[PATCH v2 0/3] kasan: support backing vmalloc space with real shadow memory

[PATCH v2 0/3] kasan: support backing vmalloc space with real shadow memory

[Daniel Axtens]

Currently, vmalloc space is backed by the early shadow page. This
means that kasan is incompatible with VMAP_STACK, and it also provides
a hurdle for architectures that do not have a dedicated module space
(like powerpc64).

This series provides a mechanism to back vmalloc space with real,
dynamically allocated memory. I have only wired up x86, because that's
the only currently supported arch I can work with easily, but it's
very easy to wire up other architectures.

This has been discussed before in the context of VMAP_STACK:
 - https://bugzilla.kernel.org/show_bug.cgi?id=202009
 - https://lkml.org/lkml/2018/7/22/198
 - https://lkml.org/lkml/2019/7/19/822

In terms of implementation details:

Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.

Instead, share backing space across multiple mappings. Allocate
a backing page the first time a mapping in vmalloc space uses a
particular page of the shadow region. Keep this page around
regardless of whether the mapping is later freed - in the mean time
the page could have become shared by another vmalloc mapping.

This can in theory lead to unbounded memory growth, but the vmalloc
allocator is pretty good at reusing addresses, so the practical memory
usage appears to grow at first but then stay fairly stable.

If we run into practical memory exhaustion issues, I'm happy to
consider hooking into the book-keeping that vmap does, but I am not
convinced that it will be an issue.

v1: https://lore.kernel.org/linux-mm/20190725055503.19507-1-dja@axtens.net/T/
v2: address review comments:
 - Patch 1: use kasan_unpoison_shadow's built-in handling of
            ranges that do not align to a full shadow byte
 - Patch 3: prepopulate pgds rather than faulting things in

Daniel Axtens (3):
  kasan: support backing vmalloc space with real shadow memory
  fork: support VMAP_STACK with KASAN_VMALLOC
  x86/kasan: support KASAN_VMALLOC

 Documentation/dev-tools/kasan.rst | 60 ++++++++++++++++++++++++++++++
 arch/Kconfig                      |  9 +++--
 arch/x86/Kconfig                  |  1 +
 arch/x86/mm/kasan_init_64.c       | 61 +++++++++++++++++++++++++++++++
 include/linux/kasan.h             | 16 ++++++++
 kernel/fork.c                     |  4 ++
 lib/Kconfig.kasan                 | 16 ++++++++
 lib/test_kasan.c                  | 26 +++++++++++++
 mm/kasan/common.c                 | 51 ++++++++++++++++++++++++++
 mm/kasan/generic_report.c         |  3 ++
 mm/kasan/kasan.h                  |  1 +
 mm/vmalloc.c                      | 15 +++++++-
 12 files changed, 258 insertions(+), 5 deletions(-)

-- 
2.20.1

[PATCH v2 1/3] kasan: support backing vmalloc space with real shadow memory

[Daniel Axtens]

Hook into vmalloc and vmap, and dynamically allocate real shadow
memory to back the mappings.

Most mappings in vmalloc space are small, requiring less than a full
page of shadow space. Allocating a full shadow page per mapping would
therefore be wasteful. Furthermore, to ensure that different mappings
use different shadow pages, mappings would have to be aligned to
KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.

Instead, share backing space across multiple mappings. Allocate
a backing page the first time a mapping in vmalloc space uses a
particular page of the shadow region. Keep this page around
regardless of whether the mapping is later freed - in the mean time
the page could have become shared by another vmalloc mapping.

This can in theory lead to unbounded memory growth, but the vmalloc
allocator is pretty good at reusing addresses, so the practical memory
usage grows at first but then stays fairly stable.

This requires architecture support to actually use: arches must stop
mapping the read-only zero page over portion of the shadow region that
covers the vmalloc space and instead leave it unmapped.

This allows KASAN with VMAP_STACK, and will be needed for architectures
that do not have a separate module space (e.g. powerpc64, which I am
currently working on).

Link: https://bugzilla.kernel.org/show_bug.cgi?id=202009
Signed-off-by: Daniel Axtens <dja@axtens.net>

---

v2: let kasan_unpoison_shadow deal with ranges that do not use a
    full shadow byte.
---
 Documentation/dev-tools/kasan.rst | 60 +++++++++++++++++++++++++++++++
 include/linux/kasan.h             | 16 +++++++++
 lib/Kconfig.kasan                 | 16 +++++++++
 lib/test_kasan.c                  | 26 ++++++++++++++
 mm/kasan/common.c                 | 51 ++++++++++++++++++++++++++
 mm/kasan/generic_report.c         |  3 ++
 mm/kasan/kasan.h                  |  1 +
 mm/vmalloc.c                      | 15 +++++++-
 8 files changed, 187 insertions(+), 1 deletion(-)

