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.
43 #include <linux/kprobes.h>
44 #include <linux/ptrace.h>
45 #include <linux/string.h>
46 #include <linux/slab.h>
47 #include <linux/hardirq.h>
48 #include <linux/preempt.h>
49 #include <linux/module.h>
50 #include <linux/kdebug.h>
51 #include <linux/kallsyms.h>
52 #include <linux/ftrace.h>
54 #include <asm/cacheflush.h>
56 #include <asm/pgtable.h>
57 #include <asm/uaccess.h>
58 #include <asm/alternative.h>
60 #include <asm/debugreg.h>
62 void jprobe_return_end(void);
64 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
65 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
67 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs))
69 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
70 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
71 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
72 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
73 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
76 * Undefined/reserved opcodes, conditional jump, Opcode Extension
77 * Groups, and some special opcodes can not boost.
79 static const u32 twobyte_is_boostable
[256 / 32] = {
80 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
81 /* ---------------------------------------------- */
82 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
83 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
84 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
85 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
86 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
87 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
88 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
89 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
90 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
91 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
92 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
93 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
94 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
95 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
96 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
97 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
98 /* ----------------------------------------------- */
99 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
103 struct kretprobe_blackpoint kretprobe_blacklist
[] = {
104 {"__switch_to", }, /* This function switches only current task, but
105 doesn't switch kernel stack.*/
106 {NULL
, NULL
} /* Terminator */
108 const int kretprobe_blacklist_size
= ARRAY_SIZE(kretprobe_blacklist
);
110 static void __kprobes
__synthesize_relative_insn(void *from
, void *to
, u8 op
)
112 struct __arch_relative_insn
{
115 } __attribute__((packed
)) *insn
;
117 insn
= (struct __arch_relative_insn
*)from
;
118 insn
->raddr
= (s32
)((long)(to
) - ((long)(from
) + 5));
122 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
123 static void __kprobes
synthesize_reljump(void *from
, void *to
)
125 __synthesize_relative_insn(from
, to
, RELATIVEJUMP_OPCODE
);
129 * Skip the prefixes of the instruction.
131 static kprobe_opcode_t
*__kprobes
skip_prefixes(kprobe_opcode_t
*insn
)
135 attr
= inat_get_opcode_attribute((insn_byte_t
)*insn
);
136 while (inat_is_legacy_prefix(attr
)) {
138 attr
= inat_get_opcode_attribute((insn_byte_t
)*insn
);
141 if (inat_is_rex_prefix(attr
))
148 * Returns non-zero if opcode is boostable.
149 * RIP relative instructions are adjusted at copying time in 64 bits mode
151 static int __kprobes
can_boost(kprobe_opcode_t
*opcodes
)
153 kprobe_opcode_t opcode
;
154 kprobe_opcode_t
*orig_opcodes
= opcodes
;
156 if (search_exception_tables((unsigned long)opcodes
))
157 return 0; /* Page fault may occur on this address. */
160 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
162 opcode
= *(opcodes
++);
164 /* 2nd-byte opcode */
165 if (opcode
== 0x0f) {
166 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
168 return test_bit(*opcodes
,
169 (unsigned long *)twobyte_is_boostable
);
172 switch (opcode
& 0xf0) {
175 goto retry
; /* REX prefix is boostable */
178 if (0x63 < opcode
&& opcode
< 0x67)
179 goto retry
; /* prefixes */
180 /* can't boost Address-size override and bound */
181 return (opcode
!= 0x62 && opcode
!= 0x67);
183 return 0; /* can't boost conditional jump */
185 /* can't boost software-interruptions */
186 return (0xc1 < opcode
&& opcode
< 0xcc) || opcode
== 0xcf;
188 /* can boost AA* and XLAT */
189 return (opcode
== 0xd4 || opcode
== 0xd5 || opcode
== 0xd7);
191 /* can boost in/out and absolute jmps */
192 return ((opcode
& 0x04) || opcode
== 0xea);
194 if ((opcode
& 0x0c) == 0 && opcode
!= 0xf1)
195 goto retry
; /* lock/rep(ne) prefix */
196 /* clear and set flags are boostable */
197 return (opcode
== 0xf5 || (0xf7 < opcode
&& opcode
< 0xfe));
199 /* segment override prefixes are boostable */
200 if (opcode
== 0x26 || opcode
== 0x36 || opcode
== 0x3e)
201 goto retry
; /* prefixes */
202 /* CS override prefix and call are not boostable */
203 return (opcode
!= 0x2e && opcode
!= 0x9a);
207 /* Recover the probed instruction at addr for further analysis. */
208 static int recover_probed_instruction(kprobe_opcode_t
*buf
, unsigned long addr
)
211 kp
= get_kprobe((void *)addr
);
216 * Basically, kp->ainsn.insn has an original instruction.
