[RFC,v6,03/40] vfs: Add MAY_DELETE_SELF and MAY_DELETE_CHILD permission flags
diff mbox

Message ID 1438689218-6921-4-git-send-email-agruenba@redhat.com
State New
Headers show

Commit Message

Andreas Grünbacher Aug. 4, 2015, 11:53 a.m. UTC
Normally, deleting a file requires write and execute access to the parent
directory.  With Richacls, a process with MAY_DELETE_SELF access to a file
may delete the file even without write access to the parent directory.

To support that, pass the MAY_DELETE_CHILD mask flag to inode_permission()
when checking for delete access inside a directory, and MAY_DELETE_SELF
when checking for delete access to a file itelf.

The MAY_DELETE_SELF permission does not override the sticky directory
check.  It probably should.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/namei.c         | 15 +++++++++++----
 include/linux/fs.h |  2 ++
 2 files changed, 13 insertions(+), 4 deletions(-)

Comments

J. Bruce Fields Aug. 28, 2015, 8:44 p.m. UTC | #1
On Tue, Aug 04, 2015 at 01:53:01PM +0200, Andreas Gruenbacher wrote:
> Normally, deleting a file requires write and execute access to the parent
> directory.  With Richacls, a process with MAY_DELETE_SELF access to a file
> may delete the file even without write access to the parent directory.
> 
> To support that, pass the MAY_DELETE_CHILD mask flag to inode_permission()
> when checking for delete access inside a directory, and MAY_DELETE_SELF
> when checking for delete access to a file itelf.
> 
> The MAY_DELETE_SELF permission does not override the sticky directory
> check.  It probably should.

I guess it would basically just let the file owner delegate permission
to delete your file to a non-owner?  Makes sense to me to allow that.

--b.

> 
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
> ---
>  fs/namei.c         | 15 +++++++++++----
>  include/linux/fs.h |  2 ++
>  2 files changed, 13 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/namei.c b/fs/namei.c
> index 3504d36..2ac759c 100644
> --- a/fs/namei.c
> +++ b/fs/namei.c
> @@ -454,7 +454,7 @@ static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
>   * changing the "normal" UIDs which are used for other things.
>   *
>   * When checking for MAY_APPEND, MAY_CREATE_FILE, MAY_CREATE_DIR,
> - * MAY_WRITE must also be set in @mask.
> + * MAY_DELETE_CHILD, MAY_DELETE_SELF, MAY_WRITE must also be set in @mask.
>   */
>  int inode_permission(struct inode *inode, int mask)
>  {
> @@ -2527,7 +2527,7 @@ static int may_delete(struct inode *dir, struct dentry *victim,
>  		      bool isdir, bool replace)
>  {
>  	struct inode *inode = d_backing_inode(victim);
> -	int error, mask = MAY_WRITE | MAY_EXEC;
> +	int error, mask = MAY_EXEC;
>  
>  	if (d_is_negative(victim))
>  		return -ENOENT;
> @@ -2537,8 +2537,15 @@ static int may_delete(struct inode *dir, struct dentry *victim,
>  	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
>  
>  	if (replace)
> -		mask |= isdir ? MAY_CREATE_DIR : MAY_CREATE_FILE;
> -	error = inode_permission(dir, mask);
> +		mask |= MAY_WRITE | (isdir ? MAY_CREATE_DIR : MAY_CREATE_FILE);
> +	error = inode_permission(dir, mask | MAY_WRITE | MAY_DELETE_CHILD);
> +	if (error && IS_RICHACL(inode)) {
> +		/* Deleting is also permitted with MAY_EXEC on the directory
> +		 * and MAY_DELETE_SELF on the inode.  */
> +		if (!inode_permission(inode, MAY_DELETE_SELF) &&
> +		    !inode_permission(dir, mask))
> +			error = 0;
> +	}
>  	if (error)
>  		return error;
>  	if (IS_APPEND(dir))
> diff --git a/include/linux/fs.h b/include/linux/fs.h
> index 9c44f27..abf5b0e 100644
> --- a/include/linux/fs.h
> +++ b/include/linux/fs.h
> @@ -83,6 +83,8 @@ typedef void (dax_iodone_t)(struct buffer_head *bh_map, int uptodate);
>  #define MAY_NOT_BLOCK		0x00000080
>  #define MAY_CREATE_FILE		0x00000100
>  #define MAY_CREATE_DIR		0x00000200
> +#define MAY_DELETE_CHILD	0x00000400
> +#define MAY_DELETE_SELF		0x00000800
>  
>  /*
>   * flags in file.f_mode.  Note that FMODE_READ and FMODE_WRITE must correspond
> -- 
> 2.5.0
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andreas Gruenbacher Aug. 28, 2015, 9:08 p.m. UTC | #2
2015-08-28 22:44 GMT+02:00 J. Bruce Fields <bfields@fieldses.org>:
>> The MAY_DELETE_SELF permission does not override the sticky
>> directory check.  It probably should.
>
> I guess it would basically just let the file owner delegate permission
> to delete your file to a non-owner?  Makes sense to me to allow that.

Yes, independent of whether or not the process has MAY_DELETE_CHILD
access on the directory but not independent of the sticky directory
check, which is a bit of a weird combination.

Andreas
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Lutomirski Aug. 28, 2015, 9:36 p.m. UTC | #3
On Tue, Aug 4, 2015 at 4:53 AM, Andreas Gruenbacher
<andreas.gruenbacher@gmail.com> wrote:
> Normally, deleting a file requires write and execute access to the parent
> directory.  With Richacls, a process with MAY_DELETE_SELF access to a file
> may delete the file even without write access to the parent directory.
>
> To support that, pass the MAY_DELETE_CHILD mask flag to inode_permission()
> when checking for delete access inside a directory, and MAY_DELETE_SELF
> when checking for delete access to a file itelf.
>
> The MAY_DELETE_SELF permission does not override the sticky directory
> check.  It probably should.

