resource: fix the case of null pointer access
authorVivek Goyal <vgoyal@redhat.com>
Fri, 29 Aug 2014 22:18:29 +0000 (15:18 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 29 Aug 2014 23:28:15 +0000 (16:28 -0700)
Richard and Daniel reported that UML is broken due to changes to
resource traversal functions.  Problem is that iomem_resource.child can
be null and new code does not consider that possibility.  Old code used
a for loop and that loop will not even execute if p was null.

Revert back to for() loop logic and bail out if p is null.

I also moved sibling_only check out of resource_lock. There is no
reason to keep it inside the lock.

Following is backtrace of the UML crash.

RIP: 0033:[<0000000060039b9f>]
RSP: 0000000081459da0  EFLAGS: 00010202
RAX: 0000000000000000 RBX: 00000000219b3fff RCX: 000000006010d1d9
RDX: 0000000000000001 RSI: 00000000602dfb94 RDI: 0000000081459df8
RBP: 0000000081459de0 R08: 00000000601b59f4 R09: ffffffff0000ff00
R10: ffffffff0000ff00 R11: 0000000081459e88 R12: 0000000081459df8
R13: 00000000219b3fff R14: 00000000602dfb94 R15: 0000000000000000
Kernel panic - not syncing: Segfault with no mm
CPU: 0 PID: 1 Comm: swapper Not tainted 3.16.0-10454-g58d08e3 #13
Stack:
 00000000 000080d0 81459df0 219b3fff
 81459e70 6010d1d9 ffffffff 6033e010
 81459e50 6003a269 81459e30 00000000
Call Trace:
 [<6010d1d9>] ? kclist_add_private+0x0/0xe7
 [<6003a269>] walk_system_ram_range+0x61/0xb7
 [<6000e859>] ? proc_kcore_init+0x0/0xf1
 [<6010d574>] kcore_update_ram+0x4c/0x168
 [<6010d72e>] ? kclist_add+0x0/0x2e
 [<6000e943>] proc_kcore_init+0xea/0xf1
 [<6000e859>] ? proc_kcore_init+0x0/0xf1
 [<6000e859>] ? proc_kcore_init+0x0/0xf1
 [<600189f0>] do_one_initcall+0x13c/0x204
 [<6004ca46>] ? parse_args+0x1df/0x2e0
 [<6004c82d>] ? parameq+0x0/0x3a
 [<601b5990>] ? strcpy+0x0/0x18
 [<60001e1a>] kernel_init_freeable+0x240/0x31e
 [<6026f1c0>] kernel_init+0x12/0x148
 [<60019fad>] new_thread_handler+0x81/0xa3

Fixes 8c86e70acead629aacb4a ("resource: provide new functions to walk
through resources").

Reported-by: Daniel Walter <sahne@0x90.at>
Tested-by: Richard Weinberger <richard@nod.at>
Tested-by: Toralf Förster <toralf.foerster@gmx.de>
Tested-by: Daniel Walter <sahne@0x90.at>
Signed-off-by: Vivek Goyal <vgoyal@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/resource.c

index da14b8d..60c5a38 100644 (file)
@@ -351,15 +351,12 @@ static int find_next_iomem_res(struct resource *res, char *name,
        end = res->end;
        BUG_ON(start >= end);
 
-       read_lock(&resource_lock);
-
-       if (first_level_children_only) {
-               p = iomem_resource.child;
+       if (first_level_children_only)
                sibling_only = true;
-       } else
-               p = &iomem_resource;
 
-       while ((p = next_resource(p, sibling_only))) {
+       read_lock(&resource_lock);
+
+       for (p = iomem_resource.child; p; p = next_resource(p, sibling_only)) {
                if (p->flags != res->flags)
                        continue;
                if (name && strcmp(p->name, name))