[kernel-hardening] [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[kernel-hardening] [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

This is the 7th version of the patch series introducing STACKLEAK to the
mainline kernel. 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.

Changes in v7
=============

1. Improved erase_kstack(). 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().

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%)

Previous changes
================

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                  |  69 +++++
 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             | 139 +++++++++
 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, 1001 insertions(+), 9 deletions(-)
 create mode 100644 drivers/misc/lkdtm_stackleak.c
 create mode 100644 scripts/gcc-plugins/stackleak_plugin.c

-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 1/6] x86/entry: Add STACKLEAK erasing the kernel stack at the end of syscalls

[Alexander Popov]

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.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 arch/Kconfig                     |  27 ++++++++++
 arch/x86/Kconfig                 |   1 +
 arch/x86/entry/entry_32.S        |  65 +++++++++++++++++++++++
 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, 241 insertions(+)

diff --git a/arch/Kconfig b/arch/Kconfig
index 400b9e1..721fdae 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -387,6 +387,13 @@ 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
@@ -517,6 +524,26 @@ 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
diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index d4fc98c..4981cb6 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -116,6 +116,7 @@ config X86
 	select HAVE_ARCH_MMAP_RND_COMPAT_BITS	if MMU && COMPAT
 	select HAVE_ARCH_COMPAT_MMAP_BASES	if MMU && COMPAT
 	select HAVE_ARCH_SECCOMP_FILTER
+	select HAVE_ARCH_STACKLEAK
 	select HAVE_ARCH_TRACEHOOK
 	select HAVE_ARCH_TRANSPARENT_HUGEPAGE
 	select HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD if X86_64
diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index ace8f32..f400d8f 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -76,6 +76,66 @@
 #endif
 .endm
 
+.macro erase_kstack
+#ifdef CONFIG_GCC_PLUGIN_STACKLEAK
+	call erase_kstack
+#endif
+.endm
+
+#ifdef CONFIG_GCC_PLUGIN_STACKLEAK
+/* For the detailed comments, see erase_kstack in entry_64.S */
+ENTRY(erase_kstack)
+	pushl	%edi
+	pushl	%ecx
+	pushl	%eax
+	pushl	%ebp
+
+	movl	PER_CPU_VAR(current_task), %ebp
+	mov	TASK_lowest_stack(%ebp), %edi
+	mov	$STACKLEAK_POISON, %eax
+	std
+
+1:
+	mov	%edi, %ecx
+	and	$THREAD_SIZE_asm - 1, %ecx
+	shr	$2, %ecx
+	repne	scasl
+	jecxz	2f
+
+	cmp	$STACKLEAK_POISON_CHECK_DEPTH / 4, %ecx
+	jc	2f
+
+	mov	$STACKLEAK_POISON_CHECK_DEPTH / 4, %ecx
+	repe	scasl
+	jecxz	2f
+	jne	1b
+
+2:
+	cld
+	or	$2 * 4, %edi
+	mov	%esp, %ecx
+	sub	%edi, %ecx
+
+	cmp	$THREAD_SIZE_asm, %ecx
+	jb	3f
+	ud2
+
+3:
+	shr	$2, %ecx
+	rep	stosl
+
+	movl	PER_CPU_VAR(cpu_current_top_of_stack), %edi
+	sub	$128, %edi
+	mov	%edi, TASK_lowest_stack(%ebp)
+
+	popl	%ebp
+	popl	%eax
+	popl	%ecx
+	popl	%edi
+	ret
+ENDPROC(erase_kstack)
+#endif
+
 /*
  * User gs save/restore
  *
@@ -286,6 +346,7 @@ ENTRY(ret_from_fork)
 	/* When we fork, we trace the syscall return in the child, too. */
 	movl    %esp, %eax
 	call    syscall_return_slowpath
+	erase_kstack
 	jmp     restore_all
 
 	/* kernel thread */
@@ -446,6 +507,8 @@ ENTRY(entry_SYSENTER_32)
 	ALTERNATIVE "testl %eax, %eax; jz .Lsyscall_32_done", \
 		    "jmp .Lsyscall_32_done", X86_FEATURE_XENPV
 
+	erase_kstack
+
 /* Opportunistic SYSEXIT */
 	TRACE_IRQS_ON			/* User mode traces as IRQs on. */
 	movl	PT_EIP(%esp), %edx	/* pt_regs->ip */
@@ -532,6 +595,8 @@ ENTRY(entry_INT80_32)
 	call	do_int80_syscall_32
 .Lsyscall_32_done:
 
+	erase_kstack
+
 restore_all:
 	TRACE_IRQS_IRET
 .Lrestore_all_notrace:
diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index f048e38..3ff0802 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -65,6 +65,112 @@ 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)
+	pushq	%rdi
+	pushq	%rcx
+	pushq	%rax
+	pushq	%r11
+
+	mov	PER_CPU_VAR(current_task), %r11
+	mov	TASK_lowest_stack(%r11), %rdi
+	mov	$STACKLEAK_POISON, %rax
+	std
+
+	/*
+	 * Let's search for the poison value in the stack.
+	 * Start from the lowest_stack and go to the bottom (see std above).
+	 */
+1:
+	mov	%edi, %ecx
+	and	$THREAD_SIZE_asm - 1, %ecx
+	shr	$3, %ecx
+	repne	scasq
+	jecxz	2f	/* Didn't find it. Go to poisoning. */
+
+	/*
+	 * Found the poison value in the stack. Go to poisoning if there is
+	 * not enough space left for the poison check.
+	 */
+	cmp	$STACKLEAK_POISON_CHECK_DEPTH / 8, %ecx
+	jb	2f
+
+	/*
+	 * 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:
+	/*
+	 * 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
+	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
+
+	/* Set the lowest_stack value to the top_of_stack - 256 */
+	mov	PER_CPU_VAR(cpu_current_top_of_stack), %rdi
+	sub	$256, %rdi
+	mov	%rdi, TASK_lowest_stack(%r11)
+
+	popq	%r11
+	popq	%rax
+	popq	%rcx
+	popq	%rdi
+	ret
+ENDPROC(erase_kstack)
+#endif
+
 /*
  * When dynamic function tracer is enabled it will add a breakpoint
  * to all locations that it is about to modify, sync CPUs, update
@@ -411,6 +517,8 @@ 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
+
 	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
 
 	popq	%rdi
@@ -748,6 +856,7 @@ 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
 
