> On Jul 25, 2019, at 12:49 AM, Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote: > >> On Thu, Jul 25, 2019 at 7:55 AM Daniel Axtens <dja@xxxxxxxxxx> wrote: >> >> In the case where KASAN directly allocates memory to back vmalloc >> space, don't map the early shadow page over it. >> >> Not mapping the early shadow page over the whole shadow space means >> that there are some pgds that are not populated on boot. Allow the >> vmalloc fault handler to also fault in vmalloc shadow as needed. >> >> Signed-off-by: Daniel Axtens <dja@xxxxxxxxxx> > > > Would it make things simpler if we pre-populate the top level page > tables for the whole vmalloc region? That would be > (16<<40)/4096/512/512*8 = 131072 bytes? > The check in vmalloc_fault in not really a big burden, so I am not > sure. Just brining as an option. I prefer pre-populating them. In particular, I have already spent far too much time debugging the awful explosions when the stack doesn’t have KASAN backing, and the vmap stack code is very careful to pre-populate the stack pgds — vmalloc_fault fundamentally can’t recover when the stack itself isn’t mapped. So the vmalloc_fault code, if it stays, needs some careful analysis to make sure it will actually survive all the various context switch cases. Or you can pre-populate it. > > Acked-by: Dmitry Vyukov <dvyukov@xxxxxxxxxx> > >> --- >> arch/x86/Kconfig | 1 + >> arch/x86/mm/fault.c | 13 +++++++++++++ >> arch/x86/mm/kasan_init_64.c | 10 ++++++++++ >> 3 files changed, 24 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/fault.c b/arch/x86/mm/fault.c >> index 6c46095cd0d9..d722230121c3 100644 >> --- a/arch/x86/mm/fault.c >> +++ b/arch/x86/mm/fault.c >> @@ -340,8 +340,21 @@ static noinline int vmalloc_fault(unsigned long address) >> pte_t *pte; >> >> /* Make sure we are in vmalloc area: */ >> +#ifndef CONFIG_KASAN_VMALLOC >> if (!(address >= VMALLOC_START && address < VMALLOC_END)) >> return -1; >> +#else >> + /* >> + * Some of the shadow mapping for the vmalloc area lives outside the >> + * pgds populated by kasan init. They are created dynamically and so >> + * we may need to fault them in. >> + * >> + * You can observe this with test_vmalloc's align_shift_alloc_test >> + */ >> + if (!((address >= VMALLOC_START && address < VMALLOC_END) || >> + (address >= KASAN_SHADOW_START && address < KASAN_SHADOW_END))) >> + return -1; >> +#endif >> >> /* >> * Copy kernel mappings over when needed. This can also >> diff --git a/arch/x86/mm/kasan_init_64.c b/arch/x86/mm/kasan_init_64.c >> index 296da58f3013..e2fe1c1b805c 100644 >> --- a/arch/x86/mm/kasan_init_64.c >> +++ b/arch/x86/mm/kasan_init_64.c >> @@ -352,9 +352,19 @@ 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. >> + */ >> +#ifdef CONFIG_KASAN_VMALLOC >> + 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 >> >> -- >> You received this message because you are subscribed to the Google Groups "kasan-dev" group. >> To unsubscribe from this group and stop receiving emails from it, send an email to kasan-dev+unsubscribe@xxxxxxxxxxxxxxxx. >> To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/20190725055503.19507-4-dja%40axtens.net.
