Introduce the STACKLEAK feature and a test for it
Subsystems (13)
v8 cover Introduce the STACKLEAK feature and a test for it
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.
STACKLEAK instrumentation statistics
====================================
These numbers were obtained for the 4th version of the patch series.
Size of vmlinux (x86_64_defconfig):
file size:
- STACKLEAK disabled: 35014784 bytes
- STACKLEAK enabled: 35044952 bytes (+0.086%)
.text section size (calculated by size utility):
- STACKLEAK disabled: 10752983
- STACKLEAK enabled: 11062221 (+2.876%)
The readelf utility shows 45602 functions in vmlinux. The STACKLEAK gcc plugin inserted 36 check_alloca() calls and 1265 track_stack() calls (42274 calls are inserted during GIMPLE pass and 41009 calls are deleted during RTL pass). So 2.853% of functions are instrumented.
STACKLEAK performance impact
============================
The STACKLEAK description in Kconfig includes:
"The tradeoff is the performance impact: on a single CPU system kernel compilation sees a 1% slowdown, other systems and workloads may vary and you are advised to test this feature on your expected workload before deploying it".
Here are the results of a brief performance test on x86_64 (for the 2nd version of the patch). The numbers are very different because the STACKLEAK performance penalty depends on the workloads.
Hardware: Intel Core i7-4770, 16 GB RAM
Test #1: building the Linux kernel with Ubuntu config (time make -j9) Result on 4.11-rc8:
real 32m14.893s
user 237m30.886s
sys 11m12.273s
Result on 4.11-rc8+stackleak:
real 32m26.881s (+0.62%)
user 238m38.926s (+0.48%)
sys 11m36.426s (+3.59%)
Test #2: hackbench -s 4096 -l 2000 -g 15 -f 25 -P Average result on 4.11-rc8: 8.71s
Average result on 4.11-rc8+stackleak: 9.08s (+4.29%)
Changelog
=========
Changes in v8
1. Renamed the STACKLEAK tests to STACKLEAK_ALLOCA and STACKLEAK_DEEP_RECURSION. Fixed some obsolete comments there.
2. Added the comments describing stack erasing on x86_32, because it differs a lot from one on x86_64 since v7.
Changes in v7
1. Improved erase_kstack() on x86_64. Now it detects which stack we are currently using. If we are on the thread stack, erase_kstack() writes the poison value up to the stack pointer. If we are not on the thread stack (we are on the trampoline stack introduced by Andy Lutomirski), erase_kstack() writes the poison value up to the cpu_current_top_of_stack.
N.B. Right now there are two situations when erase_kstack() is called on the thread stack:
- from sysret32_from_system_call in entry_64_compat.S;
- from syscall_trace_enter() (see the 3rd patch of this series).
2. Introduced STACKLEAK_POISON_CHECK_DEPTH macro and BUILD_BUG_ON() checking its consistency with CONFIG_STACKLEAK_TRACK_MIN_SIZE.
3. Added "disable" option for the STACKLEAK gcc plugin (asked by Laura Abbott).
4. Improved CONFIG_STACKLEAK_METRICS. Now /proc/<pid>/stack_depth shows the maximum kernel stack consumption for the current and previous syscalls. Although this information is not precise, it can be useful for estimating the STACKLEAK performance impact for different workloads.
5. Removed redundant erase_kstack() calls from syscall_trace_enter(). There is no need to erase the stack in syscall_trace_enter() when it returns -1 (thanks to Dmitry Levin for noticing that).
6. Introduced MIN_STACK_LEFT to get rid of hardcoded numbers in check_alloca().
Changes in v6
1. Examined syscall entry/exit paths.
- Added missing erase_kstack() call at ret_from_fork() for x86_32.
- Added missing erase_kstack() call at syscall_trace_enter().
- Solved questions previously marked with TODO.
2. Rebased onto v4.15-rc2, which includes Andy Lutomirski's entry changes. Andy removed sp0 from thread_struct for x86_64, which was the only issue during rebasing.
3. Removed the recursive BUG() in track_stack() that was caused by the code instrumentation. Instead, CONFIG_GCC_PLUGIN_STACKLEAK now implies CONFIG_VMAP_STACK and CONFIG_SCHED_STACK_END_CHECK, which seems to be an optimal solution.
4. Put stack erasing in syscall_trace_enter() into a separate patch and fixed my mistake with secure_computing() (found by Tycho Andersen).
5. After some experiments, kept the asm implementation of erase_kstack(), because it gives a full control over the stack for clearing it neatly and doesn't offend KASAN.
6. Improved the comments describing STACKLEAK.
Changes in v5 (mostly according to the feedback from Ingo Molnar)
1. Introduced the CONFIG_STACKLEAK_METRICS providing STACKLEAK information about tasks via the /proc file system. That information can be useful for estimating the STACKLEAK performance impact for different workloads. In particular, /proc/<pid>/lowest_stack shows the current lowest_stack value and its final value from the previous syscall.
2. Introduced a single check_alloca() implementation working for both x86_64 and x86_32.
3. Fixed coding style issues and did some refactoring in the STACKLEAK gcc plugin.
4. Put the gcc plugin and the kernel stack erasing into separate (working) patches.
Changes in v4
1. Introduced the CONFIG_STACKLEAK_TRACK_MIN_SIZE parameter instead of hard-coded track-lowest-sp.
2. Carefully looked into the assertions in track_stack() and check_alloca().
- Fixed the incorrect BUG() condition in track_stack(), thanks to Tycho Andersen. Also disabled that check if CONFIG_VMAP_STACK is enabled.
- Fixed the surplus and erroneous code for calculating stack_left in check_alloca() on x86_64. That code repeats the work which is already done in get_stack_info() and it misses the fact that different exception stacks on x86_64 have different size.
3. Introduced two lkdtm tests for the STACKLEAK feature (developed together with Tycho Andersen).
4. Added info about STACKLEAK to Documentation/security/self-protection.rst.
5. Improved the comments.
6. Decided not to change "unsigned long sp = (unsigned long)&sp" to current_stack_pointer. The original variant is more platform independent since current_stack_pointer has different type on x86 and arm.
Changes in v3
1. Added the detailed comments describing the STACKLEAK gcc plugin. Read the plugin from the bottom up, like you do for Linux kernel drivers. Additional information:
- gcc internals documentation available from gcc sources;
- wonderful slides by Diego Novillo
https://www.airs.com/dnovillo/200711-GCC-Internals/
- nice introduction to gcc plugins at LWN
https://lwn.net/Articles/457543/
2. Improved the commit message and Kconfig description according to the feedback from Kees Cook. Also added the original notice describing the performance impact of the STACKLEAK feature.
3. Removed the arch-specific ix86_cmodel check in stackleak_track_stack_gate(). It caused skipping the kernel code instrumentation for i386.
4. Fixed a minor mistake in stackleak_tree_instrument_execute(). First versions of the plugin used ENTRY_BLOCK_PTR_FOR_FN(cfun)->next_bb to get the basic block with the function prologue. That was not correct since the control flow graph edge from ENTRY_BLOCK_PTR doesn't always go to that basic block.
So later it was changed to single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun)), but not completely. next_bb was still used for entry_bb assignment, which could cause the wrong value of the prologue_instrumented variable.
I've reported this issue to Grsecurity and proposed the fix for it, but unfortunately didn't get any reply.
5. Introduced the STACKLEAK_POISON macro and renamed the config option according to the feedback from Kees Cook.
Ideas for further work
======================
- Think of erasing stack on the way out of exception handlers (idea by Andy Lutomirski).
- Think of erasing the kernel stack after invoking EFI runtime services (idea by Mark Rutland).
- Try to port STACKLEAK to arm64 (Laura Abbott is working on it).
Alexander Popov (6):
x86/entry: Add STACKLEAK erasing the kernel stack at the end of
syscalls
gcc-plugins: Add STACKLEAK plugin for tracking the kernel stack x86/entry: Erase kernel stack in syscall_trace_enter()
lkdtm: Add a test for STACKLEAK
fs/proc: Show STACKLEAK metrics in the /proc file system
doc: self-protection: Add information about STACKLEAK feature
Documentation/security/self-protection.rst | 23 +- arch/Kconfig | 53 ++++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 7 + arch/x86/entry/entry_32.S | 93 ++++++ arch/x86/entry/entry_64.S | 113 +++++++ arch/x86/entry/entry_64_compat.S | 11 + arch/x86/include/asm/processor.h | 7 + arch/x86/kernel/asm-offsets.c | 11 + arch/x86/kernel/dumpstack.c | 20 ++ arch/x86/kernel/process_32.c | 8 + arch/x86/kernel/process_64.c | 8 + drivers/misc/Makefile | 3 + drivers/misc/lkdtm.h | 4 + drivers/misc/lkdtm_core.c | 2 + drivers/misc/lkdtm_stackleak.c | 136 +++++++++ fs/exec.c | 33 ++ fs/proc/base.c | 18 ++ include/linux/compiler.h | 6 + scripts/Makefile.gcc-plugins | 3 + scripts/gcc-plugins/stackleak_plugin.c | 471 +++++++++++++++++++++++++++++ 21 files changed, 1022 insertions(+), 9 deletions(-) create mode 100644 drivers/misc/lkdtm_stackleak.c create mode 100644 scripts/gcc-plugins/stackleak_plugin.c
Signature · 1 line
2.7.4
RFC PATCH v8 1/6 x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls
The STACKLEAK feature erases the kernel stack before returning from syscalls. That reduces the information which kernel stack leak bugs can reveal and blocks some uninitialized stack variable attacks. Moreover, STACKLEAK provides runtime checks for kernel stack overflow detection.
This commit introduces the architecture-specific code filling the used part of the kernel stack with a poison value before returning to the userspace. Full STACKLEAK feature also contains the gcc plugin which comes in a separate commit.
The STACKLEAK feature is ported from grsecurity/PaX. More information at:
https://grsecurity.net/
https://pax.grsecurity.net/
This code is modified from Brad Spengler/PaX Team's code in the last public patch of grsecurity/PaX based on our understanding of the code. Changes or omissions from the original code are ours and don't reflect the original grsecurity/PaX code.
--- arch/Kconfig | 27 ++++++++++ arch/x86/Kconfig | 1 + arch/x86/entry/entry_32.S | 89 ++++++++++++++++++++++++++++++++ arch/x86/entry/entry_64.S | 109 +++++++++++++++++++++++++++++++++++++++ arch/x86/entry/entry_64_compat.S | 11 ++++ arch/x86/include/asm/processor.h | 4 ++ arch/x86/kernel/asm-offsets.c | 8 +++ arch/x86/kernel/process_32.c | 5 ++ arch/x86/kernel/process_64.c | 5 ++ include/linux/compiler.h | 6 +++ 10 files changed, 265 insertions(+)
arch/Kconfig +27 -0
@@ -401,6 +401,13 @@ config SECCOMP_FILTER See Documentation/prctl/seccomp_filter.txt for details. +config HAVE_ARCH_STACKLEAK + bool + help + An architecture should select this if it has the code which + fills the used part of the kernel stack with the STACKLEAK_POISON + value before returning from system calls. + config HAVE_GCC_PLUGINS bool help
@@ -531,6 +538,26 @@ config GCC_PLUGIN_RANDSTRUCT_PERFORMANCE in structures. This reduces the performance hit of RANDSTRUCT at the cost of weakened randomization. +config GCC_PLUGIN_STACKLEAK + bool "Erase the kernel stack before returning from syscalls" + depends on GCC_PLUGINS + depends on HAVE_ARCH_STACKLEAK + help + This option makes the kernel erase the kernel stack before it + returns from a system call. That reduces the information which + kernel stack leak bugs can reveal and blocks some uninitialized + stack variable attacks. This option also provides runtime checks + for kernel stack overflow detection. + + The tradeoff is the performance impact: on a single CPU system kernel + compilation sees a 1% slowdown, other systems and workloads may vary + and you are advised to test this feature on your expected workload + before deploying it. + + This plugin was ported from grsecurity/PaX. More information at: + * https://grsecurity.net/ + * https://pax.grsecurity.net/ + config HAVE_CC_STACKPROTECTOR bool help
arch/x86/Kconfig +1 -0
@@ -119,6 +119,7 @@ config X86 select HAVE_ARCH_COMPAT_MMAP_BASES if MMU && COMPAT select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_THREAD_STRUCT_WHITELIST + select HAVE_ARCH_STACKLEAK select HAVE_ARCH_TRACEHOOK select HAVE_ARCH_TRANSPARENT_HUGEPAGE select HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD if X86_64
arch/x86/entry/entry_32.S +89 -0
113 lines not on this page · raw message
arch/x86/entry/entry_64.S +109 -0
127 lines not on this page · raw message
arch/x86/entry/entry_64_compat.S +11 -0
@@ -19,6 +19,12 @@ .section .entry.text, "ax" + .macro erase_kstack +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + call erase_kstack +#endif + .endm + /* * 32-bit SYSENTER entry. *
@@ -238,6 +244,11 @@ GLOBAL(entry_SYSCALL_compat_after_hwframe) /* Opportunistic SYSRET */ sysret32_from_system_call: + /* + * We are not going to return to the userspace from the trampoline + * stack. So let's erase the thread stack right now. + */ + erase_kstack TRACE_IRQS_ON /* User mode traces as IRQs on. */ movq RBX(%rsp), %rbx /* pt_regs->rbx */ movq RBP(%rsp), %rbp /* pt_regs->rbp */
arch/x86/include/asm/processor.h +4 -0
@@ -494,6 +494,10 @@ struct thread_struct {mm_segment_t addr_limit; +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +unsigned long lowest_stack; +#endif + unsigned int sig_on_uaccess_err:1; unsigned int uaccess_err:1; /* uaccess failed */
arch/x86/kernel/asm-offsets.c +8 -0
@@ -39,6 +39,9 @@ void common(void) {BLANK ();OFFSET (TASK_TI_flags, task_struct, thread_info.flags);OFFSET (TASK_addr_limit, task_struct, thread.addr_limit); +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +OFFSET (TASK_lowest_stack, task_struct, thread.lowest_stack); +#endif BLANK ();OFFSET (crypto_tfm_ctx_offset, crypto_tfm, __crt_ctx);
@@ -75,6 +78,11 @@ void common(void) {OFFSET (PV_MMU_read_cr2, pv_mmu_ops, read_cr2);#endif +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +BLANK (); +DEFINE (THREAD_SIZE_asm,THREAD_SIZE ); +#endif +#ifdef CONFIG_XEN BLANK ();OFFSET (XEN_vcpu_info_mask, vcpu_info, evtchn_upcall_mask);
arch/x86/kernel/process_32.c +5 -0
@@ -136,6 +136,11 @@ int copy_thread_tls(unsigned long clone_flags, unsigned long sp, p->thread.sp0 = (unsigned long ) (childregs+1 );memset (p->thread.ptrace_bps,0 ,sizeof (p->thread.ptrace_bps)); +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + p->thread.lowest_stack = (unsigned long )task_stack_page (p) + +2 *sizeof (unsigned long ); +#endif +if (unlikely (p->flags &PF_KTHREAD )) {/* kernel thread */ memset (childregs,0 ,sizeof (struct pt_regs ));
arch/x86/kernel/process_64.c +5 -0
@@ -281,6 +281,11 @@ int copy_thread_tls(unsigned long clone_flags, unsigned long sp, p->thread.sp = (unsigned long ) fork_frame; p->thread.io_bitmap_ptr =NULL ; +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + p->thread.lowest_stack = (unsigned long )task_stack_page (p) + +2 *sizeof (unsigned long ); +#endif +savesegment (gs, p->thread.gsindex); p->thread.gsbase = p->thread.gsindex ?0 : me->thread.gsbase;savesegment (fs, p->thread.fsindex);
include/linux/compiler.h +6 -0
@@ -337,4 +337,10 @@ unsigned long read_word_at_a_time(const void *addr)compiletime_assert (__native_word (t), \ "Need native word sized stores/loads for atomicity.") +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +/* Poison value points to the unused hole in the virtual memory map */ +# define STACKLEAK_POISON -0xBEEF +# define STACKLEAK_POISON_CHECK_DEPTH 128 +#endif +#endif /* __LINUX_COMPILER_H */
Signature · 1 line
2.7.4
Borislav Petkov
maintainerRe: [PATCH RFC v8 1/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls
Quoted 17 lines · Alexander Popov
arch/x86/Kconfig +1
-0@@ -119,6 +119,7 @@ config X86 select HAVE_ARCH_COMPAT_MMAP_BASES if MMU && COMPAT select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_THREAD_STRUCT_WHITELIST + select HAVE_ARCH_STACKLEAK select HAVE_ARCH_TRACEHOOK select HAVE_ARCH_TRANSPARENT_HUGEPAGE select HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD if X86_64arch/x86/entry/entry_32.S +0
-0
arch/x86/entry/entry_32.S line 82
Quoted 6 lines
arch/x86/entry/entry_32.S +3
-0@@ -77,6 +77,90 @@ #endif .endm +.macro erase_kstack +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + call erase_kstack
Hmm, why do we need a macro *and* a function of the same name? Why not do
call erase_kstack
everywhere we need to?