 	SWITCH_TO_USER_CR3_STACK scratch_reg=%rdi
 
diff --git a/arch/x86/entry/entry_64_compat.S b/arch/x86/entry/entry_64_compat.S
index 98d5358..727dde2 100644
--- a/arch/x86/entry/entry_64_compat.S
+++ b/arch/x86/entry/entry_64_compat.S
@@ -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 */
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index d3a67fb..9736829 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -496,6 +496,10 @@ 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 */
 
diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
index 76417a9..ef5d260 100644
--- a/arch/x86/kernel/asm-offsets.c
+++ b/arch/x86/kernel/asm-offsets.c
@@ -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);
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 5224c60..6d256ab 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -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));
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index c754662..4345fdf 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -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);
diff --git a/include/linux/compiler.h b/include/linux/compiler.h
index 52e611a..001c3e2 100644
--- a/include/linux/compiler.h
+++ b/include/linux/compiler.h
@@ -327,4 +327,10 @@ static __always_inline void __write_once_size(volatile void *p, void *res, int s
 	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 */
-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 2/6] gcc-plugins: Add STACKLEAK plugin for tracking the kernel stack

[Alexander Popov]

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 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.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 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

diff --git a/arch/Kconfig b/arch/Kconfig
index 721fdae..6fcb632 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -528,6 +528,8 @@ 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
@@ -544,6 +546,18 @@ 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
diff --git a/arch/x86/kernel/dumpstack.c b/arch/x86/kernel/dumpstack.c
index afbecff..2fbbaa9 100644
--- a/arch/x86/kernel/dumpstack.c
+++ b/arch/x86/kernel/dumpstack.c
@@ -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
diff --git a/fs/exec.c b/fs/exec.c
index 7eb8d21..9865612 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -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 */
diff --git a/scripts/Makefile.gcc-plugins b/scripts/Makefile.gcc-plugins
index b2a95af..8d6070f 100644
--- a/scripts/Makefile.gcc-plugins
+++ b/scripts/Makefile.gcc-plugins
@@ -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
diff --git a/scripts/gcc-plugins/stackleak_plugin.c b/scripts/gcc-plugins/stackleak_plugin.c
new file mode 100644
index 0000000..6fc991c
--- /dev/null
+++ b/scripts/gcc-plugins/stackleak_plugin.c
@@ -0,0 +1,471 @@
+/*
+ * Copyright 2011-2017 by the PaX Team <pageexec@freemail.hu>
+ * Modified by Alexander Popov <alex.popov@linux.com>
+ * Licensed under the GPL v2
+ *
+ * Note: the choice of the license means that the compilation process is
+ * NOT 'eligible' as defined by gcc's library exception to the GPL v3,
+ * but for the kernel it doesn't matter since it doesn't link against
+ * any of the gcc libraries
+ *
+ * This gcc plugin is needed for tracking the lowest border of the kernel stack
+ * and checking that alloca calls don't cause stack overflow. It 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 the "track-min-size" plugin parameter.
+ *
+ * This plugin is ported from grsecurity/PaX. For more information see:
+ *   https://grsecurity.net/
+ *   https://pax.grsecurity.net/
+ *
+ * Debugging:
+ *  - use fprintf() to stderr, debug_generic_expr(), debug_gimple_stmt()
+ *     and print_rtl();
+ *  - add "-fdump-tree-all -fdump-rtl-all" to the plugin CFLAGS in
+ *     Makefile.gcc-plugins to see the verbose dumps of the gcc passes;
+ *  - use gcc -E to understand the preprocessing shenanigans;
+ *  - use gcc with enabled CFG/GIMPLE/SSA verification (--enable-checking).
+ */
+
+#include "gcc-common.h"
+
+__visible int plugin_is_GPL_compatible;
+
+static int track_frame_size = -1;
+static const char track_function[] = "track_stack";
+static const char check_function[] = "check_alloca";
+
+/*
+ * Mark these global variables (roots) for gcc garbage collector since
+ * they point to the garbage-collected memory.
+ */
+static GTY(()) tree track_function_decl;
+static GTY(()) tree check_function_decl;
+
+static struct plugin_info stackleak_plugin_info = {
+	.version = "201707101337",
+	.help = "track-min-size=nn\ttrack stack for functions with a stack frame size >= nn bytes\n"
+		"disable\t\tdo not activate the plugin\n"
+};
+
+static void stackleak_check_alloca(gimple_stmt_iterator *gsi)
+{
+	gimple stmt;
+	gcall *check_alloca;
+	tree alloca_size;
+	cgraph_node_ptr node;
+	int frequency;
+	basic_block bb;
+
+	/* Insert call to void check_alloca(unsigned long size) */
+	alloca_size = gimple_call_arg(gsi_stmt(*gsi), 0);
+	stmt = gimple_build_call(check_function_decl, 1, alloca_size);
+	check_alloca = as_a_gcall(stmt);
+	gsi_insert_before(gsi, check_alloca, GSI_SAME_STMT);
+
+	/* Update the cgraph */
+	bb = gimple_bb(check_alloca);
+	node = cgraph_get_create_node(check_function_decl);
+	gcc_assert(node);
+	frequency = compute_call_stmt_bb_frequency(current_function_decl, bb);
+	cgraph_create_edge(cgraph_get_node(current_function_decl), node,
+			check_alloca, bb->count, frequency, bb->loop_depth);
+}
+
+static void stackleak_add_instrumentation(gimple_stmt_iterator *gsi, bool after)
+{
+	gimple stmt;
+	gcall *track_stack;
