| Message ID | 1427208544-8232-3-git-send-email-a.ryabinin@samsung.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Andrey, On Tue, Mar 24, 2015 at 05:49:04PM +0300, Andrey Ryabinin wrote: > diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig > index 4085df1..10bbd71 100644 > --- a/arch/arm64/Kconfig > +++ b/arch/arm64/Kconfig > @@ -41,6 +41,7 @@ config ARM64 > select HAVE_ARCH_AUDITSYSCALL > select HAVE_ARCH_BITREVERSE > select HAVE_ARCH_JUMP_LABEL > + select HAVE_ARCH_KASAN if SPARSEMEM_VMEMMAP > select HAVE_ARCH_KGDB > select HAVE_ARCH_SECCOMP_FILTER > select HAVE_ARCH_TRACEHOOK > @@ -116,6 +117,12 @@ config GENERIC_CSUM > config GENERIC_CALIBRATE_DELAY > def_bool y > > +config KASAN_SHADOW_OFFSET > + hex > + default 0xdfff200000000000 if ARM64_VA_BITS_48 > + default 0xdffffc8000000000 if ARM64_VA_BITS_42 > + default 0xdfffff9000000000 if ARM64_VA_BITS_39 Can we compute these at build time in some C header? Or they need to be passed to gcc when compiling the kernel so that it generates the right instrumentation? I'm not familiar with KASan but is the offset address supposed to be accessible? The addresses encoded above would always generate a fault (level 0 / address size fault). > diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h > index bd5db28..f5ce010 100644 > --- a/arch/arm64/include/asm/pgtable.h > +++ b/arch/arm64/include/asm/pgtable.h > @@ -40,7 +40,7 @@ > * fixed mappings and modules > */ > #define VMEMMAP_SIZE ALIGN((1UL << (VA_BITS - PAGE_SHIFT)) * sizeof(struct page), PUD_SIZE) > -#define VMALLOC_START (UL(0xffffffffffffffff) << VA_BITS) > +#define VMALLOC_START ((UL(0xffffffffffffffff) << VA_BITS) + (UL(1) << (VA_BITS - 3))) I assume that's where you want to make room for KASan? Some comments and macros would be useful for why this is needed and how it is calculated. It also needs to be disabled when KASan is not enabled. > diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S > index 51c9811..1a99e95 100644 > --- a/arch/arm64/kernel/head.S > +++ b/arch/arm64/kernel/head.S > @@ -482,6 +482,9 @@ __mmap_switched: > str_l x21, __fdt_pointer, x5 // Save FDT pointer > str_l x24, memstart_addr, x6 // Save PHYS_OFFSET > mov x29, #0 > +#ifdef CONFIG_KASAN > + b kasan_early_init > +#endif Nitpick: tab between b and kasan_early_init. > diff --git a/arch/arm64/mm/kasan_init.c b/arch/arm64/mm/kasan_init.c > new file mode 100644 > index 0000000..df537da > --- /dev/null > +++ b/arch/arm64/mm/kasan_init.c > @@ -0,0 +1,211 @@ > +#include <linux/kasan.h> > +#include <linux/kernel.h> > +#include <linux/memblock.h> > +#include <linux/start_kernel.h> > + > +#include <asm/page.h> > +#include <asm/pgtable.h> > +#include <asm/tlbflush.h> > + > +static char kasan_zero_page[PAGE_SIZE] __page_aligned_bss; Can we not use the system's zero_page or it's not initialised yet? > +static pgd_t tmp_page_table[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); > + > +#if CONFIG_PGTABLE_LEVELS > 3 > +static pud_t kasan_zero_pud[PTRS_PER_PUD] __page_aligned_bss; > +#endif > +#if CONFIG_PGTABLE_LEVELS > 2 > +static pmd_t kasan_zero_pmd[PTRS_PER_PMD] __page_aligned_bss; > +#endif > +static pte_t kasan_zero_pte[PTRS_PER_PTE] __page_aligned_bss; > + > +static void __init init_kasan_page_tables(void) > +{ > + int i; > + > +#if CONFIG_PGTABLE_LEVELS > 3 > + for (i = 0; i < PTRS_PER_PUD; i++) > + set_pud(&kasan_zero_pud[i], __pud(__pa(kasan_zero_pmd) > + | PAGE_KERNEL)); > +#endif > +#if CONFIG_PGTABLE_LEVELS > 2 > + for (i = 0; i < PTRS_PER_PMD; i++) > + set_pmd(&kasan_zero_pmd[i], __pmd(__pa(kasan_zero_pte) > + | PAGE_KERNEL)); > +#endif These don't look right. You are setting page attributes on table entries. You should use the standard pmd_populate etc. macros here, see early_fixmap_init() as an example. > + for (i = 0; i < PTRS_PER_PTE; i++) > + set_pte(&kasan_zero_pte[i], __pte(__pa(kasan_zero_page) > + | PAGE_KERNEL)); PAGE_KERNEL is pgprot_t, so you mix the types here. Just do something like: set_pte(..., pfn_pte(zero_pfn, PAGE_KERNEL_RO)); (shouldn't it be read-only?) > +void __init kasan_map_early_shadow(pgd_t *pgdp) > +{ > + int i; > + unsigned long start = KASAN_SHADOW_START; > + unsigned long end = KASAN_SHADOW_END; > + pgd_t pgd; > + > +#if CONFIG_PGTABLE_LEVELS > 3 > + pgd = __pgd(__pa(kasan_zero_pud) | PAGE_KERNEL); > +#elif CONFIG_PGTABLE_LEVELS > 2 > + pgd = __pgd(__pa(kasan_zero_pmd) | PAGE_KERNEL); > +#else > + pgd = __pgd(__pa(kasan_zero_pte) | PAGE_KERNEL); > +#endif > + > + for (i = pgd_index(start); start < end; i++) { > + set_pgd(&pgdp[i], pgd); > + start += PGDIR_SIZE; > + } > +} Same problem as above with PAGE_KERNEL. You should just use pgd_populate(). > + > +void __init