And then do
ENTRY(erase_kstack)
#ifdef CONFIG_GCC_PLUGIN_STACKLEAK ...
to tone down the ifdeffery.
arch/x86/entry/entry_32.S line 94
Quoted 12 lines
arch/x86/entry/entry_32.S +12
-0+#endif +.endm + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +ENTRY(erase_kstack) + pushl %edi + pushl %ecx + pushl %eax + pushl %ebp + + movl PER_CPU_VAR(current_task), %ebp + mov TASK_lowest_stack(%ebp), %edi
So instead of TASK_lowest_stack and adding all the ifdeffery around, why not do this computation:
(unsigned long)task_stack_page(p) + 2 * sizeof(unsigned long);
in asm?
You only need task->stack as an offset variable. And other code might need it too, anyway so you could just as well use it instead of adding another var to thread_struct.
/me reads further.
Ah. So this lowest_stack thing changes at the end:
"Set the lowest_stack value to the top_of_stack - 256"
Why?
So that magic needs more explanation for slow people like me.
arch/x86/entry/entry_32.S line 100
Quoted 6 lines
arch/x86/entry/entry_32.S +6
-0+ mov $STACKLEAK_POISON, %eax + std + + /* + * Let's search for the poison value in the stack. + * Start from the lowest_stack and go to the bottom (see std above).
Let's write insns capitalized: "see STD above".
...
arch/x86/entry/entry_32.S line 111
Quoted 11 lines
arch/x86/entry/entry_32.S +11
-0+ /* + * Check that some further qwords contain poison. If so, the part + * of the stack below the address in %rdi is likely to be poisoned. + * Otherwise we need to search deeper. + */ + mov $STACKLEAK_POISON_CHECK_DEPTH / 8, %ecx + repe scasq + jecxz 2f /* Poison the upper part of the stack */ + jne 1b /* Search deeper */ + +2:
You could name that label
.Lpoison
and make the code more readable:
jb .Lpoison
and so on.
Quoted 35 lines
arch/x86/entry/entry_32.S +35
-0+ /* + * Two qwords at the bottom of the thread stack are reserved and + * should not be poisoned (see CONFIG_SCHED_STACK_END_CHECK). + */ + or $2 * 8, %rdi + + /* + * Check whether we are on the thread stack to prepare the counter + * for stack poisoning. + */ + mov PER_CPU_VAR(cpu_current_top_of_stack), %rcx + sub %rsp, %rcx + cmp $THREAD_SIZE_asm, %rcx + jb 3f + + /* + * We are not on the thread stack, so we can write poison between + * the address in %rdi and the stack top. + */ + mov PER_CPU_VAR(cpu_current_top_of_stack), %rcx + sub %rdi, %rcx + jmp 4f + +3: + /* + * We are on the thread stack, so we can write poison between the + * address in %rdi and the address in %rsp (not included, used memory). + */ + mov %rsp, %rcx + sub %rdi, %rcx + +4: + /* Check that the counter value is sane */ + cmp $THREAD_SIZE_asm, %rcx + jb 5f
arch/x86/entry/entry_32.S line 156
Quoted 10 lines
arch/x86/entry/entry_32.S +10
-0+ ud2 + +5: + /* + * So let's write the poison value to the kernel stack. Start from the + * address in %rdi and move up (see cld). + */ + cld + shr $3, %ecx + rep stosq
Hohumm, that's >2K loops per syscall here and stack is
0xffffc90000008010: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008020: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008030: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008040: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008050: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008060: 0xffffffffffff4111 0xffffffffffff4111
...
Quoted 21 lines
arch/x86/kernel/process_64.c +5
-0@@ -281,6 +281,11 @@ int copy_thread_tls(unsigned long clone_flags, unsigned long sp, p->thread.sp = (unsigned long ) fork_frame; p->thread.io_bitmap_ptr =NULL ; +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + p->thread.lowest_stack = (unsigned long )task_stack_page (p) + +2 *sizeof (unsigned long ); +#endif +savesegment (gs, p->thread.gsindex); p->thread.gsbase = p->thread.gsindex ?0 : me->thread.gsbase;savesegment (fs, p->thread.fsindex);include/linux/compiler.h +0
-0
include/linux/compiler.h unsigned long read_word_at_a_time(const void *addr) · line 341
Quoted 5 lines
include/linux/compiler.h +2
-0@@ -337,4 +337,10 @@ unsigned long read_word_at_a_time(const void *addr)compiletime_assert (__native_word (t), \ "Need native word sized stores/loads for atomicity.") +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +/* Poison value points to the unused hole in the virtual memory map */
There are a bunch of those there now. Please document it in Documentation/x86/x86_64/mm.txt
Quoted 8 lines
include/linux/compiler.h +4
-0+# define STACKLEAK_POISON -0xBEEF +# define STACKLEAK_POISON_CHECK_DEPTH 128 +#endif +#endif /* __LINUX_COMPILER_H */ Signature · 1 line
2.7.4
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH RFC v8 1/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls
Hello Borislav,
Thanks a lot for your review. Please see my comments below.
Quoted 31 lines · Borislav Petkov
Quoted 17 lines · Alexander Popov
arch/x86/Kconfig +1
-0@@ -119,6 +119,7 @@ config X86 select HAVE_ARCH_COMPAT_MMAP_BASES if MMU && COMPAT select HAVE_ARCH_SECCOMP_FILTER select HAVE_ARCH_THREAD_STRUCT_WHITELIST + select HAVE_ARCH_STACKLEAK select HAVE_ARCH_TRACEHOOK select HAVE_ARCH_TRANSPARENT_HUGEPAGE select HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD if X86_64arch/x86/entry/entry_32.S +0
-0arch/x86/entry/entry_32.S line 82
Quoted 6 lines
arch/x86/entry/entry_32.S +3
-0@@ -77,6 +77,90 @@ #endif .endm +.macro erase_kstack +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + call erase_kstackHmm, why do we need a macro *and* a function of the same name? Why not do
call
erase_kstackeverywhere we need to?
Quoted 7 lines
And then do
ENTRY(erase_kstack)#ifdef CONFIG_GCC_PLUGIN_STACKLEAK ...to tone down the ifdeffery.
This approach doesn't work. There is the diff:
arch/x86/entry/entry_64.S +4 -10
@@ -66,14 +66,8 @@ END(native_usergs_sysret64) TRACE_IRQS_FLAGS EFLAGS(%rsp) .endm -.macro erase_kstack -#ifdef CONFIG_GCC_PLUGIN_STACKLEAK - call erase_kstack -#endif -.endm - -#ifdef CONFIG_GCC_PLUGIN_STACKLEAK ENTRY(erase_kstack) +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK pushq %rdi pushq %rcx pushq %rax
@@ -173,8 +167,8 @@ ENTRY(erase_kstack) popq %rcx popq %rdi ret -ENDPROC(erase_kstack) #endif +ENDPROC(erase_kstack) /* * When dynamic function tracer is enabled it will add a breakpoint
@@ -455,7 +449,7 @@ syscall_return_via_sysret: * We are on the trampoline stack. All regs except RDI are live. * We can do future final exit work right here. */ - erase_kstack + call erase_kstack SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
@@ -765,7 +759,7 @@ GLOBAL(swapgs_restore_regs_and_return_to_usermode) * We are on the trampoline stack. All regs except RDI are live. * We can do future final exit work right here. */ - erase_kstack + call erase_kstack SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
It crashes the kernel with STACKLEAK disabled:
[ 1.709081] VFS: Mounted root (ext4 filesystem) readonly on device 8:0. [ 1.710202] devtmpfs: mounted [ 1.711741] Freeing unused kernel memory: 1276K [ 1.712391] Kernel memory protection disabled. [ 1.760050] PANIC: double fault, error_code: 0x0 [ 1.760627] CPU: 0 PID: 1 Comm: init Not tainted 4.16.0-rc1+ #8 [ 1.761005] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014 [ 1.761005] RIP: 0010:update_curr+0x7/0x160 [ 1.761005] RSP: 0000:fffffe0000002000 EFLAGS: 00010002 [ 1.761005] RAX: ffff88007c890100 RBX: ffff88007fc20900 RCX: 000000003da0c28c [ 1.761005] RDX: ffff88007fc20900 RSI: ffff88007c890100 RDI: ffff88007fc20900 [ 1.761005] RBP: ffff88007c890100 R08: ffffffffffffffda R09: 0000000000000001 [ 1.761005] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88007fc20900 [ 1.761005] R13: ffff88007b529d00 R14: ffff88007c890100 R15: 0000000000000000 [ 1.761005] FS: 0000000000000000(0000) GS:ffff88007fc00000(0000) knlGS:0000000000000000 [ 1.761005] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 1.761005] CR2: fffffe0000001ff8 CR3: 000000007b090000 CR4: 00000000000006f0 [ 1.761005] Call Trace: [ 1.761005] <ENTRY_TRAMPOLINE> [ 1.761005] put_prev_entity+0x31/0x580 [ 1.761005] pick_next_task_fair+0x454/0x470 [ 1.761005] __schedule+0x14b/0x6f0 [ 1.761005] schedule+0x23/0x80 [ 1.761005] exit_to_usermode_loop+0x5c/0x90 [ 1.761005] do_syscall_64+0xfa/0x110 [ 1.761005] entry_SYSCALL_64_after_hwframe+0x21/0x86 [ 1.761005] RIP: 81a00935:swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] RSP: 81a00935:ffffffff81a00935 EFLAGS: ffffffff81a00935 ORIG_RAX: ffffffffffffffda [ 1.761005] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 0000000000000000 [ 1.761005] RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffff81a00935 [ 1.761005] RBP: 0000000000000000 R08: 0000000000000000 R09: 0000000000000000 [ 1.761005] R10: 0000000000000000 R11: 0000000000000000 R12: 0000000000000000 [ 1.761005] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] ? swapgs_restore_regs_and_return_to_usermode+0x38/0x8c [ 1.761005] </ENTRY_TRAMPOLINE> [ 1.761005] Code: 02 00 00 8b 78 18 e8 29 ff ff ff 48 ba db 34 b6 d7 82 de 1b 43 48 f7 e2 48 89 d0 48 c1 e8 12 c3 0f 1f 40 00 41 56 41 55 41 54 55 <53> 48 8b 5f 40 48 8b 87 30 01 00 00 48 85 db 48 8b 80 68 09 00 [ 1.761005] Kernel panic - not syncing: Machine halted.
Actually gcc creates a strange erase_kstack():
ffffffff81a00010 <erase_kstack>:
ffffffff81a00010: 5f pop %rdi ffffffff81a00011: eb 31 jmp ffffffff81a00044 <entry_SYSCALL_64_after_hwframe> ffffffff81a00013: 0f 1f 00 nopl (%rax) ffffffff81a00016: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
ffffffff81a0001d: 00 00 00
But with the initial macro the binary doesn't have erase_kstack() at all when STACKLEAK is disabled.
Quoted 26 lines
arch/x86/entry/entry_32.S line 94
Quoted 12 lines
arch/x86/entry/entry_32.S +12
-0+#endif +.endm + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +ENTRY(erase_kstack) + pushl %edi + pushl %ecx + pushl %eax + pushl %ebp + + movl PER_CPU_VAR(current_task), %ebp + mov TASK_lowest_stack(%ebp), %ediSo instead of
TASK_lowest_stackand adding all the ifdeffery around, why not do this computation:(unsigned long)task_stack_page(p) + 2 * sizeof(unsigned long);in asm?
You only need
task->stackas an offset variable. And other code might need it too, anyway so you could just as well use it instead of adding another var tothread_struct./me reads further.
Quoted 7 lines
Ah. So this
lowest_stackthing changes at the end:"Set the
lowest_stackvalue to thetop_of_stack- 256"Why?
So that magic needs more explanation for slow people like me.
Thanks for your question.
The commit message says: "Full STACKLEAK feature also contains the gcc plugin which comes in a separate commit".
And the next commit in this series introduces that plugin. Let me quote its commit message as well:
This commit introduces the STACKLEAK gcc plugin. It is needed for:
- tracking the lowest border of the kernel stack, which is important for the code erasing the used part of the kernel stack at the end of syscalls (comes in a separate commit);
- checking that alloca calls don't cause stack overflow.
So this plugin instruments the kernel code inserting:
- the check_alloca() call before alloca and the track_stack() call
after it;
- the track_stack() call for the functions with a stack frame size greater than or equal to CONFIG_STACKLEAK_TRACK_MIN_SIZE.
Tracking the lowest border of the kernel stack with the lowest_stack variable makes STACKLEAK so efficient (please see the performance statistics in the cover letter).
Quoted 8 lines
arch/x86/entry/entry_32.S line 100
Quoted 6 lines
arch/x86/entry/entry_32.S +6
-0+ mov $STACKLEAK_POISON, %eax + std + + /* + * Let's search for the poison value in the stack. + * Start from the lowest_stack and go to the bottom (see std above).Let's write insns capitalized: "see STD above".
Ok.
Quoted 16 lines
...
arch/x86/entry/entry_32.S
Quoted 11 lines
arch/x86/entry/entry_32.S +11
-0+ /* + * Check that some further qwords contain poison. If so, the part + * of the stack below the address in %rdi is likely to be poisoned. + * Otherwise we need to search deeper. + */ + mov $STACKLEAK_POISON_CHECK_DEPTH / 8, %ecx + repe scasq + jecxz 2f /* Poison the upper part of the stack */ + jne 1b /* Search deeper */ + +2:You could name that label
Quoted 7 lines
.Lpoison
and make the code more readable:
jb .Lpoison
and so on.
Ok, thanks.
Quoted 48 lines
Quoted 35 lines
arch/x86/entry/entry_32.S +35
-0+ /* + * Two qwords at the bottom of the thread stack are reserved and + * should not be poisoned (see CONFIG_SCHED_STACK_END_CHECK). + */ + or $2 * 8, %rdi + + /* + * Check whether we are on the thread stack to prepare the counter + * for stack poisoning. + */ + mov PER_CPU_VAR(cpu_current_top_of_stack), %rcx + sub %rsp, %rcx + cmp $THREAD_SIZE_asm, %rcx + jb 3f + + /* + * We are not on the thread stack, so we can write poison between + * the address in %rdi and the stack top. + */ + mov PER_CPU_VAR(cpu_current_top_of_stack), %rcx + sub %rdi, %rcx + jmp 4f + +3: + /* + * We are on the thread stack, so we can write poison between the + * address in %rdi and the address in %rsp (not included, used memory). + */ + mov %rsp, %rcx + sub %rdi, %rcx + +4: + /* Check that the counter value is sane */ + cmp $THREAD_SIZE_asm, %rcx + jb 5farch/x86/entry/entry_32.S
Quoted 10 lines
arch/x86/entry/entry_32.S +10
-0+ ud2 + +5: + /* + * So let's write the poison value to the kernel stack. Start from the + * address in %rdi and move up (see cld). + */ + cld + shr $3, %ecx + rep stosqHohumm, that's >2K loops per syscall here and stack is
Quoted 6 lines
0xffffc90000008010: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008020: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008030: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008040: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008050: 0xffffffffffff4111 0xffffffffffff4111 0xffffc90000008060: 0xffffffffffff4111 0xffffffffffff4111
Yes, so the stack is erased. That is the actual goal.
Quoted 8 lines
...
arch/x86/entry/entry_32.S
Quoted 2 lines
arch/x86/entry/entry_32.S +2
-0+#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +/* Poison value points to the unused hole in the virtual memory map */There are a bunch of those there now. Please document it in
Documentation/x86/x86_64/mm.txtQuoted 1 line
+# define
STACKLEAK_POISON-0xBEEF
The mm.txt already has this line:
ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole
Excuse me, I didn't get what to document.
Thanks again.
Best regards,
Alexander
Borislav Petkov
maintainerRe: [PATCH RFC v8 1/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls
Quoted 8 lines · Alexander Popov
Actually gcc creates a strange
erase_kstack():
ffffffff81a00010<erase_kstack>:ffffffff81a00010: 5f pop %rdi ffffffff81a00011: eb 31 jmp ffffffff81a00044 <entry_SYSCALL_64_after_hwframe> ffffffff81a00013: 0f 1f 00 nopl (%rax) ffffffff81a00016: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
ffffffff81a0001d: 00 00 00
Something's very strange here. That code looks like:
ENTRY(entry_SYSCALL_64_stage2)
UNWIND_HINT_EMPTY
popq %rdi
jmp entry_SYSCALL_64_after_hwframeEND(entry_SYSCALL_64_stage2)
ENTRY(entry_SYSCALL_64)
...
so frankly I wonder what gcc is even doing here?!? And why isn't that erase_kstack symbol completely empty. It gives it that address ffffffff81a00010 and size and this looks really wrong.
/me experiments a bit more, notices the ENDPROC()...