217 * However, RIP-relative instruction can not do single-stepping
218 * at different place, __copy_instruction() tweaks the displacement of
219 * that instruction. In that case, we can't recover the instruction
220 * from the kp->ainsn.insn.
222 * On the other hand, kp->opcode has a copy of the first byte of
223 * the probed instruction, which is overwritten by int3. And
224 * the instruction at kp->addr is not modified by kprobes except
225 * for the first byte, we can recover the original instruction
226 * from it and kp->opcode.
228 memcpy(buf
, kp
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
233 /* Check if paddr is at an instruction boundary */
234 static int __kprobes
can_probe(unsigned long paddr
)
237 unsigned long addr
, offset
= 0;
239 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
241 if (!kallsyms_lookup_size_offset(paddr
, NULL
, &offset
))
244 /* Decode instructions */
245 addr
= paddr
- offset
;
246 while (addr
< paddr
) {
247 kernel_insn_init(&insn
, (void *)addr
);
248 insn_get_opcode(&insn
);
251 * Check if the instruction has been modified by another
252 * kprobe, in which case we replace the breakpoint by the
253 * original instruction in our buffer.
255 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
) {
256 ret
= recover_probed_instruction(buf
, addr
);
259 * Another debugging subsystem might insert
260 * this breakpoint. In that case, we can't
264 kernel_insn_init(&insn
, buf
);
266 insn_get_length(&insn
);
270 return (addr
== paddr
);
274 * Returns non-zero if opcode modifies the interrupt flag.
276 static int __kprobes
is_IF_modifier(kprobe_opcode_t
*insn
)
279 insn
= skip_prefixes(insn
);
284 case 0xcf: /* iret/iretd */
285 case 0x9d: /* popf/popfd */
293 * Copy an instruction and adjust the displacement if the instruction
294 * uses the %rip-relative addressing mode.
295 * If it does, Return the address of the 32-bit displacement word.
296 * If not, return null.
297 * Only applicable to 64-bit x86.
299 static int __kprobes
__copy_instruction(u8
*dest
, u8
*src
, int recover
)
303 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
305 kernel_insn_init(&insn
, src
);
307 insn_get_opcode(&insn
);
308 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
) {
309 ret
= recover_probed_instruction(buf
,
313 kernel_insn_init(&insn
, buf
);
316 insn_get_length(&insn
);
317 memcpy(dest
, insn
.kaddr
, insn
.length
);
320 if (insn_rip_relative(&insn
)) {
323 kernel_insn_init(&insn
, dest
);
324 insn_get_displacement(&insn
);
326 * The copied instruction uses the %rip-relative addressing
327 * mode. Adjust the displacement for the difference between
328 * the original location of this instruction and the location
329 * of the copy that will actually be run. The tricky bit here
330 * is making sure that the sign extension happens correctly in
331 * this calculation, since we need a signed 32-bit result to
332 * be sign-extended to 64 bits when it's added to the %rip
333 * value and yield the same 64-bit result that the sign-
334 * extension of the original signed 32-bit displacement would
337 newdisp
= (u8
*) src
+ (s64
) insn
.displacement
.value
-
339 BUG_ON((s64
) (s32
) newdisp
!= newdisp
); /* Sanity check. */
340 disp
= (u8
*) dest
+ insn_offset_displacement(&insn
);
341 *(s32
*) disp
= (s32
) newdisp
;
347 static void __kprobes
arch_copy_kprobe(struct kprobe
*p
)
350 * Copy an instruction without recovering int3, because it will be
351 * put by another subsystem.
