mm: kmsan: maintain KMSAN metadata for page operations

Insert KMSAN hooks that make the necessary bookkeeping changes:
- poison page shadow and origins in alloc_pages()/free_page();
- clear page shadow and origins in clear_page(), copy_user_highpage();
- copy page metadata in copy_highpage(), wp_page_copy();
- handle vmap()/vunmap()/iounmap();

Link: https://lkml.kernel.org/r/[e-mail]
Signed-off-by: Alexander Potapenko <[e-mail]>
Cc: Alexander Viro <[e-mail]>
Cc: Alexei Starovoitov <[e-mail]>
Cc: Andrey Konovalov <[e-mail]>
Cc: Andrey Konovalov <[e-mail]>
Cc: Andy Lutomirski <[e-mail]>
Cc: Arnd Bergmann <[e-mail]>
Cc: Borislav Petkov <[e-mail]>
Cc: Christoph Hellwig <[e-mail]>
Cc: Christoph Lameter <[e-mail]>
Cc: David Rientjes <[e-mail]>
Cc: Dmitry Vyukov <[e-mail]>
Cc: Eric Biggers <[e-mail]>
Cc: Eric Biggers <[e-mail]>
Cc: Eric Dumazet <[e-mail]>
Cc: Greg Kroah-Hartman <[e-mail]>
Cc: Herbert Xu <[e-mail]>
Cc: Ilya Leoshkevich <[e-mail]>
Cc: Ingo Molnar <[e-mail]>
Cc: Jens Axboe <[e-mail]>
Cc: Joonsoo Kim <[e-mail]>
Cc: Kees Cook <[e-mail]>
Cc: Marco Elver <[e-mail]>
Cc: Mark Rutland <[e-mail]>
Cc: Matthew Wilcox <[e-mail]>
Cc: Michael S. Tsirkin <[e-mail]>
Cc: Pekka Enberg <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Cc: Petr Mladek <[e-mail]>
Cc: Stephen Rothwell <[e-mail]>
Cc: Steven Rostedt <[e-mail]>
Cc: Thomas Gleixner <[e-mail]>
Cc: Vasily Gorbik <[e-mail]>
Cc: Vegard Nossum <[e-mail]>
Cc: Vlastimil Babka <[e-mail]>
Signed-off-by: Andrew Morton <[e-mail]>

arch/x86/include/asm/page_64.h

···
 #include <asm/cpufeatures.h>
 #include <asm/alternative.h>
 
+#include <linux/kmsan-checks.h>
+
 /* duplicated to the one in bootmem.h */
 extern unsigned long max_pfn;
 extern unsigned long phys_base;
···
 
 static inline void clear_page(void *page)
 {
+	/*
+	 * Clean up KMSAN metadata for the page being cleared. The assembly call
+	 * below clobbers @page, so we perform unpoisoning before it.
+	 */
+	kmsan_unpoison_memory(page, PAGE_SIZE);
 	alternative_call_2(clear_page_orig,
 			   clear_page_rep, X86_FEATURE_REP_GOOD,
 			   clear_page_erms, X86_FEATURE_ERMS,

arch/x86/mm/ioremap.c

···
 #include <linux/cc_platform.h>
 #include <linux/efi.h>
 #include <linux/pgtable.h>
+#include <linux/kmsan.h>
 
 #include <asm/set_memory.h>
 #include <asm/e820/api.h>
···
 		return;
 	}
 
+	kmsan_iounmap_page_range((unsigned long)addr,
+		(unsigned long)addr + get_vm_area_size(p));
 	memtype_free(p->phys_addr, p->phys_addr + get_vm_area_size(p));
 
 	/* Finally remove it */

include/linux/highmem.h

···
 #include <linux/kernel.h>
 #include <linux/bug.h>
 #include <linux/cacheflush.h>
+#include <linux/kmsan.h>
 #include <linux/mm.h>
 #include <linux/uaccess.h>
 #include <linux/hardirq.h>
···
 	vfrom = kmap_local_page(from);
 	vto = kmap_local_page(to);
 	copy_user_page(vto, vfrom, vaddr, to);
+	kmsan_unpoison_memory(page_address(to), PAGE_SIZE);
 	kunmap_local(vto);
 	kunmap_local(vfrom);
 }
···
 	vfrom = kmap_local_page(from);
 	vto = kmap_local_page(to);
 	copy_page(vto, vfrom);
+	kmsan_copy_page_meta(to, from);
 	kunmap_local(vto);
 	kunmap_local(vfrom);
 }