diff --git a/Documentation/dev-tools/kasan.rst b/Documentation/dev-tools/kasan.rst
index b72d07d70239..35fda484a672 100644
--- a/Documentation/dev-tools/kasan.rst
+++ b/Documentation/dev-tools/kasan.rst
@@ -215,3 +215,63 @@ brk handler is used to print bug reports.
 A potential expansion of this mode is a hardware tag-based mode, which would
 use hardware memory tagging support instead of compiler instrumentation and
 manual shadow memory manipulation.
+
+What memory accesses are sanitised by KASAN?
+--------------------------------------------
+
+The kernel maps memory in a number of different parts of the address
+space. This poses something of a problem for KASAN, which requires
+that all addresses accessed by instrumented code have a valid shadow
+region.
+
+The range of kernel virtual addresses is large: there is not enough
+real memory to support a real shadow region for every address that
+could be accessed by the kernel.
+
+By default
+~~~~~~~~~~
+
+By default, architectures only map real memory over the shadow region
+for the linear mapping (and potentially other small areas). For all
+other areas - such as vmalloc and vmemmap space - a single read-only
+page is mapped over the shadow area. This read-only shadow page
+declares all memory accesses as permitted.
+
+This presents a problem for modules: they do not live in the linear
+mapping, but in a dedicated module space. By hooking in to the module
+allocator, KASAN can temporarily map real shadow memory to cover
+them. This allows detection of invalid accesses to module globals, for
+example.
+
+This also creates an incompatibility with ``VMAP_STACK``: if the stack
+lives in vmalloc space, it will be shadowed by the read-only page, and
+the kernel will fault when trying to set up the shadow data for stack
+variables.
+
+CONFIG_KASAN_VMALLOC
+~~~~~~~~~~~~~~~~~~~~
+
+With ``CONFIG_KASAN_VMALLOC``, KASAN can cover vmalloc space at the
+cost of greater memory usage. Currently this is only supported on x86.
+
+This works by hooking into vmalloc and vmap, and dynamically
+allocating real shadow memory to back the mappings.
+
+Most mappings in vmalloc space are small, requiring less than a full
+page of shadow space. Allocating a full shadow page per mapping would
+therefore be wasteful. Furthermore, to ensure that different mappings
+use different shadow pages, mappings would have to be aligned to
+``KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE``.
+
+Instead, we share backing space across multiple mappings. We allocate
+a backing page the first time a mapping in vmalloc space uses a
+particular page of the shadow region. We keep this page around
+regardless of whether the mapping is later freed - in the mean time
+this page could have become shared by another vmalloc mapping.
+
+This can in theory lead to unbounded memory growth, but the vmalloc
+allocator is pretty good at reusing addresses, so the practical memory
+usage grows at first but then stays fairly stable.
+
+This allows ``VMAP_STACK`` support on x86, and enables support of
+architectures that do not have a fixed module region.
diff --git a/include/linux/kasan.h b/include/linux/kasan.h
index cc8a03cc9674..fcabc5a03fca 100644
--- a/include/linux/kasan.h
+++ b/include/linux/kasan.h
@@ -70,8 +70,18 @@ struct kasan_cache {
 	int free_meta_offset;
 };
 
+/*
+ * These functions provide a special case to support backing module
+ * allocations with real shadow memory. With KASAN vmalloc, the special
+ * case is unnecessary, as the work is handled in the generic case.
+ */
+#ifndef CONFIG_KASAN_VMALLOC
 int kasan_module_alloc(void *addr, size_t size);
 void kasan_free_shadow(const struct vm_struct *vm);
+#else
+static inline int kasan_module_alloc(void *addr, size_t size) { return 0; }
+static inline void kasan_free_shadow(const struct vm_struct *vm) {}
+#endif
 
 int kasan_add_zero_shadow(void *start, unsigned long size);
 void kasan_remove_zero_shadow(void *start, unsigned long size);
@@ -194,4 +204,10 @@ static inline void *kasan_reset_tag(const void *addr)
 
 #endif /* CONFIG_KASAN_SW_TAGS */
 
+#ifdef CONFIG_KASAN_VMALLOC
+void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area);
+#else
+static inline void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area) {}
+#endif
+
 #endif /* LINUX_KASAN_H */
diff --git a/lib/Kconfig.kasan b/lib/Kconfig.kasan
index 4fafba1a923b..a320dc2e9317 100644
--- a/lib/Kconfig.kasan
+++ b/lib/Kconfig.kasan
@@ -6,6 +6,9 @@ config HAVE_ARCH_KASAN
 config HAVE_ARCH_KASAN_SW_TAGS
 	bool
 
+config	HAVE_ARCH_KASAN_VMALLOC
+	bool
+
 config CC_HAS_KASAN_GENERIC
 	def_bool $(cc-option, -fsanitize=kernel-address)
 
@@ -135,6 +138,19 @@ config KASAN_S390_4_LEVEL_PAGING
 	  to 3TB of RAM with KASan enabled). This options allows to force
 	  4-level paging instead.
 
