2 * Kernel Probes (KProbes)
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 * Copyright (C) IBM Corporation, 2002, 2004
20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
21 * Probes initial implementation ( includes contributions from
23 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
24 * interface to access function arguments.
25 * 2004-Oct Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
26 * <prasanna@in.ibm.com> adapted for x86_64 from i386.
27 * 2005-Mar Roland McGrath <roland@redhat.com>
28 * Fixed to handle %rip-relative addressing mode correctly.
29 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
30 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
31 * <prasanna@in.ibm.com> added function-return probes.
32 * 2005-May Rusty Lynch <rusty.lynch@intel.com>
33 * Added function return probes functionality
34 * 2006-Feb Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added
35 * kprobe-booster and kretprobe-booster for i386.
36 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster
37 * and kretprobe-booster for x86-64
38 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven
39 * <arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com>
40 * unified x86 kprobes code.
42 #include <linux/kprobes.h>
43 #include <linux/ptrace.h>
44 #include <linux/string.h>
45 #include <linux/slab.h>
46 #include <linux/hardirq.h>
47 #include <linux/preempt.h>
48 #include <linux/extable.h>
49 #include <linux/kdebug.h>
50 #include <linux/kallsyms.h>
51 #include <linux/ftrace.h>
52 #include <linux/frame.h>
53 #include <linux/kasan.h>
55 #include <asm/text-patching.h>
56 #include <asm/cacheflush.h>
58 #include <asm/pgtable.h>
59 #include <linux/uaccess.h>
60 #include <asm/alternative.h>
62 #include <asm/debugreg.h>
66 void jprobe_return_end(void);
68 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
69 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
71 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
73 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
74 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
75 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
76 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
77 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
80 * Undefined/reserved opcodes, conditional jump, Opcode Extension
81 * Groups, and some special opcodes can not boost.
82 * This is non-const and volatile to keep gcc from statically
83 * optimizing it out, as variable_test_bit makes gcc think only
84 * *(unsigned long*) is used.
86 static volatile u32 twobyte_is_boostable
[256 / 32] = {
87 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
88 /* ---------------------------------------------- */
89 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
90 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */
91 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
92 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
93 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
94 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
95 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
96 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
97 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
98 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
99 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
100 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
101 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
102 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
103 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
104 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
105 /* ----------------------------------------------- */
106 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
110 struct kretprobe_blackpoint kretprobe_blacklist
[] = {
111 {"__switch_to", }, /* This function switches only current task, but
112 doesn't switch kernel stack.*/
113 {NULL
, NULL
} /* Terminator */
116 const int kretprobe_blacklist_size
= ARRAY_SIZE(kretprobe_blacklist
);
118 static nokprobe_inline
void
119 __synthesize_relative_insn(void *from
, void *to
, u8 op
)
121 struct __arch_relative_insn
{
126 insn
= (struct __arch_relative_insn
*)from
;
127 insn
->raddr
= (s32
)((long)(to
) - ((long)(from
) + 5));
131 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
132 void synthesize_reljump(void *from
, void *to
)
134 __synthesize_relative_insn(from
, to
, RELATIVEJUMP_OPCODE
);
136 NOKPROBE_SYMBOL(synthesize_reljump
);
138 /* Insert a call instruction at address 'from', which calls address 'to'.*/
139 void synthesize_relcall(void *from
, void *to
)
141 __synthesize_relative_insn(from
, to
, RELATIVECALL_OPCODE
);
143 NOKPROBE_SYMBOL(synthesize_relcall
);
146 * Skip the prefixes of the instruction.
148 static kprobe_opcode_t
*skip_prefixes(kprobe_opcode_t
*insn
)
152 attr
= inat_get_opcode_attribute((insn_byte_t
)*insn
);
153 while (inat_is_legacy_prefix(attr
)) {
155 attr
= inat_get_opcode_attribute((insn_byte_t
)*insn
);
158 if (inat_is_rex_prefix(attr
))
163 NOKPROBE_SYMBOL(skip_prefixes
);
166 * Returns non-zero if opcode is boostable.
