x86/mm: Implement ASLR for kernel memory regions

Randomizes the virtual address space of kernel memory regions for
x86_64. This first patch adds the infrastructure and does not randomize
any region. The following patches will randomize the physical memory
mapping, vmalloc and vmemmap regions.

This security feature mitigates exploits relying on predictable kernel
addresses. These addresses can be used to disclose the kernel modules
base addresses or corrupt specific structures to elevate privileges
bypassing the current implementation of KASLR. This feature can be
enabled with the CONFIG_RANDOMIZE_MEMORY option.

The order of each memory region is not changed. The feature looks at the
available space for the regions based on different configuration options
and randomizes the base and space between each. The size of the physical
memory mapping is the available physical memory. No performance impact
was detected while testing the feature.

Entropy is generated using the KASLR early boot functions now shared in
the lib directory (originally written by Kees Cook). Randomization is
done on PGD & PUD page table levels to increase possible addresses. The
physical memory mapping code was adapted to support PUD level virtual
addresses. This implementation on the best configuration provides 30,000
possible virtual addresses in average for each memory region.  An
additional low memory page is used to ensure each CPU can start with a
PGD aligned virtual address (for realmode).

x86/dump_pagetable was updated to correctly display each region.

Updated documentation on x86_64 memory layout accordingly.

Performance data, after all patches in the series:

Kernbench shows almost no difference (-+ less than 1%):

Before:

Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.63 (1.2695)
User Time 1034.89 (1.18115) System Time 87.056 (0.456416) Percent CPU 1092.9
(13.892) Context Switches 199805 (3455.33) Sleeps 97907.8 (900.636)

After:

Average Optimal load -j 12 Run (std deviation): Elapsed Time 102.489 (1.10636)
User Time 1034.86 (1.36053) System Time 87.764 (0.49345) Percent CPU 1095
(12.7715) Context Switches 199036 (4298.1) Sleeps 97681.6 (1031.11)

Hackbench shows 0% difference on average (hackbench 90 repeated 10 times):

attemp,before,after 1,0.076,0.069 2,0.072,0.069 3,0.066,0.066 4,0.066,0.068
5,0.066,0.067 6,0.066,0.069 7,0.067,0.066 8,0.063,0.067 9,0.067,0.065
10,0.068,0.071 average,0.0677,0.0677

Signed-off-by: Thomas Garnier <[email protected]>
Signed-off-by: Kees Cook <[email protected]>
Cc: Alexander Kuleshov <[email protected]>
Cc: Alexander Popov <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Andy Lutomirski <[email protected]>
Cc: Aneesh Kumar K.V <[email protected]>
Cc: Baoquan He <[email protected]>
Cc: Boris Ostrovsky <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Borislav Petkov <[email protected]>
Cc: Brian Gerst <[email protected]>
Cc: Christian Borntraeger <[email protected]>
Cc: Dan Williams <[email protected]>
Cc: Dave Hansen <[email protected]>
Cc: Dave Young <[email protected]>
Cc: Denys Vlasenko <[email protected]>
Cc: Dmitry Vyukov <[email protected]>
Cc: H. Peter Anvin <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Joerg Roedel <[email protected]>
Cc: Jonathan Corbet <[email protected]>
Cc: Josh Poimboeuf <[email protected]>
Cc: Juergen Gross <[email protected]>
Cc: Kirill A. Shutemov <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Lv Zheng <[email protected]>
Cc: Mark Salter <[email protected]>
Cc: Martin Schwidefsky <[email protected]>
Cc: Matt Fleming <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Stephen Smalley <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Toshi Kani <[email protected]>
Cc: Xiao Guangrong <[email protected]>
Cc: Yinghai Lu <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

Author: thgarnie

Documentation/x86/x86_64/mm.txt

@@ -39,4 +39,8 @@ memory window (this size is arbitrary, it can be raised later if needed).
 The mappings are not part of any other kernel PGD and are only available
 during EFI runtime calls.
 
+Note that if CONFIG_RANDOMIZE_MEMORY is enabled, the direct mapping of all
+physical memory, vmalloc/ioremap space and virtual memory map are randomized.
+Their order is preserved but their base will be offset early at boot time.
+
 -Andi Kleen, Jul 2004

arch/x86/Kconfig

@@ -1993,6 +1993,23 @@ config PHYSICAL_ALIGN
 
 	  Don't change this unless you know what you are doing.
 