+config KASAN_VMALLOC
+	bool "Back mappings in vmalloc space with real shadow memory"
+	depends on KASAN && HAVE_ARCH_KASAN_VMALLOC
+	help
+	  By default, the shadow region for vmalloc space is the read-only
+	  zero page. This means that KASAN cannot detect errors involving
+	  vmalloc space.
+
+	  Enabling this option will hook in to vmap/vmalloc and back those
+	  mappings with real shadow memory allocated on demand. This allows
+	  for KASAN to detect more sorts of errors (and to support vmapped
+	  stacks), but at the cost of higher memory usage.
+
 config TEST_KASAN
 	tristate "Module for testing KASAN for bug detection"
 	depends on m && KASAN
diff --git a/lib/test_kasan.c b/lib/test_kasan.c
index b63b367a94e8..d375246f5f96 100644
--- a/lib/test_kasan.c
+++ b/lib/test_kasan.c
@@ -18,6 +18,7 @@
 #include <linux/slab.h>
 #include <linux/string.h>
 #include <linux/uaccess.h>
+#include <linux/vmalloc.h>
 
 /*
  * Note: test functions are marked noinline so that their names appear in
@@ -709,6 +710,30 @@ static noinline void __init kmalloc_double_kzfree(void)
 	kzfree(ptr);
 }
 
+#ifdef CONFIG_KASAN_VMALLOC
+static noinline void __init vmalloc_oob(void)
+{
+	void *area;
+
+	pr_info("vmalloc out-of-bounds\n");
+
+	/*
+	 * We have to be careful not to hit the guard page.
+	 * The MMU will catch that and crash us.
+	 */
+	area = vmalloc(3000);
+	if (!area) {
+		pr_err("Allocation failed\n");
+		return;
+	}
+
+	((volatile char *)area)[3100];
+	vfree(area);
+}
+#else
+static void __init vmalloc_oob(void) {}
+#endif
+
 static int __init kmalloc_tests_init(void)
 {
 	/*
@@ -752,6 +777,7 @@ static int __init kmalloc_tests_init(void)
 	kasan_strings();
 	kasan_bitops();
 	kmalloc_double_kzfree();
+	vmalloc_oob();
 
 	kasan_restore_multi_shot(multishot);
 
diff --git a/mm/kasan/common.c b/mm/kasan/common.c
index 2277b82902d8..15d8f4ad581b 100644
--- a/mm/kasan/common.c
+++ b/mm/kasan/common.c
@@ -568,6 +568,7 @@ void kasan_kfree_large(void *ptr, unsigned long ip)
 	/* The object will be poisoned by page_alloc. */
 }
 
+#ifndef CONFIG_KASAN_VMALLOC
 int kasan_module_alloc(void *addr, size_t size)
 {
 	void *ret;
@@ -603,6 +604,7 @@ void kasan_free_shadow(const struct vm_struct *vm)
 	if (vm->flags & VM_KASAN)
 		vfree(kasan_mem_to_shadow(vm->addr));
 }
+#endif
 
 extern void __kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip);
 
@@ -722,3 +724,52 @@ static int __init kasan_memhotplug_init(void)
 
 core_initcall(kasan_memhotplug_init);
 #endif
+
+#ifdef CONFIG_KASAN_VMALLOC
+void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area)
+{
+	unsigned long shadow_alloc_start, shadow_alloc_end;
+	unsigned long addr;
+	unsigned long backing;
+	pgd_t *pgdp;
+	p4d_t *p4dp;
+	pud_t *pudp;
+	pmd_t *pmdp;
+	pte_t *ptep;
+	pte_t backing_pte;
+
+	shadow_alloc_start = ALIGN_DOWN(
+		(unsigned long)kasan_mem_to_shadow(area->addr),
+		PAGE_SIZE);
+	shadow_alloc_end = ALIGN(
+		(unsigned long)kasan_mem_to_shadow(area->addr + area->size),
+		PAGE_SIZE);
+
+	addr = shadow_alloc_start;
+	do {
+		pgdp = pgd_offset_k(addr);
+		p4dp = p4d_alloc(&init_mm, pgdp, addr);
+		pudp = pud_alloc(&init_mm, p4dp, addr);
+		pmdp = pmd_alloc(&init_mm, pudp, addr);
+		ptep = pte_alloc_kernel(pmdp, addr);
+
+		/*
+		 * we can validly get here if pte is not none: it means we
+		 * allocated this page earlier to use part of it for another
+		 * allocation
+		 */
+		if (pte_none(*ptep)) {
+			backing = __get_free_page(GFP_KERNEL);
+			backing_pte = pfn_pte(PFN_DOWN(__pa(backing)),
+					      PAGE_KERNEL);
+			set_pte_at(&init_mm, addr, ptep, backing_pte);
+		}
+	} while (addr += PAGE_SIZE, addr != shadow_alloc_end);
+
+	kasan_unpoison_shadow(area->addr, requested_size);
+	requested_size = round_up(requested_size, KASAN_SHADOW_SCALE_SIZE);
+	kasan_poison_shadow(area->addr + requested_size,
+			    area->size - requested_size,
+			    KASAN_VMALLOC_INVALID);
+}
+#endif
diff --git a/mm/kasan/generic_report.c b/mm/kasan/generic_report.c
index 36c645939bc9..2d97efd4954f 100644
--- a/mm/kasan/generic_report.c
+++ b/mm/kasan/generic_report.c
@@ -86,6 +86,9 @@ static const char *get_shadow_bug_type(struct kasan_access_info *info)
 	case KASAN_ALLOCA_RIGHT:
 		bug_type = "alloca-out-of-bounds";
 		break;
+	case KASAN_VMALLOC_INVALID:
+		bug_type = "vmalloc-out-of-bounds";
+		break;
 	}
 
 	return bug_type;
diff --git a/mm/kasan/kasan.h b/mm/kasan/kasan.h
index 014f19e76247..8b1f2fbc780b 100644
--- a/mm/kasan/kasan.h
+++ b/mm/kasan/kasan.h
@@ -25,6 +25,7 @@
 #endif
 