+	cgraph_node_ptr node;
+	int frequency;
+	basic_block bb;
+
+	/* Insert call to void track_stack(void) */
+	stmt = gimple_build_call(track_function_decl, 0);
+	track_stack = as_a_gcall(stmt);
+	if (after)
+		gsi_insert_after(gsi, track_stack, GSI_CONTINUE_LINKING);
+	else
+		gsi_insert_before(gsi, track_stack, GSI_SAME_STMT);
+
+	/* Update the cgraph */
+	bb = gimple_bb(track_stack);
+	node = cgraph_get_create_node(track_function_decl);
+	gcc_assert(node);
+	frequency = compute_call_stmt_bb_frequency(current_function_decl, bb);
+	cgraph_create_edge(cgraph_get_node(current_function_decl), node,
+			track_stack, bb->count, frequency, bb->loop_depth);
+}
+
+static bool is_alloca(gimple stmt)
+{
+	if (gimple_call_builtin_p(stmt, BUILT_IN_ALLOCA))
+		return true;
+
+#if BUILDING_GCC_VERSION >= 4007
+	if (gimple_call_builtin_p(stmt, BUILT_IN_ALLOCA_WITH_ALIGN))
+		return true;
+#endif
+
+	return false;
+}
+
+/*
+ * Work with the GIMPLE representation of the code.
+ * Insert the check_alloca() call before alloca and track_stack() call after
+ * it. Also insert track_stack() call into the beginning of the function
+ * if it is not instrumented.
+ */
+static unsigned int stackleak_tree_instrument_execute(void)
+{
+	basic_block bb, entry_bb;
+	bool prologue_instrumented = false, is_leaf = true;
+	gimple_stmt_iterator gsi;
+
+	/*
+	 * ENTRY_BLOCK_PTR is a basic block which represents possible entry
+	 * point of a function. This block does not contain any code and
+	 * has a CFG edge to its successor.
+	 */
+	gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+	entry_bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
+
+	/*
+	 * 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;
+
+			stmt = gsi_stmt(gsi);
+
+			/* Leaf function is a function which makes no calls */
+			if (is_gimple_call(stmt))
+				is_leaf = false;
+
+			if (!is_alloca(stmt))
+				continue;
+
+			/* 2. Insert stack overflow check before alloca call */
+			stackleak_check_alloca(&gsi);
+
+			/* 3. Insert track_stack() call after alloca call */
+			stackleak_add_instrumentation(&gsi, true);
+			if (bb == entry_bb)
+				prologue_instrumented = true;
+		}
+	}
+
+	if (prologue_instrumented)
+		return 0;
+
+	/*
+	 * 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;
+	}
+
+	if (is_leaf &&
+	    !strncmp(IDENTIFIER_POINTER(DECL_NAME(current_function_decl)),
+		     "_paravirt_", 10)) {
+		return 0;
+	}
+
+	/* 4. Insert track_stack() call at the function beginning */
+	bb = entry_bb;
+	if (!single_pred_p(bb)) {
+		/* gcc_assert(bb_loop_depth(bb) ||
+				(bb->flags & BB_IRREDUCIBLE_LOOP)); */
+		split_edge(single_succ_edge(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+		gcc_assert(single_succ_p(ENTRY_BLOCK_PTR_FOR_FN(cfun)));
+		bb = single_succ(ENTRY_BLOCK_PTR_FOR_FN(cfun));
+	}
+	gsi = gsi_after_labels(bb);
+	stackleak_add_instrumentation(&gsi, false);
+
+	return 0;
+}
+
+/*
+ * Work with the RTL representation of the code.
+ * Remove the unneeded track_stack() calls from the functions which don't
+ * call alloca and have the stack frame size less than track_frame_size.
+ */
+static unsigned int stackleak_final_execute(void)
+{
+	rtx_insn *insn, *next;
+
+	if (cfun->calls_alloca)
+		return 0;
+
+	if (get_frame_size() >= track_frame_size)
+		return 0;
+
+	/*
+	 * 1. Find track_stack() calls. Loop through the chain of insns,
+	 * which is an RTL representation of the code for a function.
+	 *
+	 * The example of a matching insn:
+	 *    (call_insn 8 4 10 2 (call (mem (symbol_ref ("track_stack")
+	 *    [flags 0x41] <function_decl 0x7f7cd3302a80 track_stack>)
+	 *    [0 track_stack S1 A8]) (0)) 675 {*call} (expr_list
+	 *    (symbol_ref ("track_stack") [flags 0x41] <function_decl
+	 *    0x7f7cd3302a80 track_stack>) (expr_list (0) (nil))) (nil))
+	 */
+	for (insn = get_insns(); insn; insn = next) {
+		rtx body;
+
+		next = NEXT_INSN(insn);
+
+		/* Check the expression code of the insn */
+		if (!CALL_P(insn))
+			continue;
+
+		/*
+		 * Check the expression code of the insn body, which is an RTL
+		 * Expression (RTX) describing the side effect performed by
+		 * that insn.
+		 */
+		body = PATTERN(insn);
+		if (GET_CODE(body) != CALL)
+			continue;
+
+		/*
+		 * Check the first operand of the call expression. It should
+		 * be a mem RTX describing the needed subroutine with a
+		 * symbol_ref RTX.
+		 */
+		body = XEXP(body, 0);
+		if (GET_CODE(body) != MEM)
+			continue;
+
+		body = XEXP(body, 0);
+		if (GET_CODE(body) != SYMBOL_REF)
+			continue;
+
+		if (SYMBOL_REF_DECL(body) != track_function_decl)
+			continue;
+
+		/* 2. Delete the track_stack() call */
+		delete_insn_and_edges(insn);
+#if BUILDING_GCC_VERSION >= 4007
+		if (GET_CODE(next) == NOTE &&
+		    NOTE_KIND(next) == NOTE_INSN_CALL_ARG_LOCATION) {
+			insn = next;
+			next = NEXT_INSN(insn);
+			delete_insn_and_edges(insn);
+		}
+#endif
+	}
+
+	/*
+	 * Uncomment the following to see the code which was cleaned at this
+	 * pass. It should not contain check_alloca() and track_stack() calls.
+	 * The stack frame size should be less than track_frame_size.
+	 *
+	 * warning(0, "Instrumentation is removed, stack frame size: %ld",
+	 * 						get_frame_size());
+	 * print_simple_rtl(stderr, get_insns());
+	 */
+
+	return 0;
+}
+
+static bool stackleak_track_stack_gate(void)
+{
+	tree section;
+
+	section = lookup_attribute("section",
+				   DECL_ATTRIBUTES(current_function_decl));
+	if (section && TREE_VALUE(section)) {
+		section = TREE_VALUE(TREE_VALUE(section));
+
+		if (!strncmp(TREE_STRING_POINTER(section), ".init.text", 10))
+			return false;
+		if (!strncmp(TREE_STRING_POINTER(section), ".devinit.text", 13))