include/linux/kmsan.h

···
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * KMSAN API for subsystems.
+ *
+ * Copyright (C) 2017-2022 Google LLC
+ * Author: Alexander Potapenko <[e-mail]>
+ *
+ */
+#ifndef _LINUX_KMSAN_H
+#define _LINUX_KMSAN_H
+
+#include <linux/gfp.h>
+#include <linux/kmsan-checks.h>
+#include <linux/types.h>
+
+struct page;
+
+#ifdef CONFIG_KMSAN
+
+/**
+ * kmsan_alloc_page() - Notify KMSAN about an alloc_pages() call.
+ * @page:  struct page pointer returned by alloc_pages().
+ * @order: order of allocated struct page.
+ * @flags: GFP flags used by alloc_pages()
+ *
+ * KMSAN marks 1<<@order pages starting at @page as uninitialized, unless
+ * @flags contain __GFP_ZERO.
+ */
+void kmsan_alloc_page(struct page *page, unsigned int order, gfp_t flags);
+
+/**
+ * kmsan_free_page() - Notify KMSAN about a free_pages() call.
+ * @page:  struct page pointer passed to free_pages().
+ * @order: order of deallocated struct page.
+ *
+ * KMSAN marks freed memory as uninitialized.
+ */
+void kmsan_free_page(struct page *page, unsigned int order);
+
+/**
+ * kmsan_copy_page_meta() - Copy KMSAN metadata between two pages.
+ * @dst: destination page.
+ * @src: source page.
+ *
+ * KMSAN copies the contents of metadata pages for @src into the metadata pages
+ * for @dst. If @dst has no associated metadata pages, nothing happens.
+ * If @src has no associated metadata pages, @dst metadata pages are unpoisoned.
+ */
+void kmsan_copy_page_meta(struct page *dst, struct page *src);
+
+/**
+ * kmsan_map_kernel_range_noflush() - Notify KMSAN about a vmap.
+ * @start:	start of vmapped range.
+ * @end:	end of vmapped range.
+ * @prot:	page protection flags used for vmap.
+ * @pages:	array of pages.
+ * @page_shift:	page_shift passed to vmap_range_noflush().
+ *
+ * KMSAN maps shadow and origin pages of @pages into contiguous ranges in
+ * vmalloc metadata address range.
+ */
+void kmsan_vmap_pages_range_noflush(unsigned long start, unsigned long end,
+				    pgprot_t prot, struct page **pages,
+				    unsigned int page_shift);
+
+/**
+ * kmsan_vunmap_kernel_range_noflush() - Notify KMSAN about a vunmap.
+ * @start: start of vunmapped range.
+ * @end:   end of vunmapped range.
+ *
+ * KMSAN unmaps the contiguous metadata ranges created by
+ * kmsan_map_kernel_range_noflush().
+ */
+void kmsan_vunmap_range_noflush(unsigned long start, unsigned long end);
+
+/**
+ * kmsan_ioremap_page_range() - Notify KMSAN about a ioremap_page_range() call.
+ * @addr:	range start.
+ * @end:	range end.
+ * @phys_addr:	physical range start.
+ * @prot:	page protection flags used for ioremap_page_range().
+ * @page_shift:	page_shift argument passed to vmap_range_noflush().
+ *
+ * KMSAN creates new metadata pages for the physical pages mapped into the
+ * virtual memory.
+ */
+void kmsan_ioremap_page_range(unsigned long addr, unsigned long end,
+			      phys_addr_t phys_addr, pgprot_t prot,
+			      unsigned int page_shift);
+
+/**
+ * kmsan_iounmap_page_range() - Notify KMSAN about a iounmap_page_range() call.
+ * @start: range start.
+ * @end:   range end.
+ *
+ * KMSAN unmaps the metadata pages for the given range and, unlike for
+ * vunmap_page_range(), also deallocates them.
+ */
+void kmsan_iounmap_page_range(unsigned long start, unsigned long end);
+
+#else
+
+static inline int kmsan_alloc_page(struct page *page, unsigned int order,
+				   gfp_t flags)
+{
+	return 0;
+}
+
+static inline void kmsan_free_page(struct page *page, unsigned int order)
+{
+}
+
+static inline void kmsan_copy_page_meta(struct page *dst, struct page *src)
+{
+}
+
+static inline void kmsan_vmap_pages_range_noflush(unsigned long start,
+						  unsigned long end,
+						  pgprot_t prot,
+						  struct page **pages,
+						  unsigned int page_shift)
+{
+}
+
+static inline void kmsan_vunmap_range_noflush(unsigned long start,
+					      unsigned long end)
+{
+}
+
+static inline void kmsan_ioremap_page_range(unsigned long start,
+					    unsigned long end,
+					    phys_addr_t phys_addr,
+					    pgprot_t prot,
+					    unsigned int page_shift)
+{
+}
+
+static inline void kmsan_iounmap_page_range(unsigned long start,
+					    unsigned long end)
+{
+}
+
+#endif
+
+#endif /* _LINUX_KMSAN_H */