kasan_early_init(void) > +{ > + init_kasan_page_tables(); > + kasan_map_early_shadow(swapper_pg_dir); > + kasan_map_early_shadow(idmap_pg_dir); > + flush_tlb_all(); > + start_kernel(); > +} Why do you need to map the kasan page tables into the idmap? > + > +static void __init clear_pgds(unsigned long start, > + unsigned long end) > +{ > + for (; start && start < end; start += PGDIR_SIZE) > + set_pgd(pgd_offset_k(start), __pgd(0)); > +} We have dedicated pgd_clear() macro. > + > +static int __init zero_pte_populate(pmd_t *pmd, unsigned long addr, > + unsigned long end) > +{ > + pte_t *pte = pte_offset_kernel(pmd, addr); > + > + while (addr + PAGE_SIZE <= end) { > + set_pte(pte, __pte(__pa(kasan_zero_page) > + | PAGE_KERNEL_RO)); See above for a pfn_pte() usage. > + addr += PAGE_SIZE; > + pte = pte_offset_kernel(pmd, addr); > + } > + return 0; > +} > + > +static int __init zero_pmd_populate(pud_t *pud, unsigned long addr, > + unsigned long end) > +{ > + int ret = 0; > + pmd_t *pmd = pmd_offset(pud, addr); > + > + while (IS_ALIGNED(addr, PMD_SIZE) && addr + PMD_SIZE <= end) { > + set_pmd(pmd, __pmd(__pa(kasan_zero_pte) > + | PAGE_KERNEL_RO)); > + addr += PMD_SIZE; > + pmd++; > + } > + > + if (addr < end) { > + if (pmd_none(*pmd)) { > + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); > + if (!p) > + return -ENOMEM; > + set_pmd(pmd, __pmd(__pa(p) | PAGE_KERNEL)); > + } > + ret = zero_pte_populate(pmd, addr, end); > + } > + return ret; > +} > + > +static int __init zero_pud_populate(pgd_t *pgd, unsigned long addr, > + unsigned long end) > +{ > + int ret = 0; > + pud_t *pud = pud_offset(pgd, addr); > + > +#if CONFIG_PGTABLE_LEVELS > 2 > + while (IS_ALIGNED(addr, PUD_SIZE) && addr + PUD_SIZE <= end) { > + set_pud(pud, __pud(__pa(kasan_zero_pmd) > + | PAGE_KERNEL_RO)); > + addr += PUD_SIZE; > + pud++; > + } > +#endif > + > + if (addr < end) { > + if (pud_none(*pud)) { > + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); > + if (!p) > + return -ENOMEM; > + set_pud(pud, __pud(__pa(p) | PAGE_KERNEL)); > + } > + ret = zero_pmd_populate(pud, addr, end); > + } > + return ret; > +} > + > +static int __init zero_pgd_populate(unsigned long addr, unsigned long end) > +{ > + int ret = 0; > + pgd_t *pgd = pgd_offset_k(addr); > + > +#if CONFIG_PGTABLE_LEVELS > 3 > + while (IS_ALIGNED(addr, PGDIR_SIZE) && addr + PGDIR_SIZE <= end) { > + set_pgd(pgd, __pgd(__pa(kasan_zero_pud) > + | PAGE_KERNEL_RO)); > + addr += PGDIR_SIZE; > + pgd++; > + } > +#endif All these PAGE_KERNEL_RO on table entries are wrong. Please use the standard pgd/pud/pmd_populate macros. As for the while loops above, we have a standard way to avoid the #ifdef's by using pgd_addr_end() etc. See __create_mapping() as an example, there are a few others throughout the kernel. > + > + if (addr < end) { > + if (pgd_none(*pgd)) { > + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); > + if (!p) > + return -ENOMEM; > + set_pgd(pgd, __pgd(__pa(p) | PAGE_KERNEL)); I'm just commenting here but it applies to the previous functions. You may be able to use functions like vmmemap_pgd_populate() which look very similar (and they also use pgd_populate instead of the set_pgd). > +static void cpu_set_ttbr1(unsigned long ttbr1) > +{ > + asm( > + " msr ttbr1_el1, %0\n" > + " isb" > + : > + : "r" (ttbr1)); > +} > + > +void __init kasan_init(void) > +{ > + struct memblock_region *reg; > + > + memcpy(tmp_page_table, swapper_pg_dir, sizeof(tmp_page_table)); > + cpu_set_ttbr1(__pa(tmp_page_table)); Why is this needed? The code lacks comments in several places but here I couldn't figure out what the point is. > + > + clear_pgds(KASAN_SHADOW_START, KASAN_SHADOW_END); > + > + populate_zero_shadow(KASAN_SHADOW_START, > + (unsigned long)kasan_mem_to_shadow((void *)MODULES_VADDR)); > + > + for_each_memblock(memory, reg) { > + void *start = (void *)__phys_to_virt(reg->base); > + void *end = (void *)__phys_to_virt(reg->base + reg->size); > + > + if (start >= end) > + break; > + > + vmemmap_populate((unsigned long)kasan_mem_to_shadow(start), > + (unsigned long)kasan_mem_to_shadow(end), > + pfn_to_nid(virt_to_pfn(start))); > + } > + > + memset(kasan_zero_page, 0, PAGE_SIZE); > + cpu_set_ttbr1(__pa(swapper_pg_dir)); > + init_task.kasan_depth = 0; > +}