353 __copy_instruction(p
->ainsn
.insn
, p
->addr
, 0);
355 if (can_boost(p
->addr
))
356 p
->ainsn
.boostable
= 0;
358 p
->ainsn
.boostable
= -1;
360 p
->opcode
= *p
->addr
;
363 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
365 if (alternatives_text_reserved(p
->addr
, p
->addr
))
368 if (!can_probe((unsigned long)p
->addr
))
370 /* insn: must be on special executable page on x86. */
371 p
->ainsn
.insn
= get_insn_slot();
378 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
380 text_poke(p
->addr
, ((unsigned char []){BREAKPOINT_INSTRUCTION
}), 1);
383 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
385 text_poke(p
->addr
, &p
->opcode
, 1);
388 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
391 free_insn_slot(p
->ainsn
.insn
, (p
->ainsn
.boostable
== 1));
392 p
->ainsn
.insn
= NULL
;
396 static void __kprobes
save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
398 kcb
->prev_kprobe
.kp
= kprobe_running();
399 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
400 kcb
->prev_kprobe
.old_flags
= kcb
->kprobe_old_flags
;
401 kcb
->prev_kprobe
.saved_flags
= kcb
->kprobe_saved_flags
;
404 static void __kprobes
restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
406 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
407 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
408 kcb
->kprobe_old_flags
= kcb
->prev_kprobe
.old_flags
;
409 kcb
->kprobe_saved_flags
= kcb
->prev_kprobe
.saved_flags
;
412 static void __kprobes
set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
413 struct kprobe_ctlblk
*kcb
)
415 __get_cpu_var(current_kprobe
) = p
;
416 kcb
->kprobe_saved_flags
= kcb
->kprobe_old_flags
417 = (regs
->flags
& (X86_EFLAGS_TF
| X86_EFLAGS_IF
));
418 if (is_IF_modifier(p
->ainsn
.insn
))
419 kcb
->kprobe_saved_flags
&= ~X86_EFLAGS_IF
;
422 static void __kprobes
clear_btf(void)
424 if (test_thread_flag(TIF_BLOCKSTEP
)) {
425 unsigned long debugctl
= get_debugctlmsr();
427 debugctl
&= ~DEBUGCTLMSR_BTF
;
428 update_debugctlmsr(debugctl
);
432 static void __kprobes
restore_btf(void)
434 if (test_thread_flag(TIF_BLOCKSTEP
)) {
435 unsigned long debugctl
= get_debugctlmsr();
437 debugctl
|= DEBUGCTLMSR_BTF
;
438 update_debugctlmsr(debugctl
);
442 void __kprobes
arch_prepare_kretprobe(struct kretprobe_instance
*ri
,
443 struct pt_regs
*regs
)
445 unsigned long *sara
= stack_addr(regs
);
447 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
449 /* Replace the return addr with trampoline addr */
450 *sara
= (unsigned long) &kretprobe_trampoline
;
453 #ifdef CONFIG_OPTPROBES
454 static int __kprobes
setup_detour_execution(struct kprobe
*p
,
455 struct pt_regs
*regs
,
458 #define setup_detour_execution(p, regs, reenter) (0)
461 static void __kprobes
setup_singlestep(struct kprobe
*p
, struct pt_regs
*regs
,
462 struct kprobe_ctlblk
*kcb
, int reenter
)
464 if (setup_detour_execution(p
, regs
, reenter
))
467 #if !defined(CONFIG_PREEMPT)
468 if (p
->ainsn
.boostable
== 1 && !p
->post_handler
) {
469 /* Boost up -- we can execute copied instructions directly */
471 reset_current_kprobe();
473 * Reentering boosted probe doesn't reset current_kprobe,
474 * nor set current_kprobe, because it doesn't use single
477 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
478 preempt_enable_no_resched();
483 save_previous_kprobe(kcb
);
484 set_current_kprobe(p
, regs
, kcb
);
485 kcb
->kprobe_status
= KPROBE_REENTER
;
487 kcb
->kprobe_status
= KPROBE_HIT_SS
;
488 /* Prepare real single stepping */
490 regs
->flags
|= X86_EFLAGS_TF
;
491 regs
->flags
&= ~X86_EFLAGS_IF
;
492 /* single step inline if the instruction is an int3 */
493 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
494 regs
->ip
= (unsigned long)p
->addr
;
496 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
500 * We have reentered the kprobe_handler(), since another probe was hit while
501 * within the handler. We save the original kprobes variables and just single
502 * step on the instruction of the new probe without calling any user handlers.
