Uprobe is a tracing mechanism for userspace programs.
Typical uprobe will incur overhead of two traps.
First trap is caused by replaced trap insn, and
the second trap is to execute the original displaced
insn in user space.
To reduce the overhead, kernel provides hooks
for architectures to emulate the original insn
and skip the second trap. In x86, emulation
is done for certain branch insns.
This patch extends the emulation to "push <reg>"
insns. These insns are typical in the beginning
of the function. For example, bcc
in https://github.com/iovisor/bcc repo provides
tools to measure funclantency, detect memleak, etc.
The tools will place uprobes in the beginning of
function and possibly uretprobes at the end of function.
This patch is able to reduce the trap overhead for
uprobe from 2 to 1.
Without this patch, uretprobe will typically incur
three traps. With this patch, if the function starts
with "push" insn, the number of traps can be
reduced from 3 to 2.
An experiment was conducted on two local VMs,
fedora 26 64-bit VM and 32-bit VM, both 4 processors
and 4GB memory, booted with latest tip repo (and this patch).
The host is MacBook with intel i7 processor.
The test program looks like:
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <sys/time.h>
static void test() __attribute__((noinline));
void test() {}
int main() {
struct timeval start, end;
gettimeofday(&start, NULL);
for (int i = 0; i < 1000000; i++) {
test();
}
gettimeofday(&end, NULL);
printf("%ld\n", ((end.tv_sec * 1000000 + end.tv_usec)
- (start.tv_sec * 1000000 + start.tv_usec)));
return 0;
}
The program is compiled without optimization, and
the first insn for function "test" is "push %rbp".
The host is relatively idle.
Before the test run, the uprobe is inserted as below for uprobe:
echo 'p <binary>:<test_func_offset>' > /sys/kernel/debug/tracing/uprobe_events
echo 1 > /sys/kernel/debug/tracing/events/uprobes/enable
and for uretprobe:
echo 'r <binary>:<test_func_offset>' > /sys/kernel/debug/tracing/uprobe_events
echo 1 > /sys/kernel/debug/tracing/events/uprobes/enable
Unit: microsecond(usec) per loop iteration
x86_64 W/ this patch W/O this patch
uprobe 1.55 3.1
uretprobe 2.0 3.6
x86_32 W/ this patch W/O this patch
uprobe 1.41 3.5
uretprobe 1.75 4.0
You can see that this patch significantly reduced the overhead,
50% for uprobe and 44% for uretprobe on x86_64, and even more
on x86_32.
Signed-off-by: Yonghong Song <yhs@fb.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: kernel-team@fb.com
Link: http://lkml.kernel.org/r/20171201001202.3706564-1-yhs@fb.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Since instruction decoder now supports EVEX-encoded instructions, two fixes
are needed to correctly handle them in uprobes.
Extended bits for MODRM.rm field need to be sanitized just like we do it
for VEX3, to avoid encoding wrong register for register-relative access.
EVEX has _two_ extended bits: b and x. Theoretically, EVEX.x should be
ignored by the CPU (since GPRs go only up to 15, not 31), but let's be
paranoid here: proper encoding for register-relative access
should have EVEX.x = 1.
Secondly, we should fetch vex.vvvv for EVEX too.