On 04/01/2015 03:28 PM, Catalin Marinas wrote: > Hi Andrey, > Hi Catalin, > On Tue, Mar 24, 2015 at 05:49:04PM +0300, Andrey Ryabinin wrote: >> diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig >> index 4085df1..10bbd71 100644 >> --- a/arch/arm64/Kconfig >> +++ b/arch/arm64/Kconfig >> @@ -41,6 +41,7 @@ config ARM64 >> select HAVE_ARCH_AUDITSYSCALL >> select HAVE_ARCH_BITREVERSE >> select HAVE_ARCH_JUMP_LABEL >> + select HAVE_ARCH_KASAN if SPARSEMEM_VMEMMAP >> select HAVE_ARCH_KGDB >> select HAVE_ARCH_SECCOMP_FILTER >> select HAVE_ARCH_TRACEHOOK >> @@ -116,6 +117,12 @@ config GENERIC_CSUM >> config GENERIC_CALIBRATE_DELAY >> def_bool y >> >> +config KASAN_SHADOW_OFFSET >> + hex >> + default 0xdfff200000000000 if ARM64_VA_BITS_48 >> + default 0xdffffc8000000000 if ARM64_VA_BITS_42 >> + default 0xdfffff9000000000 if ARM64_VA_BITS_39 > > Can we compute these at build time in some C header? Or they need to be > passed to gcc when compiling the kernel so that it generates the right > instrumentation? > Correct, this value passed to GCC. > I'm not familiar with KASan but is the offset address supposed to be > accessible? The addresses encoded above would always generate a fault > (level 0 / address size fault). > It's fine. KASAN_SHADOW_OFFSET address is shadow address that corresponds to 0 address. So KASAN_SHADOW_OFFSET could be dereferenced only if we have NULL-ptr derefernce in kernel. Shadow for kernel addresses starts from KASAN_SHADOW_START constant, which is defined in arch/arm64/include/asm/kasan.h. But since I forgot to 'git add' that file it's not present in this patch. arch/arm64/include/asm/kasan.h: /* * Compiler uses shadow offset assuming that addresses start * from 0. Kernel addresses don't start from 0, so shadow * for kernel really starts from 'compiler's shadow offset' + * ('kernel address space start' >> KASAN_SHADOW_SCALE_SHIFT) */ #define KASAN_SHADOW_START (KASAN_SHADOW_OFFSET + \ ((UL(0xffffffffffffffff) << (VA_BITS)) >> 3)) #define KASAN_SHADOW_END (KASAN_SHADOW_START + (1ULL << (VA_BITS - 3))) >> diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h >> index bd5db28..f5ce010 100644 >> --- a/arch/arm64/include/asm/pgtable.h >> +++ b/arch/arm64/include/asm/pgtable.h >> @@ -40,7 +40,7 @@ >> * fixed mappings and modules >> */ >> #define VMEMMAP_SIZE ALIGN((1UL << (VA_BITS - PAGE_SHIFT)) * sizeof(struct page), PUD_SIZE) >> -#define VMALLOC_START (UL(0xffffffffffffffff) << VA_BITS) >> +#define VMALLOC_START ((UL(0xffffffffffffffff) << VA_BITS) + (UL(1) << (VA_BITS - 3))) > > I assume that's where you want to make room for KASan? Some comments and > macros would be useful for why this is needed and how it is calculated. > It also needs to be disabled when KASan is not enabled. > Ok. >> diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S >> index 51c9811..1a99e95 100644 >> --- a/arch/arm64/kernel/head.S >> +++ b/arch/arm64/kernel/head.S >> @@ -482,6 +482,9 @@ __mmap_switched: >> str_l x21, __fdt_pointer, x5 // Save FDT pointer >> str_l x24, memstart_addr, x6 // Save PHYS_OFFSET >> mov x29, #0 >> +#ifdef CONFIG_KASAN >> + b kasan_early_init >> +#endif > > Nitpick: tab between b and kasan_early_init. > >> diff --git a/arch/arm64/mm/kasan_init.c b/arch/arm64/mm/kasan_init.c >> new file mode 100644 >> index 0000000..df537da >> --- /dev/null >> +++ b/arch/arm64/mm/kasan_init.c >> @@ -0,0 +1,211 @@ >> +#include <linux/kasan.h> >> +#include <linux/kernel.h> >> +#include <linux/memblock.h> >> +#include <linux/start_kernel.h> >> + >> +#include <asm/page.h> >> +#include <asm/pgtable.h> >> +#include <asm/tlbflush.h> >> + >> +static char kasan_zero_page[PAGE_SIZE] __page_aligned_bss; > > Can we not use the system's zero_page or it's not initialised yet? > System's zero page allocated in paging_init() and that is too late. But I could put system's zero page into bss and use it here if you ok with this. >> +static pgd_t tmp_page_table[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); >> + >> +#if CONFIG_PGTABLE_LEVELS > 3 >> +static pud_t kasan_zero_pud[PTRS_PER_PUD] __page_aligned_bss; >> +#endif >> +#if CONFIG_PGTABLE_LEVELS > 2 >> +static pmd_t kasan_zero_pmd[PTRS_PER_PMD] __page_aligned_bss; >> +#endif >> +static pte_t kasan_zero_pte[PTRS_PER_PTE] __page_aligned_bss; >> + >> +static void __init init_kasan_page_tables(void) >> +{ >> + int i; >> + >> +#if CONFIG_PGTABLE_LEVELS > 3 >> + for (i = 0; i < PTRS_PER_PUD; i++) >> + set_pud(&kasan_zero_pud[i], __pud(__pa(kasan_zero_pmd) >> + | PAGE_KERNEL)); >> +#endif >> +#if CONFIG_PGTABLE_LEVELS > 2 >> + for (i = 0; i < PTRS_PER_PMD; i++) >> + set_pmd(&kasan_zero_pmd[i], __pmd(__pa(kasan_zero_pte) >> + | PAGE_KERNEL)); >> +#endif > > These don't look right. You are setting page attributes on table > entries. You should use the standard pmd_populate etc. macros here, see > early_fixmap_init() as an example. > Right. Will fix. >> + for (i = 0; i < PTRS_PER_PTE; i++) >> + set_pte(&kasan_zero_pte[i], __pte(__pa(kasan_zero_page) >> + | PAGE_KERNEL)); > > PAGE_KERNEL is pgprot_t, so you mix the types here. Just do something > like: > > set_pte(..., pfn_pte(zero_pfn, PAGE_KERNEL_RO)); > > (shouldn't