vsprintf: refactor %pK code out of pointer()
authorTobin C. Harding <me@tobin.cc>
Wed, 22 Nov 2017 23:56:39 +0000 (10:56 +1100)
committerTobin C. Harding <me@tobin.cc>
Wed, 29 Nov 2017 01:03:24 +0000 (12:03 +1100)
Currently code to handle %pK is all within the switch statement in
pointer(). This is the wrong level of abstraction. Each of the other switch
clauses call a helper function, pK should do the same.

Refactor code out of pointer() to new function restricted_pointer().

Signed-off-by: Tobin C. Harding <me@tobin.cc>
lib/vsprintf.c

index 1746bae..8dc5cf8 100644 (file)
@@ -1343,6 +1343,59 @@ char *uuid_string(char *buf, char *end, const u8 *addr,
        return string(buf, end, uuid, spec);
 }
 
+int kptr_restrict __read_mostly;
+
+static noinline_for_stack
+char *restricted_pointer(char *buf, char *end, const void *ptr,
+                        struct printf_spec spec)
+{
+       spec.base = 16;
+       spec.flags |= SMALL;
+       if (spec.field_width == -1) {
+               spec.field_width = 2 * sizeof(ptr);
+               spec.flags |= ZEROPAD;
+       }
+
+       switch (kptr_restrict) {
+       case 0:
+               /* Always print %pK values */
+               break;
+       case 1: {
+               const struct cred *cred;
+
+               /*
+                * kptr_restrict==1 cannot be used in IRQ context
+                * because its test for CAP_SYSLOG would be meaningless.
+                */
+               if (in_irq() || in_serving_softirq() || in_nmi())
+                       return string(buf, end, "pK-error", spec);
+
+               /*
+                * Only print the real pointer value if the current
+                * process has CAP_SYSLOG and is running with the
+                * same credentials it started with. This is because
+                * access to files is checked at open() time, but %pK
+                * checks permission at read() time. We don't want to
+                * leak pointer values if a binary opens a file using
+                * %pK and then elevates privileges before reading it.
+                */
+               cred = current_cred();
+               if (!has_capability_noaudit(current, CAP_SYSLOG) ||
+                   !uid_eq(cred->euid, cred->uid) ||
+                   !gid_eq(cred->egid, cred->gid))
+                       ptr = NULL;
+               break;
+       }
+       case 2:
+       default:
+               /* Always print 0's for %pK */
+               ptr = NULL;
+               break;
+       }
+
+       return number(buf, end, (unsigned long)ptr, spec);
+}
+
 static noinline_for_stack
 char *netdev_bits(char *buf, char *end, const void *addr, const char *fmt)
 {
@@ -1591,8 +1644,6 @@ char *device_node_string(char *buf, char *end, struct device_node *dn,
        return widen_string(buf, buf - buf_start, end, spec);
 }
 
-int kptr_restrict __read_mostly;
-
 /*
  * Show a '%p' thing.  A kernel extension is that the '%p' is followed
  * by an extra set of alphanumeric characters that are extended format
@@ -1792,47 +1843,7 @@ char *pointer(const char *fmt, char *buf, char *end, void *ptr,
                        return buf;
                }
        case 'K':
-               switch (kptr_restrict) {
-               case 0:
-                       /* Always print %pK values */
-                       break;
-               case 1: {
-                       const struct cred *cred;
-
-                       /*
-                        * kptr_restrict==1 cannot be used in IRQ context
-                        * because its test for CAP_SYSLOG would be meaningless.
-                        */
-                       if (in_irq() || in_serving_softirq() || in_nmi()) {
-                               if (spec.field_width == -1)
-                                       spec.field_width = default_width;
-                               return string(buf, end, "pK-error", spec);
-                       }
-
-                       /*
-                        * Only print the real pointer value if the current
-                        * process has CAP_SYSLOG and is running with the
-                        * same credentials it started with. This is because
-                        * access to files is checked at open() time, but %pK
-                        * checks permission at read() time. We don't want to
-                        * leak pointer values if a binary opens a file using
-                        * %pK and then elevates privileges before reading it.
-                        */
-                       cred = current_cred();
-                       if (!has_capability_noaudit(current, CAP_SYSLOG) ||
-                           !uid_eq(cred->euid, cred->uid) ||
-                           !gid_eq(cred->egid, cred->gid))
-                               ptr = NULL;
-                       break;
-               }
-               case 2:
-               default:
-                       /* Always print 0's for %pK */
-                       ptr = NULL;
-                       break;
-               }
-               break;
-
+               return restricted_pointer(buf, end, ptr, spec);
        case 'N':
                return netdev_bits(buf, end, ptr, fmt);
        case 'a':