167 * RIP relative instructions are adjusted at copying time in 64 bits mode
169 int can_boost(kprobe_opcode_t
*opcodes
)
171 kprobe_opcode_t opcode
;
172 kprobe_opcode_t
*orig_opcodes
= opcodes
;
174 if (search_exception_tables((unsigned long)opcodes
))
175 return 0; /* Page fault may occur on this address. */
178 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
180 opcode
= *(opcodes
++);
182 /* 2nd-byte opcode */
183 if (opcode
== 0x0f) {
184 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
186 return test_bit(*opcodes
,
187 (unsigned long *)twobyte_is_boostable
);
190 switch (opcode
& 0xf0) {
193 goto retry
; /* REX prefix is boostable */
196 if (0x63 < opcode
&& opcode
< 0x67)
197 goto retry
; /* prefixes */
198 /* can't boost Address-size override and bound */
199 return (opcode
!= 0x62 && opcode
!= 0x67);
201 return 0; /* can't boost conditional jump */
203 /* can't boost software-interruptions */
204 return (0xc1 < opcode
&& opcode
< 0xcc) || opcode
== 0xcf;
206 /* can boost AA* and XLAT */
207 return (opcode
== 0xd4 || opcode
== 0xd5 || opcode
== 0xd7);
209 /* can boost in/out and absolute jmps */
210 return ((opcode
& 0x04) || opcode
== 0xea);
212 if ((opcode
& 0x0c) == 0 && opcode
!= 0xf1)
213 goto retry
; /* lock/rep(ne) prefix */
214 /* clear and set flags are boostable */
215 return (opcode
== 0xf5 || (0xf7 < opcode
&& opcode
< 0xfe));
217 /* segment override prefixes are boostable */
218 if (opcode
== 0x26 || opcode
== 0x36 || opcode
== 0x3e)
219 goto retry
; /* prefixes */
220 /* CS override prefix and call are not boostable */
221 return (opcode
!= 0x2e && opcode
!= 0x9a);
226 __recover_probed_insn(kprobe_opcode_t
*buf
, unsigned long addr
)
231 kp
= get_kprobe((void *)addr
);
232 faddr
= ftrace_location(addr
);
234 * Addresses inside the ftrace location are refused by
235 * arch_check_ftrace_location(). Something went terribly wrong
236 * if such an address is checked here.
238 if (WARN_ON(faddr
&& faddr
!= addr
))
241 * Use the current code if it is not modified by Kprobe
242 * and it cannot be modified by ftrace.
248 * Basically, kp->ainsn.insn has an original instruction.
249 * However, RIP-relative instruction can not do single-stepping
250 * at different place, __copy_instruction() tweaks the displacement of
251 * that instruction. In that case, we can't recover the instruction
252 * from the kp->ainsn.insn.
254 * On the other hand, in case on normal Kprobe, kp->opcode has a copy
255 * of the first byte of the probed instruction, which is overwritten
256 * by int3. And the instruction at kp->addr is not modified by kprobes
257 * except for the first byte, we can recover the original instruction
258 * from it and kp->opcode.
260 * In case of Kprobes using ftrace, we do not have a copy of
261 * the original instruction. In fact, the ftrace location might
262 * be modified at anytime and even could be in an inconsistent state.
263 * Fortunately, we know that the original code is the ideal 5-byte
266 memcpy(buf
, (void *)addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
268 memcpy(buf
, ideal_nops
[NOP_ATOMIC5
], 5);
271 return (unsigned long)buf
;
275 * Recover the probed instruction at addr for further analysis.
276 * Caller must lock kprobes by kprobe_mutex, or disable preemption
277 * for preventing to release referencing kprobes.
278 * Returns zero if the instruction can not get recovered.
280 unsigned long recover_probed_instruction(kprobe_opcode_t
*buf
, unsigned long addr
)
282 unsigned long __addr
;
284 __addr
= __recover_optprobed_insn(buf
, addr
);
288 return __recover_probed_insn(buf
, addr
);
291 /* Check if paddr is at an instruction boundary */
292 static int can_probe(unsigned long paddr
)
294 unsigned long addr
, __addr
, offset
= 0;
296 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
298 if (!kallsyms_lookup_size_offset(paddr
, NULL
, &offset
))
301 /* Decode instructions */
302 addr
= paddr
- offset
;
303 while (addr
< paddr
) {
305 * Check if the instruction has been modified by another
306 * kprobe, in which case we replace the breakpoint by the
307 * original instruction in our buffer.