it be read-only?) > It should be read-only, but only after kasan_init(). It should be writable earlier because stack instrumentation writes to shadow memory. In function's prologue compiler writes to shadow to poison redzones around stack variables. >> +void __init kasan_map_early_shadow(pgd_t *pgdp) >> +{ >> + int i; >> + unsigned long start = KASAN_SHADOW_START; >> + unsigned long end = KASAN_SHADOW_END; >> + pgd_t pgd; >> + >> +#if CONFIG_PGTABLE_LEVELS > 3 >> + pgd = __pgd(__pa(kasan_zero_pud) | PAGE_KERNEL); >> +#elif CONFIG_PGTABLE_LEVELS > 2 >> + pgd = __pgd(__pa(kasan_zero_pmd) | PAGE_KERNEL); >> +#else >> + pgd = __pgd(__pa(kasan_zero_pte) | PAGE_KERNEL); >> +#endif >> + >> + for (i = pgd_index(start); start < end; i++) { >> + set_pgd(&pgdp[i], pgd); >> + start += PGDIR_SIZE; >> + } >> +} > > Same problem as above with PAGE_KERNEL. You should just use > pgd_populate(). > Ok >> + >> +void __init kasan_early_init(void) >> +{ >> + init_kasan_page_tables(); >> + kasan_map_early_shadow(swapper_pg_dir); >> + kasan_map_early_shadow(idmap_pg_dir); >> + flush_tlb_all(); >> + start_kernel(); >> +} > > Why do you need to map the kasan page tables into the idmap? > I don't need it. This is some left-over that should be removed. >> + >> +static void __init clear_pgds(unsigned long start, >> + unsigned long end) >> +{ >> + for (; start && start < end; start += PGDIR_SIZE) >> + set_pgd(pgd_offset_k(start), __pgd(0)); >> +} > > We have dedicated pgd_clear() macro. > I need to remove references to kasan_zero_p* tables from swapper_pg_dir so pgd_clear() will not work here because it's noop on CONFIG_PGTABLE_LEVEL <= 3. [...] >> +static int __init zero_pgd_populate(unsigned long addr, unsigned long end) >> +{ >> + int ret = 0; >> + pgd_t *pgd = pgd_offset_k(addr); >> + >> +#if CONFIG_PGTABLE_LEVELS > 3 >> + while (IS_ALIGNED(addr, PGDIR_SIZE) && addr + PGDIR_SIZE <= end) { >> + set_pgd(pgd, __pgd(__pa(kasan_zero_pud) >> + | PAGE_KERNEL_RO)); >> + addr += PGDIR_SIZE; >> + pgd++; >> + } >> +#endif > > All these PAGE_KERNEL_RO on table entries are wrong. Please use the > standard pgd/pud/pmd_populate macros. > > As for the while loops above, we have a standard way to avoid the > #ifdef's by using pgd_addr_end() etc. See __create_mapping() as an > example, there are a few others throughout the kernel. > Ok. [...] >> +static void cpu_set_ttbr1(unsigned long ttbr1) >> +{ >> + asm( >> + " msr ttbr1_el1, %0\n" >> + " isb" >> + : >> + : "r" (ttbr1)); >> +} >> + >> +void __init kasan_init(void) >> +{ >> + struct memblock_region *reg; >> + >> + memcpy(tmp_page_table, swapper_pg_dir, sizeof(tmp_page_table)); >> + cpu_set_ttbr1(__pa(tmp_page_table)); > > Why is this needed? The code lacks comments in several places but here I > couldn't figure out what the point is. > To setup shadow memory properly we need to unmap early shadow first (clear_pgds() in next line). But instrumented kernel cannot run with unmaped shadow so this temporary page table with early shadow until setting up shadow in swapper_pg_dir will be finished. I'll add comment about this here. >> + >> + clear_pgds(KASAN_SHADOW_START, KASAN_SHADOW_END); >> + >> + populate_zero_shadow(KASAN_SHADOW_START, >> + (unsigned long)kasan_mem_to_shadow((void *)MODULES_VADDR)); >> + >> + for_each_memblock(memory, reg) { >> + void *start = (void *)__phys_to_virt(reg->base); >> + void *end = (void *)__phys_to_virt(reg->base + reg->size); >> + >> + if (start >= end) >> + break; >> + >> + vmemmap_populate((unsigned long)kasan_mem_to_shadow(start), >> + (unsigned long)kasan_mem_to_shadow(end), >> + pfn_to_nid(virt_to_pfn(start))); >> + } >> + >> + memset(kasan_zero_page, 0, PAGE_SIZE); >> + cpu_set_ttbr1(__pa(swapper_pg_dir)); >> + init_task.kasan_depth = 0; >> +} > Thank you for detailed review.
On Tuesday 24 March 2015 17:49:04 Andrey Ryabinin wrote: > arch/arm64/mm/kasan_init.c | 211 +++++++++++++++++++++++++++++++++++ > Just one very high-level question: as this code is clearly derived from the x86 version and nontrivial, could we move most of it out of arch/{x86,arm64} into mm/kasan/init.c and have the rest in some header file? Arnd
On 04/09/2015 11:17 PM, Arnd Bergmann wrote: > On Tuesday 24 March 2015 17:49:04 Andrey Ryabinin wrote: >> arch/arm64/mm/kasan_init.c | 211 +++++++++++++++++++++++++++++++++++ >> > > Just one very high-level question: as this code is clearly derived from > the x86 version and nontrivial, could we move most of it out of > arch/{x86,arm64} into mm/kasan/init.c and have the rest in some header > file? > I think most of this could be moved out from arch code, but not everything. E.g. kasan_init() function is too arch-specific.
