I'd rather store 'sb' pointer in erofs_buf and then do the
right thing in erofs_bread() which is the only place that seems to care
about the erofs_is_fscache_mode() distinction... Also blkszbits is then
trivially sb->s_blocksize_bits so it would all seem much more
straightforward.

								Honza

> ---
>  fs/erofs/data.c     | 18 ++++++++++++------
>  fs/erofs/internal.h |  2 ++
>  2 files changed, 14 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/erofs/data.c b/fs/erofs/data.c
> index c98aeda8abb2..bbe2fe199bf3 100644
> --- a/fs/erofs/data.c
> +++ b/fs/erofs/data.c
> @@ -32,8 +32,8 @@ void erofs_put_metabuf(struct erofs_buf *buf)
>  void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>  		  enum erofs_kmap_type type)
>  {
> -	struct inode *inode = buf->inode;
> -	erofs_off_t offset = (erofs_off_t)blkaddr << inode->i_blkbits;
> +	u8 blkszbits = buf->inode ? buf->inode->i_blkbits : buf->blkszbits;
> +	erofs_off_t offset = (erofs_off_t)blkaddr << blkszbits;
>  	pgoff_t index = offset >> PAGE_SHIFT;
>  	struct page *page = buf->page;
>  	struct folio *folio;
> @@ -43,7 +43,9 @@ void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>  		erofs_put_metabuf(buf);
>  
>  		nofs_flag = memalloc_nofs_save();
> -		folio = read_cache_folio(inode->i_mapping, index, NULL, NULL);
> +		folio = buf->inode ?
> +			read_mapping_folio(buf->inode->i_mapping, index, NULL) :
> +			bdev_read_folio(buf->bdev, offset);
>  		memalloc_nofs_restore(nofs_flag);
>  		if (IS_ERR(folio))
>  			return folio;
> @@ -67,10 +69,14 @@ void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>  
>  void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
>  {
> -	if (erofs_is_fscache_mode(sb))
> +	if (erofs_is_fscache_mode(sb)) {
>  		buf->inode = EROFS_SB(sb)->s_fscache->inode;
> -	else
> -		buf->inode = sb->s_bdev->bd_inode;
> +		buf->bdev = NULL;
> +	} else {
> +		buf->inode = NULL;
> +		buf->bdev = sb->s_bdev;
> +		buf->blkszbits = EROFS_SB(sb)->blkszbits;
> +	}
>  }
>  
>  void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
> index b0409badb017..c9206351b485 100644
> --- a/fs/erofs/internal.h
> +++ b/fs/erofs/internal.h
> @@ -224,8 +224,10 @@ enum erofs_kmap_type {
>  
>  struct erofs_buf {
>  	struct inode *inode;
> +	struct block_device *bdev;
>  	struct page *page;
>  	void *base;
> +	u8 blkszbits;
>  	enum erofs_kmap_type kmap_type;
>  };
>  #define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })
> -- 
> 2.39.2
>
Yu Kuai Jan. 4, 2024, 12:32 p.m. UTC | #2 
Hi, Jan!

在 2024/01/04 20:02, Jan Kara 写道:
> On Thu 21-12-23 16:58:26, Yu Kuai wrote:
>> From: Yu Kuai <yukuai3@huawei.com>
>>
>> Avoid to access bd_inode directly, prepare to remove bd_inode from
>> block_device.
>>
>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
> 
> I'm not erofs maintainer but IMO this is quite ugly and grows erofs_buf
> unnecessarily. I'd rather store 'sb' pointer in erofs_buf and then do the
> right thing in erofs_bread() which is the only place that seems to care
> about the erofs_is_fscache_mode() distinction... Also blkszbits is then
> trivially sb->s_blocksize_bits so it would all seem much more
> straightforward.

Thanks for your suggestion, I'll follow this unless Gao Xiang has other
suggestions.