+			return false;
+		if (!strncmp(TREE_STRING_POINTER(section), ".cpuinit.text", 13))
+			return false;
+		if (!strncmp(TREE_STRING_POINTER(section), ".meminit.text", 13))
+			return false;
+	}
+
+	return track_frame_size >= 0;
+}
+
+/* Build function declarations for track_stack() and check_alloca() */
+static void stackleak_start_unit(void *gcc_data __unused,
+				 void *user_data __unused)
+{
+	tree fntype;
+
+	/* void track_stack(void) */
+	fntype = build_function_type_list(void_type_node, NULL_TREE);
+	track_function_decl = build_fn_decl(track_function, fntype);
+	DECL_ASSEMBLER_NAME(track_function_decl); /* for LTO */
+	TREE_PUBLIC(track_function_decl) = 1;
+	TREE_USED(track_function_decl) = 1;
+	DECL_EXTERNAL(track_function_decl) = 1;
+	DECL_ARTIFICIAL(track_function_decl) = 1;
+	DECL_PRESERVE_P(track_function_decl) = 1;
+
+	/* void check_alloca(unsigned long) */
+	fntype = build_function_type_list(void_type_node,
+				long_unsigned_type_node, NULL_TREE);
+	check_function_decl = build_fn_decl(check_function, fntype);
+	DECL_ASSEMBLER_NAME(check_function_decl); /* for LTO */
+	TREE_PUBLIC(check_function_decl) = 1;
+	TREE_USED(check_function_decl) = 1;
+	DECL_EXTERNAL(check_function_decl) = 1;
+	DECL_ARTIFICIAL(check_function_decl) = 1;
+	DECL_PRESERVE_P(check_function_decl) = 1;
+}
+
+/*
+ * Pass gate function is a predicate function that gets executed before the
+ * corresponding pass. If the return value is 'true' the pass gets executed,
+ * otherwise, it is skipped.
+ */
+static bool stackleak_tree_instrument_gate(void)
+{
+	return stackleak_track_stack_gate();
+}
+
+#define PASS_NAME stackleak_tree_instrument
+#define PROPERTIES_REQUIRED PROP_gimple_leh | PROP_cfg
+#define TODO_FLAGS_START TODO_verify_ssa | TODO_verify_flow | TODO_verify_stmts
+#define TODO_FLAGS_FINISH TODO_verify_ssa | TODO_verify_stmts | TODO_dump_func \
+			| TODO_update_ssa | TODO_rebuild_cgraph_edges
+#include "gcc-generate-gimple-pass.h"
+
+static bool stackleak_final_gate(void)
+{
+	return stackleak_track_stack_gate();
+}
+
+#define PASS_NAME stackleak_final
+#define TODO_FLAGS_FINISH TODO_dump_func
+#include "gcc-generate-rtl-pass.h"
+
+/*
+ * Every gcc plugin exports a plugin_init() function that is called right
+ * after the plugin is loaded. This function is responsible for registering
+ * the plugin callbacks and doing other required initialization.
+ */
+__visible int plugin_init(struct plugin_name_args *plugin_info,
+			  struct plugin_gcc_version *version)
+{
+	const char * const plugin_name = plugin_info->base_name;
+	const int argc = plugin_info->argc;
+	const struct plugin_argument * const argv = plugin_info->argv;
+	int i = 0;
+
+	/* Extra GGC root tables describing our GTY-ed data */
+	static const struct ggc_root_tab gt_ggc_r_gt_stackleak[] = {
+		{
+			.base = &track_function_decl,
+			.nelt = 1,
+			.stride = sizeof(track_function_decl),
+			.cb = &gt_ggc_mx_tree_node,
+			.pchw = &gt_pch_nx_tree_node
+		},
+		{
+			.base = &check_function_decl,
+			.nelt = 1,
+			.stride = sizeof(check_function_decl),
+			.cb = &gt_ggc_mx_tree_node,
+			.pchw = &gt_pch_nx_tree_node
+		},
+		LAST_GGC_ROOT_TAB
+	};
+
+	/*
+	 * The stackleak_tree_instrument pass should be executed before the
+	 * "optimized" pass, which is the control flow graph cleanup that is
+	 * performed just before expanding gcc trees to the RTL. In former
+	 * versions of the plugin this new pass was inserted before the
+	 * "tree_profile" pass, which is currently called "profile".
+	 */
+	PASS_INFO(stackleak_tree_instrument, "optimized", 1,
+						PASS_POS_INSERT_BEFORE);
+
+	/*
+	 * 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);
+
+	if (!plugin_default_version_check(version, &gcc_version)) {
+		error(G_("incompatible gcc/plugin versions"));
+		return 1;
+	}
+
+	/* Parse the plugin arguments */
+	for (i = 0; i < argc; i++) {
+		if (!strcmp(argv[i].key, "disable"))
+			return 0;
+
+		if (!strcmp(argv[i].key, "track-min-size")) {
+			if (!argv[i].value) {
+				error(G_("no value supplied for option '-fplugin-arg-%s-%s'"),
+					plugin_name, argv[i].key);
+				return 1;
+			}
+
+			track_frame_size = atoi(argv[i].value);
+			if (track_frame_size < 0) {
+				error(G_("invalid option argument '-fplugin-arg-%s-%s=%s'"),
+					plugin_name, argv[i].key, argv[i].value);
+				return 1;
+			}
+		} else {
+			error(G_("unknown option '-fplugin-arg-%s-%s'"),
+					plugin_name, argv[i].key);
+			return 1;
+		}
+	}
+
+	/* Give the information about the plugin */
+	register_callback(plugin_name, PLUGIN_INFO, NULL,
+						&stackleak_plugin_info);
+
+	/* Register to be called before processing a translation unit */
+	register_callback(plugin_name, PLUGIN_START_UNIT,
+					&stackleak_start_unit, NULL);
+
+	/* Register an extra GCC garbage collector (GGC) root table */
+	register_callback(plugin_name, PLUGIN_REGISTER_GGC_ROOTS, NULL,
+					(void *)&gt_ggc_r_gt_stackleak);
+
+	/*
+	 * Hook into the Pass Manager to register new gcc passes.
+	 *
+	 * The stack frame size info is available only at the last RTL pass,
+	 * when it's too late to insert complex code like a function call.
+	 * So we register two gcc passes to instrument every function at first
+	 * and remove the unneeded instrumentation later.
+	 */
+	register_callback(plugin_name, PLUGIN_PASS_MANAGER_SETUP, NULL,
+					&stackleak_tree_instrument_pass_info);
+	register_callback(plugin_name, PLUGIN_PASS_MANAGER_SETUP, NULL,
+					&stackleak_final_pass_info);
+
+	return 0;
+}
-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 3/6] x86/entry: Erase kernel stack in syscall_trace_enter()