+config RANDOMIZE_MEMORY
+	bool "Randomize the kernel memory sections"
+	depends on X86_64
+	depends on RANDOMIZE_BASE
+	default RANDOMIZE_BASE
+	---help---
+	   Randomizes the base virtual address of kernel memory sections
+	   (physical memory mapping, vmalloc & vmemmap). This security feature
+	   makes exploits relying on predictable memory locations less reliable.
+
+	   The order of allocations remains unchanged. Entropy is generated in
+	   the same way as RANDOMIZE_BASE. Current implementation in the optimal
+	   configuration have in average 30,000 different possible virtual
+	   addresses for each memory section.
+
+	   If unsure, say N.
+
 config HOTPLUG_CPU
 	bool "Support for hot-pluggable CPUs"
 	depends on SMP

arch/x86/include/asm/kaslr.h

@@ -3,4 +3,10 @@
 
 unsigned long kaslr_get_random_long(const char *purpose);
 
+#ifdef CONFIG_RANDOMIZE_MEMORY
+void kernel_randomize_memory(void);
+#else
+static inline void kernel_randomize_memory(void) { }
+#endif /* CONFIG_RANDOMIZE_MEMORY */
+
 #endif

arch/x86/include/asm/pgtable.h

@@ -732,11 +732,16 @@ void early_alloc_pgt_buf(void);
 #ifdef CONFIG_X86_64
 /* Realmode trampoline initialization. */
 extern pgd_t trampoline_pgd_entry;
-static inline void __meminit init_trampoline(void)
+static inline void __meminit init_trampoline_default(void)
 {
 	/* Default trampoline pgd value */
 	trampoline_pgd_entry = init_level4_pgt[pgd_index(__PAGE_OFFSET)];
 }
+# ifdef CONFIG_RANDOMIZE_MEMORY
+void __meminit init_trampoline(void);
+# else
+#  define init_trampoline init_trampoline_default
+# endif
 #else
 static inline void init_trampoline(void) { }
 #endif

arch/x86/kernel/setup.c

@@ -113,6 +113,7 @@
 #include <asm/prom.h>
 #include <asm/microcode.h>
 #include <asm/mmu_context.h>
+#include <asm/kaslr.h>
 
 /*
  * max_low_pfn_mapped: highest direct mapped pfn under 4GB
@@ -942,6 +943,8 @@ void __init setup_arch(char **cmdline_p)
 
 	x86_init.oem.arch_setup();
 
+	kernel_randomize_memory();
+
 	iomem_resource.end = (1ULL << boot_cpu_data.x86_phys_bits) - 1;
 	setup_memory_map();
 	parse_setup_data();

arch/x86/mm/Makefile

@@ -37,4 +37,5 @@ obj-$(CONFIG_NUMA_EMU)		+= numa_emulation.o
 
 obj-$(CONFIG_X86_INTEL_MPX)	+= mpx.o
 obj-$(CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS) += pkeys.o
+obj-$(CONFIG_RANDOMIZE_MEMORY) += kaslr.o
 

arch/x86/mm/dump_pagetables.c

@@ -72,9 +72,9 @@ static struct addr_marker address_markers[] = {
 	{ 0, "User Space" },
 #ifdef CONFIG_X86_64
 	{ 0x8000000000000000UL, "Kernel Space" },
-	{ PAGE_OFFSET,		"Low Kernel Mapping" },
-	{ VMALLOC_START,        "vmalloc() Area" },
-	{ VMEMMAP_START,        "Vmemmap" },
+	{ 0/* PAGE_OFFSET */,   "Low Kernel Mapping" },
+	{ 0/* VMALLOC_START */, "vmalloc() Area" },
+	{ 0/* VMEMMAP_START */, "Vmemmap" },
 # ifdef CONFIG_X86_ESPFIX64
 	{ ESPFIX_BASE_ADDR,	"ESPfix Area", 16 },
 # endif
@@ -434,8 +434,16 @@ void ptdump_walk_pgd_level_checkwx(void)
 
 static int __init pt_dump_init(void)
 {
+	/*
+	 * Various markers are not compile-time constants, so assign them
+	 * here.
+	 */
+#ifdef CONFIG_X86_64
+	address_markers[LOW_KERNEL_NR].start_address = PAGE_OFFSET;
+	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
+	address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
+#endif
 #ifdef CONFIG_X86_32
-	/* Not a compile-time constant on x86-32 */
 	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
 	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
 # ifdef CONFIG_HIGHMEM

arch/x86/mm/init.c

@@ -17,6 +17,7 @@
 #include <asm/proto.h>
 #include <asm/dma.h>		/* for MAX_DMA_PFN */
 #include <asm/microcode.h>
+#include <asm/kaslr.h>
 