This is now super easy because instruction decoder populates
vex_prefix.bytes[2] for all flavors of (e)vex encodings, even for VEX2.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: linux-kernel@vger.kernel.org
Cc: <stable@vger.kernel.org> # v4.1+
Fixes: 8a764a875f ("x86/asm/decoder: Create artificial 3rd byte for 2-byte VEX")
Link: http://lkml.kernel.org/r/20160811154521.20469-1-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull x86 asm updates from Ingo Molnar:
"The main changes in this cycle were:
- MSR access API fixes and enhancements (Andy Lutomirski)
- early exception handling improvements (Andy Lutomirski)
- user-space FS/GS prctl usage fixes and improvements (Andy
Lutomirski)
- Remove the cpu_has_*() APIs and replace them with equivalents
(Borislav Petkov)
- task switch micro-optimization (Brian Gerst)
- 32-bit entry code simplification (Denys Vlasenko)
- enhance PAT handling in enumated CPUs (Toshi Kani)
... and lots of other cleanups/fixlets"
* 'x86-asm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (70 commits)
x86/arch_prctl/64: Restore accidentally removed put_cpu() in ARCH_SET_GS
x86/entry/32: Remove asmlinkage_protect()
x86/entry/32: Remove GET_THREAD_INFO() from entry code
x86/entry, sched/x86: Don't save/restore EFLAGS on task switch
x86/asm/entry/32: Simplify pushes of zeroed pt_regs->REGs
selftests/x86/ldt_gdt: Test set_thread_area() deletion of an active segment
x86/tls: Synchronize segment registers in set_thread_area()
x86/asm/64: Rename thread_struct's fs and gs to fsbase and gsbase
x86/arch_prctl/64: Remove FSBASE/GSBASE < 4G optimization
x86/segments/64: When load_gs_index fails, clear the base
x86/segments/64: When loadsegment(fs, ...) fails, clear the base
x86/asm: Make asm/alternative.h safe from assembly
x86/asm: Stop depending on ptrace.h in alternative.h
x86/entry: Rename is_{ia32,x32}_task() to in_{ia32,x32}_syscall()
x86/asm: Make sure verify_cpu() has a good stack
x86/extable: Add a comment about early exception handlers
x86/msr: Set the return value to zero when native_rdmsr_safe() fails
x86/paravirt: Make "unsafe" MSR accesses unsafe even if PARAVIRT=y
x86/paravirt: Add paravirt_{read,write}_msr()
x86/msr: Carry on after a non-"safe" MSR access fails
...
The is_ia32_task()/is_x32_task() function names are a big misnomer: they
suggests that the compat-ness of a system call is a task property, which
is not true, the compatness of a system call purely depends on how it
was invoked through the system call layer.
A task may call 32-bit and 64-bit and x32 system calls without changing
any of its kernel visible state.
This specific minomer is also actively dangerous, as it might cause kernel
developers to use the wrong kind of security checks within system calls.
So rename it to in_{ia32,x32}_syscall().
Suggested-by: Andy Lutomirski <luto@amacapital.net>
Suggested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Dmitry Safonov <dsafonov@virtuozzo.com>
[ Expanded the changelog. ]
Acked-by: Andy Lutomirski <luto@kernel.org>
Cc: 0x7f454c46@gmail.com
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: akpm@linux-foundation.org
Cc: linux-mm@kvack.org
Link: http://lkml.kernel.org/r/1460987025-30360-1-git-send-email-dsafonov@virtuozzo.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The uprobe_xol_ops structures are never modified, so declare them as const.
Done with the help of Coccinelle.
Signed-off-by: Julia Lawall <Julia.Lawall@lip6.fr>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: kernel-janitors@vger.kernel.org
Link: http://lkml.kernel.org/r/1460200649-32526-1-git-send-email-Julia.Lawall@lip6.fr
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The previous change documents that cleanup_return_instances()
can't always detect the dead frames, the stack can grow. But
there is one special case which imho worth fixing:
arch_uretprobe_is_alive() can return true when the stack didn't
actually grow, but the next "call" insn uses the already
invalidated frame.
Test-case:
#include <stdio.h>
#include <setjmp.h>
jmp_buf jmp;
int nr = 1024;
void func_2(void)
{
if (--nr == 0)
return;
longjmp(jmp, 1);
}
void func_1(void)
{
setjmp(jmp);
func_2();
}
int main(void)
{
func_1();
return 0;
}
If you ret-probe func_1() and func_2() prepare_uretprobe() hits
the MAX_URETPROBE_DEPTH limit and "return" from func_2() is not
reported.
When we know that the new call is not chained, we can do the
more strict check. In this case "sp" points to the new ret-addr,
so every frame which uses the same "sp" must be dead. The only
complication is that arch_uretprobe_is_alive() needs to know was
it chained or not, so we add the new RP_CHECK_CHAIN_CALL enum
and change prepare_uretprobe() to pass RP_CHECK_CALL only if
!chained.
Note: arch_uretprobe_is_alive() could also re-read *sp and check
if this word is still trampoline_vaddr. This could obviously
improve the logic, but I would like to avoid another
copy_from_user() especially in the case when we can't avoid the
false "alive == T" positives.
Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134028.GA4786@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86 doesn't care (so far), but as Pratyush Anand pointed
out other architectures might want why arch_uretprobe_is_alive()
was called and use different checks depending on the context.