 #define KASAN_GLOBAL_REDZONE    0xFA  /* redzone for global variable */
+#define KASAN_VMALLOC_INVALID   0xF9  /* unallocated space in vmapped page */
 
 /*
  * Stack redzone shadow values
diff --git a/mm/vmalloc.c b/mm/vmalloc.c
index 4fa8d84599b0..8cbcb5056c9b 100644
--- a/mm/vmalloc.c
+++ b/mm/vmalloc.c
@@ -2012,6 +2012,15 @@ static void setup_vmalloc_vm(struct vm_struct *vm, struct vmap_area *va,
 	va->vm = vm;
 	va->flags |= VM_VM_AREA;
 	spin_unlock(&vmap_area_lock);
+
+	/*
+	 * If we are in vmalloc space we need to cover the shadow area with
+	 * real memory. If we come here through VM_ALLOC, this is done
+	 * by a higher level function that has access to the true size,
+	 * which might not be a full page.
+	 */
+	if (is_vmalloc_addr(vm->addr) && !(vm->flags & VM_ALLOC))
+		kasan_cover_vmalloc(vm->size, vm);
 }
 
 static void clear_vm_uninitialized_flag(struct vm_struct *vm)
@@ -2483,6 +2492,8 @@ void *__vmalloc_node_range(unsigned long size, unsigned long align,
 	if (!addr)
 		return NULL;
 
+	kasan_cover_vmalloc(real_size, area);
+
 	/*
 	 * In this function, newly allocated vm_struct has VM_UNINITIALIZED
 	 * flag. It means that vm_struct is not fully initialized.
@@ -3324,9 +3335,11 @@ struct vm_struct **pcpu_get_vm_areas(const unsigned long *offsets,
 	spin_unlock(&vmap_area_lock);
 
 	/* insert all vm's */
-	for (area = 0; area < nr_vms; area++)
+	for (area = 0; area < nr_vms; area++) {
 		setup_vmalloc_vm(vms[area], vas[area], VM_ALLOC,
 				 pcpu_get_vm_areas);
+		kasan_cover_vmalloc(sizes[area], vms[area]);
+	}
 
 	kfree(vas);
 	return vms;
-- 
2.20.1

[PATCH v2 2/3] fork: support VMAP_STACK with KASAN_VMALLOC

[Daniel Axtens]

Supporting VMAP_STACK with KASAN_VMALLOC is straightforward:

 - clear the shadow region of vmapped stacks when swapping them in
 - tweak Kconfig to allow VMAP_STACK to be turned on with KASAN

Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>
---
 arch/Kconfig  | 9 +++++----
 kernel/fork.c | 4 ++++
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/arch/Kconfig b/arch/Kconfig
index a7b57dd42c26..e791196005e1 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -825,16 +825,17 @@ config HAVE_ARCH_VMAP_STACK
 config VMAP_STACK
 	default y
 	bool "Use a virtually-mapped stack"
-	depends on HAVE_ARCH_VMAP_STACK && !KASAN
+	depends on HAVE_ARCH_VMAP_STACK
+	depends on !KASAN || KASAN_VMALLOC
 	---help---
 	  Enable this if you want the use virtually-mapped kernel stacks
 	  with guard pages.  This causes kernel stack overflows to be
 	  caught immediately rather than causing difficult-to-diagnose
 	  corruption.
 
-	  This is presently incompatible with KASAN because KASAN expects
-	  the stack to map directly to the KASAN shadow map using a formula
-	  that is incorrect if the stack is in vmalloc space.
+	  To use this with KASAN, the architecture must support backing
+	  virtual mappings with real shadow memory, and KASAN_VMALLOC must
+	  be enabled.
 
 config ARCH_OPTIONAL_KERNEL_RWX
 	def_bool n
diff --git a/kernel/fork.c b/kernel/fork.c
index d8ae0f1b4148..ce3150fe8ff2 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -94,6 +94,7 @@
 #include <linux/livepatch.h>
 #include <linux/thread_info.h>
 #include <linux/stackleak.h>
+#include <linux/kasan.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -215,6 +216,9 @@ static unsigned long *alloc_thread_stack_node(struct task_struct *tsk, int node)
 		if (!s)
 			continue;
 
+		/* Clear the KASAN shadow of the stack. */
+		kasan_unpoison_shadow(s->addr, THREAD_SIZE);
+
 		/* Clear stale pointers from reused stack. */
 		memset(s->addr, 0, THREAD_SIZE);
 
-- 
2.20.1

[PATCH v2 3/3] x86/kasan: support KASAN_VMALLOC

[Daniel Axtens]

In the case where KASAN directly allocates memory to back vmalloc
space, don't map the early shadow page over it.

