Re: [PATCH 3/3] eCryptfs: Encrypted passthrough

[Andrew Morton <akpm@osdl.org>]

On Tue, 9 Jan 2007 16:23:37 -0600
Michael Halcrow <mhalcrow@us.ibm.com> wrote:

> +				page_virt = (char *)kmap(page);

Do we _have_ to use kmap here?  It's slow and theoretically deadlocky. 
kmap_atomic() is much preferred.

Can the other kmap() calls in ecryptfs be converted?

We'd actually like to remove kmap() one day.  Not much chance of that, but
it's an objective.

> +				if (!page_virt) {
> +					rc = -ENOMEM;
> +					printk(KERN_ERR "Error mapping page\n");
> +					goto out;
> +				}
> +				memset(page_virt, 0, PAGE_CACHE_SIZE);
> +				if (page->index == 0) {
> +					rc = ecryptfs_read_xattr_region(
> +						page_virt, file->f_path.dentry);

Are we assured that ecryptfs_read_xattr_region() cannot overrun the page?

> +					set_header_info(page_virt, crypt_stat);
> +				}

The kernel must always run flush_dcache_page() after modifying a pagecache
page by hand.  Please review all of ecryptfs for this.


> +				kunmap(page);