[PATCH v6 0/2] vfs: fix dedupe permission check

[PATCH v6 0/2] vfs: fix dedupe permission check

[Mark Fasheh]

Hi Andrew/Al,

Could I please have these patches put in a tree for more public testing?
They've hit fsdevel a few times now, I have links to the discussions in the
change log below.


The following patches fix a couple of issues with the permission check
we do in vfs_dedupe_file_range().

The first patch expands our check to allow dedupe of a file if the
user owns it or otherwise would be allowed to write to it.

Current behavior is that we'll allow dedupe only if:

- the user is an admin (root)
- the user has the file open for write

This makes it impossible for a user to dedupe their own file set
unless they do it as root, or ensure that all files have write
permission. There's a couple of duperemove bugs open for this:

https://github.com/markfasheh/duperemove/issues/129
https://github.com/markfasheh/duperemove/issues/86

The other problem we have is also related to forcing the user to open
target files for write - A process trying to exec a file currently
being deduped gets ETXTBUSY. The answer (as above) is to allow them to
open the targets ro - root can already do this. There was a patch from
Adam Borowski to fix this back in 2016:

https://lkml.org/lkml/2016/7/17/130

which I have incorporated into my changes.


The 2nd patch fixes our return code for permission denied to be
EPERM. For some reason we're returning EINVAL - I think that's
probably my fault. At any rate, we need to be returning something
descriptive of the actual problem, otherwise callers see EINVAL and
can't really make a valid determination of what's gone wrong.

This has also popped up in duperemove, mostly in the form of cryptic
error messages. Because this is a code returned to userspace, I did
check the other users of extent-same that I could find. Both 'bees'
and 'rust-btrfs' do the same as duperemove and simply report the error
(as they should).


One way I tested these patches was to make non-root owned files with
read-only permissions and see if I could dedupe them as the owning user. For
example, the following script fails on an unpatched kernel and succeeds with
the patches applied.

  TESTDIR=/btrfs
  USER=mfasheh

  rm -f $TESTDIR/file*

  dd if=/dev/zero of=$TESTDIR/file1 count=1024 bs=1024
  dd if=/dev/zero of=$TESTDIR/file2 count=1024 bs=1024

  chown $USER $TESTDIR/file*
  chmod 444 $TESTDIR/file*

  # open file* for read and dedupe
  sudo -u $USER duperemove -Ad $TESTDIR/file*


Lastly, I have an update to the fi_deduperange manpage to reflect these
changes. That patch is attached below.


git pull https://github.com/markfasheh/linux dedupe-perms

Thanks,
  --Mark

Changes from V5 to V6:
- Rebase and retest on 4.19-rc3
- Add a note on testing

Changes from V4 to V5:
- Rebase and retest on 4.18-rc8
- Place updated manpage patch below, CC linux-api
- V4 discussion: https://patchwork.kernel.org/patch/10530365/

Changes from V3 to V4:
- Add a patch (below) to ioctl_fideduperange.2 explaining our
  changes. I will send this patch once the kernel update is
  accepted. Thanks to Darrick Wong for this suggestion.
- V3 discussion: https://www.spinics.net/lists/linux-btrfs/msg79135.html

Changes from V2 to V3:
- Return bool from allow_file_dedupe
- V2 discussion: https://www.spinics.net/lists/linux-btrfs/msg78421.html

Changes from V1 to V2:
- Add inode_permission check as suggested by Adam Borowski
- V1 discussion: https://marc.info/?l=linux-xfs&m=152606684017965&w=2