Add the new argument to distinguish 2 callers.
Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134026.GA4779@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Add the x86 specific version of arch_uretprobe_is_alive()
helper. It returns true if the stack frame mangled by
prepare_uretprobe() is still on stack. So if it returns false,
we know that the probed function has already returned.
We add the new return_instance->stack member and change the
generic code to initialize it in prepare_uretprobe, but it
should be equally useful for other architectures.
TODO: this assumes that the probed application can't use
multiple stacks (say sigaltstack). We will try to improve
this logic later.
Tested-by: Pratyush Anand <panand@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Acked-by: Anton Arapov <arapov@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20150721134018.GA4766@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The uprobes code has a nice helper, is_64bit_mm(), that consults
both the runtime and compile-time flags for 32-bit support.
Instead of reinventing the wheel, pull it in to an x86 header so
we can use it for MPX.
I prefer passing the 'mm' around to test_thread_flag(TIF_IA32)
because it makes it explicit where the context is coming from.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave@sr71.net>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/20150607183704.F0209999@viggo.jf.intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
user_mode_vm() and user_mode() are now the same. Change all callers
of user_mode_vm() to user_mode().
The next patch will remove the definition of user_mode_vm.
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brad Spengler <spender@grsecurity.net>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/43b1f57f3df70df5a08b0925897c660725015554.1426728647.git.luto@kernel.org
[ Merged to a more recent kernel. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Enabled probing of lar, lsl, popcnt, lddqu, prefetch insns.
They should be safe to probe, they throw no exceptions.
Enabled probing of 3-byte opcodes 0f 38-3f xx - these are
vector isns, so should be safe.
Enabled probing of many currently undefined 0f xx insns.
At the rate new vector instructions are getting added,
we don't want to constantly enable more bits.
We want to only occasionally *disable* ones which
for some reason can't be probed.
This includes 0f 24,26 opcodes, which are undefined
since Pentium. On 486, they were "mov to/from test register".
Explained more fully what 0f 78,79 opcodes are.
Explained what 0f ae opcode is. (It's unclear why we don't allow
probing it, but let's not change it for now).
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1423768732-32194-3-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
This change fixes 1-byte opcode tables so that only insns
for which we have real reasons to disallow probing are marked
with unset bits.
To that end:
Set bits for all prefix bytes. Their setting is ignored anyway -
we check the bitmap against OPCODE1(insn), not against first
byte. Keeping them set to 0 only confuses code reader with
"why we don't support that opcode" question.
Thus: enable bytes c4,c5 in 64-bit mode (VEX prefixes).
Byte 62 (EVEX prefix) is not yet enabled since insn decoder
does not support that yet.
For 32-bit mode, enable probing of opcodes 63 (arpl) and d6
(salc). They don't require any special handling.
For 64-bit mode, disable 9a and ea - these undefined opcodes
were mistakenly left enabled.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1423768732-32194-2-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
After adding these, it's clear we have some awkward choices
there. Some valid instructions are prohibited from uprobing
while several invalid ones are allowed.
Hopefully future edits to the good-opcode tables will fix wrong
bits or explain why those bits are not wrong.
No actual code changes.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Jim Keniston <jkenisto@us.ibm.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1423768732-32194-1-git-send-email-dvlasenk@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The current x86 instruction decoder steps along through the
instruction stream but always ensures that it never steps farther
than the largest possible instruction size (MAX_INSN_SIZE).
The MPX code is now going to be doing some decoding of userspace
instructions. We copy those from userspace in to the kernel and
they're obviously completely untrusted coming from userspace. In
addition to the constraint that instructions can only be so long,
we also have to be aware of how long the buffer is that came in
from userspace. This _looks_ to be similar to what the perf and
kprobes is doing, but it's unclear to me whether they are
affected.
The whole reason we need this is that it is perfectly valid to be
executing an instruction within MAX_INSN_SIZE bytes of an
unreadable page. We should be able to gracefully handle short
reads in those cases.
This adds support to the decoder to record how long the buffer
being decoded is and to refuse to "validate" the instruction if
we would have gone over the end of the buffer to decode it.
The kprobes code probably needs to be looked at here a bit more
carefully. This patch still respects the MAX_INSN_SIZE limit
there but the kprobes code does look like it might be able to
be a bit more strict than it currently is.
