lkml.org 
[lkml]   [2015]   [Apr]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: linux-next: manual merge of the vfs tree with the ext4 tree
On Tue, Apr 14, 2015 at 01:00:00PM -0400, Theodore Ts'o wrote:

> > Look, either nd_get_link() points inside that page (in which case that
> > kfree() is obviously invalid), or it points at kmalloc'ed buffer. In
> > which case kfree() is correct, but WTF do you need anything _else_?
> > Such as mapped pages, etc.
>
> Yes, it's either one or the other.
>
> 1) In the case of an unencrypted symlink which is too big to fit in
> the inode, we map in the first (only) block of the symlink, and set
> the link to it.

... and that kfree() will bugger us.

> 2) In the case of an encrypted symlink, we allocate memory and decrypt
> from the first block (or the i_block[] array in the inode), and then
> release the page if necessary.

... and that should've dropped that page in ->follow_link().

> I suppose we could have gone from two struct inode_operations (for
> fast and "slow" symlinks), to four struct inodes_operations (for
> [fast, unencrypted symlinks], [fast, encrypted symlinks], [slow,
> unencrypted symlinks], and [slow, encrypted symlinks]), but it was
> simpler to use a single follow_link() and put_link() function to
> handle multiple cases.

Except that you do not handle the slow unencrypted case - you end up with
kfree() on the freshly kunmaped address.


\
 
 \ /
  Last update: 2015-04-14 19:41    [W:0.069 / U:0.080 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site