We prepopulate pgds/p4ds for the range that would otherwise be empty.
This is required to get it synced to hardware on boot, allowing the
lower levels of the page tables to be filled dynamically.

Acked-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>

---

v2: move from faulting in shadow pgds to prepopulating
---
 arch/x86/Kconfig            |  1 +
 arch/x86/mm/kasan_init_64.c | 61 +++++++++++++++++++++++++++++++++++++
 2 files changed, 62 insertions(+)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 222855cc0158..40562cc3771f 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -134,6 +134,7 @@ config X86
 	select HAVE_ARCH_JUMP_LABEL
 	select HAVE_ARCH_JUMP_LABEL_RELATIVE
 	select HAVE_ARCH_KASAN			if X86_64
+	select HAVE_ARCH_KASAN_VMALLOC		if X86_64
 	select HAVE_ARCH_KGDB
 	select HAVE_ARCH_MMAP_RND_BITS		if MMU
 	select HAVE_ARCH_MMAP_RND_COMPAT_BITS	if MMU && COMPAT
diff --git a/arch/x86/mm/kasan_init_64.c b/arch/x86/mm/kasan_init_64.c
index 296da58f3013..2f57c4ddff61 100644
--- a/arch/x86/mm/kasan_init_64.c
+++ b/arch/x86/mm/kasan_init_64.c
@@ -245,6 +245,52 @@ static void __init kasan_map_early_shadow(pgd_t *pgd)
 	} while (pgd++, addr = next, addr != end);
 }
 
+static void __init kasan_shallow_populate_p4ds(pgd_t *pgd,
+		unsigned long addr,
+		unsigned long end,
+		int nid)
+{
+	p4d_t *p4d;
+	unsigned long next;
+	void *p;
+
+	p4d = p4d_offset(pgd, addr);
+	do {
+		next = p4d_addr_end(addr, end);
+
+		if (p4d_none(*p4d)) {
+			p = early_alloc(PAGE_SIZE, nid, true);
+			p4d_populate(&init_mm, p4d, p);
+		}
+	} while (p4d++, addr = next, addr != end);
+}
+
+static void __init kasan_shallow_populate_pgds(void *start, void *end)
+{
+	unsigned long addr, next;
+	pgd_t *pgd;
+	void *p;
+	int nid = early_pfn_to_nid((unsigned long)start);
+
+	addr = (unsigned long)start;
+	pgd = pgd_offset_k(addr);
+	do {
+		next = pgd_addr_end(addr, (unsigned long)end);
+
+		if (pgd_none(*pgd)) {
+			p = early_alloc(PAGE_SIZE, nid, true);
+			pgd_populate(&init_mm, pgd, p);
+		}
+
+		/*
+		 * we need to populate p4ds to be synced when running in
+		 * four level mode - see sync_global_pgds_l4()
+		 */
+		kasan_shallow_populate_p4ds(pgd, addr, next, nid);
+	} while (pgd++, addr = next, addr != (unsigned long)end);
+}
+
+
 #ifdef CONFIG_KASAN_INLINE
 static int kasan_die_handler(struct notifier_block *self,
 			     unsigned long val,
@@ -352,9 +398,24 @@ void __init kasan_init(void)
 	shadow_cpu_entry_end = (void *)round_up(
 			(unsigned long)shadow_cpu_entry_end, PAGE_SIZE);
 
+	/*
+	 * If we're in full vmalloc mode, don't back vmalloc space with early
+	 * shadow pages. Instead, prepopulate pgds/p4ds so they are synced to
+	 * the global table and we can populate the lower levels on demand.
+	 */
+#ifdef CONFIG_KASAN_VMALLOC
+	kasan_shallow_populate_pgds(
+		kasan_mem_to_shadow((void *)PAGE_OFFSET + MAXMEM),
+		kasan_mem_to_shadow((void *)VMALLOC_END));
+
+	kasan_populate_early_shadow(
+		kasan_mem_to_shadow((void *)VMALLOC_END + 1),
+		shadow_cpu_entry_begin);
+#else
 	kasan_populate_early_shadow(
 		kasan_mem_to_shadow((void *)PAGE_OFFSET + MAXMEM),
 		shadow_cpu_entry_begin);
+#endif
 
 	kasan_populate_shadow((unsigned long)shadow_cpu_entry_begin,
 			      (unsigned long)shadow_cpu_entry_end, 0);
-- 
2.20.1

Re: [PATCH v2 1/3] kasan: support backing vmalloc space with real shadow memory

[Mark Rutland]

