Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756470AbaDHJcL (ORCPT ); Tue, 8 Apr 2014 05:32:11 -0400 Received: from smarthost01d.mail.zen.net.uk ([212.23.1.7]:40188 "EHLO smarthost01d.mail.zen.net.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756369AbaDHJcH (ORCPT ); Tue, 8 Apr 2014 05:32:07 -0400 Message-ID: <1396949519.3567.14.camel@linaro1.home> Subject: Re: [PATCH 2/2] arm: use fixmap for text patching when text is RO From: "Jon Medhurst (Tixy)" To: Kees Cook Cc: Rabin Vincent , "linux-arm-kernel@lists.infradead.org" , LKML , Laura Abbott Date: Tue, 08 Apr 2014 10:31:59 +0100 In-Reply-To: References: <1396646870-29695-1-git-send-email-rabin@rab.in> <1396646870-29695-2-git-send-email-rabin@rab.in> <1396879071.3654.43.camel@linaro1.home> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.4.4-3 Mime-Version: 1.0 Content-Transfer-Encoding: 7bit X-Originating-smarthost01d-IP: [82.69.122.217] Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2014-04-07 at 11:04 -0700, Kees Cook wrote: > On Mon, Apr 7, 2014 at 6:57 AM, Jon Medhurst (Tixy) wrote: > > On Fri, 2014-04-04 at 23:27 +0200, Rabin Vincent wrote: > >> Use fixmaps for text patching when the kernel text is read-only, > >> inspired by x86. This makes jump labels and kprobes work with the > >> currently available CONFIG_DEBUG_SET_MODULE_RONX and the upcoming > >> CONFIG_DEBUG_RODATA options. > >> > >> Signed-off-by: Rabin Vincent > >> --- > >> arch/arm/include/asm/fixmap.h | 3 ++ > >> arch/arm/kernel/jump_label.c | 2 +- > >> arch/arm/kernel/patch.c | 66 ++++++++++++++++++++++++++++++++++++++----- > >> arch/arm/kernel/patch.h | 12 +++++++- > >> 4 files changed, 74 insertions(+), 9 deletions(-) > >> > >> diff --git a/arch/arm/include/asm/fixmap.h b/arch/arm/include/asm/fixmap.h > >> index 55ed076..79c1719 100644 > >> --- a/arch/arm/include/asm/fixmap.h > >> +++ b/arch/arm/include/asm/fixmap.h > >> @@ -18,6 +18,9 @@ > >> #define FIXADDR_TOP (FIXADDR_END - PAGE_SIZE) > >> > >> enum fixed_addresses { > >> + FIX_TEXT_POKE0, > >> + FIX_TEXT_POKE1, > >> + > >> FIX_KMAP_BEGIN, > >> FIX_KMAP_END = (FIXADDR_TOP - FIXADDR_START) >> PAGE_SHIFT, > >> __end_of_fixed_addresses > >> diff --git a/arch/arm/kernel/jump_label.c b/arch/arm/kernel/jump_label.c > >> index 4ce4f78..afeeb9e 100644 > >> --- a/arch/arm/kernel/jump_label.c > >> +++ b/arch/arm/kernel/jump_label.c > >> @@ -19,7 +19,7 @@ static void __arch_jump_label_transform(struct jump_entry *entry, > >> insn = arm_gen_nop(); > >> > >> if (is_static) > >> - __patch_text(addr, insn); > >> + __patch_text_early(addr, insn); > >> else > >> patch_text(addr, insn); > >> } > >> diff --git a/arch/arm/kernel/patch.c b/arch/arm/kernel/patch.c > >> index 07314af..761c5b6 100644 > >> --- a/arch/arm/kernel/patch.c > >> +++ b/arch/arm/kernel/patch.c > >> @@ -1,8 +1,11 @@ > >> #include > >> +#include > >> #include > >> +#include > >> #include > >> > >> #include > >> +#include > >> #include > >> #include > >> > >> @@ -13,21 +16,67 @@ struct patch { > >> unsigned int insn; > >> }; > >> > >> -void __kprobes __patch_text(void *addr, unsigned int insn) > >> +static DEFINE_SPINLOCK(patch_lock); > >> + > >> +static void __kprobes *patch_map(void *addr, int fixmap, unsigned long *flags) > >> +{ > >> + unsigned int uintaddr = (uintptr_t) addr; > >> + bool module = !core_kernel_text(uintaddr); > >> + struct page *page; > >> + > >> + if (module && IS_ENABLED(CONFIG_DEBUG_SET_MODULE_RONX)) { > >> + page = vmalloc_to_page(addr); > >> + } else if (!module && IS_ENABLED(CONFIG_ARM_KERNMEM_PERMS)) { > > I think this should probably be CONFIG_DEBUG_RODATA now that I've > moved RO exclusively into that config. > > >> + page = virt_to_page(addr); > >> + } else { > >> + return addr; > >> + } > > > > I can't help but think that it'd be more maintainable to just always use > > fixmap, though that would obviously have some performance impact (not > > sure that particularly matters for text patching) and it would expose > > possible fixmap bugginess to more kernels (see my next comment...) > > > >> + > >> + if (flags) > >> + spin_lock_irqsave(&patch_lock, *flags); > >> + > >> + set_fixmap(fixmap, page_to_phys(page)); > >> + > >> + return (void *) (__fix_to_virt(fixmap) + (uintaddr & ~PAGE_MASK)); > > > > How does fixmap cope with cache colouring? Looking at the implementation > > it looks like it doesn't and so fixmap use on ARM is possibly buggy. > > > > For the text patching case where we know there are no writeable mappings > > [1] this should be OK if we used set_fixmap_nocache here, so long as we > > also invalidated the dcache later for the proper virtual address. > > > > [1] Can we know there are no writeable mappings though, the ftrace code > > modifying patches from Kees Cook have there own way of modifying text > > code permissions. > > I think performance becomes an issue for massive updates like ftrace > seems to do, so I got the sense it was good to have the "bulk" way to > do it, and then this tiny poking way to do it. Probably performance is significant, but as safely modifying kernel code is virtually impossible to get right I would think that it best to concentrate on a single method for this. There again, the whole concept of things like ftrace is fundamentally flawed, with it being impossible to determine all the functions which require notrace annotations in order to avoid kernel crashes and corruption. So perhaps a little flakiness in kprobes and ftrace is acceptable? (I hope these things aren't enabled on distro kernels and usable by non-root processes...) -- Tixy -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/