Ok, I think I know what it is:
END(erase_kstack) gives
.globl erase_kstack ; .p2align 4, 0x90 ; erase_kstack:
# 167 "arch/x86/entry/entry_64.S"
.size erase_kstack, .-erase_kstack
# 182 "arch/x86/entry/entry_64.S"
and that generates a STT_NOTYPE symbol:
67318: ffffffff81a00000 0 NOTYPE GLOBAL DEFAULT 1 erase_kstack
In that case, the symbol gets ignored by objdump as it dumps this at that address:
ffffffff81a00000 <entry_SYSCALL_64_stage2>:
ffffffff81a00000: 5f pop %rdi ffffffff81a00001: eb 31 jmp ffffffff81a00034 <entry_SYSCALL_64_after_hwframe> ffffffff81a00003: 0f 1f 00 nopl (%rax) ffffffff81a00006: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1) ffffffff81a0000d: 00 00 00
which is basically the strange erase_kstack() variant you pasted above.
while ENDPROC(erase_kstack) gives
.globl erase_kstack ; .p2align 4, 0x90 ; erase_kstack:
# 167 "arch/x86/entry/entry_64.S"
.type erase_kstack, @function ; .size erase_kstack, .-erase_kstack # 182 "arch/x86/entry/entry_64.S"
and that generates a STT_FUNC:
67318: ffffffff81a00000 0 FUNC GLOBAL DEFAULT 1 erase_kstack
Now, there's this comment over ENDPROC:
/* If symbol 'name' is treated as a subroutine (gets called, and returns)
* then please use ENDPROC to mark 'name' as STT_FUNC for the benefit of
* static analysis tools such as stack depth analyzer.
*/
#ifndef ENDPROC
and I don't think we want to analyze the stack depth of erase_kstack itself.
However, even if we did END(erase_kstack), the calls are still in the code:
ffffffff81a00111: e8 ea fe ff ff callq ffffffff81a00000 <entry_SYSCALL_64_stage2>
so macro it is. But please call the macro something else, not the same name as the function.
Quoted 5 lines
The commit message says: "Full STACKLEAK feature also contains the gcc plugin which comes in a separate commit".
And the next commit in this series introduces that plugin. Let me quote its commit message as well:
...
Quoted 3 lines
Tracking the lowest border of the kernel stack with the
lowest_stackvariable makes STACKLEAK so efficient (please see the performance statistics in the cover letter).
Aha, there it is. I guess I better continue looking through the patchset. :)
Quoted 4 lines
The mm.txt already has this line:
ffffffffffe00000- ffffffffffffffff (=2 MB) unused holeExcuse me, I didn't get what to document.
You say
/* Poison value points to the unused hole in the virtual memory map */
but we do change that memory map from time to time and there are multiple unused holes.
So do something like this so that there are no clashes when someone decides to use that unused hole:
---
Documentation/x86/x86_64/mm.txt +2 -0
@@ -24,6 +24,7 @@ ffffffffa0000000 - [fixmap start] (~1526 MB) module mapping space (variable) [fixmap start] - ffffffffff5fffff kernel-internal fixmap range ffffffffff600000 - ffffffffff600fff (=4 kB) legacy vsyscall ABI ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole +Stackleak poison value in this last hole: 0xffffffffffff4111 Virtual memory map with 5 level page tables:
@@ -50,6 +51,7 @@ ffffffffa0000000 - fffffffffeffffff (1520 MB) module mapping space [fixmap start] - ffffffffff5fffff kernel-internal fixmap range ffffffffff600000 - ffffffffff600fff (=4 kB) legacy vsyscall ABI ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole +Stackleak poison value in this last hole: 0xffffffffffff4111 Architecture defines a 64-bit virtual address. Implementations can support less. Currently supported are 48- and 57-bit virtual addresses. Bits 63
---
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Re: [PATCH RFC v8 1/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls
Hello Borislav,
Quoted 8 lines · Borislav Petkov
On Thu, Feb 22, 2018 at 12:49:44AM +0300, Alexander Popov wrote:
However, even if we didEND(erase_kstack), the calls are still in the code:
ffffffff81a00111: e8 ea fe ff ff callqffffffff81a00000<entry_SYSCALL_64_stage2>so macro it is. But please call the macro something else, not the same name as the function.
Thanks for your time spent on this! I'll call it ERASE_KSTACK and it will look like other macros.
Quoted 29 lines
Quoted 4 lines
The mm.txt already has this line:
ffffffffffe00000- ffffffffffffffff (=2 MB) unused holeExcuse me, I didn't get what to document.
You say
/* Poison value points to the unused hole in the virtual memory map */
but we do change that memory map from time to time and there are multiple unused holes.
So do something like this so that there are no clashes when someone decides to use that unused hole:
---
Documentation/x86/x86_64/mm.txt +1
-0@@ -24,6 +24,7 @@ ffffffffa0000000 - [fixmap start] (~1526 MB) module mapping space (variable) [fixmap start] - ffffffffff5fffff kernel-internal fixmap range ffffffffff600000 - ffffffffff600fff (=4 kB) legacy vsyscall ABI ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole +Stackleak poison value in this last hole: 0xffffffffffff4111 Virtual memory map with 5 level page tables:
Documentation/x86/x86_64/mm.txt ffffffffa0000000 - fffffffffeffffff (1520 MB) module mapping space · line 54
Quoted 4 lines
Documentation/x86/x86_64/mm.txt +1
-0@@ -50,6 +51,7 @@ ffffffffa0000000 - fffffffffeffffff (1520 MB) module mapping space [fixmap start] - ffffffffff5fffff kernel-internal fixmap range ffffffffff600000 - ffffffffff600fff (=4 kB) legacy vsyscall ABI ffffffffffe00000 - ffffffffffffffff (=2 MB) unused hole +Stackleak poison value in this last hole: 0xffffffffffff4111
Ok, I see. Thank you very much.
Best regards,
Alexander
RFC PATCH v8 2/6 gcc-plugins: Add STACKLEAK plugin for tracking the kernel stack
In reply to Alexander PopovThe STACKLEAK feature erases the kernel stack before returning from syscalls. That reduces the information which kernel stack leak bugs can reveal and blocks some uninitialized stack variable attacks. Moreover, STACKLEAK provides runtime checks for kernel stack overflow detection.
This commit introduces the STACKLEAK gcc plugin. It is needed for:
- tracking the lowest border of the kernel stack, which is important for the code erasing the used part of the kernel stack at the end of syscalls (comes in a separate commit);
- checking that alloca calls don't cause stack overflow.
So this plugin instruments the kernel code inserting:
- the check_alloca() call before alloca and the track_stack() call
after it;
- the track_stack() call for the functions with a stack frame size greater than or equal to CONFIG_STACKLEAK_TRACK_MIN_SIZE.
The STACKLEAK feature is ported from grsecurity/PaX. More information at:
https://grsecurity.net/
https://pax.grsecurity.net/
This code is modified from Brad Spengler/PaX Team's code in the last public patch of grsecurity/PaX based on our understanding of the code. Changes or omissions from the original code are ours and don't reflect the original grsecurity/PaX code.
--- arch/Kconfig | 14 + arch/x86/kernel/dumpstack.c | 20 ++ fs/exec.c | 33 +++ scripts/Makefile.gcc-plugins | 3 + scripts/gcc-plugins/stackleak_plugin.c | 471 +++++++++++++++++++++++++++++++++ 5 files changed, 541 insertions(+) create mode 100644 scripts/gcc-plugins/stackleak_plugin.c
arch/Kconfig +14 -0
@@ -542,6 +542,8 @@ config GCC_PLUGIN_STACKLEAK bool "Erase the kernel stack before returning from syscalls" depends on GCC_PLUGINS depends on HAVE_ARCH_STACKLEAK + imply VMAP_STACK + imply SCHED_STACK_END_CHECK help This option makes the kernel erase the kernel stack before it returns from a system call. That reduces the information which
@@ -558,6 +560,18 @@ config GCC_PLUGIN_STACKLEAK * https://grsecurity.net/ * https://pax.grsecurity.net/ +config STACKLEAK_TRACK_MIN_SIZE + int "Minimum stack frame size of functions tracked by STACKLEAK" + default 100 + range 0 4096 + depends on GCC_PLUGIN_STACKLEAK + help + The STACKLEAK gcc plugin instruments the kernel code for tracking + the lowest border of the kernel stack (and for some other purposes). + It inserts the track_stack() call for the functions with a stack + frame size greater than or equal to this parameter. + If unsure, leave the default value 100. + config HAVE_CC_STACKPROTECTOR bool help
arch/x86/kernel/dumpstack.c +20 -0
@@ -375,3 +375,23 @@ static int __init code_bytes_setup(char *s)return 1 ; }__setup ("code_bytes=" , code_bytes_setup); + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + +#define MIN_STACK_LEFT 256 + +void __used check_alloca (unsigned long size) +{ +unsigned long sp = (unsigned long )&sp; +struct stack_info stack_info = {0 }; +unsigned long visit_mask =0 ; +unsigned long stack_left; + +BUG_ON (get_stack_info (&sp, current, &stack_info, &visit_mask)); + + stack_left = sp - (unsigned long )stack_info.begin; +BUG_ON (stack_left <MIN_STACK_LEFT || + size >= stack_left -MIN_STACK_LEFT ); +} +EXPORT_SYMBOL (check_alloca); +#endif
fs/exec.c +33 -0
@@ -1962,3 +1962,36 @@ COMPAT_SYSCALL_DEFINE5(execveat, int, fd, argv, envp, flags); }#endif + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +void __used track_stack (void ) +{ +/* +* N.B. The arch-specific part of the STACKLEAK feature fills the +* kernel stack with the poison value, which has the register width. +* That code assumes that the value of thread.lowest_stack is aligned +* on the register width boundary. +* +* That is true for x86 and x86_64 because of the kernel stack +* alignment on these platforms (for details, see cc_stack_align in +* arch/x86/Makefile). Take care of that when you port STACKLEAK to +* new platforms. +*/ +unsigned long sp = (unsigned long )&sp; + +/* +* Having CONFIG_STACKLEAK_TRACK_MIN_SIZE larger than +* STACKLEAK_POISON_CHECK_DEPTH makes the poison search in +* erase_kstack() unreliable. Let's prevent that. +*/ +BUILD_BUG_ON (CONFIG_STACKLEAK_TRACK_MIN_SIZE > +STACKLEAK_POISON_CHECK_DEPTH ); + +if (sp < current->thread.lowest_stack && + sp >= (unsigned long )task_stack_page (current) + +2 *sizeof (unsigned long )) { + current->thread.lowest_stack = sp; + } +} +EXPORT_SYMBOL (track_stack); +#endif /* CONFIG_GCC_PLUGIN_STACKLEAK */
scripts/Makefile.gcc-plugins +3 -0
@@ -35,6 +35,9 @@ ifdef CONFIG_GCC_PLUGINS gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_RANDSTRUCT) += -DRANDSTRUCT_PLUGIN gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_RANDSTRUCT_PERFORMANCE) += -fplugin-arg-randomize_layout_plugin-performance-mode + gcc-plugin-$(CONFIG_GCC_PLUGIN_STACKLEAK) += stackleak_plugin.so + gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_STACKLEAK) += -DSTACKLEAK_PLUGIN -fplugin-arg-stackleak_plugin-track-min-size=$(CONFIG_STACKLEAK_TRACK_MIN_SIZE) + GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y)) export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN GCC_PLUGIN_SUBDIR
scripts/gcc-plugins/stackleak_plugin.c new file +471 -0
471 lines not on this page · raw message
Signature · 1 line
2.7.4
RFC PATCH v8 3/6 x86/entry: Erase kernel stack in syscall_trace_enter()
In reply to Alexander PopovMake STACKLEAK erase kernel stack after ptrace/seccomp/auditing not to leave any sensitive information on the stack for the syscall code.
This code is modified from Brad Spengler/PaX Team's code in the last public patch of grsecurity/PaX based on our understanding of the code. Changes or omissions from the original code are ours and don't reflect the original grsecurity/PaX code.
--- arch/x86/entry/common.c | 7 +++++++ 1 file changed, 7 insertions(+)
arch/x86/entry/common.c +7 -0
@@ -46,6 +46,12 @@ __visible inline void enter_from_user_mode(void)static inline void enter_from_user_mode (void ) {}#endif +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +asmlinkage void erase_kstack(void); +#else +static void erase_kstack (void ) {} +#endif +static void do_audit_syscall_entry (struct pt_regs *regs,u32 arch) {#ifdef CONFIG_X86_64
@@ -128,6 +134,7 @@ static long syscall_trace_enter(struct pt_regs *regs)do_audit_syscall_entry (regs, arch); +erase_kstack ();return ret ?: regs->orig_ax; }
Signature · 1 line
2.7.4
RFC PATCH v8 4/6 lkdtm: Add a test for STACKLEAK
In reply to Alexander PopovIntroduce two lkdtm tests for the STACKLEAK feature: STACKLEAK_ALLOCA and STACKLEAK_DEEP_RECURSION. Both of them check that the current task stack is properly erased (filled with STACKLEAK_POISON).
STACKLEAK_ALLOCA tests that:
- check_alloca() allows alloca calls which don't exhaust the kernel stack;
- alloca calls which exhaust/overflow the kernel stack hit BUG() in
check_alloca().
STACKLEAK_DEEP_RECURSION tests that exhausting the current task stack with a deep recursion is detected by CONFIG_VMAP_STACK (which is implied by CONFIG_GCC_PLUGIN_STACKLEAK).
--- drivers/misc/Makefile | 3 + drivers/misc/lkdtm.h | 4 ++ drivers/misc/lkdtm_core.c | 2 + drivers/misc/lkdtm_stackleak.c | 136 +++++++++++++++++++++++++++++++++++++++++ 4 files changed, 145 insertions(+) create mode 100644 drivers/misc/lkdtm_stackleak.c
drivers/misc/Makefile +3 -0
@@ -65,6 +65,9 @@ lkdtm-$(CONFIG_LKDTM) += lkdtm_perms.o lkdtm-$(CONFIG_LKDTM) += lkdtm_refcount.o lkdtm-$(CONFIG_LKDTM) += lkdtm_rodata_objcopy.o lkdtm-$(CONFIG_LKDTM) += lkdtm_usercopy.o +lkdtm-$(CONFIG_LKDTM) += lkdtm_stackleak.o + +KASAN_SANITIZE_lkdtm_stackleak.o := n KCOV_INSTRUMENT_lkdtm_rodata.o := n
drivers/misc/lkdtm.h +4 -0
@@ -83,4 +83,8 @@ void lkdtm_USERCOPY_STACK_FRAME_FROM(void);void lkdtm_USERCOPY_STACK_BEYOND (void );void lkdtm_USERCOPY_KERNEL (void ); +/* lkdtm_stackleak.c */ +void lkdtm_STACKLEAK_ALLOCA (void ); +void lkdtm_STACKLEAK_DEEP_RECURSION (void ); +#endif
drivers/misc/lkdtm_core.c +2 -0
@@ -183,6 +183,8 @@ static const struct crashtype crashtypes[] = {CRASHTYPE (USERCOPY_STACK_FRAME_FROM ),CRASHTYPE (USERCOPY_STACK_BEYOND ),CRASHTYPE (USERCOPY_KERNEL ), +CRASHTYPE (STACKLEAK_ALLOCA ), + CRASHTYPE(STACKLEAK_DEEP_RECURSION), };
drivers/misc/lkdtm_stackleak.c new file +136 -0
136 lines not on this page · raw message
Signature · 1 line
2.7.4
RFC PATCH v8 5/6 fs/proc: Show STACKLEAK metrics in the /proc file system
In reply to Alexander PopovIntroduce CONFIG_STACKLEAK_METRICS providing STACKLEAK information about tasks via the /proc file system. In particular, /proc/<pid>/stack_depth shows the maximum kernel stack consumption for the current and previous syscalls. Although this information is not precise, it can be useful for estimating the STACKLEAK performance impact for your workloads.