308 * Also, jump optimization will change the breakpoint to
309 * relative-jump. Since the relative-jump itself is
310 * normally used, we just go through if there is no kprobe.
312 __addr
= recover_probed_instruction(buf
, addr
);
315 kernel_insn_init(&insn
, (void *)__addr
, MAX_INSN_SIZE
);
316 insn_get_length(&insn
);
319 * Another debugging subsystem might insert this breakpoint.
320 * In that case, we can't recover it.
322 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
)
327 return (addr
== paddr
);
331 * Returns non-zero if opcode modifies the interrupt flag.
333 static int is_IF_modifier(kprobe_opcode_t
*insn
)
336 insn
= skip_prefixes(insn
);
341 case 0xcf: /* iret/iretd */
342 case 0x9d: /* popf/popfd */
350 * Copy an instruction and adjust the displacement if the instruction
351 * uses the %rip-relative addressing mode.
352 * If it does, Return the address of the 32-bit displacement word.
353 * If not, return null.
354 * Only applicable to 64-bit x86.
356 int __copy_instruction(u8
*dest
, u8
*src
)
359 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
361 unsigned long recovered_insn
=
362 recover_probed_instruction(buf
, (unsigned long)src
);
366 kernel_insn_init(&insn
, (void *)recovered_insn
, MAX_INSN_SIZE
);
367 insn_get_length(&insn
);
368 length
= insn
.length
;
370 /* Another subsystem puts a breakpoint, failed to recover */
371 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
)
373 memcpy(dest
, insn
.kaddr
, length
);
376 if (insn_rip_relative(&insn
)) {
379 kernel_insn_init(&insn
, dest
, length
);
380 insn_get_displacement(&insn
);
382 * The copied instruction uses the %rip-relative addressing
383 * mode. Adjust the displacement for the difference between
384 * the original location of this instruction and the location
385 * of the copy that will actually be run. The tricky bit here
386 * is making sure that the sign extension happens correctly in
387 * this calculation, since we need a signed 32-bit result to
388 * be sign-extended to 64 bits when it's added to the %rip
389 * value and yield the same 64-bit result that the sign-
390 * extension of the original signed 32-bit displacement would
393 newdisp
= (u8
*) src
+ (s64
) insn
.displacement
.value
- (u8
*) dest
;
394 if ((s64
) (s32
) newdisp
!= newdisp
) {
395 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp
);
396 pr_err("\tSrc: %p, Dest: %p, old disp: %x\n", src
, dest
, insn
.displacement
.value
);
399 disp
= (u8
*) dest
+ insn_offset_displacement(&insn
);
400 *(s32
*) disp
= (s32
) newdisp
;
406 static int arch_copy_kprobe(struct kprobe
*p
)
410 /* Copy an instruction with recovering if other optprobe modifies it.*/
411 ret
= __copy_instruction(p
->ainsn
.insn
, p
->addr
);
416 * __copy_instruction can modify the displacement of the instruction,
417 * but it doesn't affect boostable check.
