CRED: Fix regression in cap_capable() as shown up by sys_faccessat() [ver #2]

Fix a regression in cap_capable() due to:

	commit 5ff7711e635b32f0a1e558227d030c7e45b4a465
	Author: David Howells <dhowells@redhat.com>
	Date:   Wed Dec 31 02:52:28 2008 +0000

	    CRED: Differentiate objective and effective subjective credentials on a task

The problem is that the above patch allows a process to have two sets of
credentials, and for the most part uses the subjective credentials when
accessing current's creds.

There is, however, one exception: cap_capable(), and thus capable(), uses the
real/objective credentials of the target task, whether or not it is the current
task.

Ordinarily this doesn't matter, since usually the two cred pointers in current
point to the same set of creds.  However, sys_faccessat() makes use of this
facility to override the credentials of the calling process to make its test,
without affecting the creds as seen from other processes.

One of the things sys_faccessat() does is to make an adjustment to the
effective capabilities mask, which cap_capable(), as it stands, then ignores.

The affected capability check is in generic_permission():

	if (!(mask & MAY_EXEC) || execute_ok(inode))
		if (capable(CAP_DAC_OVERRIDE))
			return 0;

This change splits capable() from has_capability() down into the commoncap and
SELinux code.  The capable() security op now only deals with the current
process, and uses the current process's subjective creds.  A new security op -
task_capable() - is introduced that can check any task's objective creds.

strictly the capable() security op is superfluous with the presence of the
task_capable() op, however it should be faster to call the capable() op since
two fewer arguments need be passed down through the various layers.

This can be tested by compiling the following program from the XFS testsuite:

/*
 *  t_access_root.c - trivial test program to show permission bug.
 *
 *  Written by Michael Kerrisk - copyright ownership not pursued.
 *  Sourced from: http://linux.derkeiler.com/Mailing-Lists/Kernel/2003-10/6030.html
 */
#include <limits.h>
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/stat.h>

#define UID 500
#define GID 100
#define PERM 0
#define TESTPATH "/tmp/t_access"

static void
errExit(char *msg)
{
    perror(msg);
    exit(EXIT_FAILURE);
} /* errExit */

static void
accessTest(char *file, int mask, char *mstr)
{
    printf("access(%s, %s) returns %d\n", file, mstr, access(file, mask));
} /* accessTest */

int
main(int argc, char *argv[])
{
    int fd, perm, uid, gid;
    char *testpath;
    char cmd[PATH_MAX + 20];

    testpath = (argc > 1) ? argv[1] : TESTPATH;
    perm = (argc > 2) ? strtoul(argv[2], NULL, 8) : PERM;
    uid = (argc > 3) ? atoi(argv[3]) : UID;
    gid = (argc > 4) ? atoi(argv[4]) : GID;

    unlink(testpath);

    fd = open(testpath, O_RDWR | O_CREAT, 0);
    if (fd == -1) errExit("open");

    if (fchown(fd, uid, gid) == -1) errExit("fchown");
    if (fchmod(fd, perm) == -1) errExit("fchmod");
    close(fd);

    snprintf(cmd, sizeof(cmd), "ls -l %s", testpath);
    system(cmd);

    if (seteuid(uid) == -1) errExit("seteuid");

    accessTest(testpath, 0, "0");
    accessTest(testpath, R_OK, "R_OK");
    accessTest(testpath, W_OK, "W_OK");
    accessTest(testpath, X_OK, "X_OK");
    accessTest(testpath, R_OK | W_OK, "R_OK | W_OK");
    accessTest(testpath, R_OK | X_OK, "R_OK | X_OK");
    accessTest(testpath, W_OK | X_OK, "W_OK | X_OK");
    accessTest(testpath, R_OK | W_OK | X_OK, "R_OK | W_OK | X_OK");

    exit(EXIT_SUCCESS);
} /* main */

This can be run against an Ext3 filesystem as well as against an XFS
filesystem.  If successful, it will show:

	[root@andromeda src]# ./t_access_root /tmp/xxx 0 4043 4043
	---------- 1 dhowells dhowells 0 2008-12-31 03:00 /tmp/xxx
	access(/tmp/xxx, 0) returns 0
	access(/tmp/xxx, R_OK) returns 0
	access(/tmp/xxx, W_OK) returns 0
	access(/tmp/xxx, X_OK) returns -1
	access(/tmp/xxx, R_OK | W_OK) returns 0
	access(/tmp/xxx, R_OK | X_OK) returns -1
	access(/tmp/xxx, W_OK | X_OK) returns -1
	access(/tmp/xxx, R_OK | W_OK | X_OK) returns -1