Hi Daniel,

On Tue, Jul 30, 2019 at 12:21:06AM +1000, Daniel Axtens wrote:
> Hook into vmalloc and vmap, and dynamically allocate real shadow
> memory to back the mappings.
> 
> Most mappings in vmalloc space are small, requiring less than a full
> page of shadow space. Allocating a full shadow page per mapping would
> therefore be wasteful. Furthermore, to ensure that different mappings
> use different shadow pages, mappings would have to be aligned to
> KASAN_SHADOW_SCALE_SIZE * PAGE_SIZE.
> 
> Instead, share backing space across multiple mappings. Allocate
> a backing page the first time a mapping in vmalloc space uses a
> particular page of the shadow region. Keep this page around
> regardless of whether the mapping is later freed - in the mean time
> the page could have become shared by another vmalloc mapping.
> 
> This can in theory lead to unbounded memory growth, but the vmalloc
> allocator is pretty good at reusing addresses, so the practical memory
> usage grows at first but then stays fairly stable.
> 
> This requires architecture support to actually use: arches must stop
> mapping the read-only zero page over portion of the shadow region that
> covers the vmalloc space and instead leave it unmapped.
> 
> This allows KASAN with VMAP_STACK, and will be needed for architectures
> that do not have a separate module space (e.g. powerpc64, which I am
> currently working on).
> 
> Link: https://bugzilla.kernel.org/show_bug.cgi?id=202009
> Signed-off-by: Daniel Axtens <dja@axtens.net>

This generally looks good, but I have a few concerns below, mostly
related to concurrency.

[...]

> diff --git a/mm/kasan/common.c b/mm/kasan/common.c
> index 2277b82902d8..15d8f4ad581b 100644
> --- a/mm/kasan/common.c
> +++ b/mm/kasan/common.c
> @@ -568,6 +568,7 @@ void kasan_kfree_large(void *ptr, unsigned long ip)
>  	/* The object will be poisoned by page_alloc. */
>  }
>  
> +#ifndef CONFIG_KASAN_VMALLOC
>  int kasan_module_alloc(void *addr, size_t size)
>  {
>  	void *ret;
> @@ -603,6 +604,7 @@ void kasan_free_shadow(const struct vm_struct *vm)
>  	if (vm->flags & VM_KASAN)
>  		vfree(kasan_mem_to_shadow(vm->addr));
>  }
> +#endif

IIUC we can drop MODULE_ALIGN back to PAGE_SIZE in this case, too.

>  
>  extern void __kasan_report(unsigned long addr, size_t size, bool is_write, unsigned long ip);
>  
> @@ -722,3 +724,52 @@ static int __init kasan_memhotplug_init(void)
>  
>  core_initcall(kasan_memhotplug_init);
>  #endif
> +
> +#ifdef CONFIG_KASAN_VMALLOC
> +void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area)

Nit: I think it would be more consistent to call this
kasan_populate_vmalloc().

> +{
> +	unsigned long shadow_alloc_start, shadow_alloc_end;
> +	unsigned long addr;
> +	unsigned long backing;
> +	pgd_t *pgdp;
> +	p4d_t *p4dp;
> +	pud_t *pudp;
> +	pmd_t *pmdp;
> +	pte_t *ptep;
> +	pte_t backing_pte;

Nit: I think it would be preferable to use 'page' rather than 'backing',
and 'pte' rather than 'backing_pte', since there's no otehr namespace to
collide with here. Otherwise, using 'shadow' rather than 'backing' would
be consistent with the existing kasan code.

> +
> +	shadow_alloc_start = ALIGN_DOWN(
> +		(unsigned long)kasan_mem_to_shadow(area->addr),
> +		PAGE_SIZE);
> +	shadow_alloc_end = ALIGN(
> +		(unsigned long)kasan_mem_to_shadow(area->addr + area->size),
> +		PAGE_SIZE);
> +
> +	addr = shadow_alloc_start;
> +	do {
> +		pgdp = pgd_offset_k(addr);
> +		p4dp = p4d_alloc(&init_mm, pgdp, addr);
> +		pudp = pud_alloc(&init_mm, p4dp, addr);
> +		pmdp = pmd_alloc(&init_mm, pudp, addr);
> +		ptep = pte_alloc_kernel(pmdp, addr);
> +
> +		/*
> +		 * we can validly get here if pte is not none: it means we
> +		 * allocated this page earlier to use part of it for another
> +		 * allocation
> +		 */
> +		if (pte_none(*ptep)) {
> +			backing = __get_free_page(GFP_KERNEL);
> +			backing_pte = pfn_pte(PFN_DOWN(__pa(backing)),
> +					      PAGE_KERNEL);
> +			set_pte_at(&init_mm, addr, ptep, backing_pte);
> +		}

Does anything prevent two threads from racing to allocate the same
shadow page?

AFAICT it's possible for two threads to get down to the ptep, then both
see pte_none(*ptep)), then both try to allocate the same page.

I suspect we have to take init_mm::page_table_lock when plumbing this
in, similarly to __pte_alloc().

> +	} while (addr += PAGE_SIZE, addr != shadow_alloc_end);
> +
> +	kasan_unpoison_shadow(area->addr, requested_size);
> +	requested_size = round_up(requested_size, KASAN_SHADOW_SCALE_SIZE);
> +	kasan_poison_shadow(area->addr + requested_size,
> +			    area->size - requested_size,
> +			    KASAN_VMALLOC_INVALID);

IIUC, this could leave the final portion of an allocated page
unpoisoned.

I think it might make more sense to poison each page when it's
allocated, then plumb it into the page tables, then unpoison the object.