419 if (can_boost(p
->ainsn
.insn
))
420 p
->ainsn
.boostable
= 0;
422 p
->ainsn
.boostable
= -1;
424 /* Check whether the instruction modifies Interrupt Flag or not */
425 p
->ainsn
.if_modifier
= is_IF_modifier(p
->ainsn
.insn
);
427 /* Also, displacement change doesn't affect the first byte */
428 p
->opcode
= p
->ainsn
.insn
[0];
433 int arch_prepare_kprobe(struct kprobe
*p
)
435 if (alternatives_text_reserved(p
->addr
, p
->addr
))
438 if (!can_probe((unsigned long)p
->addr
))
440 /* insn: must be on special executable page on x86. */
441 p
->ainsn
.insn
= get_insn_slot();
445 return arch_copy_kprobe(p
);
448 void arch_arm_kprobe(struct kprobe
*p
)
450 text_poke(p
->addr
, ((unsigned char []){BREAKPOINT_INSTRUCTION
}), 1);
453 void arch_disarm_kprobe(struct kprobe
*p
)
455 text_poke(p
->addr
, &p
->opcode
, 1);
458 void arch_remove_kprobe(struct kprobe
*p
)
461 free_insn_slot(p
->ainsn
.insn
, (p
->ainsn
.boostable
== 1));
462 p
->ainsn
.insn
= NULL
;
466 static nokprobe_inline
void
467 save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
469 kcb
->prev_kprobe
.kp
= kprobe_running();
470 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
471 kcb
->prev_kprobe
.old_flags
= kcb
->kprobe_old_flags
;
472 kcb
->prev_kprobe
.saved_flags
= kcb
->kprobe_saved_flags
;
475 static nokprobe_inline
void
476 restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
478 __this_cpu_write(current_kprobe
, kcb
->prev_kprobe
.kp
);
479 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
480 kcb
->kprobe_old_flags
= kcb
->prev_kprobe
.old_flags
;
481 kcb
->kprobe_saved_flags
= kcb
->prev_kprobe
.saved_flags
;
484 static nokprobe_inline
void
485 set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
486 struct kprobe_ctlblk
*kcb
)
488 __this_cpu_write(current_kprobe
, p
);
489 kcb
->kprobe_saved_flags
= kcb
->kprobe_old_flags
490 = (regs
->flags
& (X86_EFLAGS_TF
| X86_EFLAGS_IF
));
491 if (p
->ainsn
.if_modifier
)
492 kcb
->kprobe_saved_flags
&= ~X86_EFLAGS_IF
;
495 static nokprobe_inline
void clear_btf(void)
497 if (test_thread_flag(TIF_BLOCKSTEP
)) {
498 unsigned long debugctl
= get_debugctlmsr();
500 debugctl
&= ~DEBUGCTLMSR_BTF
;
501 update_debugctlmsr(debugctl
);
505 static nokprobe_inline
void restore_btf(void)
507 if (test_thread_flag(TIF_BLOCKSTEP
)) {
508 unsigned long debugctl
= get_debugctlmsr();
510 debugctl
|= DEBUGCTLMSR_BTF
;
511 update_debugctlmsr(debugctl
);
515 void arch_prepare_kretprobe(struct kretprobe_instance
*ri
, struct pt_regs
*regs
)
517 unsigned long *sara
= stack_addr(regs
);
519 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
521 /* Replace the return addr with trampoline addr */
522 *sara
= (unsigned long) &kretprobe_trampoline
;
524 NOKPROBE_SYMBOL(arch_prepare_kretprobe
);
526 static void setup_singlestep(struct kprobe
*p
, struct pt_regs
*regs
,
527 struct kprobe_ctlblk
*kcb
, int reenter
)
529 if (setup_detour_execution(p
, regs
, reenter
))
532 #if !defined(CONFIG_PREEMPT)
533 if (p
->ainsn
.boostable
== 1 && !p
->post_handler
) {
534 /* Boost up -- we can execute copied instructions directly */
536 reset_current_kprobe();
538 * Reentering boosted probe doesn't reset current_kprobe,
539 * nor set current_kprobe, because it doesn't use single
542 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
543 preempt_enable_no_resched();
548 save_previous_kprobe(kcb
);
549 set_current_kprobe(p
, regs
, kcb
);
550 kcb
->kprobe_status
= KPROBE_REENTER
;
552 kcb
->kprobe_status
= KPROBE_HIT_SS
;
553 /* Prepare real single stepping */
555 regs
->flags
|= X86_EFLAGS_TF
;
556 regs
->flags
&= ~X86_EFLAGS_IF
;
557 /* single step inline if the instruction is an int3 */
558 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
559 regs
->ip
= (unsigned long)p
->addr
;
561 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
563 NOKPROBE_SYMBOL(setup_singlestep
);
566 * We have reentered the kprobe_handler(), since another probe was hit while
567 * within the handler. We save the original kprobes variables and just single
568 * step on the instruction of the new probe without calling any user handlers.
570 static int reenter_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
571 struct kprobe_ctlblk
*kcb
)
573 switch (kcb
->kprobe_status
) {
574 case KPROBE_HIT_SSDONE
:
575 case KPROBE_HIT_ACTIVE
:
577 kprobes_inc_nmissed_count(p
);
578 setup_singlestep(p
, regs
, kcb
, 1);
581 /* A probe has been hit in the codepath leading up to, or just
582 * after, single-stepping of a probed instruction. This entire
583 * codepath should strictly reside in .kprobes.text section.
