diff mbox

[RFC,3/8] ext4: decrypt all contiguous blocks in a page

Message ID 20180112141129.27507-4-chandan@linux.vnet.ibm.com (mailing list archive)
State New, archived
Headers show

Commit Message

Chandan Rajendra Jan. 12, 2018, 2:11 p.m. UTC
With blocksize < pagesize, a page can contain more than one block. Hence
this commit changes completion_pages() to invoke fscrypt_decrypt_page()
in order to decrypt all the contiguous blocks mapped by the page.

Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
---
 fs/crypto/bio.c | 19 +++++++++++++++++--
 1 file changed, 17 insertions(+), 2 deletions(-)

Comments

Eric Biggers Jan. 17, 2018, 2:18 a.m. UTC | #1
Hi Chandan,

On Fri, Jan 12, 2018 at 07:41:24PM +0530, Chandan Rajendra wrote:
> With blocksize < pagesize, a page can contain more than one block. Hence
> this commit changes completion_pages() to invoke fscrypt_decrypt_page()
> in order to decrypt all the contiguous blocks mapped by the page.
> 
> Signed-off-by: Chandan Rajendra <chandan@linux.vnet.ibm.com>
> ---
>  fs/crypto/bio.c | 19 +++++++++++++++++--
>  1 file changed, 17 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/crypto/bio.c b/fs/crypto/bio.c
> index 0d5e6a5..eb6e06a 100644
> --- a/fs/crypto/bio.c
> +++ b/fs/crypto/bio.c
> @@ -40,8 +40,23 @@ static void completion_pages(struct work_struct *work)
>  
>  	bio_for_each_segment_all(bv, bio, i) {
>  		struct page *page = bv->bv_page;
> -		int ret = fscrypt_decrypt_page(page->mapping->host, page,
> -				PAGE_SIZE, 0, page->index);
> +		struct inode *inode = page->mapping->host;
> +		const unsigned long blocksize = inode->i_sb->s_blocksize;
> +		const unsigned blkbits = inode->i_blkbits;
> +		int page_blk = page->index << (PAGE_SHIFT - blkbits);
> +		int blk = page_blk + (bv->bv_offset >> blkbits);

Use 'u64' for the block number:

	u64 page_blk = (u64)page->index << (PAGE_SHIFT - blkbits);
	u64 blk = page_blk + (bv->bv_offset >> blkbits);

> +		int nr_blks = bv->bv_len >> blkbits;
> +		int ret = 0;
> +		int j;
> +
> +		for (j = 0; j < nr_blks; j++, blk++) {
> +			ret = fscrypt_decrypt_page(page->mapping->host,
> +						page, blocksize,
> +						bv->bv_offset + (j << blkbits),
> +						blk);
> +			if (ret)
> +				break;
> +		}
>  
>  		if (ret) {
>  			WARN_ON_ONCE(1);

Since that we'll now actually be operating on blocks rather than pages, some
renaming seems to be in order, otherwise things will get very confusing.  e.g.:

	fscrypt_decrypt_page() -> fscrypt_decrypt_block()
	fscrypt_encrypt_page() -> fscrypt_encrypt_block()
	completion_pages() -> completion_blocks()
	fscrypt_decrypt_bio_pages() -> fscrypt_decrypt_bio_blocks()

Please also update the comment for completion_pages() / completion_blocks() to
clarify that it is decrypting *blocks*, not *pages*.

(Yes, we should have named all these functions as *_block() originally.  But
this is a good time to fix it!)

Eric
diff mbox

Patch

diff --git a/fs/crypto/bio.c b/fs/crypto/bio.c
index 0d5e6a5..eb6e06a 100644
--- a/fs/crypto/bio.c
+++ b/fs/crypto/bio.c
@@ -40,8 +40,23 @@  static void completion_pages(struct work_struct *work)
 
 	bio_for_each_segment_all(bv, bio, i) {
 		struct page *page = bv->bv_page;
-		int ret = fscrypt_decrypt_page(page->mapping->host, page,
-				PAGE_SIZE, 0, page->index);
+		struct inode *inode = page->mapping->host;
+		const unsigned long blocksize = inode->i_sb->s_blocksize;
+		const unsigned blkbits = inode->i_blkbits;
+		int page_blk = page->index << (PAGE_SHIFT - blkbits);
+		int blk = page_blk + (bv->bv_offset >> blkbits);
+		int nr_blks = bv->bv_len >> blkbits;
+		int ret = 0;
+		int j;
+
+		for (j = 0; j < nr_blks; j++, blk++) {
+			ret = fscrypt_decrypt_page(page->mapping->host,
+						page, blocksize,
+						bv->bv_offset + (j << blkbits),
+						blk);
+			if (ret)
+				break;
+		}
 
 		if (ret) {
 			WARN_ON_ONCE(1);