 /*
  * We need to define the tracepoints somewhere, and tlb.c

arch/x86/mm/kaslr.c

@@ -0,0 +1,152 @@
+/*
+ * This file implements KASLR memory randomization for x86_64. It randomizes
+ * the virtual address space of kernel memory regions (physical memory
+ * mapping, vmalloc & vmemmap) for x86_64. This security feature mitigates
+ * exploits relying on predictable kernel addresses.
+ *
+ * Entropy is generated using the KASLR early boot functions now shared in
+ * the lib directory (originally written by Kees Cook). Randomization is
+ * done on PGD & PUD page table levels to increase possible addresses. The
+ * physical memory mapping code was adapted to support PUD level virtual
+ * addresses. This implementation on the best configuration provides 30,000
+ * possible virtual addresses in average for each memory region. An additional
+ * low memory page is used to ensure each CPU can start with a PGD aligned
+ * virtual address (for realmode).
+ *
+ * The order of each memory region is not changed. The feature looks at
+ * the available space for the regions based on different configuration
+ * options and randomizes the base and space between each. The size of the
+ * physical memory mapping is the available physical memory.
+ */
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/random.h>
+
+#include <asm/pgalloc.h>
+#include <asm/pgtable.h>
+#include <asm/setup.h>
+#include <asm/kaslr.h>
+
+#include "mm_internal.h"
+
+#define TB_SHIFT 40
+
+/*
+ * Virtual address start and end range for randomization. The end changes base
+ * on configuration to have the highest amount of space for randomization.
+ * It increases the possible random position for each randomized region.
+ *
+ * You need to add an if/def entry if you introduce a new memory region
+ * compatible with KASLR. Your entry must be in logical order with memory
+ * layout. For example, ESPFIX is before EFI because its virtual address is
+ * before. You also need to add a BUILD_BUG_ON in kernel_randomize_memory to
+ * ensure that this order is correct and won't be changed.
+ */
+static const unsigned long vaddr_start;
+static const unsigned long vaddr_end;
+
+/*
+ * Memory regions randomized by KASLR (except modules that use a separate logic
+ * earlier during boot). The list is ordered based on virtual addresses. This
+ * order is kept after randomization.
+ */
+static __initdata struct kaslr_memory_region {
+	unsigned long *base;
+	unsigned long size_tb;
+} kaslr_regions[] = {
+};
+
+/* Get size in bytes used by the memory region */
+static inline unsigned long get_padding(struct kaslr_memory_region *region)
+{
+	return (region->size_tb << TB_SHIFT);
+}
+
+/*
+ * Apply no randomization if KASLR was disabled at boot or if KASAN
+ * is enabled. KASAN shadow mappings rely on regions being PGD aligned.
+ */
+static inline bool kaslr_memory_enabled(void)
+{
+	return kaslr_enabled() && !config_enabled(CONFIG_KASAN);
+}
+
+/* Initialize base and padding for each memory region randomized with KASLR */
+void __init kernel_randomize_memory(void)
+{
+	size_t i;
+	unsigned long vaddr = vaddr_start;
+	unsigned long rand;
+	struct rnd_state rand_state;
+	unsigned long remain_entropy;
+
+	if (!kaslr_memory_enabled())
+		return;
+
+	/* Calculate entropy available between regions */
+	remain_entropy = vaddr_end - vaddr_start;
+	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++)
+		remain_entropy -= get_padding(&kaslr_regions[i]);
+
+	prandom_seed_state(&rand_state, kaslr_get_random_long("Memory"));
+
+	for (i = 0; i < ARRAY_SIZE(kaslr_regions); i++) {
+		unsigned long entropy;
+
+		/*
+		 * Select a random virtual address using the extra entropy
+		 * available.
+		 */
+		entropy = remain_entropy / (ARRAY_SIZE(kaslr_regions) - i);
+		prandom_bytes_state(&rand_state, &rand, sizeof(rand));
+		entropy = (rand % (entropy + 1)) & PUD_MASK;
+		vaddr += entropy;
+		*kaslr_regions[i].base = vaddr;
+
+		/*
+		 * Jump the region and add a minimum padding based on
+		 * randomization alignment.
+		 */
+		vaddr += get_padding(&kaslr_regions[i]);
+		vaddr = round_up(vaddr + 1, PUD_SIZE);
+		remain_entropy -= entropy;
+	}
+}
+
+/*
+ * Create PGD aligned trampoline table to allow real mode initialization
+ * of additional CPUs. Consume only 1 low memory page.
+ */
+void __meminit init_trampoline(void)
+{
+	unsigned long paddr, paddr_next;
+	pgd_t *pgd;
+	pud_t *pud_page, *pud_page_tramp;
+	int i;
+
+	if (!kaslr_memory_enabled()) {
+		init_trampoline_default();
+		return;
+	}
+
+	pud_page_tramp = alloc_low_page();
+
+	paddr = 0;
+	pgd = pgd_offset_k((unsigned long)__va(paddr));
+	pud_page = (pud_t *) pgd_page_vaddr(*pgd);
+
+	for (i = pud_index(paddr); i < PTRS_PER_PUD; i++, paddr = paddr_next) {
+		pud_t *pud, *pud_tramp;
+		unsigned long vaddr = (unsigned long)__va(paddr);
+
+		pud_tramp = pud_page_tramp + pud_index(paddr);
+		pud = pud_page + pud_index(vaddr);
+		paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
+
+		*pud_tramp = *pud;
+	}
+
+	set_pgd(&trampoline_pgd_entry,
+		__pgd(_KERNPG_TABLE | __pa(pud_page_tramp)));
+}