Signed-off-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: x86@kernel.org
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Ananth N Mavinakayanahalli <ananth@in.ibm.com>
Cc: Anil S Keshavamurthy <anil.s.keshavamurthy@intel.com>
Cc: "David S. Miller" <davem@davemloft.net>
Link: http://lkml.kernel.org/r/20141114153957.E6B01535@viggo.jf.intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Purely cosmetic, no changes in .o,
1. As Jim pointed out arch_uprobe->def looks ambiguous, rename it to
->defparam.
2. Add the comment into default_post_xol_op() to explain "regs->sp +=".
3. Remove the stale part of the comment in arch_uprobe_analyze_insn().
Suggested-by: Jim Keniston <jkenisto@us.ibm.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Before this patch, instructions such as div, mul, shifts with count
in CL, cmpxchg are mishandled.
This patch adds vex prefix handling. In particular, it avoids colliding
with register operand encoded in vex.vvvv field.
Since we need to avoid two possible register operands, the selection of
scratch register needs to be from at least three registers.
After looking through a lot of CPU docs, it looks like the safest choice
is SI,DI,BX. Selecting BX needs care to not collide with implicit use of
BX by cmpxchg8b.
Test-case:
#include <stdio.h>
static const char *const pass[] = { "FAIL", "pass" };
long two = 2;
void test1(void)
{
long ax = 0, dx = 0;
asm volatile("\n"
" xor %%edx,%%edx\n"
" lea 2(%%edx),%%eax\n"
// We divide 2 by 2. Result (in eax) should be 1:
" probe1: .globl probe1\n"
" divl two(%%rip)\n"
// If we have a bug (eax mangled on entry) the result will be 2,
// because eax gets restored by probe machinery.
: "=a" (ax), "=d" (dx) /*out*/
: "0" (ax), "1" (dx) /*in*/
: "memory" /*clobber*/
);
dprintf(2, "%s: %s\n", __func__,
pass[ax == 1]
);
}
long val2 = 0;
void test2(void)
{
long old_val = val2;
long ax = 0, dx = 0;
asm volatile("\n"
" mov val2,%%eax\n" // eax := val2
" lea 1(%%eax),%%edx\n" // edx := eax+1
// eax is equal to val2. cmpxchg should store edx to val2:
" probe2: .globl probe2\n"
" cmpxchg %%edx,val2(%%rip)\n"
// If we have a bug (eax mangled on entry), val2 will stay unchanged
: "=a" (ax), "=d" (dx) /*out*/
: "0" (ax), "1" (dx) /*in*/
: "memory" /*clobber*/
);
dprintf(2, "%s: %s\n", __func__,
pass[val2 == old_val + 1]
);
}
long val3[2] = {0,0};
void test3(void)
{
long old_val = val3[0];
long ax = 0, dx = 0;
asm volatile("\n"
" mov val3,%%eax\n" // edx:eax := val3
" mov val3+4,%%edx\n"
" mov %%eax,%%ebx\n" // ecx:ebx := edx:eax + 1
" mov %%edx,%%ecx\n"
" add $1,%%ebx\n"
" adc $0,%%ecx\n"
// edx:eax is equal to val3. cmpxchg8b should store ecx:ebx to val3:
" probe3: .globl probe3\n"
" cmpxchg8b val3(%%rip)\n"
// If we have a bug (edx:eax mangled on entry), val3 will stay unchanged.
// If ecx:edx in mangled, val3 will get wrong value.
: "=a" (ax), "=d" (dx) /*out*/
: "0" (ax), "1" (dx) /*in*/
: "cx", "bx", "memory" /*clobber*/
);
dprintf(2, "%s: %s\n", __func__,
pass[val3[0] == old_val + 1 && val3[1] == 0]
);
}
int main(int argc, char **argv)
{
test1();
test2();
test3();
return 0;
}
Before this change all tests fail if probe{1,2,3} are probed.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
It is possible to replace rip-relative addressing mode with addressing
mode of the same length: (reg+disp32). This eliminates the need to fix
up immediate and correct for changing instruction length.
And we can kill arch_uprobe->def.riprel_target.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Ignoring the "correction" logic riprel_pre_xol() and riprel_post_xol()
are very similar but look quite differently.