584 * Raise a BUG or we'll continue in an endless reentering loop
585 * and eventually a stack overflow.
587 printk(KERN_WARNING
"Unrecoverable kprobe detected at %p.\n",
592 /* impossible cases */
599 NOKPROBE_SYMBOL(reenter_kprobe
);
602 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
603 * remain disabled throughout this function.
605 int kprobe_int3_handler(struct pt_regs
*regs
)
607 kprobe_opcode_t
*addr
;
609 struct kprobe_ctlblk
*kcb
;
614 addr
= (kprobe_opcode_t
*)(regs
->ip
- sizeof(kprobe_opcode_t
));
616 * We don't want to be preempted for the entire
617 * duration of kprobe processing. We conditionally
618 * re-enable preemption at the end of this function,
619 * and also in reenter_kprobe() and setup_singlestep().
623 kcb
= get_kprobe_ctlblk();
624 p
= get_kprobe(addr
);
627 if (kprobe_running()) {
628 if (reenter_kprobe(p
, regs
, kcb
))
631 set_current_kprobe(p
, regs
, kcb
);
632 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
635 * If we have no pre-handler or it returned 0, we
636 * continue with normal processing. If we have a
637 * pre-handler and it returned non-zero, it prepped
638 * for calling the break_handler below on re-entry
639 * for jprobe processing, so get out doing nothing
642 if (!p
->pre_handler
|| !p
->pre_handler(p
, regs
))
643 setup_singlestep(p
, regs
, kcb
, 0);
646 } else if (*addr
!= BREAKPOINT_INSTRUCTION
) {
648 * The breakpoint instruction was removed right
649 * after we hit it. Another cpu has removed
650 * either a probepoint or a debugger breakpoint
651 * at this address. In either case, no further
652 * handling of this interrupt is appropriate.
653 * Back up over the (now missing) int3 and run
654 * the original instruction.
656 regs
->ip
= (unsigned long)addr
;
657 preempt_enable_no_resched();
659 } else if (kprobe_running()) {
660 p
= __this_cpu_read(current_kprobe
);
661 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
662 if (!skip_singlestep(p
, regs
, kcb
))
663 setup_singlestep(p
, regs
, kcb
, 0);
666 } /* else: not a kprobe fault; let the kernel handle it */
668 preempt_enable_no_resched();
671 NOKPROBE_SYMBOL(kprobe_int3_handler
);
674 * When a retprobed function returns, this code saves registers and
675 * calls trampoline_handler() runs, which calls the kretprobe's handler.
678 ".global kretprobe_trampoline\n"
679 ".type kretprobe_trampoline, @function\n"
680 "kretprobe_trampoline:\n"
682 /* We don't bother saving the ss register */
687 " call trampoline_handler\n"
688 /* Replace saved sp with true return address. */
689 " movq %rax, 152(%rsp)\n"
696 " call trampoline_handler\n"
697 /* Move flags to cs */
698 " movl 56(%esp), %edx\n"
699 " movl %edx, 52(%esp)\n"
700 /* Replace saved flags with true return address. */
701 " movl %eax, 56(%esp)\n"
706 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"
708 NOKPROBE_SYMBOL(kretprobe_trampoline
);
709 STACK_FRAME_NON_STANDARD(kretprobe_trampoline
);
712 * Called from kretprobe_trampoline
714 __visible __used
void *trampoline_handler(struct pt_regs
*regs
)
716 struct kretprobe_instance
*ri
= NULL
;
717 struct hlist_head
*head
, empty_rp
;
718 struct hlist_node
*tmp
;
719 unsigned long flags
, orig_ret_address
= 0;
720 unsigned long trampoline_address
= (unsigned long)&kretprobe_trampoline
;
721 kprobe_opcode_t
*correct_ret_addr
= NULL
;
723 INIT_HLIST_HEAD(&empty_rp
);
724 kretprobe_hash_lock(current
, &head
, &flags
);
725 /* fixup registers */
727 regs
->cs
= __KERNEL_CS
;
729 regs
->cs
= __KERNEL_CS
| get_kernel_rpl();
732 regs
->ip
= trampoline_address
;
733 regs
->orig_ax
= ~0UL;
736 * It is possible to have multiple instances associated with a given
737 * task either because multiple functions in the call path have
738 * return probes installed on them, and/or more than one
739 * return probe was registered for a target function.