--- arch/Kconfig | 12 ++++++++++++ arch/x86/entry/entry_32.S | 4 ++++ arch/x86/entry/entry_64.S | 4 ++++ arch/x86/include/asm/processor.h | 3 +++ arch/x86/kernel/asm-offsets.c | 3 +++ arch/x86/kernel/process_32.c | 3 +++ arch/x86/kernel/process_64.c | 3 +++ fs/proc/base.c | 18 ++++++++++++++++++ 8 files changed, 50 insertions(+)
arch/Kconfig +12 -0
@@ -572,6 +572,18 @@ config STACKLEAK_TRACK_MIN_SIZE frame size greater than or equal to this parameter. If unsure, leave the default value 100. +config STACKLEAK_METRICS + bool "Show STACKLEAK metrics in the /proc file system" + depends on GCC_PLUGIN_STACKLEAK + depends on PROC_FS + help + If this is set, STACKLEAK metrics for every task are available in + the /proc file system. In particular, /proc/<pid>/stack_depth + shows the maximum kernel stack consumption for the current and + previous syscalls. Although this information is not precise, it + can be useful for estimating the STACKLEAK performance impact for + your workloads. + config HAVE_CC_STACKPROTECTOR bool help
arch/x86/entry/entry_32.S +4 -0
@@ -134,6 +134,10 @@ ENTRY(erase_kstack) mov %esp, %ecx sub %edi, %ecx +#ifdef CONFIG_STACKLEAK_METRICS + mov %edi, TASK_prev_lowest_stack(%ebp) +#endif + /* Check that the counter value is sane. */ cmp $THREAD_SIZE_asm, %ecx jb 3f
arch/x86/entry/entry_64.S +4 -0
@@ -119,6 +119,10 @@ ENTRY(erase_kstack) */ or $2 * 8, %rdi +#ifdef CONFIG_STACKLEAK_METRICS + mov %rdi, TASK_prev_lowest_stack(%r11) +#endif + /* * Check whether we are on the thread stack to prepare the counter * for stack poisoning.
arch/x86/include/asm/processor.h +3 -0
@@ -496,6 +496,9 @@ struct thread_struct {#ifdef CONFIG_GCC_PLUGIN_STACKLEAK unsigned long lowest_stack; +# ifdef CONFIG_STACKLEAK_METRICS +unsigned long prev_lowest_stack; +# endif #endif unsigned int sig_on_uaccess_err:1;
arch/x86/kernel/asm-offsets.c +3 -0
@@ -41,6 +41,9 @@ void common(void) {OFFSET (TASK_addr_limit, task_struct, thread.addr_limit);#ifdef CONFIG_GCC_PLUGIN_STACKLEAK OFFSET (TASK_lowest_stack, task_struct, thread.lowest_stack); +# ifdef CONFIG_STACKLEAK_METRICS +OFFSET (TASK_prev_lowest_stack, task_struct, thread.prev_lowest_stack); +# endif #endif BLANK ();
arch/x86/kernel/process_32.c +3 -0
@@ -139,6 +139,9 @@ int copy_thread_tls(unsigned long clone_flags, unsigned long sp,#ifdef CONFIG_GCC_PLUGIN_STACKLEAK p->thread.lowest_stack = (unsigned long )task_stack_page (p) +2 *sizeof (unsigned long ); +# ifdef CONFIG_STACKLEAK_METRICS + p->thread.prev_lowest_stack = p->thread.lowest_stack; +# endif #endif if (unlikely (p->flags &PF_KTHREAD )) {
arch/x86/kernel/process_64.c +3 -0
@@ -284,6 +284,9 @@ int copy_thread_tls(unsigned long clone_flags, unsigned long sp,#ifdef CONFIG_GCC_PLUGIN_STACKLEAK p->thread.lowest_stack = (unsigned long )task_stack_page (p) +2 *sizeof (unsigned long ); +# ifdef CONFIG_STACKLEAK_METRICS + p->thread.prev_lowest_stack = p->thread.lowest_stack; +# endif #endif savesegment (gs, p->thread.gsindex);
fs/proc/base.c +18 -0
@@ -2914,6 +2914,21 @@ static int proc_pid_patch_state(struct seq_file *m, struct pid_namespace *ns, }#endif /* CONFIG_LIVEPATCH */ +#ifdef CONFIG_STACKLEAK_METRICS +static int proc_stack_depth (struct seq_file *m,struct pid_namespace *ns, +struct pid *pid,struct task_struct *task) +{ +unsigned long prev_depth =THREAD_SIZE - + (task->thread.prev_lowest_stack & (THREAD_SIZE -1 )); +unsigned long depth =THREAD_SIZE - + (task->thread.lowest_stack & (THREAD_SIZE -1 )); + +seq_printf (m,"previous stack depth: %lu\nstack depth: %lu\n" , + prev_depth, depth); +return 0 ; +} +#endif /* CONFIG_STACKLEAK_METRICS */ +/* * Thread groups */
@@ -3018,6 +3033,9 @@ static const struct pid_entry tgid_base_stuff[] = {#ifdef CONFIG_LIVEPATCH ONE("patch_state", S_IRUSR, proc_pid_patch_state),#endif +#ifdef CONFIG_STACKLEAK_METRICS + ONE("stack_depth", S_IRUGO, proc_stack_depth), +#endif };static int proc_tgid_base_readdir (struct file *file,struct dir_context *ctx)
Signature · 1 line
2.7.4
RFC PATCH v8 6/6 doc: self-protection: Add information about STACKLEAK feature
In reply to Alexander PopovAdd information about STACKLEAK feature to "Stack depth overflow" and "Memory poisoning" sections of self-protection.rst.
--- Documentation/security/self-protection.rst | 23 ++++++++++++++--------- 1 file changed, 14 insertions(+), 9 deletions(-)
Documentation/security/self-protection.rst +14 -9
@@ -165,10 +165,15 @@ Stack depth overflow A less well understood attack is using a bug that triggers the kernel to consume stack memory with deep function calls or large stack allocations. With this attack it is possible to write beyond the end of -the kernel's preallocated stack space and into sensitive structures. Two -important changes need to be made for better protections: moving the -sensitive thread_info structure elsewhere, and adding a faulting memory -hole at the bottom of the stack to catch these overflows. +the kernel's preallocated stack space and into sensitive structures. +The combination of the following measures gives better protection: + +* moving the sensitive thread_info structure off the stack + (``CONFIG_THREAD_INFO_IN_TASK``); +* adding a faulting memory hole at the bottom of the stack to catch + these overflows (``CONFIG_VMAP_STACK``); +* runtime checking that alloca() calls don't overstep the stack boundary + (``CONFIG_GCC_PLUGIN_STACKLEAK``). Heap memory integrity ---------------------
@@ -302,11 +307,11 @@ sure structure holes are cleared. Memory poisoning ---------------- -When releasing memory, it is best to poison the contents (clear stack on -syscall return, wipe heap memory on a free), to avoid reuse attacks that -rely on the old contents of memory. This frustrates many uninitialized -variable attacks, stack content exposures, heap content exposures, and -use-after-free attacks. +When releasing memory, it is best to poison the contents, to avoid reuse +attacks that rely on the old contents of memory. E.g., clear stack on a +syscall return (``CONFIG_GCC_PLUGIN_STACKLEAK``), wipe heap memory on a +free. This frustrates many uninitialized variable attacks, stack content +exposures, heap content exposures, and use-after-free attacks. Destination tracking --------------------
Signature · 1 line
2.7.4
Quoted 9 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.
Hello! I've just tested STACKLEAK against the recent CVE-2017-17712 exploit:
http://openwall.com/lists/oss-security/2018/02/20/1
This vulnerability is a race condition in raw_sendmsg() in net/ipv4/raw.c. It leads to uninitialized stack pointer usage which can be used for a local privilege escalation.
CVE-2017-17712 was discovered and fixed by Mohamed Ghannam (kudos to him!). He also provided a stable PoC exploit for it.
With STACKLEAK, the uninitialized stack pointer is set to STACKLEAK_POISON (-0xBEEF) and points to the unused hole in the virtual memory map. That blocks the stack spraying needed for CVE-2017-17712 exploit (similar to CVE-2010-2963).
Best regards,
Alexander
Quoted 4 lines · Alexander Popov
Quoted 3 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
If we can borrow some of luto or tglx's time, I think that'd be best:
they've been looking at the entry code a lot lately. :) Regardless, I think the addition to the entry code is clean (especially now that the fast path is gone *sob*). :P
Quoted 5 lines
Quoted 5 lines
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.
I've added this series to my kernel.org trees, which means 0-day will start grinding on it too now:
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git
kspp/gcc-plugin/stackleak
The LKDTM tests look great and check out for me. I think the code is clear, so I'd like to get it into -next, but I want to be sure I'm not stepping on x86 toes first.
Laura, how does arm64 look for this? Would it be possible to add it to this series (at least on kernel.org for build/run testing)?
Quoted 7 lines
Hello! I've just tested STACKLEAK against the recent CVE-2017-17712 exploit:
http://openwall.com/lists/oss-security/2018/02/20/1This vulnerability is a race condition in
raw_sendmsg()innet/ipv4/raw.c. It leads to uninitialized stack pointer usage which can be used for a local privilege escalation.
Quoted 6 lines
CVE-2017-17712 was discovered and fixed by Mohamed Ghannam (kudos to him!). He also provided a stable PoC exploit for it.
With STACKLEAK, the uninitialized stack pointer is set to
STACKLEAK_POISON(-0xBEEF) and points to the unused hole in the virtual memory map. That blocks the stack spraying needed for CVE-2017-17712 exploit (similar to CVE-2010-2963).
Nice check! I wonder if the the byref structleak also solves it? Regardless, I think stackleak has better performance and greater coverage.
Thanks!
-Kees
Signature · 2 lines
Kees Cook Pixel Security
Quoted 22 lines · Kees Cook
Quoted 4 lines · Alexander Popov
Quoted 3 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
If we can borrow some of luto or tglx's time, I think that'd be best:
they've been looking at the entry code a lot lately. :) Regardless, I think the addition to the entry code is clean (especially now that the fast path is gone *sob*). :PQuoted 5 lines
Quoted 5 lines
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.I've added this series to my kernel.org trees, which means 0-day will start grinding on it too now:
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git
kspp/gcc-plugin/stackleak
Quoted 7 lines
The LKDTM tests look great and check out for me. I think the code is clear, so I'd like to get it into -next, but I want to be sure I'm not stepping on x86 toes first.
Laura, how does arm64 look for this? Would it be possible to add it to this series (at least on kernel.org for build/run testing)?
I fell behind on rebasing/testing so I need to bring it up to date. Assuming the arm folks are okay with the approach, we can bring it in for kernel.org testing once I'm finished.
Thanks,
Laura
cover Stackleak for arm64
This is the arm64 version of the STACKLEAK plugin originall from grsecurity. See
https://marc.info/?l=kernel-hardening&m=151880470609808 for the full x86 version. This is based on top of Kees' branch for stackleak and has been cleaned up to use a few macros from that branch.
Comments welcome, if there are no major objections Kees will queue this up to get some CI testing. This passed both of the LKDTM tests.
Laura Abbott (2):
stackleak: Update for arm64
arm64: Clear the stack
arch/arm64/Kconfig | 1 + arch/arm64/include/asm/processor.h | 6 ++ arch/arm64/kernel/asm-offsets.c | 3 + arch/arm64/kernel/entry.S | 108 +++++++++++++++++++++++++++++++++ arch/arm64/kernel/process.c | 16 +++++ drivers/firmware/efi/libstub/Makefile | 3 +- scripts/Makefile.gcc-plugins | 5 +- scripts/gcc-plugins/stackleak_plugin.c | 5 ++ 8 files changed, 145 insertions(+), 2 deletions(-)
Signature · 1 line
2.14.3
PATCH 1/2 stackleak: Update for arm64
arm64 has another layer of indirection in the RTL. Account for this in the plugin.
--- scripts/gcc-plugins/stackleak_plugin.c | 5 +++++ 1 file changed, 5 insertions(+)
scripts/gcc-plugins/stackleak_plugin.c +5 -0
@@ -244,6 +244,11 @@ static unsigned int stackleak_final_execute(void)* that insn. */ body =PATTERN (insn); +/* arm64 is different */ +if (GET_CODE (body) ==PARALLEL ) { + body =XEXP (body,0 ); + body =XEXP (body,0 ); + }if (GET_CODE (body) !=CALL )continue ;
Signature · 1 line
2.14.3
Re: [PATCH 1/2] stackleak: Update for arm64
Hi Laura,
Quoted 14 lines · Laura Abbott
arm64 has another layer of indirection in the RTL. Account for this in the plugin.
Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/stackleak_plugin.c | 5 +++++ 1 file changed, 5 insertions(+)scripts/gcc-plugins/stackleak_plugin.c +0
-0
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 251
Quoted 8 lines
scripts/gcc-plugins/stackleak_plugin.c +5
-0@@ -244,6 +244,11 @@ static unsigned int stackleak_final_execute(void)* that insn. */ body =PATTERN (insn); +/* arm64 is different */ +if (GET_CODE (body) ==PARALLEL ) { + body =XEXP (body,0 ); + body =XEXP (body,0 ); + }
Like most kernel developers, I don't know the first thing about GCC internals so I asked our GCC team and Richard (CC'd) reckons this should be:
if (GET_CODE(body) == PARALLEL) body = XVECEXP(body, 0, 0);
instead of the hunk above. Can you give that a go instead, please?
Cheers,
Will
Re: [PATCH 1/2] stackleak: Update for arm64
Hello Will, Richard and GCC folks!
Quoted 27 lines · Will Deacon
Quoted 14 lines · Laura Abbott
arm64 has another layer of indirection in the RTL. Account for this in the plugin.
Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/stackleak_plugin.c | 5 +++++ 1 file changed, 5 insertions(+)scripts/gcc-plugins/stackleak_plugin.c +0
-0scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 251
Quoted 8 lines
scripts/gcc-plugins/stackleak_plugin.c +5
-0@@ -244,6 +244,11 @@ static unsigned int stackleak_final_execute(void)* that insn. */ body =PATTERN (insn); +/* arm64 is different */ +if (GET_CODE (body) ==PARALLEL ) { + body =XEXP (body,0 ); + body =XEXP (body,0 ); + }Like most kernel developers, I don't know the first thing about GCC internals so I asked our GCC team and Richard (CC'd) reckons this should be:
Quoted 4 lines
if (GET_CODE(body) == PARALLEL) body = XVECEXP(body, 0, 0);instead of the hunk above. Can you give that a go instead, please?
Thanks a lot!
Would you be so kind to take a look at the whole STACKLEAK plugin?
http://www.openwall.com/lists/kernel-hardening/2018/02/16/4
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/gcc-plugin/stackleak&id=57a0a6763b12e82dd462593d0f42be610e93cdc9
It's not very big. I documented it in detail. I would be really grateful for the review!
Best regards,
Alexander
Re: [PATCH 1/2] stackleak: Update for arm64
Hi Alexander,
Sorry for the slow reply, been caught up in an office move.
Quoted 35 lines · Alexander Popov
Hello Will, Richard and GCC folks!
Quoted 27 lines · Will Deacon
Quoted 14 lines · Laura Abbott
arm64 has another layer of indirection in the RTL. Account for this in the plugin.
Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/stackleak_plugin.c | 5 +++++ 1 file changed, 5 insertions(+)scripts/gcc-plugins/stackleak_plugin.c +0
-0scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 251
Quoted 8 lines
scripts/gcc-plugins/stackleak_plugin.c +5
-0@@ -244,6 +244,11 @@ static unsigned int stackleak_final_execute(void)* that insn. */ body =PATTERN (insn); +/* arm64 is different */ +if (GET_CODE (body) ==PARALLEL ) { + body =XEXP (body,0 ); + body =XEXP (body,0 ); + }Like most kernel developers, I don't know the first thing about GCC internals so I asked our GCC team and Richard (CC'd) reckons this should be:
Quoted 4 lines
if (GET_CODE(body) == PARALLEL) body = XVECEXP(body, 0, 0);instead of the hunk above. Can you give that a go instead, please?
Quoted 8 lines
Thanks a lot!
Would you be so kind to take a look at the whole STACKLEAK plugin?
http://www.openwall.com/lists/kernel-hardening/2018/02/16/4
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/gcc-plugin/stackleak&id=57a0a6763b12e82dd462593d0f42be610e93cdc9It's not very big. I documented it in detail. I would be really grateful for the review!
Looks good to me FWIW. Just a couple of minor things:
Quoted 4 lines
(quoted diff) +4
-0+ /* + * 1. Loop through the GIMPLE statements in each of cfun basic blocks. + * cfun is a global variable which represents the function that is + * currently processed.
Review
Quoted 10 lines
(quoted diff) +10
-0+ */ + FOR_EACH_BB_FN(bb, cfun) { + for (gsi = gsi_start_bb(bb); !gsi_end_p(gsi); gsi_next(&gsi)) { + gimple stmt; + + stmt = gsi_stmt(gsi); + + /* Leaf function is a function which makes no calls */ + if (is_gimple_call(stmt)) + is_leaf = false;
It's probably not going to matter in practice, but it might be worth emphasising in the comments that this leafness is only approximate. It will sometimes be a false positive, because we could still end up creating calls to libgcc functions from non-call statements (or for target-specific reasons). It can also be a false negative, since call statements can be to built-in or internal functions that end up being open-coded.
Review
Quoted 5 lines
(quoted diff) +5
-0+ /* + * The stackleak_final pass should be executed before the "final" pass, + * which turns the RTL (Register Transfer Language) into assembly. + */ + PASS_INFO(stackleak_final, "final", 1, PASS_POS_INSERT_BEFORE);
This might be too late, since it happens e.g. after addresses have been calculated for branch ranges, and after machine-specific passes (e.g. bundling on ia64).
The stack size is final after reload, so inserting the pass after that might be better.
Thanks,
Richard
Re: [PATCH 1/2] stackleak: Update for arm64
Quoted 3 lines · Richard Sandiford
Hi Alexander,
Sorry for the slow reply, been caught up in an office move.
Thank you very much for the review, Richard!