1. Add the "UPROBE_FIX_RIP_AX | UPROBE_FIX_RIP_CX" check at the start
of riprel_pre_xol(), like the same check in riprel_post_xol().
2. Add the trivial scratch_reg() helper which returns the address of
scratch register pre_xol/post_xol need to change.
3. Change these functions to use the new helper and avoid copy-and-paste
under if/else branches.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
default_pre_xol_op() passes ¤t->utask->autask to riprel_pre_xol()
and this is just ugly because it still needs to load current->utask to
read ->vaddr.
Remove this argument, change riprel_pre_xol() to use current->utask.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
handle_riprel_insn(), pre_xol_rip_insn() and handle_riprel_post_xol()
look confusing and inconsistent. Rename them into riprel_analyze(),
riprel_pre_xol(), and riprel_post_xol() respectively.
No changes in compiled code.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Now that UPROBE_FIX_IP/UPROBE_FIX_CALL are mutually exclusive we can
use a single "fix_ip_or_call" enum instead of 2 fix_* booleans. This
way the logic looks more understandable and clean to me.
While at it, join "case 0xea" with other "ip is correct" ret/lret cases.
Also change default_post_xol_op() to use "else if" for the same reason.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
The only insn which could have both UPROBE_FIX_IP and UPROBE_FIX_CALL
was 0xe8 "call relative", and now it is handled by branch_xol_ops.
So we can change default_post_xol_op(UPROBE_FIX_CALL) to simply push
the address of next insn == utask->vaddr + insn.length, just we need
to record insn.length into the new auprobe->def.ilen member.
Note: if/when we teach branch_xol_ops to support jcxz/loopz we can
remove the "correction" logic, UPROBE_FIX_IP can use the same address.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Extract the "push return address" code from branch_emulate_op() into
the new simple helper, push_ret_address(). It will have more users.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
handle_riprel_insn() assumes that nobody else could modify ->fixups
before. This is correct but fragile, change it to use "|=".
Also make ->fixups u8, we are going to add the new members into the
union. It is not clear why UPROBE_FIX_RIP_.X lived in the upper byte,
redefine them so that they can fit into u8.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Finally we can move arch_uprobe->fixups/rip_rela_target_address
into the new "def" struct and place this struct in the union, they
are only used by default_xol_ops paths.
The patch also renames rip_rela_target_address to riprel_target just
to make this name shorter.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
UPROBE_FIX_SETF is only needed to handle "popf" correctly but it is
processed by the generic arch_uprobe_post_xol() code. This doesn't
allows us to make ->fixups private for default_xol_ops.
1 Change default_post_xol_op(UPROBE_FIX_SETF) to set ->saved_tf = T.
"popf" always reads the flags from stack, it doesn't matter if TF
was set or not before single-step. Ignoring the naming, this is
even more logical, "saved_tf" means "owned by application" and we
do not own this flag after "popf".
2. Change arch_uprobe_post_xol() to save ->saved_tf into the local
"bool send_sigtrap" before ->post_xol().
3. Change arch_uprobe_post_xol() to ignore UPROBE_FIX_SETF and just
check ->saved_tf after ->post_xol().
With this patch ->fixups and ->rip_rela_target_address are only used
by default_xol_ops hooks, we are ready to remove them from the common
part of arch_uprobe.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
014940bad8 "uprobes/x86: Send SIGILL if arch_uprobe_post_xol() fails"
changed arch_uprobe_post_xol() to use arch_uprobe_abort_xol() if ->post_xol
fails. This was correct and helped to avoid the additional complications,
we need to clear X86_EFLAGS_TF in this case.
However, now that we have uprobe_xol_ops->abort() hook it would be better
to avoid arch_uprobe_abort_xol() here. ->post_xol() should likely do what
->abort() does anyway, we should not do the same work twice. Currently only
handle_riprel_post_xol() can be called twice, this is unnecessary but safe.
Still this is not clean and can lead to the problems in future.
Change arch_uprobe_post_xol() to clear X86_EFLAGS_TF and restore ->ip by
hand and avoid arch_uprobe_abort_xol(). This temporary uglifies the usage
of autask.saved_tf, we will cleanup this later.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
arch_uprobe_abort_xol() calls handle_riprel_post_xol() even if
auprobe->ops != default_xol_ops. This is fine correctness wise, only
default_pre_xol_op() can set UPROBE_FIX_RIP_AX|UPROBE_FIX_RIP_CX and
otherwise handle_riprel_post_xol() is nop.