If unsuccessful, it will show:

	[root@andromeda src]# ./t_access_root /tmp/xxx 0 4043 4043
	---------- 1 dhowells dhowells 0 2008-12-31 02:56 /tmp/xxx
	access(/tmp/xxx, 0) returns 0
	access(/tmp/xxx, R_OK) returns -1
	access(/tmp/xxx, W_OK) returns -1
	access(/tmp/xxx, X_OK) returns -1
	access(/tmp/xxx, R_OK | W_OK) returns -1
	access(/tmp/xxx, R_OK | X_OK) returns -1
	access(/tmp/xxx, W_OK | X_OK) returns -1
	access(/tmp/xxx, R_OK | W_OK | X_OK) returns -1

I've also tested the fix with the SELinux and syscalls LTP testsuites.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: James Morris <jmorris@namei.org>

1 files changed, 29 insertions, 13 deletions

diff --git a/security/commoncap.c b/security/commoncap.c
index 79713545cd63..7f0b2a68717d 100644
--- a/security/commoncap.c
+++ b/security/commoncap.c
@@ -43,28 +43,44 @@ int cap_netlink_recv(struct sk_buff *skb, int cap)
 EXPORT_SYMBOL(cap_netlink_recv);
 
 /**
- * cap_capable - Determine whether a task has a particular effective capability
- * @tsk: The task to query
+ * cap_capable - Determine whether current has a particular effective capability
  * @cap: The capability to check for
  * @audit: Whether to write an audit message or not
  *
  * Determine whether the nominated task has the specified capability amongst
- * its effective set, returning 0 if it does, -ve if it does not.
+ * its effective set, returning 0 if it does, -ve if it does not.  Note that
+ * this uses current's subjective/effective credentials.
  *
  * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
  * function.  That is, it has the reverse semantics: cap_capable() returns 0
  * when a task has a capability, but the kernel's capable() returns 1 for this
  * case.
  */
-int cap_capable(struct task_struct *tsk, int cap, int audit)
+int cap_capable(int cap, int audit)
 {
-        __u32 cap_raised;
+        return cap_raised(current_cap(), cap) ? 0 : -EPERM;
+}
 
-        /* Derived from include/linux/sched.h:capable. */
-        rcu_read_lock();
-        cap_raised = cap_raised(__task_cred(tsk)->cap_effective, cap);
-        rcu_read_unlock();
-        return cap_raised ? 0 : -EPERM;
+/**
+ * cap_has_capability - Determine whether a task has a particular effective capability
+ * @tsk: The task to query
+ * @cred: The credentials to use
+ * @cap: The capability to check for
+ * @audit: Whether to write an audit message or not
+ *
+ * Determine whether the nominated task has the specified capability amongst
+ * its effective set, returning 0 if it does, -ve if it does not.  Note that
+ * this uses the task's objective/real credentials.
+ *
+ * NOTE WELL: cap_has_capability() cannot be used like the kernel's
+ * has_capability() function.  That is, it has the reverse semantics:
+ * cap_has_capability() returns 0 when a task has a capability, but the
+ * kernel's has_capability() returns 1 for this case.
+ */
+int cap_task_capable(struct task_struct *tsk, const struct cred *cred, int cap,
+                     int audit)
+{
+        return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM;
 }
 
 /**
@@ -160,7 +176,7 @@ static inline int cap_inh_is_capped(void)
         /* they are so limited unless the current task has the CAP_SETPCAP
          * capability
          */
-        if (cap_capable(current, CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
+        if (cap_capable(CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0)
                 return 0;
 #endif
         return 1;
@@ -869,7 +885,7 @@ int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
                      & (new->securebits ^ arg2))                        /*[1]*/
                     || ((new->securebits & SECURE_ALL_LOCKS & ~arg2))   /*[2]*/
                     || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS))   /*[3]*/
-                    || (cap_capable(current, CAP_SETPCAP, SECURITY_CAP_AUDIT) != 0) /*[4]*/
+                    || (cap_capable(CAP_SETPCAP, SECURITY_CAP_AUDIT) != 0) /*[4]*/
                         /*
                          * [1] no changing of bits that are locked
                          * [2] no unlocking of locks
@@ -950,7 +966,7 @@ int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 {
         int cap_sys_admin = 0;
 
-        if (cap_capable(current, CAP_SYS_ADMIN, SECURITY_CAP_NOAUDIT) == 0)
+        if (cap_capable(CAP_SYS_ADMIN, SECURITY_CAP_NOAUDIT) == 0)
                 cap_sys_admin = 1;
         return __vm_enough_memory(mm, pages, cap_sys_admin);
 }