Re: [PATCH 1/2] ecryptfs: Improve metatdata read failure logging

[Tyler Hicks]

On 2012-01-11 18:00:41, Tim Gardner wrote:
> There are 3 read failure cases in ecryptfs_read_metadata(), but only 2
> of them are uniquely noted by kernel log messages. This patch
> identifies and logs each read failure case. It also correctly interprets
> a negative return value from ecryptfs_read_lower().

I'm not sure that the negative return value was incorrectly interpreted.
See below.

> 
> Removes unnecessary variable initialization.
> 
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> Cc: stable@xxxxxxxxxxxxxxx
> Cc: Tyler Hicks <tyler.hicks@xxxxxxxxxxxxx>
> Signed-off-by: Tim Gardner <tim.gardner@xxxxxxxxxxxxx>
> ---
>  fs/ecryptfs/crypto.c |   17 +++++++++++------
>  1 files changed, 11 insertions(+), 6 deletions(-)
> 
> diff --git a/fs/ecryptfs/crypto.c b/fs/ecryptfs/crypto.c
> index d3c8776..ac063bd 100644
> --- a/fs/ecryptfs/crypto.c
> +++ b/fs/ecryptfs/crypto.c
> @@ -1590,8 +1590,8 @@ int ecryptfs_read_and_validate_xattr_region(struct dentry *dentry,
>   */
>  int ecryptfs_read_metadata(struct dentry *ecryptfs_dentry)
>  {
> -	int rc = 0;
> -	char *page_virt = NULL;
> +	int rc;
> +	char *page_virt;
>  	struct inode *ecryptfs_inode = ecryptfs_dentry->d_inode;
>  	struct ecryptfs_crypt_stat *crypt_stat =
>  	    &ecryptfs_inode_to_private(ecryptfs_inode)->crypt_stat;
> @@ -1611,10 +1611,15 @@ int ecryptfs_read_metadata(struct dentry *ecryptfs_dentry)
>  	}
>  	rc = ecryptfs_read_lower(page_virt, 0, crypt_stat->extent_size,
>  				 ecryptfs_inode);
> -	if (rc >= 0)
> -		rc = ecryptfs_read_headers_virt(page_virt, crypt_stat,
> -						ecryptfs_dentry,
> -						ECRYPTFS_VALIDATE_HEADER_SIZE);
> +	if (rc < 0) {
> +		printk(KERN_ERR "%s: Could not read %u bytes\n",
> +			__func__, crypt_stat->extent_size);
> +		goto out;
> +	}

This may break things a bit for users that use the
ecryptfs_xattr_metadata mount option (which stores the metadata in an
xattr, rather than the header of the file). A failure to read the lower
file here doesn't matter because the metadata is likely stored in an
xattr, which will be found with the ecryptfs_read_xattr_region() call
below.

I've never liked that ecryptfs potentially looks in both the header and
the xattr for metadata, but that's how it has been since the very early
days of eCryptfs. I've wanted to change this but there is the potential
to break things for users who have a mixture of files with metadata in
the header and in the header.

Maybe we just go whole hog for 3.3 and only look in either the header or
xattr for metadata, depending on whether or not ecryptfs_xattr_metadata
is used?

I could live with that, but I would definitely not want something like
that to go to the stable kernel. Any thoughts?

Tyler

> +
> +	rc = ecryptfs_read_headers_virt(page_virt, crypt_stat,
> +					ecryptfs_dentry,
> +					ECRYPTFS_VALIDATE_HEADER_SIZE);
>  	if (rc) {
>  		memset(page_virt, 0, PAGE_CACHE_SIZE);
>  		rc = ecryptfs_read_xattr_region(page_virt, ecryptfs_inode);
> -- 
> 1.7.8.3
>

Attachment: signature.asc
Description: Digital signature