Unify migrate_pages and move_pages access checks
authorOtto Ebeling <otto.ebeling@iki.fi>
Thu, 16 Nov 2017 01:38:14 +0000 (17:38 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 16 Nov 2017 02:21:06 +0000 (18:21 -0800)
Commit 197e7e521384 ("Sanitize 'move_pages()' permission checks") fixed
a security issue I reported in the move_pages syscall, and made it so
that you can't act on set-uid processes unless you have the
CAP_SYS_PTRACE capability.

Unify the access check logic of migrate_pages to match the new behavior
of move_pages.  We discussed this a bit in the security@ list and
thought it'd be good for consistency even though there's no evident
security impact.  The NUMA node access checks are left intact and
require CAP_SYS_NICE as before.

Link: http://lkml.kernel.org/r/alpine.DEB.2.11.1710011830320.6333@lakka.kapsi.fi
Signed-off-by: Otto Ebeling <otto.ebeling@iki.fi>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Willy Tarreau <w@1wt.eu>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/mempolicy.c

index a2af6d5..dad166b 100644 (file)
@@ -85,6 +85,7 @@
 #include <linux/interrupt.h>
 #include <linux/init.h>
 #include <linux/compat.h>
+#include <linux/ptrace.h>
 #include <linux/swap.h>
 #include <linux/seq_file.h>
 #include <linux/proc_fs.h>
@@ -1365,7 +1366,6 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
                const unsigned long __user *, old_nodes,
                const unsigned long __user *, new_nodes)
 {
-       const struct cred *cred = current_cred(), *tcred;
        struct mm_struct *mm = NULL;
        struct task_struct *task;
        nodemask_t task_nodes;
@@ -1401,15 +1401,10 @@ SYSCALL_DEFINE4(migrate_pages, pid_t, pid, unsigned long, maxnode,
        err = -EINVAL;
 
        /*
-        * Check if this process has the right to modify the specified
-        * process. The right exists if the process has administrative
-        * capabilities, superuser privileges or the same
-        * userid as the target process.
+        * Check if this process has the right to modify the specified process.
+        * Use the regular "ptrace_may_access()" checks.
         */
-       tcred = __task_cred(task);
-       if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
-           !uid_eq(cred->uid,  tcred->suid) && !uid_eq(cred->uid,  tcred->uid) &&
-           !capable(CAP_SYS_NICE)) {
+       if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
                rcu_read_unlock();
                err = -EPERM;
                goto out_put;