mm/internal.h

···
 }
 #endif
 
+int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
+			       pgprot_t prot, struct page **pages,
+			       unsigned int page_shift);
+
 void vunmap_range_noflush(unsigned long start, unsigned long end);
+
+void __vunmap_range_noflush(unsigned long start, unsigned long end);
 
 int numa_migrate_prep(struct page *page, struct vm_area_struct *vma,
 		      unsigned long addr, int page_nid, int *flags);

mm/kmsan/hooks.c

···
 
 #include <linux/cacheflush.h>
 #include <linux/gfp.h>
+#include <linux/kmsan.h>
 #include <linux/mm.h>
 #include <linux/mm_types.h>
 #include <linux/slab.h>
···
  * kmsan_enter_runtime()/kmsan_leave_runtime(), because this will lead to
  * skipping effects of functions like memset() inside instrumented code.
  */
+
+static unsigned long vmalloc_shadow(unsigned long addr)
+{
+	return (unsigned long)kmsan_get_metadata((void *)addr,
+						 KMSAN_META_SHADOW);
+}
+
+static unsigned long vmalloc_origin(unsigned long addr)
+{
+	return (unsigned long)kmsan_get_metadata((void *)addr,
+						 KMSAN_META_ORIGIN);
+}
+
+void kmsan_vunmap_range_noflush(unsigned long start, unsigned long end)
+{
+	__vunmap_range_noflush(vmalloc_shadow(start), vmalloc_shadow(end));
+	__vunmap_range_noflush(vmalloc_origin(start), vmalloc_origin(end));
+	flush_cache_vmap(vmalloc_shadow(start), vmalloc_shadow(end));
+	flush_cache_vmap(vmalloc_origin(start), vmalloc_origin(end));
+}
+
+/*
+ * This function creates new shadow/origin pages for the physical pages mapped
+ * into the virtual memory. If those physical pages already had shadow/origin,
+ * those are ignored.
+ */
+void kmsan_ioremap_page_range(unsigned long start, unsigned long end,
+			      phys_addr_t phys_addr, pgprot_t prot,
+			      unsigned int page_shift)
+{
+	gfp_t gfp_mask = GFP_KERNEL | __GFP_ZERO;
+	struct page *shadow, *origin;
+	unsigned long off = 0;
+	int nr;
+
+	if (!kmsan_enabled || kmsan_in_runtime())
+		return;
+
+	nr = (end - start) / PAGE_SIZE;
+	kmsan_enter_runtime();
+	for (int i = 0; i < nr; i++, off += PAGE_SIZE) {
+		shadow = alloc_pages(gfp_mask, 1);
+		origin = alloc_pages(gfp_mask, 1);
+		__vmap_pages_range_noflush(
+			vmalloc_shadow(start + off),
+			vmalloc_shadow(start + off + PAGE_SIZE), prot, &shadow,
+			PAGE_SHIFT);
+		__vmap_pages_range_noflush(
+			vmalloc_origin(start + off),
+			vmalloc_origin(start + off + PAGE_SIZE), prot, &origin,
+			PAGE_SHIFT);
+	}
+	flush_cache_vmap(vmalloc_shadow(start), vmalloc_shadow(end));
+	flush_cache_vmap(vmalloc_origin(start), vmalloc_origin(end));
+	kmsan_leave_runtime();
+}
+
+void kmsan_iounmap_page_range(unsigned long start, unsigned long end)
+{
+	unsigned long v_shadow, v_origin;
+	struct page *shadow, *origin;
+	int nr;
+
+	if (!kmsan_enabled || kmsan_in_runtime())
+		return;
+
+	nr = (end - start) / PAGE_SIZE;
+	kmsan_enter_runtime();
+	v_shadow = (unsigned long)vmalloc_shadow(start);
+	v_origin = (unsigned long)vmalloc_origin(start);
+	for (int i = 0; i < nr;
+	     i++, v_shadow += PAGE_SIZE, v_origin += PAGE_SIZE) {
+		shadow = kmsan_vmalloc_to_page_or_null((void *)v_shadow);
+		origin = kmsan_vmalloc_to_page_or_null((void *)v_origin);
+		__vunmap_range_noflush(v_shadow, vmalloc_shadow(end));
+		__vunmap_range_noflush(v_origin, vmalloc_origin(end));
+		if (shadow)
+			__free_pages(shadow, 1);
+		if (origin)
+			__free_pages(origin, 1);
+	}
+	flush_cache_vmap(vmalloc_shadow(start), vmalloc_shadow(end));
+	flush_cache_vmap(vmalloc_origin(start), vmalloc_origin(end));
+	kmsan_leave_runtime();
+}
 