Quoted 19 lines
Quoted 6 lines · Alexander Popov
Would you be so kind to take a look at the whole STACKLEAK plugin?
http://www.openwall.com/lists/kernel-hardening/2018/02/16/4
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/gcc-plugin/stackleak&id=57a0a6763b12e82dd462593d0f42be610e93cdc9It's not very big. I documented it in detail. I would be really grateful for the review!
Looks good to me FWIW. Just a couple of minor things:
Quoted 9 lines
(quoted diff) +9
-0+ /* + * 1. Loop through the GIMPLE statements in each of cfun basic blocks. + * cfun is a global variable which represents the function that is + * currently processed. + */ + FOR_EACH_BB_FN(bb, cfun) { + for (gsi = gsi_start_bb(bb); !gsi_end_p(gsi); gsi_next(&gsi)) { + gimple stmt; +
Quoted 8 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ stmt = gsi_stmt(gsi); + + /* Leaf function is a function which makes no calls */ + if (is_gimple_call(stmt)) + is_leaf = false;It's probably not going to matter in practice, but it might be worth emphasising in the comments that this leafness is only approximate.
That's important, thank you! May I ask why you think it's not going to matter in practice?
Quoted 5 lines
It will sometimes be a false positive, because we could still end up creating calls to libgcc functions from non-call statements (or for target-specific reasons). It can also be a false negative, since call statements can be to built-in or internal functions that end up being open-coded.
Oh, that raises the question: how does this leafness inaccuracy affect in my particular case?
is_leaf is currently used for finding the special cases to skip the track_stack() call insertion:
/*
* Special cases to skip the instrumentation.
*
* Taking the address of static inline functions materializes them,
* but we mustn't instrument some of them as the resulting stack
* alignment required by the function call ABI will break other
* assumptions regarding the expected (but not otherwise enforced)
* register clobbering ABI.
*
* Case in point: native_save_fl on amd64 when optimized for size
* clobbers rdx if it were instrumented here.
*
* TODO: any more special cases?
*/
if (is_leaf &&
!TREE_PUBLIC(current_function_decl) &&
DECL_DECLARED_INLINE_P(current_function_decl)) {
return 0;
}And now it seems to me that the stackleak plugin should not instrument all static inline functions, regardless of is_leaf. Do you agree?
Quoted 10 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ /* + * The stackleak_final pass should be executed before the "final" pass, + * which turns the RTL (Register Transfer Language) into assembly. + */ + PASS_INFO(stackleak_final, "final", 1, PASS_POS_INSERT_BEFORE);This might be too late, since it happens e.g. after addresses have been calculated for branch ranges, and after machine-specific passes (e.g. bundling on ia64).
Quoted 2 lines
The stack size is final after reload, so inserting the pass after that might be better.
Thanks for that notice. May I ask for the additional clarification?
I specified -fdump-passes and see a lot of passes between reload and final:
...
rtl-sched1 : OFF
rtl-ira : ON
rtl-reload : ON
rtl-vzeroupper : OFF
*all-postreload : OFF
rtl-postreload : OFF
rtl-gcse2 : OFF
rtl-split2 : ON
rtl-ree : ON
rtl-cmpelim : OFF
rtl-btl1 : OFF
rtl-pro_and_epilogue : ON
rtl-dse2 : ON
rtl-csa : ON
rtl-jump2 : ON
rtl-compgotos : ON
rtl-sched_fusion : OFF
rtl-peephole2 : ON
rtl-ce3 : ON
rtl-rnreg : OFF
rtl-cprop_hardreg : ON
rtl-rtl_dce : ON
rtl-bbro : ON
rtl-btl2 : OFF
*leaf_regs : ON
rtl-split4 : ON
rtl-sched2 : ON
*stack_regs : ON
rtl-split3 : OFF
rtl-stack : ON
*all-late_compilation : OFF
rtl-alignments : ON
rtl-vartrack : ON
*free_cfg : ON
rtl-mach : ON
rtl-barriers : ON
rtl-dbr : OFF
rtl-split5 : OFF
rtl-eh_ranges : OFF
rtl-shorten : ON
rtl-nothrow : ON
rtl-dwarf2 : ON
rtl-stackleak_final : ON
rtl-final : ON
rtl-dfinish : ON
clean_state : ONWhere exactly would you recommend me to insert the stackleak_final pass, which removes the unneeded track_stack() calls?
Best regards,
Alexander
Re: [PATCH 1/2] stackleak: Update for arm64
Quoted 36 lines · Alexander Popov
Quoted 3 lines · Richard Sandiford
Hi Alexander,
Sorry for the slow reply, been caught up in an office move.
Thank you very much for the review, Richard!
Quoted 20 lines
Quoted 7 lines · Alexander Popov
Would you be so kind to take a look at the whole STACKLEAK plugin?
http://www.openwall.com/lists/kernel-hardening/2018/02/16/4
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/gcc-plugin/stackleak&id=57a0a6763b12e82dd462593d0f42be610e93cdc9It's not very big. I documented it in detail. I would be really grateful for the
review!Looks good to me FWIW. Just a couple of minor things:
Quoted 9 lines
(quoted diff) +9
-0+ /* + * 1. Loop through the GIMPLE statements in each of cfun basic blocks. + * cfun is a global variable which represents the function that is + * currently processed. + */ + FOR_EACH_BB_FN(bb, cfun) { + for (gsi = gsi_start_bb(bb); !gsi_end_p(gsi); gsi_next(&gsi)) { + gimple stmt; +Quoted 8 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ stmt = gsi_stmt(gsi); + + /* Leaf function is a function which makes no calls */ + if (is_gimple_call(stmt)) + is_leaf = false;It's probably not going to matter in practice, but it might be worth emphasising in the comments that this leafness is only approximate.
Quoted 2 lines
That's important, thank you! May I ask why you think it's not going to matter in practice?
I just thought the kind of calls it misses are going to have very shallow frames, but from what you said later I guess that isn't the point. It also might be a bit too hand-wavy for something like this :-)
Quoted 27 lines
Quoted 5 lines
It will sometimes be a false positive, because we could still end up creating calls to libgcc functions from non-call statements (or for target-specific reasons). It can also be a false negative, since call statements can be to built-in or internal functions that end up being open-coded.
Oh, that raises the question: how does this leafness inaccuracy affect in my particular case?
is_leafis currently used for finding the special cases to skip thetrack_stack()call insertion:/*
* Special cases to skip the instrumentation.
*
* Taking the address of static inline functions materializes them,
* but we mustn't instrument some of them as the resulting stack
* alignment required by the function call ABI will break other
* assumptions regarding the expected (but not otherwise enforced)
* register clobbering ABI.
*
* Case in point:native_save_flon amd64 when optimized for size
* clobbers rdx if it were instrumented here.
*
* TODO: any more special cases?
*/
if (is_leaf&&
Quoted 8 lines
!TREE_PUBLIC(current_function_decl) && DECL_DECLARED_INLINE_P(current_function_decl)) { return 0; }And now it seems to me that the stackleak plugin should not instrument all static inline functions, regardless of
is_leaf. Do you agree?
OK. I'd missed that this was just a heuristic to detect certain kinds of linux function, so it's probably fine as it is.
Not sure whether it's safe to punt for general static inline functions. E.g. couldn't you have a static inline function that just provides a more convenient interface to another function? But I guess it's a linux-specific heuristic, so I can't really say.
TBH the paravirt save_fl stuff seems like dancing on the edge, but that's another story. :-)
Quoted 64 lines
Quoted 10 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ /* + * The stackleak_final pass should be executed before the "final" pass, + * which turns the RTL (Register Transfer Language) into assembly. + */ + PASS_INFO(stackleak_final, "final", 1, PASS_POS_INSERT_BEFORE);This might be too late, since it happens e.g. after addresses have been calculated for branch ranges, and after machine-specific passes (e.g. bundling on ia64).
Quoted 2 lines
The stack size is final after reload, so inserting the pass after that might be better.
Thanks for that notice. May I ask for the additional clarification?
I specified -fdump-passes and see a lot of passes between reload and final:
...rtl-sched1 : OFF rtl-ira : ON rtl-reload : ON rtl-vzeroupper : OFF *all-postreload : OFF rtl-postreload : OFF rtl-gcse2 : OFF rtl-split2 : ON rtl-ree : ON rtl-cmpelim : OFF rtl-btl1 : OFF rtl-pro_and_epilogue : ON rtl-dse2 : ON rtl-csa : ON rtl-jump2 : ON rtl-compgotos : ON rtl-sched_fusion : OFF rtl-peephole2 : ON rtl-ce3 : ON rtl-rnreg : OFF rtl-cprop_hardreg : ON rtl-rtl_dce : ON rtl-bbro : ON rtl-btl2 : OFF *leaf_regs : ON rtl-split4 : ON rtl-sched2 : ON *stack_regs : ON rtl-split3 : OFF rtl-stack : ON *all-late_compilation : OFF rtl-alignments : ON rtl-vartrack : ON *free_cfg : ON rtl-mach : ON rtl-barriers : ON rtl-dbr : OFF rtl-split5 : OFF rtl-eh_ranges : OFF rtl-shorten : ON rtl-nothrow : ON rtl-dwarf2 : ON rtl-stackleak_final : ON rtl-final : ON rtl-dfinish : ON clean_state : ON
Quoted 2 lines
Where exactly would you recommend me to insert the
stackleak_finalpass, which removes the unneededtrack_stack()calls?
Directly after rtl-reload seems best. That's the first point at which the frame size is final, and reload is one of the few rtl passes that always runs. Doing it there could also help with things like shrink wrapping (part of rtl-pro_and_epilogue).
Thanks,
Richard
Re: [PATCH 1/2] stackleak: Update for arm64
Thanks for your reply, Richard!
Quoted 74 lines · Richard Sandiford
Quoted 30 lines · Alexander Popov
Quoted 20 lines · Richard Sandiford
Quoted 7 lines · Alexander Popov
Would you be so kind to take a look at the whole STACKLEAK plugin?
http://www.openwall.com/lists/kernel-hardening/2018/02/16/4
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git/commit/?h=kspp/gcc-plugin/stackleak&id=57a0a6763b12e82dd462593d0f42be610e93cdc9It's not very big. I documented it in detail. I would be really grateful for the
review!Looks good to me FWIW. Just a couple of minor things:
Quoted 9 lines
(quoted diff) +9
-0+ /* + * 1. Loop through the GIMPLE statements in each of cfun basic blocks. + * cfun is a global variable which represents the function that is + * currently processed. + */ + FOR_EACH_BB_FN(bb, cfun) { + for (gsi = gsi_start_bb(bb); !gsi_end_p(gsi); gsi_next(&gsi)) { + gimple stmt; +Quoted 8 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ stmt = gsi_stmt(gsi); + + /* Leaf function is a function which makes no calls */ + if (is_gimple_call(stmt)) + is_leaf = false;It's probably not going to matter in practice, but it might be worth emphasising in the comments that this leafness is only approximate.
Quoted 2 lines
That's important, thank you! May I ask why you think it's not going to matter in practice?
I just thought the kind of calls it misses are going to have very shallow frames, but from what you said later I guess that isn't the point. It also might be a bit too hand-wavy for something like this :-)
Quoted 27 lines
Quoted 5 lines
It will sometimes be a false positive, because we could still end up creating calls to libgcc functions from non-call statements (or for target-specific reasons). It can also be a false negative, since call statements can be to built-in or internal functions that end up being open-coded.
Oh, that raises the question: how does this leafness inaccuracy affect in my particular case?
is_leafis currently used for finding the special cases to skip thetrack_stack()call insertion:/*
* Special cases to skip the instrumentation.
*
* Taking the address of static inline functions materializes them,
* but we mustn't instrument some of them as the resulting stack
* alignment required by the function call ABI will break other
* assumptions regarding the expected (but not otherwise enforced)
* register clobbering ABI.
*
* Case in point:native_save_flon amd64 when optimized for size
* clobbers rdx if it were instrumented here.
*
* TODO: any more special cases?
*/
if (is_leaf&&Quoted 8 lines
!TREE_PUBLIC(current_function_decl) && DECL_DECLARED_INLINE_P(current_function_decl)) { return 0; }And now it seems to me that the stackleak plugin should not instrument all static inline functions, regardless of
is_leaf. Do you agree?
Quoted 7 lines
OK. I'd missed that this was just a heuristic to detect certain kinds of linux function, so it's probably fine as it is.
Not sure whether it's safe to punt for general static inline functions. E.g. couldn't you have a static inline function that just provides a more convenient interface to another function? But I guess it's a linux-specific heuristic, so I can't really say.
Huh, I got the insight! I think that the current approach (originally by PaX Team) should work fine despite the false positives which you described:
If some static inline function already does explicit calls (so is_leaf is false), adding the track_stack() call will not introduce anything special that can break the aforementioned register clobbering ABI in that function.
Does it sound reasonable?
However, I don't know what to with false negatives.
Quoted 2 lines
TBH the paravirt
save_flstuff seems like dancing on the edge, but that's another story. :-)
That's interesting. Could you elaborate on that?
Quoted 16 lines
Quoted 16 lines
Quoted 10 lines
Review
Quoted 5 lines
(quoted diff) +5
-0+ /* + * The stackleak_final pass should be executed before the "final" pass, + * which turns the RTL (Register Transfer Language) into assembly. + */ + PASS_INFO(stackleak_final, "final", 1, PASS_POS_INSERT_BEFORE);This might be too late, since it happens e.g. after addresses have been calculated for branch ranges, and after machine-specific passes (e.g. bundling on ia64).
Quoted 2 lines
The stack size is final after reload, so inserting the pass after that might be better.
Thanks for that notice. May I ask for the additional clarification?
I specified -fdump-passes and see a lot of passes between reload and final:
...
Quoted 8 lines
Quoted 3 lines
Where exactly would you recommend me to insert the
stackleak_finalpass, which removes the unneededtrack_stack()calls?Directly after rtl-reload seems best. That's the first point at which the frame size is final, and reload is one of the few rtl passes that always runs. Doing it there could also help with things like shrink wrapping (part of rtl-pro_and_epilogue).
Thanks a lot for your detailed answer. I'll follow your advice in the next version of the patch series.
Best regards,
Alexander
Re: [PATCH 1/2] stackleak: Update for arm64
In reply to Will DeaconQuoted 29 lines · Will Deacon
Hi Laura,
Quoted 14 lines · Laura Abbott
arm64 has another layer of indirection in the RTL. Account for this in the plugin.
Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/stackleak_plugin.c | 5 +++++ 1 file changed, 5 insertions(+)scripts/gcc-plugins/stackleak_plugin.c +0
-0scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 251
Quoted 8 lines
scripts/gcc-plugins/stackleak_plugin.c +5
-0@@ -244,6 +244,11 @@ static unsigned int stackleak_final_execute(void)* that insn. */ body =PATTERN (insn); +/* arm64 is different */ +if (GET_CODE (body) ==PARALLEL ) { + body =XEXP (body,0 ); + body =XEXP (body,0 ); + }Like most kernel developers, I don't know the first thing about GCC internals so I asked our GCC team and Richard (CC'd) reckons this should be:
Quoted 9 lines
if (GET_CODE(body) == PARALLEL) body = XVECEXP(body, 0, 0);instead of the hunk above. Can you give that a go instead, please?
Cheers,
Will
Yep, seems to work fine and makes sense from my understanding of gcc internals. I'll fix it up for the next version. Thanks for the review!