[Alexander Popov]

Make 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.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 arch/x86/entry/common.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index d7d3cc2..cd38727 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -45,6 +45,12 @@ __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
@@ -127,6 +133,7 @@ static long syscall_trace_enter(struct pt_regs *regs)
 
 	do_audit_syscall_entry(regs, arch);
 
+	erase_kstack();
 	return ret ?: regs->orig_ax;
 }
 
-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 4/6] lkdtm: Add a test for STACKLEAK

[Alexander Popov]

Introduce two lkdtm tests for the STACKLEAK feature: STACKLEAK_CHECK_ALLOCA
and STACKLEAK_TRACK_STACK. Both of them check that the current task stack
is properly erased (filled with STACKLEAK_POISON).

STACKLEAK_CHECK_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_TRACK_STACK tests that exhausting the thread stack with a
recursion hits the guard page provided by CONFIG_VMAP_STACK.

Signed-off-by: Tycho Andersen <tycho@docker.com>
Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 drivers/misc/Makefile          |   3 +
 drivers/misc/lkdtm.h           |   4 ++
 drivers/misc/lkdtm_core.c      |   2 +
 drivers/misc/lkdtm_stackleak.c | 139 +++++++++++++++++++++++++++++++++++++++++
 4 files changed, 148 insertions(+)
 create mode 100644 drivers/misc/lkdtm_stackleak.c

diff --git a/drivers/misc/Makefile b/drivers/misc/Makefile
index 5ca5f64..66c3605 100644
--- a/drivers/misc/Makefile
+++ b/drivers/misc/Makefile
@@ -63,6 +63,9 @@ 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
 
diff --git a/drivers/misc/lkdtm.h b/drivers/misc/lkdtm.h
index 687a0db..2a0cd0d 100644
--- a/drivers/misc/lkdtm.h
+++ b/drivers/misc/lkdtm.h
@@ -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_CHECK_ALLOCA(void);
+void lkdtm_STACKLEAK_TRACK_STACK(void);
+
 #endif
diff --git a/drivers/misc/lkdtm_core.c b/drivers/misc/lkdtm_core.c
index ba92291..3a22ffa 100644
--- a/drivers/misc/lkdtm_core.c
+++ b/drivers/misc/lkdtm_core.c
@@ -183,6 +183,8 @@ static const struct crashtype crashtypes[] = {
 	CRASHTYPE(USERCOPY_STACK_FRAME_FROM),
 	CRASHTYPE(USERCOPY_STACK_BEYOND),
 	CRASHTYPE(USERCOPY_KERNEL),
+	CRASHTYPE(STACKLEAK_CHECK_ALLOCA),
+	CRASHTYPE(STACKLEAK_TRACK_STACK),
 };
 