 /* Functions from kmsan-checks.h follow. */
 void kmsan_poison_memory(const void *address, size_t size, gfp_t flags)

mm/kmsan/shadow.c

···
 
 	return (is_origin ? origin_ptr_for(page) : shadow_ptr_for(page)) + off;
 }
+
+void kmsan_copy_page_meta(struct page *dst, struct page *src)
+{
+	if (!kmsan_enabled || kmsan_in_runtime())
+		return;
+	if (!dst || !page_has_metadata(dst))
+		return;
+	if (!src || !page_has_metadata(src)) {
+		kmsan_internal_unpoison_memory(page_address(dst), PAGE_SIZE,
+					       /*checked*/ false);
+		return;
+	}
+
+	kmsan_enter_runtime();
+	__memcpy(shadow_ptr_for(dst), shadow_ptr_for(src), PAGE_SIZE);
+	__memcpy(origin_ptr_for(dst), origin_ptr_for(src), PAGE_SIZE);
+	kmsan_leave_runtime();
+}
+
+void kmsan_alloc_page(struct page *page, unsigned int order, gfp_t flags)
+{
+	bool initialized = (flags & __GFP_ZERO) || !kmsan_enabled;
+	struct page *shadow, *origin;
+	depot_stack_handle_t handle;
+	int pages = 1 << order;
+
+	if (!page)
+		return;
+
+	shadow = shadow_page_for(page);
+	origin = origin_page_for(page);
+
+	if (initialized) {
+		__memset(page_address(shadow), 0, PAGE_SIZE * pages);
+		__memset(page_address(origin), 0, PAGE_SIZE * pages);
+		return;
+	}
+
+	/* Zero pages allocated by the runtime should also be initialized. */
+	if (kmsan_in_runtime())
+		return;
+
+	__memset(page_address(shadow), -1, PAGE_SIZE * pages);
+	kmsan_enter_runtime();
+	handle = kmsan_save_stack_with_flags(flags, /*extra_bits*/ 0);
+	kmsan_leave_runtime();
+	/*
+	 * Addresses are page-aligned, pages are contiguous, so it's ok
+	 * to just fill the origin pages with @handle.
+	 */
+	for (int i = 0; i < PAGE_SIZE * pages / sizeof(handle); i++)
+		((depot_stack_handle_t *)page_address(origin))[i] = handle;
+}
+
+void kmsan_free_page(struct page *page, unsigned int order)
+{
+	if (!kmsan_enabled || kmsan_in_runtime())
+		return;
+	kmsan_enter_runtime();
+	kmsan_internal_poison_memory(page_address(page),
+				     PAGE_SIZE << compound_order(page),
+				     GFP_KERNEL,
+				     KMSAN_POISON_CHECK | KMSAN_POISON_FREE);
+	kmsan_leave_runtime();
+}
+
+void kmsan_vmap_pages_range_noflush(unsigned long start, unsigned long end,
+				    pgprot_t prot, struct page **pages,
+				    unsigned int page_shift)
+{
+	unsigned long shadow_start, origin_start, shadow_end, origin_end;
+	struct page **s_pages, **o_pages;
+	int nr, mapped;
+
+	if (!kmsan_enabled)
+		return;
+
+	shadow_start = vmalloc_meta((void *)start, KMSAN_META_SHADOW);
+	shadow_end = vmalloc_meta((void *)end, KMSAN_META_SHADOW);
+	if (!shadow_start)
+		return;
+
+	nr = (end - start) / PAGE_SIZE;
+	s_pages = kcalloc(nr, sizeof(*s_pages), GFP_KERNEL);
+	o_pages = kcalloc(nr, sizeof(*o_pages), GFP_KERNEL);
+	if (!s_pages || !o_pages)
+		goto ret;
+	for (int i = 0; i < nr; i++) {
+		s_pages[i] = shadow_page_for(pages[i]);
+		o_pages[i] = origin_page_for(pages[i]);
+	}
+	prot = __pgprot(pgprot_val(prot) | _PAGE_NX);
+	prot = PAGE_KERNEL;
+
+	origin_start = vmalloc_meta((void *)start, KMSAN_META_ORIGIN);
+	origin_end = vmalloc_meta((void *)end, KMSAN_META_ORIGIN);
+	kmsan_enter_runtime();
+	mapped = __vmap_pages_range_noflush(shadow_start, shadow_end, prot,
+					    s_pages, page_shift);
+	KMSAN_WARN_ON(mapped);
+	mapped = __vmap_pages_range_noflush(origin_start, origin_end, prot,
+					    o_pages, page_shift);
+	KMSAN_WARN_ON(mapped);
+	kmsan_leave_runtime();
+	flush_tlb_kernel_range(shadow_start, shadow_end);
+	flush_tlb_kernel_range(origin_start, origin_end);
+	flush_cache_vmap(shadow_start, shadow_end);
+	flush_cache_vmap(origin_start, origin_end);
+
+ret:
+	kfree(s_pages);
+	kfree(o_pages);
+}