But this doesn't look clean and this doesn't allow us to move ->fixups
into the union in arch_uprobe. Move this handle_riprel_post_xol() call
into the new default_abort_op() hook and change arch_uprobe_abort_xol()
accordingly.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Currently this doesn't matter, the only ->pre_xol() hook can't fail,
but we need to fix arch_uprobe_pre_xol() anyway. If ->pre_xol() fails
we should not change regs->ip/flags, we should just return the error
to make restart actually possible.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
is_64bit_mm() assumes that mm->context.ia32_compat means the 32-bit
instruction set, this is not true if the task is TIF_X32.
Change set_personality_ia32() to initialize mm->context.ia32_compat
by TIF_X32 or TIF_IA32 instead of 1. This allows to fix is_64bit_mm()
without affecting other users, they all treat ia32_compat as "bool".
TIF_ in ->ia32_compat looks a bit strange, but this is grep-friendly
and avoids the new define's.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Add the suitable ifdef's around good_insns_* arrays. We do not want
to add the ugly ifdef's into their only user, uprobe_init_insn(), so
the "#else" branch simply defines them as NULL. This doesn't generate
the extra code, gcc is smart enough, although the code is fine even if
it could not detect that (without CONFIG_IA32_EMULATION) is_64bit_mm()
is __builtin_constant_p().
The patch looks more complicated because it also moves good_insns_64
up close to good_insns_32.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Change uprobe_init_insn() to make insn_complete() == T, this makes
other insn_get_*() calls unnecessary.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
1. Extract the ->ia32_compat check from 64bit validate_insn_bits()
into the new helper, is_64bit_mm(), it will have more users.
TODO: this checks is actually wrong if mm owner is X32 task,
we need another fix which changes set_personality_ia32().
TODO: even worse, the whole 64-or-32-bit logic is very broken
and the fix is not simple, we need the nontrivial changes in
the core uprobes code.
2. Kill validate_insn_bits() and change its single caller to use
uprobe_init_insn(is_64bit_mm(mm).
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
validate_insn_32bits() and validate_insn_64bits() are very similar,
turn them into the single uprobe_init_insn() which has the additional
"bool x86_64" argument which can be passed to insn_init() and used to
choose between good_insns_64/good_insns_32.
Also kill UPROBE_FIX_NONE, it has no users.
Note: the current code doesn't use ifdef's consistently, good_insns_64
depends on CONFIG_X86_64 but good_insns_32 is unconditional. This patch
removes ifdef around good_insns_64, we will add it back later along with
the similar one for good_insns_32.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
All branch insns on x86 can be prefixed with the operand-size
override prefix, 0x66. It was only ever useful for performing
jumps to 32-bit offsets in 16-bit code segments.
In 32-bit code, such instructions are useless since
they cause IP truncation to 16 bits, and in case of call insns,
they save only 16 bits of return address and misalign
the stack pointer as a "bonus".
In 64-bit code, such instructions are treated differently by Intel
and AMD CPUs: Intel ignores the prefix altogether,
AMD treats them the same as in 32-bit mode.
Before this patch, the emulation code would execute
the instructions as if they have no 0x66 prefix.
With this patch, we refuse to attach uprobes to such insns.
Signed-off-by: Denys Vlasenko <dvlasenk@redhat.com>
Acked-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Change branch_setup_xol_ops() to simply use opc1 = OPCODE2(insn) - 0x10
if OPCODE1() == 0x0f; this matches the "short" jmp which checks the same
condition.
Thanks to lib/insn.c, it does the rest correctly. branch->ilen/offs are
correct no matter if this jmp is "near" or "short".
Reported-by: Jonathan Lebon <jlebon@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Teach branch_emulate_op() to emulate the conditional "short" jmp's which
check regs->flags.
Note: this doesn't support jcxz/jcexz, loope/loopz, and loopne/loopnz.
They all are rel8 and thus they can't trigger the problem, but perhaps
we will add the support in future just for completeness.
Reported-by: Jonathan Lebon <jlebon@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
See the previous "Emulate unconditional relative jmp's" which explains
why we can not execute "jmp" out-of-line, the same applies to "call".
