Re: [RFC PATCH v2 1/5] fs: add O_ENCODED open flag

On Sat, Oct 19, 2019 at 03:50:57PM +1100, Aleksa Sarai wrote:

On 2019-10-15, Omar Sandoval [off-list ref] wrote:

quoted

From: Omar Sandoval <redacted>

The upcoming RWF_ENCODED operation introduces some security concerns:

1. Compressed writes will pass arbitrary data to decompression
   algorithms in the kernel.
2. Compressed reads can leak truncated/hole punched data.

Therefore, we need to require privilege for RWF_ENCODED. It's not
possible to do the permissions checks at the time of the read or write
because, e.g., io_uring submits IO from a worker thread. So, add an open
flag which requires CAP_SYS_ADMIN. It can also be set and cleared with
fcntl(). The flag is not cleared in any way on fork or exec; it should
probably be used with O_CLOEXEC in most cases.

Note that the usual issue that unknown open flags are ignored doesn't
really matter for O_ENCODED; if the kernel doesn't support O_ENCODED,
then it doesn't support RWF_ENCODED, either.

Signed-off-by: Omar Sandoval <redacted>
---
 fs/fcntl.c                       | 10 ++++++++--
 fs/namei.c                       |  4 ++++
 include/linux/fcntl.h            |  2 +-
 include/uapi/asm-generic/fcntl.h |  4 ++++
 4 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/fs/fcntl.c b/fs/fcntl.c
index 3d40771e8e7c..45ebc6df078e 100644
--- a/fs/fcntl.c
+++ b/fs/fcntl.c

@@ -30,7 +30,8 @@
 #include <asm/siginfo.h>
 #include <linux/uaccess.h>
 
-#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
+#define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME | \
+		    O_ENCODED)
 
 static int setfl(int fd, struct file * filp, unsigned long arg)
 {

@@ -49,6 +50,11 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 		if (!inode_owner_or_capable(inode))
 			return -EPERM;
 
+	/* O_ENCODED can only be set by superuser */
+	if ((arg & O_ENCODED) && !(filp->f_flags & O_ENCODED) &&
+	    !capable(CAP_SYS_ADMIN))
+		return -EPERM;

I have a feeling the error should probably be an EACCES and not EPERM.

Shrug, I wanted to make this consistent with O_NOATIME, which uses
EPERM. EACCES seems more appropriate for lacking permissions for a
particular path rather than for an operation, but the lines are blurry.

quoted

+
 	/* required for strict SunOS emulation */
 	if (O_NONBLOCK != O_NDELAY)
 	       if (arg & O_NDELAY)

@@ -1031,7 +1037,7 @@ static int __init fcntl_init(void)
 	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
 	 * is defined as O_NONBLOCK on some platforms and not on others.
 	 */
-	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
+	BUILD_BUG_ON(22 - 1 /* for O_RDONLY being 0 */ !=
 		HWEIGHT32(
 			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
 			__FMODE_EXEC | __FMODE_NONOTIFY));

diff --git a/fs/namei.c b/fs/namei.c
index 671c3c1a3425..ae86b125888a 100644
--- a/fs/namei.c
+++ b/fs/namei.c

@@ -2978,6 +2978,10 @@ static int may_open(const struct path *path, int acc_mode, int flag)
 	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
 		return -EPERM;
 
+	/* O_ENCODED can only be set by superuser */
+	if ((flag & O_ENCODED) && !capable(CAP_SYS_ADMIN))
+		return -EPERM;

I would suggest that this check be put into build_open_flags() rather
than putting it this late in open(). Also, same nit about the error
return as above.

This is where we check permissions for O_NOATIME, shouldn't we keep all
of those permission checks in the same place? build_open_flags() only
checks for flag validity.

quoted

+
 	return 0;
 }
 

diff --git a/include/linux/fcntl.h b/include/linux/fcntl.h
index d019df946cb2..5fac02479639 100644
--- a/include/linux/fcntl.h
+++ b/include/linux/fcntl.h

@@ -9,7 +9,7 @@
 	(O_RDONLY | O_WRONLY | O_RDWR | O_CREAT | O_EXCL | O_NOCTTY | O_TRUNC | \
 	 O_APPEND | O_NDELAY | O_NONBLOCK | O_NDELAY | __O_SYNC | O_DSYNC | \
 	 FASYNC	| O_DIRECT | O_LARGEFILE | O_DIRECTORY | O_NOFOLLOW | \
-	 O_NOATIME | O_CLOEXEC | O_PATH | __O_TMPFILE)
+	 O_NOATIME | O_CLOEXEC | O_PATH | __O_TMPFILE | O_ENCODED)
 
 #ifndef force_o_largefile
 #define force_o_largefile() (!IS_ENABLED(CONFIG_ARCH_32BIT_OFF_T))

diff --git a/include/uapi/asm-generic/fcntl.h b/include/uapi/asm-generic/fcntl.h
index 9dc0bf0c5a6e..8c5cbd5942e3 100644
--- a/include/uapi/asm-generic/fcntl.h
+++ b/include/uapi/asm-generic/fcntl.h

@@ -97,6 +97,10 @@
 #define O_NDELAY	O_NONBLOCK
 #endif
 
+#ifndef O_ENCODED
+#define O_ENCODED	040000000
+#endif

You should also define this for all of the architectures which don't use
the generic O_* flag values. On alpha, O_PATH is equal to the value you
picked (just be careful on sparc -- 0x4000000 is the next free bit, but
it's used by FMODE_NONOTIFY.)

Good catch, I'll fix that. Thanks!