741 * We can handle this because:
742 * - instances are always pushed into the head of the list
743 * - when multiple return probes are registered for the same
744 * function, the (chronologically) first instance's ret_addr
745 * will be the real return address, and all the rest will
746 * point to kretprobe_trampoline.
748 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
749 if (ri
->task
!= current
)
750 /* another task is sharing our hash bucket */
753 orig_ret_address
= (unsigned long)ri
->ret_addr
;
755 if (orig_ret_address
!= trampoline_address
)
757 * This is the real return address. Any other
758 * instances associated with this task are for
759 * other calls deeper on the call stack
764 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
766 correct_ret_addr
= ri
->ret_addr
;
767 hlist_for_each_entry_safe(ri
, tmp
, head
, hlist
) {
768 if (ri
->task
!= current
)
769 /* another task is sharing our hash bucket */
772 orig_ret_address
= (unsigned long)ri
->ret_addr
;
773 if (ri
->rp
&& ri
->rp
->handler
) {
774 __this_cpu_write(current_kprobe
, &ri
->rp
->kp
);
775 get_kprobe_ctlblk()->kprobe_status
= KPROBE_HIT_ACTIVE
;
776 ri
->ret_addr
= correct_ret_addr
;
777 ri
->rp
->handler(ri
, regs
);
778 __this_cpu_write(current_kprobe
, NULL
);
781 recycle_rp_inst(ri
, &empty_rp
);
783 if (orig_ret_address
!= trampoline_address
)
785 * This is the real return address. Any other
786 * instances associated with this task are for
787 * other calls deeper on the call stack
792 kretprobe_hash_unlock(current
, &flags
);
794 hlist_for_each_entry_safe(ri
, tmp
, &empty_rp
, hlist
) {
795 hlist_del(&ri
->hlist
);
798 return (void *)orig_ret_address
;
800 NOKPROBE_SYMBOL(trampoline_handler
);
803 * Called after single-stepping. p->addr is the address of the
804 * instruction whose first byte has been replaced by the "int 3"
805 * instruction. To avoid the SMP problems that can occur when we
806 * temporarily put back the original opcode to single-step, we
807 * single-stepped a copy of the instruction. The address of this
808 * copy is p->ainsn.insn.
810 * This function prepares to return from the post-single-step
811 * interrupt. We have to fix up the stack as follows:
813 * 0) Except in the case of absolute or indirect jump or call instructions,
814 * the new ip is relative to the copied instruction. We need to make
815 * it relative to the original instruction.
817 * 1) If the single-stepped instruction was pushfl, then the TF and IF
818 * flags are set in the just-pushed flags, and may need to be cleared.
820 * 2) If the single-stepped instruction was a call, the return address
821 * that is atop the stack is the address following the copied instruction.
822 * We need to make it the address following the original instruction.
824 * If this is the first time we've single-stepped the instruction at
825 * this probepoint, and the instruction is boostable, boost it: add a
826 * jump instruction after the copied instruction, that jumps to the next
827 * instruction after the probepoint.
829 static void resume_execution(struct kprobe
*p
, struct pt_regs
*regs
,
830 struct kprobe_ctlblk
*kcb
)
832 unsigned long *tos
= stack_addr(regs
);
833 unsigned long copy_ip
= (unsigned long)p
->ainsn
.insn
;
834 unsigned long orig_ip
= (unsigned long)p
->addr
;
835 kprobe_opcode_t
*insn
= p
->ainsn
.insn
;
838 insn
= skip_prefixes(insn
);
840 regs
->flags
&= ~X86_EFLAGS_TF
;
842 case 0x9c: /* pushfl */
843 *tos
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_IF
);
844 *tos
|= kcb
->kprobe_old_flags
;
846 case 0xc2: /* iret/ret/lret */
851 case 0xea: /* jmp absolute -- ip is correct */
852 /* ip is already adjusted, no more changes required */
853 p
->ainsn
.boostable
= 1;
855 case 0xe8: /* call relative - Fix return addr */
856 *tos
= orig_ip
+ (*tos
- copy_ip
);
859 case 0x9a: /* call absolute -- same as call absolute, indirect */
860 *tos
= orig_ip
+ (*tos
- copy_ip
);
864 if ((insn
[1] & 0x30) == 0x10) {
866 * call absolute, indirect
867 * Fix return addr; ip is correct.