From: Mark Fasheh <mfasheh@suse.de>

[PATCH] ioctl_fideduperange.2: clarify permission requirements

dedupe permission checks were recently relaxed - update our man page to
reflect those changes.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
---
 man2/ioctl_fideduperange.2 | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/man2/ioctl_fideduperange.2 b/man2/ioctl_fideduperange.2
index 84d20a276..4040ee064 100644
--- a/man2/ioctl_fideduperange.2
+++ b/man2/ioctl_fideduperange.2
@@ -105,9 +105,12 @@ The field
 must be zero.
 During the call,
 .IR src_fd
-must be open for reading and
+must be open for reading.
 .IR dest_fd
-must be open for writing.
+can be open for writing, or reading.
+If
+.IR dest_fd
+is open for reading, the user must have write access to the file.
 The combined size of the struct
 .IR file_dedupe_range
 and the struct
@@ -185,8 +188,8 @@ This can appear if the filesystem does not support deduplicating either file
 descriptor, or if either file descriptor refers to special inodes.
 .TP
 .B EPERM
-.IR dest_fd
-is immutable.
+This will be returned if the user lacks permission to dedupe the file referenced by
+.IR dest_fd .
 .TP
 .B ETXTBSY
 One of the files is a swap file.
-- 
2.15.1

[PATCH 1/2] vfs: allow dedupe of user owned read-only files

[Mark Fasheh]

The permission check in vfs_dedupe_file_range_one() is too coarse - We only
allow dedupe of the destination file if the user is root, or they have the
file open for write.

This effectively limits a non-root user from deduping their own read-only
files. In addition, the write file descriptor that the user is forced to
hold open can prevent execution of files. As file data during a dedupe
does not change, the behavior is unexpected and this has caused a number of
issue reports. For an example, see:

https://github.com/markfasheh/duperemove/issues/129

So change the check so we allow dedupe on the target if:

- the root or admin is asking for it
- the process has write access
- the owner of the file is asking for the dedupe
- the process could get write access

That way users can open read-only and still get dedupe.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
---
 fs/read_write.c | 16 +++++++++++++++-
 1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/fs/read_write.c b/fs/read_write.c
index 39b4a21dd933..be0e8723a049 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -1964,6 +1964,20 @@ int vfs_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
 }
 EXPORT_SYMBOL(vfs_dedupe_file_range_compare);
 
+/* Check whether we are allowed to dedupe the destination file */
+static bool allow_file_dedupe(struct file *file)
+{
+	if (capable(CAP_SYS_ADMIN))
+		return true;
+	if (file->f_mode & FMODE_WRITE)
+		return true;
+	if (uid_eq(current_fsuid(), file_inode(file)->i_uid))
+		return true;
+	if (!inode_permission(file_inode(file), MAY_WRITE))
+		return true;
+	return false;
+}
+
 int vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
 			      struct file *dst_file, loff_t dst_pos, u64 len)
 {
@@ -1978,7 +1992,7 @@ int vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
 		goto out_drop_write;
 
 	ret = -EINVAL;
-	if (!(capable(CAP_SYS_ADMIN) || (dst_file->f_mode & FMODE_WRITE)))
+	if (!allow_file_dedupe(dst_file))
 		goto out_drop_write;
 
 	ret = -EXDEV;
-- 
2.15.1

[PATCH 2/2] vfs: dedupe should return EPERM if permission is not granted

[Mark Fasheh]

Right now we return EINVAL if a process does not have permission to dedupe a
file. This was an oversight on my part. EPERM gives a true description of
the nature of our error, and EINVAL is already used for the case that the
filesystem does not support dedupe.

Signed-off-by: Mark Fasheh <mfasheh@suse.de>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Acked-by: David Sterba <dsterba@suse.com>
---
 fs/read_write.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/read_write.c b/fs/read_write.c
index be0e8723a049..c734bc2880a5 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -1991,7 +1991,7 @@ int vfs_dedupe_file_range_one(struct file *src_file, loff_t src_pos,
 	if (ret < 0)
 		goto out_drop_write;
 
-	ret = -EINVAL;
+	ret = -EPERM;
 	if (!allow_file_dedupe(dst_file))
 		goto out_drop_write;
 
-- 
2.15.1