504 static int __kprobes
reenter_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
505 struct kprobe_ctlblk
*kcb
)
507 switch (kcb
->kprobe_status
) {
508 case KPROBE_HIT_SSDONE
:
509 case KPROBE_HIT_ACTIVE
:
510 kprobes_inc_nmissed_count(p
);
511 setup_singlestep(p
, regs
, kcb
, 1);
514 /* A probe has been hit in the codepath leading up to, or just
515 * after, single-stepping of a probed instruction. This entire
516 * codepath should strictly reside in .kprobes.text section.
517 * Raise a BUG or we'll continue in an endless reentering loop
518 * and eventually a stack overflow.
520 printk(KERN_WARNING
"Unrecoverable kprobe detected at %p.\n",
525 /* impossible cases */
534 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
535 * remain disabled throughout this function.
537 static int __kprobes
kprobe_handler(struct pt_regs
*regs
)
539 kprobe_opcode_t
*addr
;
541 struct kprobe_ctlblk
*kcb
;
543 addr
= (kprobe_opcode_t
*)(regs
->ip
- sizeof(kprobe_opcode_t
));
545 * We don't want to be preempted for the entire
546 * duration of kprobe processing. We conditionally
547 * re-enable preemption at the end of this function,
548 * and also in reenter_kprobe() and setup_singlestep().
552 kcb
= get_kprobe_ctlblk();
553 p
= get_kprobe(addr
);
556 if (kprobe_running()) {
557 if (reenter_kprobe(p
, regs
, kcb
))
560 set_current_kprobe(p
, regs
, kcb
);
561 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
564 * If we have no pre-handler or it returned 0, we
565 * continue with normal processing. If we have a
566 * pre-handler and it returned non-zero, it prepped
567 * for calling the break_handler below on re-entry
568 * for jprobe processing, so get out doing nothing
571 if (!p
->pre_handler
|| !p
->pre_handler(p
, regs
))
572 setup_singlestep(p
, regs
, kcb
, 0);
575 } else if (*addr
!= BREAKPOINT_INSTRUCTION
) {
577 * The breakpoint instruction was removed right
578 * after we hit it. Another cpu has removed
579 * either a probepoint or a debugger breakpoint
580 * at this address. In either case, no further
581 * handling of this interrupt is appropriate.
582 * Back up over the (now missing) int3 and run
583 * the original instruction.
585 regs
->ip
= (unsigned long)addr
;
586 preempt_enable_no_resched();
588 } else if (kprobe_running()) {
589 p
= __get_cpu_var(current_kprobe
);
590 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
591 setup_singlestep(p
, regs
, kcb
, 0);
594 } /* else: not a kprobe fault; let the kernel handle it */
596 preempt_enable_no_resched();
601 #define SAVE_REGS_STRING \
602 /* Skip cs, ip, orig_ax. */ \
603 " subq $24, %rsp\n" \
619 #define RESTORE_REGS_STRING \
635 /* Skip orig_ax, ip, cs */ \
638 #define SAVE_REGS_STRING \
639 /* Skip cs, ip, orig_ax and gs. */ \
640 " subl $16, %esp\n" \
651 #define RESTORE_REGS_STRING \
659 /* Skip ds, es, fs, gs, orig_ax, and ip. Note: don't pop cs here*/\
664 * When a retprobed function returns, this code saves registers and
665 * calls trampoline_handler() runs, which calls the kretprobe's handler.