868 * But this is not boostable
870 *tos
= orig_ip
+ (*tos
- copy_ip
);
872 } else if (((insn
[1] & 0x31) == 0x20) ||
873 ((insn
[1] & 0x31) == 0x21)) {
875 * jmp near and far, absolute indirect
876 * ip is correct. And this is boostable
878 p
->ainsn
.boostable
= 1;
885 if (p
->ainsn
.boostable
== 0) {
886 if ((regs
->ip
> copy_ip
) &&
887 (regs
->ip
- copy_ip
) + 5 < MAX_INSN_SIZE
) {
889 * These instructions can be executed directly if it
890 * jumps back to correct address.
892 synthesize_reljump((void *)regs
->ip
,
893 (void *)orig_ip
+ (regs
->ip
- copy_ip
));
894 p
->ainsn
.boostable
= 1;
896 p
->ainsn
.boostable
= -1;
900 regs
->ip
+= orig_ip
- copy_ip
;
905 NOKPROBE_SYMBOL(resume_execution
);
908 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
909 * remain disabled throughout this function.
911 int kprobe_debug_handler(struct pt_regs
*regs
)
913 struct kprobe
*cur
= kprobe_running();
914 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
919 resume_execution(cur
, regs
, kcb
);
920 regs
->flags
|= kcb
->kprobe_saved_flags
;
922 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
923 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
924 cur
->post_handler(cur
, regs
, 0);
927 /* Restore back the original saved kprobes variables and continue. */
928 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
929 restore_previous_kprobe(kcb
);
932 reset_current_kprobe();
934 preempt_enable_no_resched();
937 * if somebody else is singlestepping across a probe point, flags
938 * will have TF set, in which case, continue the remaining processing
939 * of do_debug, as if this is not a probe hit.
941 if (regs
->flags
& X86_EFLAGS_TF
)
946 NOKPROBE_SYMBOL(kprobe_debug_handler
);
948 int kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
950 struct kprobe
*cur
= kprobe_running();
951 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
953 if (unlikely(regs
->ip
== (unsigned long)cur
->ainsn
.insn
)) {
954 /* This must happen on single-stepping */
955 WARN_ON(kcb
->kprobe_status
!= KPROBE_HIT_SS
&&
956 kcb
->kprobe_status
!= KPROBE_REENTER
);
958 * We are here because the instruction being single
959 * stepped caused a page fault. We reset the current
960 * kprobe and the ip points back to the probe address
961 * and allow the page fault handler to continue as a
964 regs
->ip
= (unsigned long)cur
->addr
;
966 * Trap flag (TF) has been set here because this fault
967 * happened where the single stepping will be done.
968 * So clear it by resetting the current kprobe:
970 regs
->flags
&= ~X86_EFLAGS_TF
;
973 * If the TF flag was set before the kprobe hit,
976 regs
->flags
|= kcb
->kprobe_old_flags
;
978 if (kcb
->kprobe_status
== KPROBE_REENTER
)
979 restore_previous_kprobe(kcb
);
981 reset_current_kprobe();
982 preempt_enable_no_resched();
983 } else if (kcb
->kprobe_status
== KPROBE_HIT_ACTIVE
||
984 kcb
->kprobe_status
== KPROBE_HIT_SSDONE
) {
986 * We increment the nmissed count for accounting,
987 * we can also use npre/npostfault count for accounting
988 * these specific fault cases.
990 kprobes_inc_nmissed_count(cur
);
993 * We come here because instructions in the pre/post
994 * handler caused the page_fault, this could happen
995 * if handler tries to access user space by
996 * copy_from_user(), get_user() etc. Let the
997 * user-specified handler try to fix it first.
999 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
1003 * In case the user-specified fault handler returned
1004 * zero, try to fix up.
1006 if (fixup_exception(regs
, trapnr
))
1010 * fixup routine could not handle it,
1011 * Let do_page_fault() fix it.
1017 NOKPROBE_SYMBOL(kprobe_fault_handler
);
1020 * Wrapper routine for handling exceptions.
