Re: [PATCH -next RFC 01/14] block: add some bdev apis

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



> +void invalidate_bdev_range(struct block_device *bdev, pgoff_t start,
> +			   pgoff_t end)
> +{
> +	invalidate_mapping_pages(bdev->bd_inode->i_mapping, start, end);
> +}
> +EXPORT_SYMBOL_GPL(invalidate_bdev_range);

All these could probably use kerneldoc comments.

For this one I really don't like it existing at all, but we'll have to
discuss that in the btrfs patch.

> +loff_t bdev_size(struct block_device *bdev)
> +{
> +	loff_t size;
> +
> +	spin_lock(&bdev->bd_size_lock);
> +	size = i_size_read(bdev->bd_inode);
> +	spin_unlock(&bdev->bd_size_lock);
> +
> +	return size;
> +}
> +EXPORT_SYMBOL_GPL(bdev_size);

No need for this one.  The callers can simply use bdev_nr_bytes.

> +struct folio *bdev_read_folio(struct block_device *bdev, pgoff_t index)
> +{
> +	return read_mapping_folio(bdev->bd_inode->i_mapping, index, NULL);
> +}
> +EXPORT_SYMBOL_GPL(bdev_read_folio);
> +
> +struct folio *bdev_read_folio_gfp(struct block_device *bdev, pgoff_t index,
> +				  gfp_t gfp)
> +{
> +	return mapping_read_folio_gfp(bdev->bd_inode->i_mapping, index, gfp);
> +}
> +EXPORT_SYMBOL_GPL(bdev_read_folio_gfp);

I think we can just drop bdev_read_folio_gfp. Half of the callers simply
pass GPK_KERNEL, and the other half passes GFP_NOFS and could just use
memalloc_nofs_save().

> +void bdev_balance_dirty_pages_ratelimited(struct block_device *bdev)
> +{
> +	return balance_dirty_pages_ratelimited(bdev->bd_inode->i_mapping);
> +}
> +EXPORT_SYMBOL_GPL(bdev_balance_dirty_pages_ratelimited);

Hmm, this is just used for block2mtd, and feels a little too low-level
to me, as block2mtd really should be using the normal fileread/write
APIs.  I guess we'll have to live with it for now if we want to expedite
killing off bd_inode.

> +void bdev_correlate_mapping(struct block_device *bdev,
> +			    struct address_space *mapping)
> +{
> +	mapping->host = bdev->bd_inode;
> +}
> +EXPORT_SYMBOL_GPL(bdev_correlate_mapping);

Maybe associated insted of correlate?  Either way this basically
fully exposes the bdev inode again :(

> +gfp_t bdev_gfp_constraint(struct block_device *bdev, gfp_t gfp)
> +{
> +	return mapping_gfp_constraint(bdev->bd_inode->i_mapping, gfp);
> +}
> +EXPORT_SYMBOL_GPL(bdev_gfp_constraint);

The right fix here is to:

 - use memalloc_nofs_save in extet instead of using
   mapping_gfp_constraint to clear it from the mapping flags
 - remove __ext4_sb_bread_gfp and just have buffer.c helper that does
   the right thing (either by changing the calling conventions of an
   existing one, or adding a new one).

> +/*
> + * The del_gendisk() function uninitializes the disk-specific data
> + * structures, including the bdi structure, without telling anyone
> + * else.  Once this happens, any attempt to call mark_buffer_dirty()
> + * (for example, by ext4_commit_super), will cause a kernel OOPS.
> + * This is a kludge to prevent these oops until we can put in a proper
> + * hook in del_gendisk() to inform the VFS and file system layers.
> + */
> +int bdev_ejected(struct block_device *bdev)
> +{
> +	struct backing_dev_info *bdi = inode_to_bdi(bdev->bd_inode);
> +
> +	return bdi->dev == NULL;
> +}
> +EXPORT_SYMBOL_GPL(bdev_ejected);

And this code in ext4 should just go away entirely.  The bdi should
always be valid for a live bdev for years.

> --- a/block/bio.c
> +++ b/block/bio.c
> @@ -1119,6 +1119,7 @@ void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
>  	WARN_ON_ONCE(off > UINT_MAX);
>  	__bio_add_page(bio, &folio->page, len, off);
>  }
> +EXPORT_SYMBOL_GPL(bio_add_folio_nofail);

How is this realted?  The export is fine, but really should be a
separate, well-documented commit.

>  
> +static inline u8 block_bits(struct block_device *bdev)
> +{
> +	return bdev->bd_inode->i_blkbits;
> +}

Not sure we should need this.  i_blkbits comes from the blocksize
the fs set, so it should have other ways to get at it.




[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]

  Powered by Linux