That way, we can rely on any shadow allocated by another thread having
been initialized to KASAN_VMALLOC_INVALID, and only need mutual
exclusion when allocating the shadow, rather than when poisoning
objects.

Thanks,
Mark.

Re: [PATCH v2 1/3] kasan: support backing vmalloc space with real shadow memory

[Daniel Axtens]

Hi Mark,

Thanks for your email - I'm very new to mm stuff and the feedback is
very helpful.

>> +#ifndef CONFIG_KASAN_VMALLOC
>>  int kasan_module_alloc(void *addr, size_t size)
>>  {
>>  	void *ret;
>> @@ -603,6 +604,7 @@ void kasan_free_shadow(const struct vm_struct *vm)
>>  	if (vm->flags & VM_KASAN)
>>  		vfree(kasan_mem_to_shadow(vm->addr));
>>  }
>> +#endif
>
> IIUC we can drop MODULE_ALIGN back to PAGE_SIZE in this case, too.

Yes, done.

>>  core_initcall(kasan_memhotplug_init);
>>  #endif
>> +
>> +#ifdef CONFIG_KASAN_VMALLOC
>> +void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area)
>
> Nit: I think it would be more consistent to call this
> kasan_populate_vmalloc().
>

Absolutely. I didn't love the name but just didn't 'click' that populate
would be a better verb.

>> +{
>> +	unsigned long shadow_alloc_start, shadow_alloc_end;
>> +	unsigned long addr;
>> +	unsigned long backing;
>> +	pgd_t *pgdp;
>> +	p4d_t *p4dp;
>> +	pud_t *pudp;
>> +	pmd_t *pmdp;
>> +	pte_t *ptep;
>> +	pte_t backing_pte;
>
> Nit: I think it would be preferable to use 'page' rather than 'backing',
> and 'pte' rather than 'backing_pte', since there's no otehr namespace to
> collide with here. Otherwise, using 'shadow' rather than 'backing' would
> be consistent with the existing kasan code.

Not a problem, done.

>> +	addr = shadow_alloc_start;
>> +	do {
>> +		pgdp = pgd_offset_k(addr);
>> +		p4dp = p4d_alloc(&init_mm, pgdp, addr);
>> +		pudp = pud_alloc(&init_mm, p4dp, addr);
>> +		pmdp = pmd_alloc(&init_mm, pudp, addr);
>> +		ptep = pte_alloc_kernel(pmdp, addr);
>> +
>> +		/*
>> +		 * we can validly get here if pte is not none: it means we
>> +		 * allocated this page earlier to use part of it for another
>> +		 * allocation
>> +		 */
>> +		if (pte_none(*ptep)) {
>> +			backing = __get_free_page(GFP_KERNEL);
>> +			backing_pte = pfn_pte(PFN_DOWN(__pa(backing)),
>> +					      PAGE_KERNEL);
>> +			set_pte_at(&init_mm, addr, ptep, backing_pte);
>> +		}
>
> Does anything prevent two threads from racing to allocate the same
> shadow page?
>
> AFAICT it's possible for two threads to get down to the ptep, then both
> see pte_none(*ptep)), then both try to allocate the same page.
>
> I suspect we have to take init_mm::page_table_lock when plumbing this
> in, similarly to __pte_alloc().

Good catch. I think you're right, I'll add the lock.

>> +	} while (addr += PAGE_SIZE, addr != shadow_alloc_end);
>> +
>> +	kasan_unpoison_shadow(area->addr, requested_size);
>> +	requested_size = round_up(requested_size, KASAN_SHADOW_SCALE_SIZE);
>> +	kasan_poison_shadow(area->addr + requested_size,
>> +			    area->size - requested_size,
>> +			    KASAN_VMALLOC_INVALID);
>
> IIUC, this could leave the final portion of an allocated page
> unpoisoned.
>
> I think it might make more sense to poison each page when it's
> allocated, then plumb it into the page tables, then unpoison the object.
>
> That way, we can rely on any shadow allocated by another thread having
> been initialized to KASAN_VMALLOC_INVALID, and only need mutual
> exclusion when allocating the shadow, rather than when poisoning
> objects.

Yes, that makes sense, will do.

Thanks again,
Daniel

Re: [PATCH v2 1/3] kasan: support backing vmalloc space with real shadow memory

[Daniel Axtens]

Daniel Axtens <dja@axtens.net> writes:

> Hi Mark,
>
> Thanks for your email - I'm very new to mm stuff and the feedback is
> very helpful.
>
>>> +#ifndef CONFIG_KASAN_VMALLOC
>>>  int kasan_module_alloc(void *addr, size_t size)
>>>  {
>>>  	void *ret;
>>> @@ -603,6 +604,7 @@ void kasan_free_shadow(const struct vm_struct *vm)
>>>  	if (vm->flags & VM_KASAN)
>>>  		vfree(kasan_mem_to_shadow(vm->addr));
>>>  }
>>> +#endif
>>
>> IIUC we can drop MODULE_ALIGN back to PAGE_SIZE in this case, too.
>
> Yes, done.
>
>>>  core_initcall(kasan_memhotplug_init);
>>>  #endif
>>> +
>>> +#ifdef CONFIG_KASAN_VMALLOC
>>> +void kasan_cover_vmalloc(unsigned long requested_size, struct vm_struct *area)
>>
>> Nit: I think it would be more consistent to call this
>> kasan_populate_vmalloc().
>>
>
> Absolutely. I didn't love the name but just didn't 'click' that populate
> would be a better verb.
>
>>> +{
>>> +	unsigned long shadow_alloc_start, shadow_alloc_end;
>>> +	unsigned long addr;
>>> +	unsigned long backing;
>>> +	pgd_t *pgdp;
>>> +	p4d_t *p4dp;
>>> +	pud_t *pudp;
>>> +	pmd_t *pmdp;
>>> +	pte_t *ptep;
>>> +	pte_t backing_pte;
>>
>> Nit: I think it would be preferable to use 'page' rather than 'backing',
>> and 'pte' rather than 'backing_pte', since there's no otehr namespace to
>> collide with here. Otherwise, using 'shadow' rather than 'backing' would
>> be consistent with the existing kasan code.
>
> Not a problem, done.
>
>>> +	addr = shadow_alloc_start;
>>> +	do {
>>> +		pgdp = pgd_offset_k(addr);
>>> +		p4dp = p4d_alloc(&init_mm, pgdp, addr);
>>> +		pudp = pud_alloc(&init_mm, p4dp, addr);
>>> +		pmdp = pmd_alloc(&init_mm, pudp, addr);
>>> +		ptep = pte_alloc_kernel(pmdp, addr);
>>> +
>>> +		/*
>>> +		 * we can validly get here if pte is not none: it means we
>>> +		 * allocated this page earlier to use part of it for another
>>> +		 * allocation
>>> +		 */
>>> +		if (pte_none(*ptep)) {
>>> +			backing = __get_free_page(GFP_KERNEL);
>>> +			backing_pte = pfn_pte(PFN_DOWN(__pa(backing)),
>>> +					      PAGE_KERNEL);
>>> +			set_pte_at(&init_mm, addr, ptep, backing_pte);
>>> +		}
>>
>> Does anything prevent two threads from racing to allocate the same
>> shadow page?
>>
>> AFAICT it's possible for two threads to get down to the ptep, then both
>> see pte_none(*ptep)), then both try to allocate the same page.
>>
>> I suspect we have to take init_mm::page_table_lock when plumbing this
>> in, similarly to __pte_alloc().
>
> Good catch. I think you're right, I'll add the lock.
>
>>> +	} while (addr += PAGE_SIZE, addr != shadow_alloc_end);
>>> +
>>> +	kasan_unpoison_shadow(area->addr, requested_size);
>>> +	requested_size = round_up(requested_size, KASAN_SHADOW_SCALE_SIZE);
>>> +	kasan_poison_shadow(area->addr + requested_size,
>>> +			    area->size - requested_size,
>>> +			    KASAN_VMALLOC_INVALID);
>>
>> IIUC, this could leave the final portion of an allocated page
>> unpoisoned.
>>
>> I think it might make more sense to poison each page when it's
>> allocated, then plumb it into the page tables, then unpoison the object.
>>
>> That way, we can rely on any shadow allocated by another thread having
>> been initialized to KASAN_VMALLOC_INVALID, and only need mutual
>> exclusion when allocating the shadow, rather than when poisoning
>> objects.

I've come a bit unstuck on this one. If a vmalloc address range is
reused, we can end up with the following sequence:

 - p = vmalloc(PAGE_SIZE) allocates ffffc90000000000

 - kasan_populate_shadow allocates a shadow page, fills it with
   KASAN_VMALLOC_INVALID, and then unpoisions
   PAGE_SIZE >> KASAN_SHADOW_SHIFT_SIZE bytes

 - vfree(p)

 - p = vmalloc(3000) also allocates ffffc90000000000 because of address
   reuse in vmalloc.

 - Now kasan_populate_shadow doesn't allocate a page, so does no
   poisioning.

 - kasan_populate_shadow unpoisions 3000 >> KASAN_SHADOW_SHIFT_SIZE
   bytes, but the PAGE_SIZE-3000 extra bytes are still unpoisioned, so
   accesses that are out-of-bounds for the 3000 byte allocation are
   missed.

So I think we do need to poision the shadow of the [requested_size,
area->size) region each time. However, I don't think we need mutual
exclusion to be able to do this safely. I think the safety is guaranteed
by vmalloc not giving the same page to multiple allocations. Because no
two threads are going to get overlapping vmalloc/vmap allocations, their
shadow ranges are not going to overlap, and so they're not going to
trample over each other.

I think it's probably still worth poisioning the pages on allocation:
for one thing, you are right that part of the shadow page will not be
poisioned otherwise, and secondly it means you migh get a kasan splat
before you get a page-not-present fault if you access beyond an
allocation, at least if the shadow happens to fall helpfully within an
already-allocated page.

v3 to come soon.

Regards,
Daniel

>
> Yes, that makes sense, will do.
>
> Thanks again,
> Daniel