mm/memory.c

···
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
 #include <linux/memremap.h>
+#include <linux/kmsan.h>
 #include <linux/ksm.h>
 #include <linux/rmap.h>
 #include <linux/export.h>
···
 			delayacct_wpcopy_end();
 			return 0;
 		}
+		kmsan_copy_page_meta(new_page, old_page);
 	}
 
 	if (mem_cgroup_charge(page_folio(new_page), mm, GFP_KERNEL))

mm/page_alloc.c

···
 #include <linux/compiler.h>
 #include <linux/kernel.h>
 #include <linux/kasan.h>
+#include <linux/kmsan.h>
 #include <linux/module.h>
 #include <linux/suspend.h>
 #include <linux/pagevec.h>
···
 	VM_BUG_ON_PAGE(PageTail(page), page);
 
 	trace_mm_page_free(page, order);
+	kmsan_free_page(page, order);
 
 	if (unlikely(PageHWPoison(page)) && !order) {
 		/*
···
 /*
  * Allocate a page from the given zone. Use pcplists for order-0 allocations.
  */
+
+/*
+ * Do not instrument rmqueue() with KMSAN. This function may call
+ * __msan_poison_alloca() through a call to set_pfnblock_flags_mask().
+ * If __msan_poison_alloca() attempts to allocate pages for the stack depot, it
+ * may call rmqueue() again, which will result in a deadlock.
+ */
+__no_sanitize_memory
 static inline
 struct page *rmqueue(struct zone *preferred_zone,
 			struct zone *zone, unsigned int order,
···
 	}
 
 	trace_mm_page_alloc(page, order, alloc_gfp, ac.migratetype);
+	kmsan_alloc_page(page, order, alloc_gfp);
 
 	return page;
 }

mm/vmalloc.c

···
 	err = vmap_range_noflush(addr, end, phys_addr, pgprot_nx(prot),
 				 ioremap_max_page_shift);
 	flush_cache_vmap(addr, end);
+	if (!err)
+		kmsan_ioremap_page_range(addr, end, phys_addr, prot,
+					 ioremap_max_page_shift);
 	return err;
 }
 
···
  *
  * This is an internal function only. Do not use outside mm/.
  */
-void vunmap_range_noflush(unsigned long start, unsigned long end)
+void __vunmap_range_noflush(unsigned long start, unsigned long end)
 {
 	unsigned long next;
 	pgd_t *pgd;
···
 
 	if (mask & ARCH_PAGE_TABLE_SYNC_MASK)
 		arch_sync_kernel_mappings(start, end);
+}
+
+void vunmap_range_noflush(unsigned long start, unsigned long end)
+{
+	kmsan_vunmap_range_noflush(start, end);
+	__vunmap_range_noflush(start, end);
 }
 
 /**
···
  *
  * This is an internal function only. Do not use outside mm/.
  */
-int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
+int __vmap_pages_range_noflush(unsigned long addr, unsigned long end,
 		pgprot_t prot, struct page **pages, unsigned int page_shift)
 {
 	unsigned int i, nr = (end - addr) >> PAGE_SHIFT;
···
 	}
 
 	return 0;
+}
+
+int vmap_pages_range_noflush(unsigned long addr, unsigned long end,
+		pgprot_t prot, struct page **pages, unsigned int page_shift)
+{
+	kmsan_vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
+	return __vmap_pages_range_noflush(addr, end, prot, pages, page_shift);
 }
 
 /**