Laura
PATCH 2/2 arm64: Clear the stack
In reply to Laura AbbottImplementation of stackleak based heavily on the x86 version
--- arch/arm64/Kconfig | 1 + arch/arm64/include/asm/processor.h | 6 ++ arch/arm64/kernel/asm-offsets.c | 3 + arch/arm64/kernel/entry.S | 108 ++++++++++++++++++++++++++++++++++ arch/arm64/kernel/process.c | 16 +++++ drivers/firmware/efi/libstub/Makefile | 3 +- scripts/Makefile.gcc-plugins | 5 +- 7 files changed, 140 insertions(+), 2 deletions(-)
arch/arm64/Kconfig +1 -0
@@ -92,6 +92,7 @@ config ARM64 select HAVE_ARCH_MMAP_RND_BITS select HAVE_ARCH_MMAP_RND_COMPAT_BITS if COMPAT select HAVE_ARCH_SECCOMP_FILTER + select HAVE_ARCH_STACKLEAK select HAVE_ARCH_THREAD_STRUCT_WHITELIST select HAVE_ARCH_TRACEHOOK select HAVE_ARCH_TRANSPARENT_HUGEPAGE
arch/arm64/include/asm/processor.h +6 -0
@@ -114,6 +114,12 @@ struct thread_struct {unsigned long fault_address;/* fault info */ unsigned long fault_code;/* ESR_EL1 value */ struct debug_info debug;/* debugging */ +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +unsigned long lowest_stack; +#ifdef CONFIG_STACKLEAK_METRICS +unsigned long prev_lowest_stack; +#endif +#endif };/*
arch/arm64/kernel/asm-offsets.c +3 -0
@@ -45,6 +45,9 @@ int main(void) DEFINE(TSK_TI_TTBR0, offsetof(struct task_struct, thread_info.ttbr0));#endif DEFINE (TSK_STACK , offsetof(struct task_struct , stack)); +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + DEFINE(TSK_TI_LOWEST_STACK, offsetof(struct task_struct, thread.lowest_stack)); +#endif BLANK (); DEFINE(THREAD_CPU_CONTEXT, offsetof(struct task_struct, thread.cpu_context));BLANK ();
arch/arm64/kernel/entry.S +108 -0
123 lines not on this page · raw message
arch/arm64/kernel/process.c +16 -0
@@ -357,6 +357,9 @@ int copy_thread(unsigned long clone_flags, unsigned long stack_start, p->thread.cpu_context.pc = (unsigned long )ret_from_fork; p->thread.cpu_context.sp = (unsigned long )childregs; +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + p->thread.lowest_stack = (unsigned long )task_stack_page (p); +#endif ptrace_hw_copy_thread (p);return 0 ;
@@ -486,3 +489,16 @@ void arch_setup_new_exec(void) { current->mm->context.flags =is_compat_task () ?MMCF_AARCH32 :0 ; } + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +void __used check_alloca (unsigned long size) +{ +unsigned long sp, stack_left; + + sp = current_stack_pointer; + + stack_left = sp & (THREAD_SIZE -1 ); +BUG_ON (stack_left <256 || size >= stack_left -256 ); +} +EXPORT_SYMBOL (check_alloca); +#endif
drivers/firmware/efi/libstub/Makefile +2 -1
@@ -20,7 +20,8 @@ cflags-$(CONFIG_EFI_ARMSTUB) += -I$(srctree)/scripts/dtc/libfdt KBUILD_CFLAGS := $(cflags-y) -DDISABLE_BRANCH_PROFILING \ -D__NO_FORTIFY \ $(call cc-option,-ffreestanding) \ - $(call cc-option,-fno-stack-protector) + $(call cc-option,-fno-stack-protector) \ + $(DISABLE_STACKLEAK_PLUGIN) GCOV_PROFILE := n KASAN_SANITIZE := n
scripts/Makefile.gcc-plugins +4 -1
@@ -37,11 +37,14 @@ ifdef CONFIG_GCC_PLUGINS gcc-plugin-$(CONFIG_GCC_PLUGIN_STACKLEAK) += stackleak_plugin.so gcc-plugin-cflags-$(CONFIG_GCC_PLUGIN_STACKLEAK) += -DSTACKLEAK_PLUGIN -fplugin-arg-stackleak_plugin-track-min-size=$(CONFIG_STACKLEAK_TRACK_MIN_SIZE) + ifdef CONFIG_GCC_PLUGIN_STACKLEAK + DISABLE_STACKLEAK_PLUGIN += -fplugin-arg-stackleak_plugin-disable + endif GCC_PLUGINS_CFLAGS := $(strip $(addprefix -fplugin=$(objtree)/scripts/gcc-plugins/, $(gcc-plugin-y)) $(gcc-plugin-cflags-y)) export PLUGINCC GCC_PLUGINS_CFLAGS GCC_PLUGIN GCC_PLUGIN_SUBDIR - export SANCOV_PLUGIN DISABLE_LATENT_ENTROPY_PLUGIN + export SANCOV_PLUGIN DISABLE_LATENT_ENTROPY_PLUGIN DISABLE_STACKLEAK_PLUGIN ifneq ($(PLUGINCC),) # SANCOV_PLUGIN can be only in CFLAGS_KCOV because avoid duplication.
Signature · 1 line
2.14.3
Mark Rutland
reviewerRe: [PATCH 2/2] arm64: Clear the stack
Hi Laura,
Quoted 1 line · Laura Abbott
Implementation of stackleak based heavily on the x86 version
Neat!
Quoted 17 lines
arch/arm64/kernel/entry.S +5
-0@@ -401,6 +401,11 @@ tsk .req x28 // current thread_info .text + .macro erase_kstack +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + bl __erase_kstack +#endif + .endm /* * Exception vectors. */
arch/arm64/kernel/entry.S work_pending: · line 909
Quoted 4 lines
arch/arm64/kernel/entry.S +1
-0@@ -901,6 +906,7 @@ work_pending: */ ret_to_user: disable_daif + erase_kstack
I *think* this should happen in finish_ret_to_user a few lines down, since we can call C code if we branch to work_pending, dirtying the stack.
Quoted 7 lines
arch/arm64/kernel/entry.S +0
-0ldr x1, [tsk, #TSK_TI_FLAGS] and x2, x1, #_TIF_WORK_MASK cbnz x2, work_pending@@ -1337,3 +1343,105 @@ alternative_else_nop_endif ENDPROC(__sdei_asm_handler) NOKPROBE(__sdei_asm_handler) #endif /* CONFIG_ARM_SDE_INTERFACE */
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1355
Quoted 10 lines
arch/arm64/kernel/entry.S +10
-0+ +/* + * This is what the stack looks like + * + * +---+ <- task_stack_page(p) + THREAD_SIZE + * | | + * +---+ <- task_stack_page(p) + THREAD_START_SP + * | | + * | | + * +---+ <- task_pt_regs(p)
THREAD_START_SP got killed off in commit 34be98f4944f9907 as part of the VMAP_STACK rework, so this can be:
+---+ <- task_stack_page(p) + THREAD_SIZE
| |
| |
+---+ <- task_pt_regs(p)
...Quoted 9 lines
arch/arm64/kernel/entry.S +9
-0+ * | | + * | | + * | | <- current_sp + * ~~~~~ + * + * ~~~~~ + * | | <- lowest_stack + * | | + * | |
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1374
Quoted 10 lines
arch/arm64/kernel/entry.S +10
-0+ * +---+ <- task_stack_page(p) + * + * This function is desgned to poison the memory between the lowest_stack + * and the current stack pointer. After clearing the stack, the lowest + * stack is reset. + */ + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +ENTRY(__erase_kstack) + mov x10, x0 // save x0 for the fast path
AFAICT, we only call this from ret_to_user, where x0 doesn't need to be preserved.
Is that for ret_fast_syscall? In some cases, ret_fast_syscall can bypass ret_to_user and calls kernel_exit directly, so we might need a call there.
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1387
Quoted 13 lines
arch/arm64/kernel/entry.S +13
-0+ + get_thread_info x0 + ldr x1, [x0, #TSK_TI_LOWEST_STACK] + + /* get the number of bytes to check for lowest stack */ + mov x3, x1 + and x3, x3, #THREAD_SIZE - 1 + lsr x3, x3, #3 + + /* generate addresses from the bottom of the stack */ + mov x4, sp + movn x2, #THREAD_SIZE - 1 + and x1, x4, x2
Can we replace the MOVN;AND with a single instruction to clear the low bits? e.g.
mov x4, sp bic x1, x4, #THREAD_SIZE - 1
... IIUC BIC is an alias for the bitfield instructions, though I can't recall exactly which one(s).
Quoted 34 lines
arch/arm64/kernel/entry.S +34
-0+ + mov x2, #STACKLEAK_POISON + + mov x5, #0 +1: + /* + * As borrowed from the x86 logic, start from the lowest_stack + * and go to the bottom to find the poison value. + * The check of 16 is to hopefully avoid false positives. + */ + cbz x3, 4f + ldr x4, [x1, x3, lsl #3] + cmp x4, x2 + csinc x5, xzr, x5, ne + tbnz x5, #STACKLEAK_POISON_CHECK_DEPTH/4, 4f // found 16 poisons? + sub x3, x3, #1 + b 1b + +4: + /* total number of bytes to poison */ + add x5, x1, x3, lsl #3 + mov x4, sp + sub x8, x4, x5 + + cmp x8, #THREAD_SIZE // sanity check the range + b.lo 5f + ASM_BUG() + +5: + /* + * We may have hit a path where the stack did not get used, + * no need to do anything here + */ + cbz x8, 7f
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1431
Quoted 10 lines
arch/arm64/kernel/entry.S +10
-0+ + sub x8, x8, #1 // don't poison the current stack pointer + + lsr x8, x8, #3 + add x3, x3, x8 + + /* + * The logic of this loop ensures the last stack word isn't + * ovewritten. + */
Is that to ensure that we don't clobber the word at the current sp value?
Quoted 16 lines
arch/arm64/kernel/entry.S +16
-0+6: + cbz x8, 7f + str x2, [x1, x3, lsl #3] + sub x3, x3, #1 + sub x8, x8, #1 + b 6b + + /* Reset the lowest stack to the top of the stack */ +7: + mov x1, sp + str x1, [x0, #TSK_TI_LOWEST_STACK] + + mov x0, x10 + ret +ENDPROC(__erase_kstack) +#endif
[...]
drivers/firmware/efi/libstub/Makefile cflags-$(CONFIG_EFI_ARMSTUB) += -I$(srctree)/scripts/dtc/libfdt · line 24
Quoted 11 lines
drivers/firmware/efi/libstub/Makefile +2
-1@@ -20,7 +20,8 @@ cflags-$(CONFIG_EFI_ARMSTUB) += -I$(srctree)/scripts/dtc/libfdt KBUILD_CFLAGS := $(cflags-y) -DDISABLE_BRANCH_PROFILING \ -D__NO_FORTIFY \ $(call cc-option,-ffreestanding) \ - $(call cc-option,-fno-stack-protector) + $(call cc-option,-fno-stack-protector) \ + $(DISABLE_STACKLEAK_PLUGIN)
I believe the KVM hyp code will also need to opt-out of this.
Thanks,
Mark.
Re: [PATCH 2/2] arm64: Clear the stack
Quoted 15 lines · Mark Rutland
Hi Laura,
Quoted 1 line · Laura Abbott
Implementation of stackleak based heavily on the x86 version
Neat!
Quoted 8 lines
arch/arm64/kernel/entry.S +0
-0@@ -401,6 +401,11 @@ tsk .req x28 // current thread_info .text
Quoted 17 lines
arch/arm64/kernel/entry.S work_pending: · line 909
Quoted 13 lines
arch/arm64/kernel/entry.S +6
-0+ .macro erase_kstack +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK + bl __erase_kstack +#endif + .endm /* * Exception vectors. */@@ -901,6 +906,7 @@ work_pending: */ ret_to_user: disable_daif + erase_kstackI *think* this should happen in
finish_ret_to_usera few lines down, since we can call C code if we branch towork_pending, dirtying the stack.
I think you're right but this didn't immediately work when I tried it. I'll have to dig into this some more.
Quoted 21 lines
Quoted 7 lines
arch/arm64/kernel/entry.S +0
-0ldr x1, [tsk, #TSK_TI_FLAGS] and x2, x1, #_TIF_WORK_MASK cbnz x2, work_pending@@ -1337,3 +1343,105 @@ alternative_else_nop_endif ENDPROC(__sdei_asm_handler) NOKPROBE(__sdei_asm_handler) #endif /* CONFIG_ARM_SDE_INTERFACE */arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1355
Quoted 10 lines
arch/arm64/kernel/entry.S +10
-0+ +/* + * This is what the stack looks like + * + * +---+ <- task_stack_page(p) + THREAD_SIZE + * | | + * +---+ <- task_stack_page(p) + THREAD_START_SP + * | | + * | | + * +---+ <- task_pt_regs(p)
THREAD_START_SPgot killed off in commit34be98f4944f9907as part of theVMAP_STACKrework, so this can be:
Quoted 6 lines
+---+ <- task_stack_page(p) + THREAD_SIZE
| |
| |
+---+ <- task_pt_regs(p)
...Good point.
Quoted 23 lines
Quoted 9 lines
arch/arm64/kernel/entry.S +9
-0+ * | | + * | | + * | | <- current_sp + * ~~~~~ + * + * ~~~~~ + * | | <- lowest_stack + * | | + * | |arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1374
Quoted 10 lines
arch/arm64/kernel/entry.S +10
-0+ * +---+ <- task_stack_page(p) + * + * This function is desgned to poison the memory between the lowest_stack + * and the current stack pointer. After clearing the stack, the lowest + * stack is reset. + */ + +#ifdef CONFIG_GCC_PLUGIN_STACKLEAK +ENTRY(__erase_kstack) + mov x10, x0 // save x0 for the fast pathAFAICT, we only call this from
ret_to_user, where x0 doesn't need to be preserved.
Quoted 3 lines
Is that for
ret_fast_syscall? In some cases,ret_fast_syscallcan bypassret_to_userand callskernel_exitdirectly, so we might need a call there.
This was a hold over when I was experimenting with calling erase_kstack more places, one of which came through ret_fast_syscall. I realized later that the erase was unnecessary but accidentally kept the saving in. I'll see about removing it assuming we don't decide later to put a call on the fast path.
Quoted 17 lines
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1387
Quoted 13 lines
arch/arm64/kernel/entry.S +13
-0+ + get_thread_info x0 + ldr x1, [x0, #TSK_TI_LOWEST_STACK] + + /* get the number of bytes to check for lowest stack */ + mov x3, x1 + and x3, x3, #THREAD_SIZE - 1 + lsr x3, x3, #3 + + /* generate addresses from the bottom of the stack */ + mov x4, sp + movn x2, #THREAD_SIZE - 1 + and x1, x4, x2Can we replace the MOVN;AND with a single instruction to clear the low bits? e.g.
Quoted 6 lines
mov x4, sp bic x1, x4, #THREAD_SIZE - 1... IIUC BIC is an alias for the bitfield instructions, though I can't recall exactly which one(s).
Good suggestion.
Quoted 38 lines
Quoted 38 lines
arch/arm64/kernel/entry.S +38
-0+ + mov x2, #STACKLEAK_POISON + + mov x5, #0 +1: + /* + * As borrowed from the x86 logic, start from the lowest_stack + * and go to the bottom to find the poison value. + * The check of 16 is to hopefully avoid false positives. + */ + cbz x3, 4f + ldr x4, [x1, x3, lsl #3] + cmp x4, x2 + csinc x5, xzr, x5, ne + tbnz x5, #STACKLEAK_POISON_CHECK_DEPTH/4, 4f // found 16 poisons? + sub x3, x3, #1 + b 1b + +4: + /* total number of bytes to poison */ + add x5, x1, x3, lsl #3 + mov x4, sp + sub x8, x4, x5 + + cmp x8, #THREAD_SIZE // sanity check the range + b.lo 5f + ASM_BUG() + +5: + /* + * We may have hit a path where the stack did not get used, + * no need to do anything here + */ + cbz x8, 7f + + sub x8, x8, #1 // don't poison the current stack pointer + + lsr x8, x8, #3
Quoted 9 lines
arch/arm64/kernel/entry.S alternative_else_nop_endif · line 1431
Quoted 6 lines
arch/arm64/kernel/entry.S +6
-0+ add x3, x3, x8 + + /* + * The logic of this loop ensures the last stack word isn't + * ovewritten. + */Is that to ensure that we don't clobber the word at the current sp value?
Correct.
Quoted 19 lines
Quoted 16 lines
arch/arm64/kernel/entry.S +16
-0+6: + cbz x8, 7f + str x2, [x1, x3, lsl #3] + sub x3, x3, #1 + sub x8, x8, #1 + b 6b + + /* Reset the lowest stack to the top of the stack */ +7: + mov x1, sp + str x1, [x0, #TSK_TI_LOWEST_STACK] + + mov x0, x10 + ret +ENDPROC(__erase_kstack) +#endif[...]
Quoted 14 lines
drivers/firmware/efi/libstub/Makefile cflags-$(CONFIG_EFI_ARMSTUB) += -I$(srctree)/scripts/dtc/libfdt · line 24
Quoted 11 lines
drivers/firmware/efi/libstub/Makefile +2
-1@@ -20,7 +20,8 @@ cflags-$(CONFIG_EFI_ARMSTUB) += -I$(srctree)/scripts/dtc/libfdt KBUILD_CFLAGS := $(cflags-y) -DDISABLE_BRANCH_PROFILING \ -D__NO_FORTIFY \ $(call cc-option,-ffreestanding) \ - $(call cc-option,-fno-stack-protector) + $(call cc-option,-fno-stack-protector) \ + $(DISABLE_STACKLEAK_PLUGIN)I believe the KVM hyp code will also need to opt-out of this.
I'll double check that.
Quoted 3 lines
Thanks,
Mark.
Thanks,
Laura
Re: [PATCH 2/2] arm64: Clear the stack
Quoted 6 lines · Laura Abbott
Quoted 3 lines
I *think* this should happen in
finish_ret_to_usera few lines down, since we can call C code if we branch towork_pending, dirtying the stack.I think you're right but this didn't immediately work when I tried it. I'll have to dig into this some more.
Okay I figured this out. Not corrupting registers works wonders.
Re: [PATCH 0/2] Stackleak for arm64
In reply to Laura AbbottQuoted 8 lines · Laura Abbott
This is the arm64 version of the STACKLEAK plugin originall from grsecurity. See
https://marc.info/?l=kernel-hardening&m=151880470609808 for the full x86 version. This is based on top of Kees' branch for stackleak and has been cleaned up to use a few macros from that branch.Comments welcome, if there are no major objections Kees will queue this up to get some CI testing. This passed both of the LKDTM tests.