667 static void __used __kprobes
kretprobe_trampoline_holder(void)
670 ".global kretprobe_trampoline\n"
671 "kretprobe_trampoline: \n"
673 /* We don't bother saving the ss register */
678 " call trampoline_handler\n"
679 /* Replace saved sp with true return address. */
680 " movq %rax, 152(%rsp)\n"
687 " call trampoline_handler\n"
688 /* Move flags to cs */
689 " movl 56(%esp), %edx\n"
690 " movl %edx, 52(%esp)\n"
691 /* Replace saved flags with true return address. */
692 " movl %eax, 56(%esp)\n"
700 * Called from kretprobe_trampoline
702 static __used __kprobes
void *trampoline_handler(struct pt_regs
*regs
)
704 struct kretprobe_instance
*ri
= NULL
;
705 struct hlist_head
*head
, empty_rp
;
706 struct hlist_node
*node
, *tmp
;
707 unsigned long flags
, orig_ret_address
= 0;
708 unsigned long trampoline_address
= (unsigned long)&kretprobe_trampoline
;
709 kprobe_opcode_t
*correct_ret_addr
= NULL
;
711 INIT_HLIST_HEAD(&empty_rp
);
712 kretprobe_hash_lock(current
, &head
, &flags
);
713 /* fixup registers */
715 regs
->cs
= __KERNEL_CS
;
717 regs
->cs
= __KERNEL_CS
| get_kernel_rpl();
720 regs
->ip
= trampoline_address
;
721 regs
->orig_ax
= ~0UL;
724 * It is possible to have multiple instances associated with a given
725 * task either because multiple functions in the call path have
726 * return probes installed on them, and/or more than one
727 * return probe was registered for a target function.
729 * We can handle this because:
730 * - instances are always pushed into the head of the list
731 * - when multiple return probes are registered for the same
732 * function, the (chronologically) first instance's ret_addr
733 * will be the real return address, and all the rest will
734 * point to kretprobe_trampoline.
736 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
737 if (ri
->task
!= current
)
738 /* another task is sharing our hash bucket */
741 orig_ret_address
= (unsigned long)ri
->ret_addr
;
743 if (orig_ret_address
!= trampoline_address
)
745 * This is the real return address. Any other
746 * instances associated with this task are for
747 * other calls deeper on the call stack
752 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
754 correct_ret_addr
= ri
->ret_addr
;
755 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
756 if (ri
->task
!= current
)
757 /* another task is sharing our hash bucket */
760 orig_ret_address
= (unsigned long)ri
->ret_addr
;
761 if (ri
->rp
&& ri
->rp
->handler
) {
762 __get_cpu_var(current_kprobe
) = &ri
->rp
->kp
;
763 get_kprobe_ctlblk()->kprobe_status
= KPROBE_HIT_ACTIVE
;
764 ri
->ret_addr
= correct_ret_addr
;
765 ri
->rp
->handler(ri
, regs
);
766 __get_cpu_var(current_kprobe
) = NULL
;
769 recycle_rp_inst(ri
, &empty_rp
);
771 if (orig_ret_address
!= trampoline_address
)
773 * This is the real return address. Any other
774 * instances associated with this task are for
775 * other calls deeper on the call stack
780 kretprobe_hash_unlock(current
, &flags
);
782 hlist_for_each_entry_safe(ri
, node
, tmp
, &empty_rp
, hlist
) {
783 hlist_del(&ri
->hlist
);
786 return (void *)orig_ret_address
;
790 * Called after single-stepping. p->addr is the address of the
791 * instruction whose first byte has been replaced by the "int 3"
792 * instruction. To avoid the SMP problems that can occur when we
793 * temporarily put back the original opcode to single-step, we
794 * single-stepped a copy of the instruction. The address of this
795 * copy is p->ainsn.insn.
797 * This function prepares to return from the post-single-step
798 * interrupt. We have to fix up the stack as follows:
800 * 0) Except in the case of absolute or indirect jump or call instructions,
801 * the new ip is relative to the copied instruction. We need to make
802 * it relative to the original instruction.
804 * 1) If the single-stepped instruction was pushfl, then the TF and IF
805 * flags are set in the just-pushed flags, and may need to be cleared.