Emulating of rip-relative call is trivial, we only need to additionally
push the ret-address. If this fails, we execute this instruction out of
line and this should trigger the trap, the probed application should die
or the same insn will be restarted if a signal handler expands the stack.
We do not even need ->post_xol() for this case.
But there is a corner (and almost theoretical) case: another thread can
expand the stack right before we execute this insn out of line. In this
case it hit the same problem we are trying to solve. So we simply turn
the probed insn into "call 1f; 1:" and add ->post_xol() which restores
->sp and restarts.
Many thanks to Jonathan who finally found the standalone reproducer,
otherwise I would never resolve the "random SIGSEGV's under systemtap"
bug-report. Now that the problem is clear we can write the simplified
test-case:
void probe_func(void), callee(void);
int failed = 1;
asm (
".text\n"
".align 4096\n"
".globl probe_func\n"
"probe_func:\n"
"call callee\n"
"ret"
);
/*
* This assumes that:
*
* - &probe_func = 0x401000 + a_bit, aligned = 0x402000
*
* - xol_vma->vm_start = TASK_SIZE_MAX - PAGE_SIZE = 0x7fffffffe000
* as xol_add_vma() asks; the 1st slot = 0x7fffffffe080
*
* so we can target the non-canonical address from xol_vma using
* the simple math below, 100 * 4096 is just the random offset
*/
asm (".org . + 0x800000000000 - 0x7fffffffe080 - 5 - 1 + 100 * 4096\n");
void callee(void)
{
failed = 0;
}
int main(void)
{
probe_func();
return failed;
}
It SIGSEGV's if you probe "probe_func" (although this is not very reliable,
randomize_va_space/etc can change the placement of xol area).
Note: as Denys Vlasenko pointed out, amd and intel treat "callw" (0x66 0xe8)
differently. This patch relies on lib/insn.c and thus implements the intel's
behaviour: 0x66 is simply ignored. Fortunately nothing sane should ever use
this insn, so we postpone the fix until we decide what should we do; emulate
or not, support or not, etc.
Reported-by: Jonathan Lebon <jlebon@redhat.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Finally we can kill the ugly (and very limited) code in __skip_sstep().
Just change branch_setup_xol_ops() to treat "nop" as jmp to the next insn.
Thanks to lib/insn.c, it is clever enough. OPCODE1() == 0x90 includes
"(rep;)+ nop;" at least, and (afaics) much more.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Currently we always execute all insns out-of-line, including relative
jmp's and call's. This assumes that even if regs->ip points to nowhere
after the single-step, default_post_xol_op(UPROBE_FIX_IP) logic will
update it correctly.
However, this doesn't work if this regs->ip == xol_vaddr + insn_offset
is not canonical. In this case CPU generates #GP and general_protection()
kills the task which tries to execute this insn out-of-line.
Now that we have uprobe_xol_ops we can teach uprobes to emulate these
insns and solve the problem. This patch adds branch_xol_ops which has
a single branch_emulate_op() hook, so far it can only handle rel8/32
relative jmp's.
TODO: move ->fixup into the union along with rip_rela_target_address.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Jonathan Lebon <jlebon@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
1. Add the trivial sizeof_long() helper and change other callers of
is_ia32_task() to use it.
TODO: is_ia32_task() is not what we actually want, TS_COMPAT does
not necessarily mean 32bit. Fortunately syscall-like insns can't be
probed so it actually works, but it would be better to rename and
use is_ia32_frame().
2. As Jim pointed out "ncopied" in arch_uretprobe_hijack_return_addr()
and adjust_ret_addr() should be named "nleft". And in fact only the
last copy_to_user() in arch_uretprobe_hijack_return_addr() actually
needs to inspect the non-zero error code.
TODO: adjust_ret_addr() should die. We can always calculate the value
we need to write into *regs->sp, just UPROBE_FIX_CALL should record
insn->length.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
SIGILL after the failed arch_uprobe_post_xol() should only be used as
a last resort, we should try to restart the probed insn if possible.
Currently only adjust_ret_addr() can fail, and this can only happen if
another thread unmapped our stack after we executed "call" out-of-line.
Most probably the application if buggy, but even in this case it can
have a handler for SIGSEGV/etc. And in theory it can be even correct
and do something non-trivial with its memory.