Hello, Laura,
Thank you. I'll take some time to learn your patches and test them on my LeMaker HiKey board. I'll return with the feedback.
Best regards,
Alexander
Borislav Petkov
maintainerQuoted 7 lines · Kees Cook
Quoted 4 lines · Alexander Popov
Quoted 3 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
If we can borrow some of luto or tglx's time, I think that'd be best:
Btw, someone just noticed: why hasn't this patchset ever been sent to lkml for a wider review?
Signature · 4 lines
Regards/Gruss,
Boris.
Good mailing practices for 400: avoid top-posting and trim the reply.Quoted 11 lines · Borislav Petkov
Quoted 7 lines · Kees Cook
Quoted 4 lines · Alexander Popov
Quoted 3 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
If we can borrow some of luto or tglx's time, I think that'd be best:
Btw, someone just noticed: why hasn't this patchset ever been sent to lkml for a wider review?
Thanks for that notice.
I just realized that I'm resending this patch series to people who I chose approximately 8 month ago, when I started to work on STACKLEAK.
I'll use scripts/get_maintainer.pl for the next version.
Best regards,
Alexander
Hello Kees,
Quoted 17 lines · Kees Cook
Quoted 4 lines · Alexander Popov
Quoted 3 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
If we can borrow some of luto or tglx's time, I think that'd be best:
they've been looking at the entry code a lot lately. :) Regardless, I think the addition to the entry code is clean (especially now that the fast path is gone *sob*). :PQuoted 5 lines
Quoted 5 lines
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.
Quoted 4 lines
I've added this series to my kernel.org trees, which means 0-day will start grinding on it too now:
https://git.kernel.org/pub/scm/linux/kernel/git/kees/linux.git
kspp/gcc-plugin/stackleak
Great, thanks!
Quoted 3 lines
The LKDTM tests look great and check out for me. I think the code is clear, so I'd like to get it into -next, but I want to be sure I'm not stepping on x86 toes first.
Yes, I see. Thanks for your review anyway.
Quoted 13 lines
Laura, how does arm64 look for this? Would it be possible to add it to this series (at least on kernel.org for build/run testing)?
Quoted 10 lines
Hello! I've just tested STACKLEAK against the recent CVE-2017-17712 exploit:
http://openwall.com/lists/oss-security/2018/02/20/1This vulnerability is a race condition in
raw_sendmsg()innet/ipv4/raw.c. It leads to uninitialized stack pointer usage which can be used for a local privilege escalation.CVE-2017-17712 was discovered and fixed by Mohamed Ghannam (kudos to him!). He also provided a stable PoC exploit for it.
Quoted 7 lines
Quoted 3 lines
With STACKLEAK, the uninitialized stack pointer is set to
STACKLEAK_POISON(-0xBEEF) and points to the unused hole in the virtual memory map. That blocks the stack spraying needed for CVE-2017-17712 exploit (similar to CVE-2010-2963).Nice check! I wonder if the the byref structleak also solves it? Regardless, I think stackleak has better performance and greater coverage.
I disabled CONFIG_GCC_PLUGIN_STACKLEAK and instead enabled CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL. The msg pointer is now 0. So the access to msg->msg_iter now gives the following:
[ 7.324333] BUG: unable to handle kernel NULL pointer dereference at 0000000000000010 [ 7.325067] IP: csum_and_copy_from_iter_full+0x1e/0x400 [ 7.325067] PGD 800000007db6c067 P4D 800000007db6c067 PUD 7d34a067 PMD 0 [ 7.325067] Oops: 0000 [#1] SMP PTI [ 7.325067] Dumping ftrace buffer: [ 7.325067] (ftrace buffer empty) [ 7.325067] Modules linked in: [ 7.325067] CPU: 0 PID: 2708 Comm: poc Not tainted 4.16.0-rc1+ #5 [ 7.325067] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Ubuntu-1.8.2-1ubuntu1 04/01/2014 [ 7.325067] RIP: 0010:csum_and_copy_from_iter_full+0x1e/0x400 [ 7.325067] RSP: 0018:ffffc90000d839d0 EFLAGS: 00010246 [ 7.325067] RAX: 0000000000000000 RBX: ffff88003d0887c0 RCX: 0000000000000010 [ 7.325067] RDX: ffffc90000d83a64 RSI: 0000000000006400 RDI: ffff88003b828024 [ 7.325067] RBP: 0000000000000000 R08: 0000000000000000 R09: ffff88003d0887c0 [ 7.325067] R10: ffff88003b828024 R11: 0000000000000000 R12: 0000000000006400 [ 7.325067] R13: 0000000000006400 R14: ffff88003de53640 R15: ffff88007d90c110 [ 7.325067] FS: 00007f52585d0700(0000) GS:ffff88003ec00000(0000) knlGS:0000000000000000 [ 7.325067] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 7.325067] CR2: 0000000000000010 CR3: 000000007db70000 CR4: 00000000000006f0 [ 7.325067] Call Trace: [ 7.325067] ? __kmalloc_reserve.isra.41+0x29/0x80 [ 7.325067] ip_generic_getfrag+0x73/0xc0 [ 7.325067] __ip_append_data.isra.48+0x68b/0x890 [ 7.325067] ? raw_destroy+0x20/0x20 [ 7.325067] ? raw_destroy+0x20/0x20 [ 7.325067] ip_append_data.part.50+0x67/0xc0 [ 7.325067] raw_sendmsg+0x486/0xa50 [ 7.325067] ? __kmalloc+0x141/0x1a0 [ 7.325067] sock_sendmsg+0x31/0x40 [ 7.325067] ___sys_sendmsg+0x277/0x2d0 [ 7.325067] ? __sys_sendmsg+0x62/0xa0 [ 7.325067] __sys_sendmsg+0x62/0xa0 [ 7.325067] do_syscall_64+0x63/0x120 [ 7.325067] entry_SYSCALL_64_after_hwframe+0x21/0x86 [ 7.325067] RIP: 0033:0x7f525c186e90 [ 7.325067] RSP: 002b:00007f52585cff00 EFLAGS: 00000293 ORIG_RAX: 000000000000002e [ 7.325067] RAX: ffffffffffffffda RBX: 0000000000000003 RCX: 00007f525c186e90 [ 7.325067] RDX: 0000000000000000 RSI: 00000000011bd010 RDI: 0000000000000003 [ 7.325067] RBP: 00000000011bd010 R08: 0000000000000000 R09: 00007f52585d0700 [ 7.325067] R10: 00007f52585cff40 R11: 0000000000000293 R12: 0000000000000000 [ 7.325067] R13: 00007fffde2c70bf R14: 0000000000000000 R15: 00007f525c5b7040 [ 7.325067] Code: 4c 89 c3 eb f6 49 01 d8 e9 7a ff ff ff 41 57 41 56 41 55 41 54 55 53 48 83 ec 48 65 48 8b 04 25 28 00 00 00 48 89 44 24 40 31 c0 <8b> 01 48 89 14 24 a8 08 0f 85 e5 00 00 00 48 39 71 10 49 89 f6 [ 7.325067] RIP: csum_and_copy_from_iter_full+0x1e/0x400 RSP: ffffc90000d839d0 [ 7.325067] CR2: 0000000000000010 [ 7.382645] ---[ end trace f6927758360fb538 ]---
I think that would block the stack spraying as well.
Best regards,
Alexander
Quoted 237 lines · Alexander Popov
This is the 8th version of the patch series introducing STACKLEAK to the mainline kernel. I've made some minor improvements while waiting for the next review by x86 maintainers.
STACKLEAK is a security feature developed by Grsecurity/PaX (kudos to them), which:
- reduces the information that can be revealed through kernel stack leak bugs;
- blocks some uninitialized stack variable attacks (e.g. CVE-2010-2963);
- introduces some runtime checks for kernel stack overflow detection.STACKLEAK instrumentation statistics
====================================These numbers were obtained for the 4th version of the patch series.
Size of vmlinux (
x86_64_defconfig):
file size:
- STACKLEAK disabled: 35014784 bytes
- STACKLEAK enabled: 35044952 bytes (+0.086%)
.text section size (calculated by size utility):
- STACKLEAK disabled: 10752983
- STACKLEAK enabled: 11062221 (+2.876%)The readelf utility shows 45602 functions in vmlinux. The STACKLEAK gcc plugin inserted 36
check_alloca()calls and 1265track_stack()calls (42274 calls are inserted during GIMPLE pass and 41009 calls are deleted during RTL pass). So 2.853% of functions are instrumented.STACKLEAK performance impact
============================The STACKLEAK description in
Kconfigincludes:
"The tradeoff is the performance impact: on a single CPU system kernel compilation sees a 1% slowdown, other systems and workloads may vary and you are advised to test this feature on your expected workload before deploying it".Here are the results of a brief performance test on
x86_64(for the 2nd version of the patch). The numbers are very different because the STACKLEAK performance penalty depends on the workloads.Hardware: Intel Core i7-4770, 16 GB RAM
Test #1: building the Linux kernel with Ubuntu config (time make -j9) Result on 4.11-rc8:
real 32m14.893s user 237m30.886s sys 11m12.273s Result on 4.11-rc8+stackleak: real 32m26.881s (+0.62%) user 238m38.926s (+0.48%) sys 11m36.426s (+3.59%)Test #2: hackbench -s 4096 -l 2000 -g 15 -f 25 -P Average result on 4.11-rc8: 8.71s
Average result on 4.11-rc8+stackleak: 9.08s (+4.29%)Changelog
=========Changes in v8
1. Renamed the STACKLEAK tests to
STACKLEAK_ALLOCAandSTACKLEAK_DEEP_RECURSION. Fixed some obsolete comments there.2. Added the comments describing stack erasing on
x86_32, because it differs a lot from one onx86_64since v7.Changes in v7
1. Improved
erase_kstack()onx86_64. Now it detects which stack we are currently using. If we are on the thread stack,erase_kstack()writes the poison value up to the stack pointer. If we are not on the thread stack (we are on the trampoline stack introduced by Andy Lutomirski),erase_kstack()writes the poison value up to thecpu_current_top_of_stack.N.B. Right now there are two situations when
erase_kstack()is called on the thread stack:
- fromsysret32_from_system_callinentry_64_compat.S;
- fromsyscall_trace_enter()(see the 3rd patch of this series).2. Introduced
STACKLEAK_POISON_CHECK_DEPTHmacro andBUILD_BUG_ON()checking its consistency withCONFIG_STACKLEAK_TRACK_MIN_SIZE.3. Added "disable" option for the STACKLEAK gcc plugin (asked by Laura Abbott).
4. Improved
CONFIG_STACKLEAK_METRICS. Now /proc/<pid>/stack_depth shows the maximum kernel stack consumption for the current and previous syscalls. Although this information is not precise, it can be useful for estimating the STACKLEAK performance impact for different workloads.5. Removed redundant
erase_kstack()calls fromsyscall_trace_enter(). There is no need to erase the stack insyscall_trace_enter()when it returns -1 (thanks to Dmitry Levin for noticing that).6. Introduced
MIN_STACK_LEFTto get rid of hardcoded numbers incheck_alloca().Changes in v6
1. Examined syscall entry/exit paths.
- Added missingerase_kstack()call atret_from_fork()forx86_32.
- Added missingerase_kstack()call atsyscall_trace_enter().
- Solved questions previously marked with TODO.2. Rebased onto v4.15-rc2, which includes Andy Lutomirski's entry changes. Andy removed sp0 from
thread_structforx86_64, which was the only issue during rebasing.3. Removed the recursive
BUG()intrack_stack()that was caused by the code instrumentation. Instead,CONFIG_GCC_PLUGIN_STACKLEAKnow impliesCONFIG_VMAP_STACKandCONFIG_SCHED_STACK_END_CHECK, which seems to be an optimal solution.4. Put stack erasing in
syscall_trace_enter()into a separate patch and fixed my mistake withsecure_computing()(found by Tycho Andersen).5. After some experiments, kept the asm implementation of
erase_kstack(), because it gives a full control over the stack for clearing it neatly and doesn't offend KASAN.6. Improved the comments describing STACKLEAK.
Changes in v5 (mostly according to the feedback from Ingo Molnar)
1. Introduced the
CONFIG_STACKLEAK_METRICSproviding STACKLEAK information about tasks via the /proc file system. That information can be useful for estimating the STACKLEAK performance impact for different workloads. In particular, /proc/<pid>/lowest_stack shows the currentlowest_stackvalue and its final value from the previous syscall.2. Introduced a single
check_alloca()implementation working for bothx86_64 and x86_32.3. Fixed coding style issues and did some refactoring in the STACKLEAK gcc plugin.
4. Put the gcc plugin and the kernel stack erasing into separate (working) patches.
Changes in v4
1. Introduced the
CONFIG_STACKLEAK_TRACK_MIN_SIZEparameter instead of hard-coded track-lowest-sp.2. Carefully looked into the assertions in
track_stack()andcheck_alloca().
- Fixed the incorrectBUG()condition intrack_stack(), thanks to Tycho Andersen. Also disabled that check ifCONFIG_VMAP_STACKis enabled.
- Fixed the surplus and erroneous code for calculatingstack_leftincheck_alloca()onx86_64. That code repeats the work which is already done inget_stack_info()and it misses the fact that different exception stacks onx86_64have different size.3. Introduced two lkdtm tests for the STACKLEAK feature (developed together with Tycho Andersen).
4. Added info about STACKLEAK to
Documentation/security/self-protection.rst.5. Improved the comments.
6. Decided not to change "unsigned long sp = (unsigned long)&sp" to
current_stack_pointer. The original variant is more platform independent sincecurrent_stack_pointerhas different type on x86 and arm.Changes in v3
1. Added the detailed comments describing the STACKLEAK gcc plugin. Read the plugin from the bottom up, like you do for Linux kernel drivers. Additional information:
- gcc internals documentation available from gcc sources;
- wonderful slides by Diego Novillo
https://www.airs.com/dnovillo/200711-GCC-Internals/
- nice introduction to gcc plugins at LWN
https://lwn.net/Articles/457543/2. Improved the commit message and
Kconfigdescription according to the feedback from Kees Cook. Also added the original notice describing the performance impact of the STACKLEAK feature.3. Removed the arch-specific
ix86_cmodelcheck instackleak_track_stack_gate(). It caused skipping the kernel code instrumentation for i386.4. Fixed a minor mistake in
stackleak_tree_instrument_execute(). First versions of the plugin used ENTRY_BLOCK_PTR_FOR_FN(cfun)->next_bb to get the basic block with the function prologue. That was not correct since the control flow graph edge fromENTRY_BLOCK_PTRdoesn't always go to that basic block.So later it was changed to single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun)), but not completely.
next_bbwas still used forentry_bbassignment, which could cause the wrong value of theprologue_instrumentedvariable.I've reported this issue to Grsecurity and proposed the fix for it, but unfortunately didn't get any reply.
5. Introduced the
STACKLEAK_POISONmacro and renamed the config option according to the feedback from Kees Cook.Ideas for further work
======================- Think of erasing stack on the way out of exception handlers (idea by Andy Lutomirski).
- Think of erasing the kernel stack after invoking EFI runtime services (idea by Mark Rutland).
- Try to port STACKLEAK to arm64 (Laura Abbott is working on it).Alexander Popov (6):
x86/entry: Add STACKLEAK erasing the kernel stack at the end of
syscalls
gcc-plugins: Add STACKLEAK plugin for tracking the kernel stack x86/entry: Erase kernel stack insyscall_trace_enter()
lkdtm: Add a test for STACKLEAK
fs/proc: Show STACKLEAK metrics in the /proc file system
doc: self-protection: Add information about STACKLEAK featureDocumentation/security/self-protection.rst | 23 +- arch/Kconfig | 53 ++++ arch/x86/Kconfig | 1 + arch/x86/entry/common.c | 7 + arch/x86/entry/entry_32.S | 93 ++++++ arch/x86/entry/entry_64.S | 113 +++++++ arch/x86/entry/entry_64_compat.S | 11 + arch/x86/include/asm/processor.h | 7 + arch/x86/kernel/asm-offsets.c | 11 + arch/x86/kernel/dumpstack.c | 20 ++ arch/x86/kernel/process_32.c | 8 + arch/x86/kernel/process_64.c | 8 + drivers/misc/Makefile | 3 + drivers/misc/lkdtm.h | 4 + drivers/misc/lkdtm_core.c | 2 + drivers/misc/lkdtm_stackleak.c | 136 +++++++++ fs/exec.c | 33 ++ fs/proc/base.c | 18 ++ include/linux/compiler.h | 6 + scripts/Makefile.gcc-plugins | 3 + scripts/gcc-plugins/stackleak_plugin.c | 471 +++++++++++++++++++++++++++++ 21 files changed, 1022 insertions(+), 9 deletions(-)
Quoted 3 lines
create mode 100644
drivers/misc/lkdtm_stackleak.ccreate mode 100644scripts/gcc-plugins/stackleak_plugin.c
This doesn't work with gcc-8. One of the errors is a minor API change which can be fixed up but a bigger problem is they changed get_frame_size
/* Return size needed for stack frame based on slots so far allocated.