 
diff --git a/drivers/misc/lkdtm_stackleak.c b/drivers/misc/lkdtm_stackleak.c
new file mode 100644
index 0000000..1f95ee3
--- /dev/null
+++ b/drivers/misc/lkdtm_stackleak.c
@@ -0,0 +1,139 @@
+/*
+ * This file tests a few aspects of the STACKLEAK feature:
+ *   - the current task stack is properly erased (filled with STACKLEAK_POISON);
+ *   - 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();
+ *   - exhausting the thread stack with a recursion is detected: it hits the
+ *    BUG() in track_stack() if CONFIG_VMAP_STACK is not enabled, or hits the
+ *    guard page otherwise.
+ *
+ * Copyright (C) Docker, Inc. 2017
+ *
+ * Authors:
+ *   Tycho Andersen <tycho@docker.com>
+ *   Alexander Popov <alex.popov@linux.com>
+ */
+
+#include "lkdtm.h"
+#include <linux/sched.h>
+#include <linux/compiler.h>
+
+#ifndef CONFIG_GCC_PLUGIN_STACKLEAK
+# define STACKLEAK_POISON -0xBEEF
+# define CONFIG_STACKLEAK_TRACK_MIN_SIZE 100
+#endif
+
+static noinline bool stack_is_erased(void)
+{
+	unsigned long *sp, left, found, i;
+
+	/*
+	 * For the details about the alignment of the poison values, see
+	 * the comment in track_stack().
+	 */
+	sp = PTR_ALIGN(&i, sizeof(unsigned long));
+
+	left = ((unsigned long)sp & (THREAD_SIZE - 1)) / sizeof(unsigned long);
+	sp--;
+
+	/*
+	 * Two unsigned long ints at the bottom of the thread stack are
+	 * reserved and not poisoned.
+	 */
+	if (left <= 2)
+		return false;
+
+	left -= 2;
+	pr_info("checking unused part of the thread stack (%lu bytes)...\n",
+					left * sizeof(unsigned long));
+
+	/* Search for 17 poison values in a row (like erase_kstack() does) */
+	for (i = 0, found = 0; i < left && found < 17; i++) {
+		if (*(sp - i) == STACKLEAK_POISON)
+			found++;
+		else
+			found = 0;
+	}
+
+	if (found < 17) {
+		pr_err("FAIL: thread stack is not erased (checked %lu bytes)\n",
+						i * sizeof(unsigned long));
+		return false;
+	}
+
+	pr_info("first %lu bytes are unpoisoned\n",
+				(i - found) * sizeof(unsigned long));
+
+	/* The rest of thread stack should be erased */
+	for (; i < left; i++) {
+		if (*(sp - i) != STACKLEAK_POISON) {
+			pr_err("FAIL: thread stack is NOT properly erased\n");
+			return false;
+		}
+	}
+
+	pr_info("the rest of the thread stack is properly erased\n");
+	return true;
+}
+
+static noinline void do_alloca(unsigned long size)
+{
+	char buf[size];
+
+	/* So this doesn't get inlined or optimized out */
+	snprintf(buf, size, "hello world\n");
+}
+
+void lkdtm_STACKLEAK_CHECK_ALLOCA(void)
+{
+	unsigned long left = (unsigned long)&left & (THREAD_SIZE - 1);
+
+	if (!stack_is_erased())
+		return;
+
+	/* Try a small alloca to see if it works */
+	pr_info("try a small alloca of 16 bytes...\n");
+	do_alloca(16);
+	pr_info("small alloca is successful\n");
+
+	/* Try to hit the BUG() in check_alloca() */
+	pr_info("try a large alloca of %lu bytes (stack overflow)...\n", left);
+	do_alloca(left);
+	pr_err("FAIL: large alloca overstepped the thread stack boundary\n");
+}
+
+/*
+ * The stack frame size of recursion() is bigger than the
+ * CONFIG_STACKLEAK_TRACK_MIN_SIZE, hence that function is instrumented
+ * by the STACKLEAK gcc plugin and it calls track_stack() at the beginning.
+ */
+static noinline unsigned long recursion(unsigned long prev_sp)
+{
+	char buf[CONFIG_STACKLEAK_TRACK_MIN_SIZE + 42];
+	unsigned long sp = (unsigned long)&sp;
+
+	snprintf(buf, sizeof(buf), "hello world\n");
+
+	if (prev_sp < sp + THREAD_SIZE)
+		sp = recursion(prev_sp);
+
+	return sp;
+}
+
+void lkdtm_STACKLEAK_TRACK_STACK(void)
+{
+	unsigned long sp = (unsigned long)&sp;
+
+	if (!stack_is_erased())
+		return;
+
+	/*
+	 * Exhaust the thread stack with a recursion. It should hit the
+	 * BUG() in track_stack() if CONFIG_VMAP_STACK is not enabled, or
+	 * access the guard page otherwise.
+	 */
+	pr_info("try to exhaust the thread stack with the recursion...\n");
+	pr_err("FAIL: thread stack exhaustion (%lu bytes) is not detected\n",
+							sp - recursion(sp));
+}
-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 5/6] fs/proc: Show STACKLEAK metrics in the /proc file system

[Alexander Popov]

Introduce 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.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 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(+)

diff --git a/arch/Kconfig b/arch/Kconfig
index 6fcb632..0f5a59d 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -558,6 +558,18 @@ 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
diff --git a/arch/x86/entry/entry_32.S b/arch/x86/entry/entry_32.S
index f400d8f..c22d190 100644
--- a/arch/x86/entry/entry_32.S
+++ b/arch/x86/entry/entry_32.S
@@ -116,6 +116,10 @@ ENTRY(erase_kstack)
 	mov	%esp, %ecx
 	sub	%edi, %ecx
 
+#ifdef CONFIG_STACKLEAK_METRICS
+	mov	%edi, TASK_prev_lowest_stack(%ebp)
+#endif
+
 	cmp	$THREAD_SIZE_asm, %ecx
 	jb	3f
 	ud2
diff --git a/arch/x86/entry/entry_64.S b/arch/x86/entry/entry_64.S
index 3ff0802..40e4ba5 100644
--- a/arch/x86/entry/entry_64.S
+++ b/arch/x86/entry/entry_64.S
@@ -118,6 +118,10 @@ 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.
diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
index 9736829..4741f26 100644
--- a/arch/x86/include/asm/processor.h
+++ b/arch/x86/include/asm/processor.h
@@ -498,6 +498,9 @@ 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;
diff --git a/arch/x86/kernel/asm-offsets.c b/arch/x86/kernel/asm-offsets.c
index ef5d260..f48197a 100644
--- a/arch/x86/kernel/asm-offsets.c
+++ b/arch/x86/kernel/asm-offsets.c
@@ -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();
diff --git a/arch/x86/kernel/process_32.c b/arch/x86/kernel/process_32.c
index 6d256ab..48993fe 100644
--- a/arch/x86/kernel/process_32.c
+++ b/arch/x86/kernel/process_32.c
@@ -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)) {
diff --git a/arch/x86/kernel/process_64.c b/arch/x86/kernel/process_64.c
index 4345fdf..faf9fae 100644
--- a/arch/x86/kernel/process_64.c
+++ b/arch/x86/kernel/process_64.c
@@ -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);
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 60316b5..3106265 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -2884,6 +2884,21 @@ 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
  */
@@ -2988,6 +3003,9 @@ 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)
-- 
2.7.4

[kernel-hardening] [PATCH RFC v7 6/6] doc: self-protection: Add information about STACKLEAK feature

[Alexander Popov]

Add information about STACKLEAK feature to "Stack depth overflow" and
"Memory poisoning" sections of self-protection.rst.

Signed-off-by: Alexander Popov <alex.popov@linux.com>
---
 Documentation/security/self-protection.rst | 23 ++++++++++++++---------
 1 file changed, 14 insertions(+), 9 deletions(-)

diff --git a/Documentation/security/self-protection.rst b/Documentation/security/self-protection.rst
index 60c8bd8..9693a90 100644
--- a/Documentation/security/self-protection.rst
+++ b/Documentation/security/self-protection.rst
@@ -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
 ---------------------
@@ -287,11 +292,11 @@ 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
 --------------------
-- 
2.7.4

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Kees Cook]

On Fri, Jan 12, 2018 at 6:19 AM, Alexander Popov <alex.popov@linux.com> wrote:
> This is the 7th version of the patch series introducing STACKLEAK to the
> mainline kernel. 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 think this is really looking good. I had some thoughts while reading
through the patches:

There are really three features in this series, and it might make
sense to separate them a bit more clearly (at least with CONFIG
choices):

1) stack clearing (with depth searching)

2) runtime stack depth tracking (making 1 much more efficient)

3) alloca checking (an additional feature, not strictly part of
clearing, but needs the same plugin infrastructure)

It seems like it should be possible to get 1 without 2 and 3 (both of
which happen in the gcc plugin), and might be good to separate for
builds that don't have gcc plugins.