807 * 2) If the single-stepped instruction was a call, the return address
808 * that is atop the stack is the address following the copied instruction.
809 * We need to make it the address following the original instruction.
811 * If this is the first time we've single-stepped the instruction at
812 * this probepoint, and the instruction is boostable, boost it: add a
813 * jump instruction after the copied instruction, that jumps to the next
814 * instruction after the probepoint.
816 static void __kprobes
resume_execution(struct kprobe
*p
,
817 struct pt_regs
*regs
, struct kprobe_ctlblk
*kcb
)
819 unsigned long *tos
= stack_addr(regs
);
820 unsigned long copy_ip
= (unsigned long)p
->ainsn
.insn
;
821 unsigned long orig_ip
= (unsigned long)p
->addr
;
822 kprobe_opcode_t
*insn
= p
->ainsn
.insn
;
825 insn
= skip_prefixes(insn
);
827 regs
->flags
&= ~X86_EFLAGS_TF
;
829 case 0x9c: /* pushfl */
830 *tos
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_IF
);
831 *tos
|= kcb
->kprobe_old_flags
;
833 case 0xc2: /* iret/ret/lret */
838 case 0xea: /* jmp absolute -- ip is correct */
839 /* ip is already adjusted, no more changes required */
840 p
->ainsn
.boostable
= 1;
842 case 0xe8: /* call relative - Fix return addr */
843 *tos
= orig_ip
+ (*tos
- copy_ip
);
846 case 0x9a: /* call absolute -- same as call absolute, indirect */
847 *tos
= orig_ip
+ (*tos
- copy_ip
);
851 if ((insn
[1] & 0x30) == 0x10) {
853 * call absolute, indirect
854 * Fix return addr; ip is correct.
855 * But this is not boostable
857 *tos
= orig_ip
+ (*tos
- copy_ip
);
859 } else if (((insn
[1] & 0x31) == 0x20) ||
860 ((insn
[1] & 0x31) == 0x21)) {
862 * jmp near and far, absolute indirect
863 * ip is correct. And this is boostable
865 p
->ainsn
.boostable
= 1;
872 if (p
->ainsn
.boostable
== 0) {
873 if ((regs
->ip
> copy_ip
) &&
874 (regs
->ip
- copy_ip
) + 5 < MAX_INSN_SIZE
) {
876 * These instructions can be executed directly if it
877 * jumps back to correct address.
879 synthesize_reljump((void *)regs
->ip
,
880 (void *)orig_ip
+ (regs
->ip
- copy_ip
));
881 p
->ainsn
.boostable
= 1;
883 p
->ainsn
.boostable
= -1;
887 regs
->ip
+= orig_ip
- copy_ip
;
894 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
895 * remain disabled throughout this function.
897 static int __kprobes
post_kprobe_handler(struct pt_regs
*regs
)
899 struct kprobe
*cur
= kprobe_running();
900 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
905 resume_execution(cur
, regs
, kcb
);
906 regs
->flags
|= kcb
->kprobe_saved_flags
;
908 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
909 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
910 cur
->post_handler(cur
, regs
, 0);
913 /* Restore back the original saved kprobes variables and continue. */
914 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
915 restore_previous_kprobe(kcb
);
918 reset_current_kprobe();
920 preempt_enable_no_resched();
923 * if somebody else is singlestepping across a probe point, flags
924 * will have TF set, in which case, continue the remaining processing
925 * of do_debug, as if this is not a probe hit.
927 if (regs
->flags
& X86_EFLAGS_TF
)
933 int __kprobes
kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
935 struct kprobe
*cur
= kprobe_running();
936 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
938 switch (kcb
->kprobe_status
) {
942 * We are here because the instruction being single
943 * stepped caused a page fault. We reset the current
944 * kprobe and the ip points back to the probe address
945 * and allow the page fault handler to continue as a
948 regs
->ip
= (unsigned long)cur
->addr
;
949 regs
->flags
|= kcb
->kprobe_old_flags
;
950 if (kcb
->kprobe_status
== KPROBE_REENTER
)
951 restore_previous_kprobe(kcb
);
953 reset_current_kprobe();
954 preempt_enable_no_resched();
956 case KPROBE_HIT_ACTIVE
:
957 case KPROBE_HIT_SSDONE
:
959 * We increment the nmissed count for accounting,
960 * we can also use npre/npostfault count for accounting
961 * these specific fault cases.