This size counts from zero. It is not rounded to STACK_BOUNDARY; the caller may have to do that. */
extern poly_int64 get_frame_size (void);
This is described at https://gcc.gnu.org/onlinedocs//gccint/poly_005fint.html
but I'm still trying to come up with a good solution. Feel free to beat me to fixing it...
Thanks,
Laura
cover Update stackleak for gcc-8
Okay I _think_ this should be the solution for gcc-8. I've split this into two patches, one for the generic framework and the other for the stackleak plugin itself. Kees, I'll let you figure out how to apply these. Feel free to fold in as appropriate.
Laura Abbott (2):
gcc-plugins: Update cgraph_create_edge for gcc-8
gcc-plugins: stackleak: Update for gcc-8
scripts/gcc-plugins/gcc-common.h | 6 ++++++ scripts/gcc-plugins/stackleak_plugin.c | 24 +++++++++++++++++++++++- 2 files changed, 29 insertions(+), 1 deletion(-)
Signature · 1 line
2.14.3
PATCH 1/2 gcc-plugins: Update cgraph_create_edge for gcc-8
gcc-8 changed the API for cgraph_create_edge. Update accordingly.
--- scripts/gcc-plugins/gcc-common.h | 6 ++++++ 1 file changed, 6 insertions(+)
scripts/gcc-plugins/gcc-common.h +6 -0
@@ -723,8 +723,14 @@ static inline const char *get_decl_section_name(const_tree decl)#define varpool_get_node (decl) varpool_node::get(decl)#define dump_varpool_node (file, node) (node)->dump(file) +#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) + +#endif #define cgraph_create_edge_including_clones (caller, callee, old_call_stmt, call_stmt, count, freq, nest, reason) \ (caller)->create_edge_including_clones((callee), (old_call_stmt), (call_stmt), (count), (freq), (reason))
Signature · 1 line
2.14.3
Re: [PATCH 1/2] gcc-plugins: Update cgraph_create_edge for gcc-8
Quoted 14 lines · Laura Abbott
gcc-8 changed the API for
cgraph_create_edge. Update accordingly.Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/gcc-common.h | 6 ++++++ 1 file changed, 6 insertions(+)scripts/gcc-plugins/gcc-common.h +0
-0@@ -723,8 +723,14 @@ static inline const char *get_decl_section_name(const_tree decl)#define varpool_get_node (decl) varpool_node::get(decl)
scripts/gcc-plugins/gcc-common.h static inline const char *get_decl_section_name(const_tree decl) · line 733
Quoted 10 lines
scripts/gcc-plugins/gcc-common.h +6
-0#define dump_varpool_node (file, node) (node)->dump(file) +#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) + +#endif
Since gcc 8 "throws away" the freq argument, could this just be updated to do the same? i.e., leave "freq" as an argument, and just don't use it? That way the plugin caller doesn't need to be updated.
+#if BUILDING_GCC_VERSION >= 8000
+#define cgraph_create_edge(caller, callee, call_stmt, count, freq, nest) \
+ (caller)->create_edge((callee), (call_stmt), (count))
+#else
#define cgraph_create_edge(caller, callee, call_stmt, count, freq, nest) \
(caller)->create_edge((callee), (call_stmt), (count), (freq))+#endif
-Kees
Signature · 2 lines
Kees Cook Pixel Security
Re: [PATCH 1/2] gcc-plugins: Update cgraph_create_edge for gcc-8
Quoted 31 lines · Kees Cook
Quoted 14 lines · Laura Abbott
gcc-8 changed the API for
cgraph_create_edge. Update accordingly.Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/gcc-common.h | 6 ++++++ 1 file changed, 6 insertions(+)scripts/gcc-plugins/gcc-common.h +0
-0@@ -723,8 +723,14 @@ static inline const char *get_decl_section_name(const_tree decl)#define varpool_get_node (decl) varpool_node::get(decl)scripts/gcc-plugins/gcc-common.h static inline const char *get_decl_section_name(const_tree decl) · line 733
Quoted 10 lines
scripts/gcc-plugins/gcc-common.h +6
-0#define dump_varpool_node (file, node) (node)->dump(file) +#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) + +#endif Since gcc 8 "throws away" the freq argument, could this just be updated to do the same? i.e., leave "freq" as an argument, and just don't use it? That way the plugin caller doesn't need to be updated.
scripts/gcc-plugins/gcc-common.h static inline const char *get_decl_section_name(const_tree decl)
Quoted 10 lines
scripts/gcc-plugins/gcc-common.h +5
-1+#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) +#endif -Kees
Yeah, I think I started out that way then ended up with this for debugging reasons. The only concern I might have is ending up with the freq argument flagged as unused by gcc but I'm indifferent overall.
Thanks,
Laura
Re: [PATCH 1/2] gcc-plugins: Update cgraph_create_edge for gcc-8
Quoted 31 lines · Laura Abbott
Quoted 30 lines · Kees Cook
Quoted 14 lines · Laura Abbott
gcc-8 changed the API for
cgraph_create_edge. Update accordingly.Signed-off-by Laura Abbott <[email protected]>--- scripts/gcc-plugins/gcc-common.h | 6 ++++++ 1 file changed, 6 insertions(+)scripts/gcc-plugins/gcc-common.h +0
-0@@ -723,8 +723,14 @@ static inline const char *get_decl_section_name(const_tree decl)#define varpool_get_node (decl) varpool_node::get(decl)scripts/gcc-plugins/gcc-common.h static inline const char *get_decl_section_name(const_tree decl) · line 733
Quoted 10 lines
scripts/gcc-plugins/gcc-common.h +6
-0#define dump_varpool_node (file, node) (node)->dump(file) +#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) + +#endif Since gcc 8 "throws away" the freq argument, could this just be updated to do the same? i.e., leave "freq" as an argument, and just don't use it? That way the plugin caller doesn't need to be updated.
Quoted 13 lines
scripts/gcc-plugins/gcc-common.h static inline const char *get_decl_section_name(const_tree decl) · line 733
Quoted 8 lines
scripts/gcc-plugins/gcc-common.h +5
-0+#if BUILDING_GCC_VERSION >=8000 +#define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ + (caller)->create_edge((callee), (call_stmt), (count)) +#else #define cgraph_create_edge (caller, callee, call_stmt, count, freq, nest) \ (caller)->create_edge((callee), (call_stmt), (count), (freq)) +#endif Yeah, I think I started out that way then ended up with this for debugging reasons. The only concern I might have is ending up with the freq argument flagged as unused by gcc but I'm indifferent overall.
Hello Laura and Kees,
Laura, thanks a lot for your work! I double-checked gcc changelog and agree with you. I will add this fix for gcc-8 as a separate commit to the next version of STACKLEAK.
And I'll do some macro cleanup as well. All the cgraph_create_edge* macros in gcc-common.h look strange.
At line 395 we have these redefinitions for dropping "nest":
#if BUILDING_GCC_VERSION >= 4007 && BUILDING_GCC_VERSION <= 4009
#define cgraph_create_edge(caller, callee, call_stmt, count, freq, nest) \
cgraph_create_edge((caller), (callee), (call_stmt), (count), (freq))
#define cgraph_create_edge_including_clones(caller, callee, old_call_stmt, call_stmt, count, freq, nest, reason) \
cgraph_create_edge_including_clones((caller), (callee), (old_call_stmt), (call_stmt), (count), (freq), (reason))
#endif
But later at line 726 "nest" is not used as well:
#define cgraph_create_edge(caller, callee, call_stmt, count, freq, nest) \
(caller)->create_edge((callee), (call_stmt), (count), (freq))
#define cgraph_create_edge_including_clones(caller, callee, old_call_stmt, call_stmt, count, freq, nest, reason) \
(caller)->create_edge_including_clones((callee), (old_call_stmt), (call_stmt), (count), (freq), (reason))
So I'm going to drop "nest" from cgraph_create_edge* macros and the plugin code.
Thanks again!
Best regards,
Alexander
PATCH 2/2 gcc-plugins: stackleak: Update for gcc-8
In reply to Laura Abbott- Use the new cgraph_create_edge API
- Account for the change in type for get_frame_size
--- scripts/gcc-plugins/stackleak_plugin.c | 24 +++++++++++++++++++++++- 1 file changed, 23 insertions(+), 1 deletion(-)
scripts/gcc-plugins/stackleak_plugin.c +23 -1
@@ -69,8 +69,13 @@ static void stackleak_check_alloca(gimple_stmt_iterator *gsi) node =cgraph_get_create_node (check_function_decl);gcc_assert (node); frequency =compute_call_stmt_bb_frequency (current_function_decl, bb); +#if BUILDING_GCC_VERSION >=8000 +cgraph_create_edge (cgraph_get_node (current_function_decl), node, + check_alloca, bb->count, bb->loop_depth); +#else cgraph_create_edge (cgraph_get_node (current_function_decl), node, check_alloca, bb->count, frequency, bb->loop_depth); +#endif }static void stackleak_add_instrumentation (gimple_stmt_iterator *gsi,bool after)
@@ -94,8 +99,13 @@ static void stackleak_add_instrumentation(gimple_stmt_iterator *gsi, bool after) node =cgraph_get_create_node (track_function_decl);gcc_assert (node); frequency =compute_call_stmt_bb_frequency (current_function_decl, bb); +#if BUILDING_GCC_VERSION >=8000 +cgraph_create_edge (cgraph_get_node (current_function_decl), node, + track_stack, bb->count, bb->loop_depth); +#else cgraph_create_edge (cgraph_get_node (current_function_decl), node, track_stack, bb->count, frequency, bb->loop_depth); +#endif }static bool is_alloca (gimple stmt)
@@ -203,6 +213,18 @@ static unsigned int stackleak_tree_instrument_execute(void)return 0 ; } + +#if BUILDING_GCC_VERSION >=8000 +bool check_frame_size () +{ +return maybe_ge (get_frame_size (), track_frame_size); +} +#else +bool check_frame_size () +{ +return get_frame_size () >= track_frame_size; +} +#endif /* * Work with the RTL representation of the code. * Remove the unneeded track_stack() calls from the functions which don't
@@ -215,7 +237,7 @@ static unsigned int stackleak_final_execute(void)if (cfun->calls_alloca)return 0 ; -if (get_frame_size () >= track_frame_size) +if (check_frame_size ())return 0 ;/*
Signature · 1 line
2.14.3
Re: [PATCH 2/2] gcc-plugins: stackleak: Update for gcc-8
Hello Laura,
Thanks for the cooperation!
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_tree_instrument_execute(void) · line 220
Quoted 4 lines · Laura Abbott
scripts/gcc-plugins/stackleak_plugin.c +4
-0+#if BUILDING_GCC_VERSION >=8000 +bool check_frame_size () +{ +return maybe_ge (get_frame_size (), track_frame_size);
After looking through this guide
https://gcc.gnu.org/onlinedocs//gccint/Guidelines-for-using-poly_005fint.html#Guidelines-for-using-poly_005fint
it seems to me that we should better use something like that:
poly_int64 frame_size = get_frame_size();
if (frame_size.to_constant() >= track_frame_size)
return 0;
May I ask for your opinion?
Quoted 11 lines
scripts/gcc-plugins/stackleak_plugin.c +7
-0+} +#else +bool check_frame_size () +{ +return get_frame_size () >= track_frame_size; +} +#endif /* * Work with the RTL representation of the code. * Remove the unneeded track_stack() calls from the functions which don't
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 242
Quoted 7 lines
scripts/gcc-plugins/stackleak_plugin.c +1
-1@@ -215,7 +237,7 @@ static unsigned int stackleak_final_execute(void)if (cfun->calls_alloca)return 0 ; -if (get_frame_size () >= track_frame_size) +if (check_frame_size ())return 0 ;
Best regards,
Alexander
Re: [PATCH 2/2] gcc-plugins: stackleak: Update for gcc-8
Quoted 14 lines · Alexander Popov
Hello Laura,
Thanks for the cooperation!
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_tree_instrument_execute(void) · line 220
Quoted 4 lines · Laura Abbott
scripts/gcc-plugins/stackleak_plugin.c +4
-0+#if BUILDING_GCC_VERSION >=8000 +bool check_frame_size () +{ +return maybe_ge (get_frame_size (), track_frame_size);After looking through this guide
https://gcc.gnu.org/onlinedocs//gccint/Guidelines-for-using-poly_005fint.html#Guidelines-for-using-poly_005fint
it seems to me that we should better use something like that:
Quoted 7 lines
poly_int64frame_size=get_frame_size();if (
frame_size.to_constant()>=track_frame_size)return 0;May I ask for your opinion?
I was a bit wary of using to_constant() because I wasn't 100% sure I could make the assertion that it could actually be a constant so using maybe_ge seemed like the conservative approach. This also getting into compiler internals so it's possible I'm misunderstanding how poly_ints are supposed to work.
Thanks,
Laura
Quoted 22 lines
Quoted 11 lines
scripts/gcc-plugins/stackleak_plugin.c +7
-0+} +#else +bool check_frame_size () +{ +return get_frame_size () >= track_frame_size; +} +#endif /* * Work with the RTL representation of the code. * Remove the unneeded track_stack() calls from the functions which don't scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_final_execute(void) · line 242
Quoted 7 lines
scripts/gcc-plugins/stackleak_plugin.c +1
-1@@ -215,7 +237,7 @@ static unsigned int stackleak_final_execute(void)if (cfun->calls_alloca)return 0 ; -if (get_frame_size () >= track_frame_size) +if (check_frame_size ())return 0 ;Best regards,
Alexander
Re: [PATCH 2/2] gcc-plugins: stackleak: Update for gcc-8
Quoted 17 lines · Laura Abbott
Quoted 16 lines · Alexander Popov
Hello Laura,
Thanks for the cooperation!
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_tree_instrument_execute(void) · line 220
Quoted 4 lines · Laura Abbott
scripts/gcc-plugins/stackleak_plugin.c +4
-0+#if BUILDING_GCC_VERSION >=8000 +bool check_frame_size () +{ +return maybe_ge (get_frame_size (), track_frame_size);After looking through this guide
https://gcc.gnu.org/onlinedocs//gccint/Guidelines-for-using-poly_005fint.html#Guidelines-for-using-poly_005fint
it seems to me that we should better use something like that:
poly_int64frame_size=get_frame_size();
Quoted 8 lines
Quoted 5 lines
if (
frame_size.to_constant()>=track_frame_size)return 0;May I ask for your opinion?
I was a bit wary of using
to_constant()because I wasn't 100% sure I could make the assertion that it could actually be a constant so using
Yeah, the assert would fire for SVE functions that spill SVE registers to the stack.
Quoted 3 lines
maybe_geseemed like the conservative approach. This also getting into compiler internals so it's possible I'm misunderstanding howpoly_intsare supposed to work.
The maybe_ge patch looks good to me FWIW.
Thanks,
Richard
Re: [PATCH 2/2] gcc-plugins: stackleak: Update for gcc-8
Quoted 27 lines · Richard Sandiford
Quoted 17 lines · Laura Abbott
Quoted 16 lines · Alexander Popov
Hello Laura,
Thanks for the cooperation!
scripts/gcc-plugins/stackleak_plugin.c static unsigned int stackleak_tree_instrument_execute(void) · line 220
Quoted 4 lines · Laura Abbott
scripts/gcc-plugins/stackleak_plugin.c +4
-0+#if BUILDING_GCC_VERSION >=8000 +bool check_frame_size () +{ +return maybe_ge (get_frame_size (), track_frame_size);After looking through this guide
https://gcc.gnu.org/onlinedocs//gccint/Guidelines-for-using-poly_005fint.html#Guidelines-for-using-poly_005fint
it seems to me that we should better use something like that:
poly_int64frame_size=get_frame_size();Quoted 8 lines
Quoted 5 lines
if (
frame_size.to_constant()>=track_frame_size)return 0;May I ask for your opinion?
I was a bit wary of using
to_constant()because I wasn't 100% sure I could make the assertion that it could actually be a constant so using
Quoted 8 lines
Yeah, the assert would fire for SVE functions that spill SVE registers to the stack.
Quoted 3 lines
maybe_geseemed like the conservative approach. This also getting into compiler internals so it's possible I'm misunderstanding howpoly_intsare supposed to work.The
maybe_gepatch looks good to me FWIW.
Thanks, Laura and Richard!
I see the rationale, I will use maybe_ge() in the next version of the patch.
Best regards,
Alexander
Re: [PATCH 0/2] Update stackleak for gcc-8
In reply to Laura AbbottQuoted 4 lines · Laura Abbott
Okay I _think_ this should be the solution for gcc-8. I've split this into two patches, one for the generic framework and the other for the stackleak plugin itself.
Nice! Thanks for solving this!
Quoted 2 lines
Kees, I'll let you figure out how to apply
these. Feel free to fold in as appropriate.
Other than the suggestion to have fewer ifdef in stackleak, this looks fine. Alexander, feel free to take these both (with adjustments) for the stackleak series: stackleak is the only user of cgraph_create_edge() so I think it's fine to adjust gcc-common.h in the series.
-Kees
Signature · 2 lines
Kees Cook Pixel Security