Once compilers are doing alloca checking (or all VLAs are removed from
the kernel), it'd be nice to be able to avoid the redundancy of 3.

Thanks for continuing to work on this!

-Kees

>
> Changes in v7
> =============
>
> 1. Improved erase_kstack(). 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().
>
> 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%)
>
> Previous changes
> ================
>
> 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                  |  69 +++++
>  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             | 139 +++++++++
>  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, 1001 insertions(+), 9 deletions(-)
>  create mode 100644 drivers/misc/lkdtm_stackleak.c
>  create mode 100644 scripts/gcc-plugins/stackleak_plugin.c
>
> --
> 2.7.4
>



-- 
Kees Cook
Pixel Security

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

Hello Kees,

Thanks for the review.

On 15.01.2018 22:59, Kees Cook wrote:
> On Fri, Jan 12, 2018 at 6:19 AM, Alexander Popov <alex.popov@linux.com> wrote:
>> This is the 7th version of the patch series introducing STACKLEAK to the
>> mainline kernel. 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 think this is really looking good. I had some thoughts while reading
> through the patches:
> 
> There are really three features in this series, and it might make
> sense to separate them a bit more clearly (at least with CONFIG
> choices):
> 
> 1) stack clearing (with depth searching)
> 
> 2) runtime stack depth tracking (making 1 much more efficient)
> 
> 3) alloca checking (an additional feature, not strictly part of
> clearing, but needs the same plugin infrastructure)
> 
> It seems like it should be possible to get 1 without 2 and 3 (both of
> which happen in the gcc plugin), and might be good to separate for
> builds that don't have gcc plugins.
> 
> Once compilers are doing alloca checking (or all VLAs are removed from
> the kernel), it'd be nice to be able to avoid the redundancy of 3.

Agree with your point. I'll make this separation in the next version.

Moreover, I'll create separate tests for these 3 features. Currently
lkdtm_stackleak combines the stack clearing check with the alloca test and the
deep recursion test.

> Thanks for continuing to work on this!

Doing my best!

Hope that x86 maintainers will have a look at this patch series too (when they
are less busy).

Best regards,
Alexander

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

Hello Kees,

On 17.01.2018 14:37, Alexander Popov wrote:
> On 15.01.2018 22:59, Kees Cook wrote:
>> On Fri, Jan 12, 2018 at 6:19 AM, Alexander Popov <alex.popov@linux.com> wrote:
>>> This is the 7th version of the patch series introducing STACKLEAK to the
>>> mainline kernel. 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 think this is really looking good. I had some thoughts while reading
>> through the patches:
>>
>> There are really three features in this series, and it might make
>> sense to separate them a bit more clearly (at least with CONFIG
>> choices):
>>
>> 1) stack clearing (with depth searching)
>>
>> 2) runtime stack depth tracking (making 1 much more efficient)
>>
>> 3) alloca checking (an additional feature, not strictly part of
>> clearing, but needs the same plugin infrastructure)
>>
>> It seems like it should be possible to get 1 without 2 and 3 (both of
>> which happen in the gcc plugin), and might be good to separate for
>> builds that don't have gcc plugins.
>>
>> Once compilers are doing alloca checking (or all VLAs are removed from
>> the kernel), it'd be nice to be able to avoid the redundancy of 3.
> 
> Agree with your point. I'll make this separation in the next version.

I have more thoughts about this separation.

Splitting (1) from (2)
----------------------

It makes the stack erasing not only slow but also unreliable. For example, an
attacker can craft some STACKLEAK_POISON values on the thread stack to deceive
the poison search during the erasing.

Of course, I can increase STACKLEAK_POISON_CHECK_DEPTH or change the default
value of lowest_stack. It will make the stack erasing even slower, but will not
give guarantees.

So I don't think that (1) without (2) is actually a good feature. I would
propose to refrain from separating the stack erasing and the lowest_stack tracking.

Splitting (2) and (3)
---------------------

The STACKLEAK gcc plugin needs to search for alloca anyway (for the correct
lowest_stack tracking). But I can introduce the "no-check-alloca" plugin
parameter for disabling the check_alloca() call insertion.

Is it worth providing something like CONFIG_GCC_PLUGIN_STACKLEAK_NO_CHECK_ALLOCA
in the Kconfig?

Thanks!

Best regards,
Alexander

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Kees Cook]

On Thu, Jan 18, 2018 at 5:09 AM, Alexander Popov <alex.popov@linux.com> wrote:
> Hello Kees,
>
> On 17.01.2018 14:37, Alexander Popov wrote:
>> On 15.01.2018 22:59, Kees Cook wrote:
>>> On Fri, Jan 12, 2018 at 6:19 AM, Alexander Popov <alex.popov@linux.com> wrote:
>>>> This is the 7th version of the patch series introducing STACKLEAK to the
>>>> mainline kernel. 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 think this is really looking good. I had some thoughts while reading
>>> through the patches:
>>>
>>> There are really three features in this series, and it might make
>>> sense to separate them a bit more clearly (at least with CONFIG
>>> choices):
>>>
>>> 1) stack clearing (with depth searching)
>>>
>>> 2) runtime stack depth tracking (making 1 much more efficient)
>>>
>>> 3) alloca checking (an additional feature, not strictly part of
>>> clearing, but needs the same plugin infrastructure)
>>>
>>> It seems like it should be possible to get 1 without 2 and 3 (both of
>>> which happen in the gcc plugin), and might be good to separate for
>>> builds that don't have gcc plugins.
>>>
>>> Once compilers are doing alloca checking (or all VLAs are removed from
>>> the kernel), it'd be nice to be able to avoid the redundancy of 3.
>>
>> Agree with your point. I'll make this separation in the next version.
>
> I have more thoughts about this separation.
>
> Splitting (1) from (2)
> ----------------------
>
> It makes the stack erasing not only slow but also unreliable. For example, an
> attacker can craft some STACKLEAK_POISON values on the thread stack to deceive
> the poison search during the erasing.
>
> Of course, I can increase STACKLEAK_POISON_CHECK_DEPTH or change the default
> value of lowest_stack. It will make the stack erasing even slower, but will not
> give guarantees.

