lkml.org 
[lkml]   [2016]   [May]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] ext4 crypto: migrate into vfs's crypto engine
Hi Eric,

Thank you for the review.

On Fri, May 06, 2016 at 09:31:02PM -0500, Eric Biggers wrote:
> Hi Jaegeuk,
>
> On Mon, Apr 25, 2016 at 05:15:36PM -0700, Jaegeuk Kim wrote:
> > This patch removes the most parts of internal crypto codes.
> > And then, it modifies and adds some ext4-specific crypt codes to use the generic
> > facility.
>
> Except for the key name prefix issue that Ted pointed out, this overall seems
> good, although I didn't read into every detail and haven't yet tested the code.
> A few comments:
>
> There are compiler errors and warnings in the function 'dx_show_leaf()', which
> is not compiled by default.

Fixed.

>
> In ext4_lookup():
> > /*
> > * DCACHE_ENCRYPTED_WITH_KEY is set if the dentry is
> > * created while the directory was encrypted and we
> > * don't have access to the key.
> > */
> > if (fscrypt_has_encryption_key(dir))
> > fscrypt_set_encrypted_dentry(dentry);
>
> Shouldn't this say "and we have access to the key"? Or is the code wrong?

Done.

>
> In ext4_empty_dir():
> > bool err = false;
>
> Since this is a bool it should not be called "err". Maybe call it "empty"
> instead.

Removed and worked around it.

>
> In ext4_finish_bio():
> > if (!page->mapping) {
> > /* The bounce data pages are unmapped. */
> > data_page = page;
> > fscrypt_pullback_bio_page(&page, false);
> > }
> ...
> >#ifdef CONFIG_EXT4_FS_ENCRYPTION
> > if (data_page)
> > fscrypt_restore_control_page(data_page);
> >#endif
>
> Does this always do the same thing as the previous code? Does !page->mapping
> always imply that the page was involved in encrypted I/O?

I think so.

>
> In ext4_encrypted_get_link():
> > if ((cstr.len +
> > sizeof(struct fscrypt_symlink_data) - 1) >
> > max_size) {
>
> Make this one line?

Done.

>
> In ext4_file_mmap()
> > int err = fscrypt_get_encryption_info(inode);
> > if (err)
> > return 0;
>
> Should the error code be propagated to the caller?

This patch tries to keep existing flow as much as possible.

>
> In ext4_ioctl():
> > case EXT4_IOC_GET_ENCRYPTION_POLICY: {
> >#ifdef CONFIG_EXT4_FS_ENCRYPTION
> > struct fscrypt_policy policy;
> > int err = 0;
> >
> > if (!ext4_encrypted_inode(inode))
> > return -ENOENT;
>
> This is existing code and I do not know if it can be changed, but I feel that
> ENOENT is a not good error code here. If the ext4_encrypted_inode() check were
> to be removed, the implementation would match f2fs and the error code would be
> ENODATA instead.

ditto.

>
> - Eric

\
 
 \ /
  Last update: 2016-05-07 21:01    [W:0.071 / U:0.556 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site