1022 int kprobe_exceptions_notify(struct notifier_block
*self
, unsigned long val
,
1025 struct die_args
*args
= data
;
1026 int ret
= NOTIFY_DONE
;
1028 if (args
->regs
&& user_mode(args
->regs
))
1031 if (val
== DIE_GPF
) {
1033 * To be potentially processing a kprobe fault and to
1034 * trust the result from kprobe_running(), we have
1035 * be non-preemptible.
1037 if (!preemptible() && kprobe_running() &&
1038 kprobe_fault_handler(args
->regs
, args
->trapnr
))
1043 NOKPROBE_SYMBOL(kprobe_exceptions_notify
);
1045 int setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1047 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1049 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1051 kcb
->jprobe_saved_regs
= *regs
;
1052 kcb
->jprobe_saved_sp
= stack_addr(regs
);
1053 addr
= (unsigned long)(kcb
->jprobe_saved_sp
);
1056 * As Linus pointed out, gcc assumes that the callee
1057 * owns the argument space and could overwrite it, e.g.
1058 * tailcall optimization. So, to be absolutely safe
1059 * we also save and restore enough stack bytes to cover
1060 * the argument area.
1061 * Use __memcpy() to avoid KASAN stack out-of-bounds reports as we copy
1062 * raw stack chunk with redzones:
1064 __memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
, MIN_STACK_SIZE(addr
));
1065 regs
->flags
&= ~X86_EFLAGS_IF
;
1066 trace_hardirqs_off();
1067 regs
->ip
= (unsigned long)(jp
->entry
);
1070 * jprobes use jprobe_return() which skips the normal return
1071 * path of the function, and this messes up the accounting of the
1072 * function graph tracer to get messed up.
1074 * Pause function graph tracing while performing the jprobe function.
1076 pause_graph_tracing();
1079 NOKPROBE_SYMBOL(setjmp_pre_handler
);
1081 void jprobe_return(void)
1083 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1085 /* Unpoison stack redzones in the frames we are going to jump over. */
1086 kasan_unpoison_stack_above_sp_to(kcb
->jprobe_saved_sp
);
1089 #ifdef CONFIG_X86_64
1090 " xchg %%rbx,%%rsp \n"
1092 " xchgl %%ebx,%%esp \n"
1095 " .globl jprobe_return_end\n"
1096 " jprobe_return_end: \n"
1098 (kcb
->jprobe_saved_sp
):"memory");
1100 NOKPROBE_SYMBOL(jprobe_return
);
1101 NOKPROBE_SYMBOL(jprobe_return_end
);
1103 int longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1105 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1106 u8
*addr
= (u8
*) (regs
->ip
- 1);
1107 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1108 void *saved_sp
= kcb
->jprobe_saved_sp
;
1110 if ((addr
> (u8
*) jprobe_return
) &&
1111 (addr
< (u8
*) jprobe_return_end
)) {
1112 if (stack_addr(regs
) != saved_sp
) {
1113 struct pt_regs
*saved_regs
= &kcb
->jprobe_saved_regs
;
1115 "current sp %p does not match saved sp %p\n",
1116 stack_addr(regs
), saved_sp
);
1117 printk(KERN_ERR
"Saved registers for jprobe %p\n", jp
);
1118 show_regs(saved_regs
);
1119 printk(KERN_ERR
"Current registers\n");
1123 /* It's OK to start function graph tracing again */
1124 unpause_graph_tracing();
1125 *regs
= kcb
->jprobe_saved_regs
;
1126 __memcpy(saved_sp
, kcb
->jprobes_stack
, MIN_STACK_SIZE(saved_sp
));
1127 preempt_enable_no_resched();
1132 NOKPROBE_SYMBOL(longjmp_break_handler
);
1134 bool arch_within_kprobe_blacklist(unsigned long addr
)
1136 return (addr
>= (unsigned long)__kprobes_text_start
&&
1137 addr
< (unsigned long)__kprobes_text_end
) ||
1138 (addr
>= (unsigned long)__entry_text_start
&&
1139 addr
< (unsigned long)__entry_text_end
);
1142 int __init
arch_init_kprobes(void)
1147 int arch_trampoline_kprobe(struct kprobe
*p
)