Yeah, that's true. I was hoping the depth check was sufficient.

> So I don't think that (1) without (2) is actually a good feature. I would
> propose to refrain from separating the stack erasing and the lowest_stack tracking.

How about an option to clear the _entire_ stack, then, when the plugin
isn't available? That gives us a range of options and provides an easy
way to compare the performance of the tracking. i.e. can compare off,
full, and smart.

> Splitting (2) and (3)
> ---------------------
>
> The STACKLEAK gcc plugin needs to search for alloca anyway (for the correct
> lowest_stack tracking). But I can introduce the "no-check-alloca" plugin
> parameter for disabling the check_alloca() call insertion.
>
> Is it worth providing something like CONFIG_GCC_PLUGIN_STACKLEAK_NO_CHECK_ALLOCA
> in the Kconfig?

Nah, we can cross that bridge when the compilers have sane alloca checking.

-Kees

-- 
Kees Cook
Pixel Security

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

On 19.01.2018 00:13, Kees Cook wrote:
> On Thu, Jan 18, 2018 at 5:09 AM, Alexander Popov <alex.popov@linux.com> wrote:
>> So I don't think that (1) without (2) is actually a good feature. I would
>> propose to refrain from separating the stack erasing and the lowest_stack tracking.
> 
> How about an option to clear the _entire_ stack, then, when the plugin
> isn't available? That gives us a range of options and provides an easy
> way to compare the performance of the tracking. i.e. can compare off,
> full, and smart.

Yes, I should try it. I'll return with the results of the performance tests.
We'll discuss them; if full stack erasing is not too slow, I'll introduce it in
the 8'th version of the patch series.

Thanks!

Best regards,
Alexander

[kernel-hardening] Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

On 20.01.2018 13:13, Alexander Popov wrote:
> On 19.01.2018 00:13, Kees Cook wrote:
>> On Thu, Jan 18, 2018 at 5:09 AM, Alexander Popov <alex.popov@linux.com> wrote:
>>> So I don't think that (1) without (2) is actually a good feature. I would
>>> propose to refrain from separating the stack erasing and the lowest_stack tracking.
>>
>> How about an option to clear the _entire_ stack, then, when the plugin
>> isn't available? That gives us a range of options and provides an easy
>> way to compare the performance of the tracking. i.e. can compare off,
>> full, and smart.
> 
> Yes, I should try it. I'll return with the results of the performance tests.
> We'll discuss them; if full stack erasing is not too slow, I'll introduce it in
> the 8'th version of the patch series.

I've made a brief performance test on x86_64 (similar to the test described in
the cover letter). I guess there might be workloads with higher performance penalty.

Hardware: Intel Core i7-4770, 16 GB RAM
Test: hackbench -s 4096 -l 2000 -g 15 -f 25 -P

Average time on v4.14.15: 9.194s
Average time on v4.14.15-stackleak-with-plugin: 9.490 (+3.22%)
Average time on v4.14.15-stackleak-full-erasing: 12.149 (+32.14%)

Honestly, I think, it is not worth having this full stack erasing as a separate
feature. Moreover, it brings some #ifdef conditionals to the erase_kstack()
code, which don't look nice.

May I ask for your opinion?

Best regards,
Alexander

Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Alexander Popov]

On 15.01.2018 22:59, Kees Cook wrote:
> On Fri, Jan 12, 2018 at 6:19 AM, Alexander Popov <alex.popov@linux.com> wrote:
>> This is the 7th version of the patch series introducing STACKLEAK to the
>> mainline kernel. 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 think this is really looking good.

Hello! May I ask whether dear x86 maintainers think so too?

Best regards,
Alexander

Re: [PATCH RFC v7 0/6] Introduce the STACKLEAK feature and a test for it

[Kees Cook]

On Thu, Jan 25, 2018 at 7:13 AM, Alexander Popov <alex.popov@linux.com> wrote:
> On 20.01.2018 13:13, Alexander Popov wrote:
>> On 19.01.2018 00:13, Kees Cook wrote:
>>> On Thu, Jan 18, 2018 at 5:09 AM, Alexander Popov <alex.popov@linux.com> wrote:
>>>> So I don't think that (1) without (2) is actually a good feature. I would
>>>> propose to refrain from separating the stack erasing and the lowest_stack tracking.
>>>
>>> How about an option to clear the _entire_ stack, then, when the plugin
>>> isn't available? That gives us a range of options and provides an easy
>>> way to compare the performance of the tracking. i.e. can compare off,
>>> full, and smart.
>>
>> Yes, I should try it. I'll return with the results of the performance tests.
>> We'll discuss them; if full stack erasing is not too slow, I'll introduce it in
>> the 8'th version of the patch series.
>
> I've made a brief performance test on x86_64 (similar to the test described in
> the cover letter). I guess there might be workloads with higher performance penalty.
>
> Hardware: Intel Core i7-4770, 16 GB RAM
> Test: hackbench -s 4096 -l 2000 -g 15 -f 25 -P
>
> Average time on v4.14.15: 9.194s
> Average time on v4.14.15-stackleak-with-plugin: 9.490 (+3.22%)
> Average time on v4.14.15-stackleak-full-erasing: 12.149 (+32.14%)

Yeeeowch. Okay, I'm convinced. :)

> Honestly, I think, it is not worth having this full stack erasing as a separate
> feature. Moreover, it brings some #ifdef conditionals to the erase_kstack()
> code, which don't look nice.
>
> May I ask for your opinion?
>
> Best regards,
> Alexander

-Kees

-- 
Kees Cook
Pixel Security