On Friday 10 April 2015 13:48:52 Andrey Ryabinin wrote: > On 04/09/2015 11:17 PM, Arnd Bergmann wrote: > > On Tuesday 24 March 2015 17:49:04 Andrey Ryabinin wrote: > >> arch/arm64/mm/kasan_init.c | 211 +++++++++++++++++++++++++++++++++++ > >> > > > > Just one very high-level question: as this code is clearly derived from > > the x86 version and nontrivial, could we move most of it out of > > arch/{x86,arm64} into mm/kasan/init.c and have the rest in some header > > file? > > > > I think most of this could be moved out from arch code, but not everything. > E.g. kasan_init() function is too arch-specific. Right, makes sense. So presumably, populate_zero_shadow could become a global function by another name, and possibly also handle registering the die handler, so you can call it from an architecture specific kasan_init() function, right? Arnd
On 04/10/2015 04:02 PM, Arnd Bergmann wrote: > On Friday 10 April 2015 13:48:52 Andrey Ryabinin wrote: >> On 04/09/2015 11:17 PM, Arnd Bergmann wrote: >>> On Tuesday 24 March 2015 17:49:04 Andrey Ryabinin wrote: >>>> arch/arm64/mm/kasan_init.c | 211 +++++++++++++++++++++++++++++++++++ >>>> >>> >>> Just one very high-level question: as this code is clearly derived from >>> the x86 version and nontrivial, could we move most of it out of >>> arch/{x86,arm64} into mm/kasan/init.c and have the rest in some header >>> file? >>> >> >> I think most of this could be moved out from arch code, but not everything. >> E.g. kasan_init() function is too arch-specific. > > Right, makes sense. So presumably, populate_zero_shadow could become a global > function by another name, and possibly also handle registering the die > handler, so you can call it from an architecture specific kasan_init() > function, right? > Yep, you are right. > Arnd >
>>> + pgd = __pgd(__pa(kasan_zero_pmd) | PAGE_KERNEL); >>> +#else >>> + pgd = __pgd(__pa(kasan_zero_pte) | PAGE_KERNEL); >>> +#endif >>> + >>> + for (i = pgd_index(start); start < end; i++) { >>> + set_pgd(&pgdp[i], pgd); >>> + start += PGDIR_SIZE; >>> + } >>> +} >> >> Same problem as above with PAGE_KERNEL. You should just use >> pgd_populate(). Any suggestion what the correct flag setting would be here for a 4K mapping? I tried fixing this by changing this to pud and setting the PMD_TYPE_TABLE flag for kasan_zero_pmd. However the MMU doesn't like it and I get a first level address translation fault. If you have any updated patches to share I'd be glad to try them out.
On 04/15/2015 05:37 AM, David Keitel wrote: >>>> + pgd = __pgd(__pa(kasan_zero_pmd) | PAGE_KERNEL); >>>> +#else >>>> + pgd = __pgd(__pa(kasan_zero_pte) | PAGE_KERNEL); >>>> +#endif >>>> + >>>> + for (i = pgd_index(start); start < end; i++) { >>>> + set_pgd(&pgdp[i], pgd); >>>> + start += PGDIR_SIZE; >>>> + } >>>> +} >>> >>> Same problem as above with PAGE_KERNEL. You should just use >>> pgd_populate(). > > Any suggestion what the correct flag setting would be here for a 4K mapping? > > I tried fixing this by changing this to pud and setting the PMD_TYPE_TABLE flag for kasan_zero_pmd. However the MMU doesn't like it and I get a first level address translation fault. > > If you have any updated patches to share I'd be glad to try them out. > Sorry, I didn't have much time on work on this yet. I've pushed the most fresh thing that I have in git: git://github.com/aryabinin/linux.git kasan/arm64v1 It's the same patches with two simple but important fixes on top of it.
On 04/15/2015 11:04 AM, Andrey Ryabinin wrote: > I've pushed the most fresh thing that I have in git: > git://github.com/aryabinin/linux.git kasan/arm64v1 > > It's the same patches with two simple but important fixes on top of it. Thanks, the two commits do fix compilation issues that I've had worked around to get to my mapping question. I've addressed the mapping problem using __create_page_tables in arch/arm64/head.S as an example. The next roadblock I hit was running into kasan_report_error calls in cgroups_early_init. After a short investigation it does seem to be a false positive due the the kasan_zero_page size and tracking bytes being reused for different memory regions. I worked around that by enabling kasan error reporting only after the kasan_init is run. This let me get to the shell with some real KAsan reports along the way. There were some other fixes and hacks to get there. I'll backtrack to evaluate which ones warrant an RFC. - David