Kuai
> 
> 								Honza
> 
>> ---
>>   fs/erofs/data.c     | 18 ++++++++++++------
>>   fs/erofs/internal.h |  2 ++
>>   2 files changed, 14 insertions(+), 6 deletions(-)
>>
>> diff --git a/fs/erofs/data.c b/fs/erofs/data.c
>> index c98aeda8abb2..bbe2fe199bf3 100644
>> --- a/fs/erofs/data.c
>> +++ b/fs/erofs/data.c
>> @@ -32,8 +32,8 @@ void erofs_put_metabuf(struct erofs_buf *buf)
>>   void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>>   		  enum erofs_kmap_type type)
>>   {
>> -	struct inode *inode = buf->inode;
>> -	erofs_off_t offset = (erofs_off_t)blkaddr << inode->i_blkbits;
>> +	u8 blkszbits = buf->inode ? buf->inode->i_blkbits : buf->blkszbits;
>> +	erofs_off_t offset = (erofs_off_t)blkaddr << blkszbits;
>>   	pgoff_t index = offset >> PAGE_SHIFT;
>>   	struct page *page = buf->page;
>>   	struct folio *folio;
>> @@ -43,7 +43,9 @@ void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>>   		erofs_put_metabuf(buf);
>>   
>>   		nofs_flag = memalloc_nofs_save();
>> -		folio = read_cache_folio(inode->i_mapping, index, NULL, NULL);
>> +		folio = buf->inode ?
>> +			read_mapping_folio(buf->inode->i_mapping, index, NULL) :
>> +			bdev_read_folio(buf->bdev, offset);
>>   		memalloc_nofs_restore(nofs_flag);
>>   		if (IS_ERR(folio))
>>   			return folio;
>> @@ -67,10 +69,14 @@ void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
>>   
>>   void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
>>   {
>> -	if (erofs_is_fscache_mode(sb))
>> +	if (erofs_is_fscache_mode(sb)) {
>>   		buf->inode = EROFS_SB(sb)->s_fscache->inode;
>> -	else
>> -		buf->inode = sb->s_bdev->bd_inode;
>> +		buf->bdev = NULL;
>> +	} else {
>> +		buf->inode = NULL;
>> +		buf->bdev = sb->s_bdev;
>> +		buf->blkszbits = EROFS_SB(sb)->blkszbits;
>> +	}
>>   }
>>   
>>   void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
>> diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
>> index b0409badb017..c9206351b485 100644
>> --- a/fs/erofs/internal.h
>> +++ b/fs/erofs/internal.h
>> @@ -224,8 +224,10 @@ enum erofs_kmap_type {
>>   
>>   struct erofs_buf {
>>   	struct inode *inode;
>> +	struct block_device *bdev;
>>   	struct page *page;
>>   	void *base;
>> +	u8 blkszbits;
>>   	enum erofs_kmap_type kmap_type;
>>   };
>>   #define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })
>> -- 
>> 2.39.2
>>
Gao Xiang Jan. 5, 2024, 4:43 a.m. UTC | #3 
On 2024/1/4 20:32, Yu Kuai wrote:
> Hi, Jan!
> 
> 在 2024/01/04 20:02, Jan Kara 写道:
>> On Thu 21-12-23 16:58:26, Yu Kuai wrote:
>>> From: Yu Kuai <yukuai3@huawei.com>
>>>
>>> Avoid to access bd_inode directly, prepare to remove bd_inode from
>>> block_device.
>>>
>>> Signed-off-by: Yu Kuai <yukuai3@huawei.com>
>>
>> I'm not erofs maintainer but IMO this is quite ugly and grows erofs_buf
>> unnecessarily. I'd rather store 'sb' pointer in erofs_buf and then do the
>> right thing in erofs_bread() which is the only place that seems to care
>> about the erofs_is_fscache_mode() distinction... Also blkszbits is then
>> trivially sb->s_blocksize_bits so it would all seem much more
>> straightforward.
> 
> Thanks for your suggestion, I'll follow this unless Gao Xiang has other
> suggestions.

Yes, that would be better, I'm fine with that.  Yet in the future we
may support a seperate large dirblocksize more than block size, but
we could revisit later.

Thanks,
Gao Xiang

> 
> Kuai
>>
>>                                 Honza
>>
diff mbox series

Patch

diff --git a/fs/erofs/data.c b/fs/erofs/data.c
index c98aeda8abb2..bbe2fe199bf3 100644
--- a/fs/erofs/data.c
+++ b/fs/erofs/data.c
@@ -32,8 +32,8 @@  void erofs_put_metabuf(struct erofs_buf *buf)
 void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
 		  enum erofs_kmap_type type)
 {
-	struct inode *inode = buf->inode;
-	erofs_off_t offset = (erofs_off_t)blkaddr << inode->i_blkbits;
+	u8 blkszbits = buf->inode ? buf->inode->i_blkbits : buf->blkszbits;
+	erofs_off_t offset = (erofs_off_t)blkaddr << blkszbits;
 	pgoff_t index = offset >> PAGE_SHIFT;
 	struct page *page = buf->page;
 	struct folio *folio;
@@ -43,7 +43,9 @@  void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
 		erofs_put_metabuf(buf);
 
 		nofs_flag = memalloc_nofs_save();
-		folio = read_cache_folio(inode->i_mapping, index, NULL, NULL);
+		folio = buf->inode ?
+			read_mapping_folio(buf->inode->i_mapping, index, NULL) :
+			bdev_read_folio(buf->bdev, offset);
 		memalloc_nofs_restore(nofs_flag);
 		if (IS_ERR(folio))
 			return folio;
@@ -67,10 +69,14 @@  void *erofs_bread(struct erofs_buf *buf, erofs_blk_t blkaddr,
 
 void erofs_init_metabuf(struct erofs_buf *buf, struct super_block *sb)
 {
-	if (erofs_is_fscache_mode(sb))
+	if (erofs_is_fscache_mode(sb)) {
 		buf->inode = EROFS_SB(sb)->s_fscache->inode;
-	else
-		buf->inode = sb->s_bdev->bd_inode;
+		buf->bdev = NULL;
+	} else {
+		buf->inode = NULL;
+		buf->bdev = sb->s_bdev;
+		buf->blkszbits = EROFS_SB(sb)->blkszbits;
+	}
 }
 
 void *erofs_read_metabuf(struct erofs_buf *buf, struct super_block *sb,
diff --git a/fs/erofs/internal.h b/fs/erofs/internal.h
index b0409badb017..c9206351b485 100644
--- a/fs/erofs/internal.h
+++ b/fs/erofs/internal.h
@@ -224,8 +224,10 @@  enum erofs_kmap_type {
 
 struct erofs_buf {
 	struct inode *inode;
+	struct block_device *bdev;
 	struct page *page;
 	void *base;
+	u8 blkszbits;
 	enum erofs_kmap_type kmap_type;
 };
 #define __EROFS_BUF_INITIALIZER	((struct erofs_buf){ .page = NULL })