arm64: kasan: Fix zero shadow mapping overriding kernel image shadow
authorCatalin Marinas <catalin.marinas@arm.com>
Thu, 10 Mar 2016 18:41:16 +0000 (18:41 +0000)
committerCatalin Marinas <catalin.marinas@arm.com>
Fri, 11 Mar 2016 11:03:35 +0000 (11:03 +0000)
With the 16KB and 64KB page size configurations, SWAPPER_BLOCK_SIZE is
PAGE_SIZE and ARM64_SWAPPER_USES_SECTION_MAPS is 0. Since
kimg_shadow_end is not page aligned (_end shifted by
KASAN_SHADOW_SCALE_SHIFT), the edges of previously mapped kernel image
shadow via vmemmap_populate() may be overridden by subsequent calls to
kasan_populate_zero_shadow(), leading to kernel panics like below:

------------------------------------------------------------------------------
Unable to handle kernel paging request at virtual address fffffc100135068c
pgd = fffffc8009ac0000
[fffffc100135068c] *pgd=00000009ffee0003, *pud=00000009ffee0003, *pmd=00000009ffee0003, *pte=00e0000081a00793
Internal error: Oops: 9600004f [#1] PREEMPT SMP
Modules linked in:
CPU: 0 PID: 1 Comm: swapper/0 Not tainted 4.5.0-rc4+ #1984
Hardware name: Juno (DT)
task: fffffe09001a0000 ti: fffffe0900200000 task.ti: fffffe0900200000
PC is at __memset+0x4c/0x200
LR is at kasan_unpoison_shadow+0x34/0x50
pc : [<fffffc800846f1cc>] lr : [<fffffc800821ff54>] pstate: 00000245
sp : fffffe0900203db0
x29: fffffe0900203db0 x28: 0000000000000000
x27: 0000000000000000 x26: 0000000000000000
x25: fffffc80099b69d0 x24: 0000000000000001
x23: 0000000000000000 x22: 0000000000002000
x21: dffffc8000000000 x20: 1fffff9001350a8c
x19: 0000000000002000 x18: 0000000000000008
x17: 0000000000000147 x16: ffffffffffffffff
x15: 79746972100e041d x14: ffffff0000000000
x13: ffff000000000000 x12: 0000000000000000
x11: 0101010101010101 x10: 1fffffc11c000000
x9 : 0000000000000000 x8 : fffffc100135068c
x7 : 0000000000000000 x6 : 000000000000003f
x5 : 0000000000000040 x4 : 0000000000000004
x3 : fffffc100134f651 x2 : 0000000000000400
x1 : 0000000000000000 x0 : fffffc100135068c

Process swapper/0 (pid: 1, stack limit = 0xfffffe0900200020)
Call trace:
[<fffffc800846f1cc>] __memset+0x4c/0x200
[<fffffc8008220044>] __asan_register_globals+0x5c/0xb0
[<fffffc8008a09d34>] _GLOBAL__sub_I_65535_1_sunrpc_cache_lookup+0x1c/0x28
[<fffffc8008f20d28>] kernel_init_freeable+0x104/0x274
[<fffffc80089e1948>] kernel_init+0x10/0xf8
[<fffffc8008093a00>] ret_from_fork+0x10/0x50
------------------------------------------------------------------------------

This patch aligns kimg_shadow_start and kimg_shadow_end to
SWAPPER_BLOCK_SIZE in all configurations.

Fixes: f9040773b7bb ("arm64: move kernel image to base of vmalloc area")
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
arch/arm64/mm/kasan_init.c

index a164183..757009d 100644 (file)
@@ -159,15 +159,12 @@ void __init kasan_init(void)
         * vmemmap_populate() has populated the shadow region that covers the
         * kernel image with SWAPPER_BLOCK_SIZE mappings, so we have to round
         * the start and end addresses to SWAPPER_BLOCK_SIZE as well, to prevent
-        * kasan_populate_zero_shadow() from replacing the PMD block mappings
-        * with PMD table mappings at the edges of the shadow region for the
-        * kernel image.
+        * kasan_populate_zero_shadow() from replacing the page table entries
+        * (PMD or PTE) at the edges of the shadow region for the kernel
+        * image.
         */
-       if (ARM64_SWAPPER_USES_SECTION_MAPS) {
-               kimg_shadow_start = round_down(kimg_shadow_start,
-                                              SWAPPER_BLOCK_SIZE);
-               kimg_shadow_end = round_up(kimg_shadow_end, SWAPPER_BLOCK_SIZE);
-       }
+       kimg_shadow_start = round_down(kimg_shadow_start, SWAPPER_BLOCK_SIZE);
+       kimg_shadow_end = round_up(kimg_shadow_end, SWAPPER_BLOCK_SIZE);
 
        kasan_populate_zero_shadow((void *)KASAN_SHADOW_START,
                                   (void *)mod_shadow_start);