On 04/17/2015 09:39 PM, David Keitel wrote: > On 04/15/2015 11:04 AM, Andrey Ryabinin wrote: >> I've pushed the most fresh thing that I have in git: >> git://github.com/aryabinin/linux.git kasan/arm64v1 >> >> It's the same patches with two simple but important fixes on top of it. > > Thanks, the two commits do fix compilation issues that I've had worked around to get to my mapping question. > > I've addressed the mapping problem using __create_page_tables in arch/arm64/head.S as an example. > > The next roadblock I hit was running into kasan_report_error calls in cgroups_early_init. After a short investigation it does seem to be a false positive due the the kasan_zero_page size and tracking bytes being reused for different memory regions. > > I worked around that by enabling kasan error reporting only after the kasan_init is run. This let me get to the shell with some real KAsan reports along the way. Reporting already disabled before kasan_init() and the last thing that kasan_init() is enable error reports. So, how did you managed to get kasan's report before kasan_init()? > There were some other fixes and hacks to get there. I'll backtrack to evaluate which ones warrant an RFC. > > - David >
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 4085df1..10bbd71 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -41,6 +41,7 @@ config ARM64 select HAVE_ARCH_AUDITSYSCALL select HAVE_ARCH_BITREVERSE select HAVE_ARCH_JUMP_LABEL + select HAVE_ARCH_KASAN if SPARSEMEM_VMEMMAP select HAVE_ARCH_KGDB select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_TRACEHOOK @@ -116,6 +117,12 @@ config GENERIC_CSUM config GENERIC_CALIBRATE_DELAY def_bool y +config KASAN_SHADOW_OFFSET + hex + default 0xdfff200000000000 if ARM64_VA_BITS_48 + default 0xdffffc8000000000 if ARM64_VA_BITS_42 + default 0xdfffff9000000000 if ARM64_VA_BITS_39 + config ZONE_DMA def_bool y diff --git a/arch/arm64/include/asm/pgtable.h b/arch/arm64/include/asm/pgtable.h index bd5db28..f5ce010 100644 --- a/arch/arm64/include/asm/pgtable.h +++ b/arch/arm64/include/asm/pgtable.h @@ -40,7 +40,7 @@ * fixed mappings and modules */ #define VMEMMAP_SIZE ALIGN((1UL << (VA_BITS - PAGE_SHIFT)) * sizeof(struct page), PUD_SIZE) -#define VMALLOC_START (UL(0xffffffffffffffff) << VA_BITS) +#define VMALLOC_START ((UL(0xffffffffffffffff) << VA_BITS) + (UL(1) << (VA_BITS - 3))) #define VMALLOC_END (PAGE_OFFSET - PUD_SIZE - VMEMMAP_SIZE - SZ_64K) #define vmemmap ((struct page *)(VMALLOC_END + SZ_64K)) @@ -73,6 +73,7 @@ extern void __pgd_error(const char *file, int line, unsigned long val); #define PAGE_KERNEL __pgprot(_PAGE_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY | PTE_WRITE) #define PAGE_KERNEL_EXEC __pgprot(_PAGE_DEFAULT | PTE_UXN | PTE_DIRTY | PTE_WRITE) +#define PAGE_KERNEL_RO __pgprot(_PAGE_DEFAULT | PTE_PXN | PTE_UXN | PTE_DIRTY) #define PAGE_HYP __pgprot(_PAGE_DEFAULT | PTE_HYP) #define PAGE_HYP_DEVICE __pgprot(PROT_DEVICE_nGnRE | PTE_HYP) diff --git a/arch/arm64/include/asm/string.h b/arch/arm64/include/asm/string.h index 64d2d48..bff522c 100644 --- a/arch/arm64/include/asm/string.h +++ b/arch/arm64/include/asm/string.h @@ -36,17 +36,33 @@ extern __kernel_size_t strnlen(const char *, __kernel_size_t); #define __HAVE_ARCH_MEMCPY extern void *memcpy(void *, const void *, __kernel_size_t); +extern void *__memcpy(void *, const void *, __kernel_size_t); #define __HAVE_ARCH_MEMMOVE extern void *memmove(void *, const void *, __kernel_size_t); +extern void *__memmove(void *, const void *, __kernel_size_t); #define __HAVE_ARCH_MEMCHR extern void *memchr(const void *, int, __kernel_size_t); #define __HAVE_ARCH_MEMSET extern void *memset(void *, int, __kernel_size_t); +extern void *__memset(void *, int, __kernel_size_t); #define __HAVE_ARCH_MEMCMP extern int memcmp(const void *, const void *, size_t); + +#if defined(CONFIG_KASAN) && !defined(__SANITIZE_ADDRESS__) + +/* + * For files that not instrumented (e.g. mm/slub.c) we + * should use not instrumented version of mem* functions. + */ + +#define memcpy(dst, src, len) __memcpy(dst, src, len) +#define memmove(dst, src, len) __memmove(dst, src, len) +#define memset(s, c, n) __memset(s, c, n) +#endif + #endif diff --git a/arch/arm64/include/asm/thread_info.h b/arch/arm64/include/asm/thread_info.h index 702e1e6..4a2c11e 100644 --- a/arch/arm64/include/asm/thread_info.h +++ b/arch/arm64/include/asm/thread_info.h @@ -24,10 +24,18 @@ #include <linux/compiler.h> #ifndef CONFIG_ARM64_64K_PAGES +#ifndef CONFIG_KASAN #define THREAD_SIZE_ORDER 2 +#else +#define THREAD_SIZE_ORDER 3 +#endif #endif +#ifndef CONFIG_KASAN #define THREAD_SIZE 16384 +#else +#define THREAD_SIZE 32768 +#endif #define THREAD_START_SP (THREAD_SIZE - 16) #ifndef __ASSEMBLY__ diff --git a/arch/arm64/kernel/head.S b/arch/arm64/kernel/head.S index 51c9811..1a99e95 100644 --- a/arch/arm64/kernel/head.S +++ b/arch/arm64/kernel/head.S @@ -482,6 +482,9 @@ __mmap_switched: str_l x21, __fdt_pointer, x5 // Save FDT pointer str_l x24, memstart_addr, x6 // Save PHYS_OFFSET mov x29, #0 +#ifdef CONFIG_KASAN + b kasan_early_init +#endif b start_kernel ENDPROC(__mmap_switched) diff --git a/arch/arm64/kernel/module.c b/arch/arm64/kernel/module.c index 67bf410..7d90c0f 100644 --- a/arch/arm64/kernel/module.c +++ b/arch/arm64/kernel/module.c @@ -21,6 +21,7 @@ #include <linux/bitops.h> #include <linux/elf.h> #include <linux/gfp.h> +#include <linux/kasan.h> #include <linux/kernel.h> #include <linux/mm.h> #include <linux/moduleloader.h> @@ -34,9 +35,18 @@ void *module_alloc(unsigned