Of course we can't restart unconditionally, so arch_uprobe_post_xol()
does this only if ->post_xol() returns -ERESTART even if currently this
is the only possible error.
default_post_xol_op(UPROBE_FIX_CALL) can always restart, but as Jim
pointed out it should not forget to pop off the return address pushed
by this insn executed out-of-line.
Note: this is not "perfect", we do not want the extra handler_chain()
after restart, but I think this is the best solution we can realistically
do without too much uglifications.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Currently the error from arch_uprobe_post_xol() is silently ignored.
This doesn't look good and this can lead to the hard-to-debug problems.
1. Change handle_singlestep() to loudly complain and send SIGILL.
Note: this only affects x86, ppc/arm can't fail.
2. Change arch_uprobe_post_xol() to call arch_uprobe_abort_xol() and
avoid TF games if it is going to return an error.
This can help to to analyze the problem, if nothing else we should
not report ->ip = xol_slot in the core-file.
Note: this means that handle_riprel_post_xol() can be called twice,
but this is fine because it is idempotent.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
arch_uprobe_analyze_insn() calls handle_riprel_insn() at the start,
but only "0xff" and "default" cases need the UPROBE_FIX_RIP_ logic.
Move the callsite into "default" case and change the "0xff" case to
fall-through.
We are going to add the various hooks to handle the rip-relative
jmp/call instructions (and more), we need this change to enforce the
fact that the new code can not conflict with is_riprel_insn() logic
which, after this change, can only be used by default_xol_ops.
Note: arch_uprobe_abort_xol() still calls handle_riprel_post_xol()
directly. This is fine unless another _xol_ops we may add later will
need to reuse "UPROBE_FIX_RIP_AX|UPROBE_FIX_RIP_CX" bits in ->fixup.
In this case we can add uprobe_xol_ops->abort() hook, which (perhaps)
we will need anyway in the long term.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Introduce arch_uprobe->ops pointing to the "struct uprobe_xol_ops",
move the current UPROBE_FIX_{RIP*,IP,CALL} code into the default
set of methods and change arch_uprobe_pre/post_xol() accordingly.
This way we can add the new uprobe_xol_ops's to handle the insns
which need the special processing (rip-relative jmp/call at least).
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
No functional changes. Preparation to simplify the review of the next
change. Just reorder the code in arch_uprobe_pre/post_xol() functions
so that UPROBE_FIX_{RIP_*,IP,CALL} logic goes to the end.
Also change arch_uprobe_pre_xol() to use utask instead of autask, to
make the code more symmetrical with arch_uprobe_post_xol().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cosmetic. Move pre_xol_rip_insn() and handle_riprel_post_xol() up to
the closely related handle_riprel_insn(). This way it is simpler to
read and understand this code, and this lessens the number of ifdef's.
While at it, update the comment in handle_riprel_post_xol() as Jim
suggested.
TODO: rename them somehow to make the naming consistent.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Kill the "mm->context.ia32_compat" check in handle_riprel_insn(), if
it is true insn_rip_relative() must return false. validate_insn_bits()
passed "ia32_compat" as !x86_64 to insn_init(), and insn_rip_relative()
checks insn->x86_64.
Also, remove the no longer needed "struct mm_struct *mm" argument and
the unnecessary "return" at the end.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
No functional changes, preparation.
Shift the code from prepare_fixups() to arch_uprobe_analyze_insn()
with the following modifications:
- Do not call insn_get_opcode() again, it was already called
by validate_insn_bits().
- Move "case 0xea" up. This way "case 0xff" can fall through
to default case.
- change "case 0xff" to use the nested "switch (MODRM_REG)",
this way the code looks a bit simpler.
- Make the comments look consistent.
While at it, kill the initialization of rip_rela_target_address and
->fixups, we can rely on kzalloc(). We will add the new members into
arch_uprobe, it would be better to assume that everything is zero by
default.
TODO: cleanup/fix the mess in validate_insn_bits() paths:
- validate_insn_64bits() and validate_insn_32bits() should be
unified.
- "ifdef" is not used consistently; if good_insns_64 depends
on CONFIG_X86_64, then probably good_insns_32 should depend
on CONFIG_X86_32/EMULATION
- the usage of mm->context.ia32_compat looks wrong if the task
is TIF_X32.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Reviewed-by: Jim Keniston <jkenisto@us.ibm.com>
Acked-by: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