Silly question from the peanut gallery: is there any such thing as
opening an fd pointing at a file such that the "open file description"
(i.e. the struct file) captures the right to delete the file?

IOW do we need FMODE_DELETE_SELF?

--Andy
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andreas Grünbacher Aug. 28, 2015, 9:54 p.m. UTC | #4
2015-08-28 23:36 GMT+02:00 Andy Lutomirski <luto@amacapital.net>:
> Silly question from the peanut gallery: is there any such thing as
> opening an fd pointing at a file such that the "open file description"
> (i.e. the struct file) captures the right to delete the file?
>
> IOW do we need FMODE_DELETE_SELF?

When would that permission be checked, what syscall would you use to
unlink an open file descriptor?

Thanks,
Andreas
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
J. Bruce Fields Aug. 28, 2015, 9:57 p.m. UTC | #5
On Fri, Aug 28, 2015 at 02:36:15PM -0700, Andy Lutomirski wrote:
> On Tue, Aug 4, 2015 at 4:53 AM, Andreas Gruenbacher
> <andreas.gruenbacher@gmail.com> wrote:
> > Normally, deleting a file requires write and execute access to the parent
> > directory.  With Richacls, a process with MAY_DELETE_SELF access to a file
> > may delete the file even without write access to the parent directory.
> >
> > To support that, pass the MAY_DELETE_CHILD mask flag to inode_permission()
> > when checking for delete access inside a directory, and MAY_DELETE_SELF
> > when checking for delete access to a file itelf.
> >
> > The MAY_DELETE_SELF permission does not override the sticky directory
> > check.  It probably should.
> 
> Silly question from the peanut gallery: is there any such thing as
> opening an fd pointing at a file such that the "open file description"
> (i.e. the struct file) captures the right to delete the file?
> 
> IOW do we need FMODE_DELETE_SELF?

I guess FMODE_READ and _WRITE make sense because we pass file
descriptors to read() and write().  But we don't have a way to pass a
file descriptor to an operation that deletes a file.

(I think Windows may be different in both respects, it might be
interesting to compare, but I really don't understand how it works...).

--b.
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andy Lutomirski Aug. 29, 2015, 1:04 a.m. UTC | #6
On Aug 28, 2015 2:54 PM, "Andreas Grünbacher"
<andreas.gruenbacher@gmail.com> wrote:
>
> 2015-08-28 23:36 GMT+02:00 Andy Lutomirski <luto@amacapital.net>:
> > Silly question from the peanut gallery: is there any such thing as
> > opening an fd pointing at a file such that the "open file description"
> > (i.e. the struct file) captures the right to delete the file?
> >
> > IOW do we need FMODE_DELETE_SELF?
>
> When would that permission be checked, what syscall would you use to
> unlink an open file descriptor?

Good point.  It's remotely plausible that there's some trick with bind
mounts, it's likely possible to unlink a directory by fd (using
unlinkat), and you can *link* a file (with linkat or /proc), but
unlinkat doesn't appear to allow you to unlink a file by fd.

--Andy

>
> Thanks,
> Andreas
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch
diff mbox

diff --git a/fs/namei.c b/fs/namei.c
index 3504d36..2ac759c 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -454,7 +454,7 @@  static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
  * changing the "normal" UIDs which are used for other things.
  *
  * When checking for MAY_APPEND, MAY_CREATE_FILE, MAY_CREATE_DIR,
- * MAY_WRITE must also be set in @mask.
+ * MAY_DELETE_CHILD, MAY_DELETE_SELF, MAY_WRITE must also be set in @mask.
  */
 int inode_permission(struct inode *inode, int mask)
 {
@@ -2527,7 +2527,7 @@  static int may_delete(struct inode *dir, struct dentry *victim,
 		      bool isdir, bool replace)
 {
 	struct inode *inode = d_backing_inode(victim);
-	int error, mask = MAY_WRITE | MAY_EXEC;
+	int error, mask = MAY_EXEC;
 
 	if (d_is_negative(victim))
 		return -ENOENT;
@@ -2537,8 +2537,15 @@  static int may_delete(struct inode *dir, struct dentry *victim,
 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
 
 	if (replace)
-		mask |= isdir ? MAY_CREATE_DIR : MAY_CREATE_FILE;
-	error = inode_permission(dir, mask);
+		mask |= MAY_WRITE | (isdir ? MAY_CREATE_DIR : MAY_CREATE_FILE);
+	error = inode_permission(dir, mask | MAY_WRITE | MAY_DELETE_CHILD);
+	if (error && IS_RICHACL(inode)) {
+		/* Deleting is also permitted with MAY_EXEC on the directory
+		 * and MAY_DELETE_SELF on the inode.  */
+		if (!inode_permission(inode, MAY_DELETE_SELF) &&
+		    !inode_permission(dir, mask))
+			error = 0;
+	}
 	if (error)
 		return error;
 	if (IS_APPEND(dir))
diff --git a/include/linux/fs.h b/include/linux/fs.h
index 9c44f27..abf5b0e 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -83,6 +83,8 @@  typedef void (dax_iodone_t)(struct buffer_head *bh_map, int uptodate);
 #define MAY_NOT_BLOCK		0x00000080
 #define MAY_CREATE_FILE		0x00000100
 #define MAY_CREATE_DIR		0x00000200
+#define MAY_DELETE_CHILD	0x00000400
+#define MAY_DELETE_SELF		0x00000800
 
 /*
  * flags in file.f_mode.  Note that FMODE_READ and FMODE_WRITE must correspond