long size) { - return __vmalloc_node_range(size, 1, MODULES_VADDR, MODULES_END, - GFP_KERNEL, PAGE_KERNEL_EXEC, 0, - NUMA_NO_NODE, __builtin_return_address(0)); + void *p; + + p = __vmalloc_node_range(size, MODULE_ALIGN, MODULES_VADDR, MODULES_END, + GFP_KERNEL, PAGE_KERNEL_EXEC, 0, + NUMA_NO_NODE, __builtin_return_address(0)); + + if (p && (kasan_module_alloc(p, size) < 0)) { + vfree(p); + return NULL; + } + + return p; } enum aarch64_reloc_op { diff --git a/arch/arm64/kernel/setup.c b/arch/arm64/kernel/setup.c index 51ef972..f197f45 100644 --- a/arch/arm64/kernel/setup.c +++ b/arch/arm64/kernel/setup.c @@ -31,6 +31,7 @@ #include <linux/screen_info.h> #include <linux/init.h> #include <linux/kexec.h> +#include <linux/kasan.h> #include <linux/crash_dump.h> #include <linux/root_dev.h> #include <linux/clk-provider.h> @@ -396,6 +397,7 @@ void __init setup_arch(char **cmdline_p) arm64_memblock_init(); paging_init(); + kasan_init(); request_standard_resources(); early_ioremap_reset(); diff --git a/arch/arm64/lib/memcpy.S b/arch/arm64/lib/memcpy.S index 8a9a96d..845e40a 100644 --- a/arch/arm64/lib/memcpy.S +++ b/arch/arm64/lib/memcpy.S @@ -56,6 +56,8 @@ C_h .req x12 D_l .req x13 D_h .req x14 +.weak memcpy +ENTRY(__memcpy) ENTRY(memcpy) mov dst, dstin cmp count, #16 @@ -199,3 +201,4 @@ ENTRY(memcpy) b.ne .Ltail63 ret ENDPROC(memcpy) +ENDPROC(__memcpy) diff --git a/arch/arm64/lib/memmove.S b/arch/arm64/lib/memmove.S index 57b19ea..48074d2 100644 --- a/arch/arm64/lib/memmove.S +++ b/arch/arm64/lib/memmove.S @@ -57,12 +57,14 @@ C_h .req x12 D_l .req x13 D_h .req x14 +.weak memmove +ENTRY(__memmove) ENTRY(memmove) cmp dstin, src - b.lo memcpy + b.lo __memcpy add tmp1, src, count cmp dstin, tmp1 - b.hs memcpy /* No overlap. */ + b.hs __memcpy /* No overlap. */ add dst, dstin, count add src, src, count @@ -195,3 +197,4 @@ ENTRY(memmove) b.ne .Ltail63 ret ENDPROC(memmove) +ENDPROC(__memmove) diff --git a/arch/arm64/lib/memset.S b/arch/arm64/lib/memset.S index 7c72dfd..4ab2594 100644 --- a/arch/arm64/lib/memset.S +++ b/arch/arm64/lib/memset.S @@ -54,6 +54,8 @@ dst .req x8 tmp3w .req w9 tmp3 .req x9 +.weak memset +ENTRY(__memset) ENTRY(memset) mov dst, dstin /* Preserve return value. */ and A_lw, val, #255 @@ -214,3 +216,4 @@ ENTRY(memset) b.ne .Ltail_maybe_long ret ENDPROC(memset) +ENDPROC(__memset) diff --git a/arch/arm64/mm/Makefile b/arch/arm64/mm/Makefile index 773d37a..e17703c 100644 --- a/arch/arm64/mm/Makefile +++ b/arch/arm64/mm/Makefile @@ -4,3 +4,6 @@ obj-y := dma-mapping.o extable.o fault.o init.o \ context.o proc.o pageattr.o obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o obj-$(CONFIG_ARM64_PTDUMP) += dump.o + +KASAN_SANITIZE_kasan_init.o := n +obj-$(CONFIG_KASAN) += kasan_init.o diff --git a/arch/arm64/mm/kasan_init.c b/arch/arm64/mm/kasan_init.c new file mode 100644 index 0000000..df537da --- /dev/null +++ b/arch/arm64/mm/kasan_init.c @@ -0,0 +1,211 @@ +#include <linux/kasan.h> +#include <linux/kernel.h> +#include <linux/memblock.h> +#include <linux/start_kernel.h> + +#include <asm/page.h> +#include <asm/pgtable.h> +#include <asm/tlbflush.h> + +static char kasan_zero_page[PAGE_SIZE] __page_aligned_bss; +static pgd_t tmp_page_table[PTRS_PER_PGD] __initdata __aligned(PAGE_SIZE); + +#if CONFIG_PGTABLE_LEVELS > 3 +static pud_t kasan_zero_pud[PTRS_PER_PUD] __page_aligned_bss; +#endif +#if CONFIG_PGTABLE_LEVELS > 2 +static pmd_t kasan_zero_pmd[PTRS_PER_PMD] __page_aligned_bss; +#endif +static pte_t kasan_zero_pte[PTRS_PER_PTE] __page_aligned_bss; + +static void __init init_kasan_page_tables(void) +{ + int i; + +#if CONFIG_PGTABLE_LEVELS > 3 + for (i = 0; i < PTRS_PER_PUD; i++) + set_pud(&kasan_zero_pud[i], __pud(__pa(kasan_zero_pmd) + | PAGE_KERNEL)); +#endif +#if CONFIG_PGTABLE_LEVELS > 2 + for (i = 0; i < PTRS_PER_PMD; i++) + set_pmd(&kasan_zero_pmd[i], __pmd(__pa(kasan_zero_pte) + | PAGE_KERNEL)); +#endif + for (i = 0; i < PTRS_PER_PTE; i++) + set_pte(&kasan_zero_pte[i], __pte(__pa(kasan_zero_page) + | PAGE_KERNEL)); +} + +void __init kasan_map_early_shadow(pgd_t *pgdp) +{ + int i; + unsigned long start = KASAN_SHADOW_START; + unsigned long end = KASAN_SHADOW_END; + pgd_t pgd; + +#if CONFIG_PGTABLE_LEVELS > 3 + pgd = __pgd(__pa(kasan_zero_pud) | PAGE_KERNEL); +#elif CONFIG_PGTABLE_LEVELS > 2 + pgd = __pgd(__pa(kasan_zero_pmd) | PAGE_KERNEL); +#else + pgd = __pgd(__pa(kasan_zero_pte) | PAGE_KERNEL); +#endif + + for (i = pgd_index(start); start < end; i++) { + set_pgd(&pgdp[i], pgd); + start += PGDIR_SIZE; + } +} + +void __init kasan_early_init(void) +{ + init_kasan_page_tables(); + kasan_map_early_shadow(swapper_pg_dir); + kasan_map_early_shadow(idmap_pg_dir); + flush_tlb_all(); + start_kernel(); +} + +static void __init clear_pgds(unsigned long start, + unsigned long end) +{ + for (; start && start < end; start += PGDIR_SIZE) + set_pgd(pgd_offset_k(start), __pgd(0)); +} + +static int __init zero_pte_populate(pmd_t *pmd, unsigned long addr, + unsigned long end) +{ + pte_t *pte = pte_offset_kernel(pmd, addr); + + while (addr + PAGE_SIZE <= end) { + set_pte(pte, __pte(__pa(kasan_zero_page) + | PAGE_KERNEL_RO)); + addr += PAGE_SIZE; + pte = pte_offset_kernel(pmd, addr); + } + return 0; +} + +static int __init zero_pmd_populate(pud_t *pud, unsigned long addr, + unsigned long end) +{ + int ret = 0; + pmd_t *pmd = pmd_offset(pud, addr); + + while (IS_ALIGNED(addr, PMD_SIZE) && addr + PMD_SIZE <= end) { + set_pmd(pmd, __pmd(__pa(kasan_zero_pte) + | PAGE_KERNEL_RO)); + addr += PMD_SIZE; + pmd++; + } + + if (addr < end) { + if (pmd_none(*pmd)) { + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); + if (!p) + return -ENOMEM; + set_pmd(pmd, __pmd(__pa(p) | PAGE_KERNEL)); + } + ret = zero_pte_populate(pmd, addr, end); + } + return ret; +} + +static int __init zero_pud_populate(pgd_t *pgd, unsigned long addr, + unsigned long end) +{ + int ret = 0; + pud_t *pud = pud_offset(pgd, addr); + +#if CONFIG_PGTABLE_LEVELS > 2 + while (IS_ALIGNED(addr, PUD_SIZE) && addr + PUD_SIZE <= end) { + set_pud(pud, __pud(__pa(kasan_zero_pmd) + | PAGE_KERNEL_RO)); + addr += PUD_SIZE; + pud++; + } +#endif + + if (addr < end) { + if (pud_none(*pud)) { + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); + if (!p) + return -ENOMEM; + set_pud(pud, __pud(__pa(p) | PAGE_KERNEL)); + } + ret = zero_pmd_populate(pud, addr, end); + } + return ret; +} + +static int __init zero_pgd_populate(unsigned long addr, unsigned long end) +{ + int ret = 0; + pgd_t *pgd = pgd_offset_k(addr); + +#if CONFIG_PGTABLE_LEVELS > 3 + while (IS_ALIGNED(addr, PGDIR_SIZE) && addr + PGDIR_SIZE <= end) { + set_pgd(pgd, __pgd(__pa(kasan_zero_pud) + | PAGE_KERNEL_RO)); + addr += PGDIR_SIZE; + pgd++; + } +#endif + + if (addr < end) { + if (pgd_none(*pgd)) { + void *p = vmemmap_alloc_block(PAGE_SIZE, NUMA_NO_NODE); + if (!p) + return -ENOMEM; + set_pgd(pgd, __pgd(__pa(p) | PAGE_KERNEL)); + } + ret = zero_pud_populate(pgd, addr, end); + } + return ret; +} + +static void __init populate_zero_shadow(unsigned long start, unsigned long end) +{ + if (zero_pgd_populate(start, end)) + panic("kasan: unable to map zero shadow!"); +} + +static void cpu_set_ttbr1(unsigned long ttbr1) +{ + asm( + " msr ttbr1_el1, %0\n" + " isb" + : + : "r" (ttbr1)); +} + +void __init kasan_init(void) +{ + struct memblock_region *reg; + + memcpy(tmp_page_table, swapper_pg_dir, sizeof(tmp_page_table)); + cpu_set_ttbr1(__pa(tmp_page_table)); + + clear_pgds(KASAN_SHADOW_START, KASAN_SHADOW_END); + + populate_zero_shadow(KASAN_SHADOW_START, + (unsigned long)kasan_mem_to_shadow((void *)MODULES_VADDR)); + + for_each_memblock(memory, reg) { + void *start = (void *)__phys_to_virt(reg->base); + void *end = (void *)__phys_to_virt(reg->base + reg->size); + + if (start >= end) + break; + + vmemmap_populate((unsigned long)kasan_mem_to_shadow(start), + (unsigned long)kasan_mem_to_shadow(end), + pfn_to_nid(virt_to_pfn(start))); + } + + memset(kasan_zero_page, 0, PAGE_SIZE); + cpu_set_ttbr1(__pa(swapper_pg_dir)); + init_task.kasan_depth = 0; +}
This patch adds arch specific code for kernel address sanitizer (see Documentation/kasan.txt). 1/8 of kernel addresses reserved for shadow memory. There was no big enough hole for this, so virtual addresses for shadow were stolen from vmalloc area. At early boot stage the whole shadow region populated with just one physical page (kasan_zero_page). Later, this page reused as readonly zero shadow for some memory that KASan currently don't track (vmalloc). After mapping the physical memory, pages for shadow memory are allocated and mapped. KASan's stack instrumentation significantly increases stack's consumption, so CONFIG_KASAN doubles THREAD_SIZE. Functions like memset/memmove/memcpy do a lot of memory accesses. If bad pointer passed to one of these function it is important to catch this. Compiler's instrumentation cannot do this since these functions are written in assembly. KASan replaces memory functions with manually instrumented variants. Original functions declared as weak symbols so strong definitions in mm/kasan/kasan.c could replace them. Original functions have aliases with '__' prefix in name, so we could call non-instrumented variant if needed. Some files built without kasan instrumentation (e.g. mm/slub.c). Original mem* function replaced (via #define) with prefixed variants to disable memory access checks for such files. Signed-off-by: Andrey Ryabinin <a.ryabinin@samsung.com> --- arch/arm64/Kconfig | 7 ++ arch/arm64/include/asm/pgtable.h | 3 +- arch/arm64/include/asm/string.h | 16 +++ arch/arm64/include/asm/thread_info.h | 8 ++ arch/arm64/kernel/head.S | 3 + arch/arm64/kernel/module.c | 16 ++- arch/arm64/kernel/setup.c | 2 + arch/arm64/lib/memcpy.S | 3 + arch/arm64/lib/memmove.S | 7 +- arch/arm64/lib/memset.S | 3 + arch/arm64/mm/Makefile | 3 + arch/arm64/mm/kasan_init.c | 211 +++++++++++++++++++++++++++++++++++ 12 files changed, 276 insertions(+), 6 deletions(-) create mode 100644 arch/arm64/mm/kasan_init.c