963 kprobes_inc_nmissed_count(cur
);
966 * We come here because instructions in the pre/post
967 * handler caused the page_fault, this could happen
968 * if handler tries to access user space by
969 * copy_from_user(), get_user() etc. Let the
970 * user-specified handler try to fix it first.
972 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
976 * In case the user-specified fault handler returned
977 * zero, try to fix up.
979 if (fixup_exception(regs
))
983 * fixup routine could not handle it,
984 * Let do_page_fault() fix it.
994 * Wrapper routine for handling exceptions.
996 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
997 unsigned long val
, void *data
)
999 struct die_args
*args
= data
;
1000 int ret
= NOTIFY_DONE
;
1002 if (args
->regs
&& user_mode_vm(args
->regs
))
1007 if (kprobe_handler(args
->regs
))
1011 if (post_kprobe_handler(args
->regs
)) {
1013 * Reset the BS bit in dr6 (pointed by args->err) to
1014 * denote completion of processing
1016 (*(unsigned long *)ERR_PTR(args
->err
)) &= ~DR_STEP
;
1022 * To be potentially processing a kprobe fault and to
1023 * trust the result from kprobe_running(), we have
1024 * be non-preemptible.
1026 if (!preemptible() && kprobe_running() &&
1027 kprobe_fault_handler(args
->regs
, args
->trapnr
))
1036 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1038 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1040 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1042 kcb
->jprobe_saved_regs
= *regs
;
1043 kcb
->jprobe_saved_sp
= stack_addr(regs
);
1044 addr
= (unsigned long)(kcb
->jprobe_saved_sp
);
1047 * As Linus pointed out, gcc assumes that the callee
1048 * owns the argument space and could overwrite it, e.g.
1049 * tailcall optimization. So, to be absolutely safe
1050 * we also save and restore enough stack bytes to cover
1051 * the argument area.
1053 memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
,
1054 MIN_STACK_SIZE(addr
));
1055 regs
->flags
&= ~X86_EFLAGS_IF
;
1056 trace_hardirqs_off();
1057 regs
->ip
= (unsigned long)(jp
->entry
);
1061 void __kprobes
jprobe_return(void)
1063 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1066 #ifdef CONFIG_X86_64
1067 " xchg %%rbx,%%rsp \n"
1069 " xchgl %%ebx,%%esp \n"
1072 " .globl jprobe_return_end\n"
1073 " jprobe_return_end: \n"
1075 (kcb
->jprobe_saved_sp
):"memory");
1078 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1080 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1081 u8
*addr
= (u8
*) (regs
->ip
- 1);
1082 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1084 if ((addr
> (u8
*) jprobe_return
) &&
1085 (addr
< (u8
*) jprobe_return_end
)) {
1086 if (stack_addr(regs
) != kcb
->jprobe_saved_sp
) {
1087 struct pt_regs
*saved_regs
= &kcb
->jprobe_saved_regs
;
1089 "current sp %p does not match saved sp %p\n",
1090 stack_addr(regs
), kcb
->jprobe_saved_sp
);
1091 printk(KERN_ERR
"Saved registers for jprobe %p\n", jp
);
1092 show_registers(saved_regs
);
1093 printk(KERN_ERR
"Current registers\n");
1094 show_registers(regs
);
1097 *regs
= kcb
->jprobe_saved_regs
;
1098 memcpy((kprobe_opcode_t
*)(kcb
->jprobe_saved_sp
),
1100 MIN_STACK_SIZE(kcb
->jprobe_saved_sp
));
1101 preempt_enable_no_resched();
1108 #ifdef CONFIG_OPTPROBES
1110 /* Insert a call instruction at address 'from', which calls address 'to'.*/
1111 static void __kprobes
synthesize_relcall(void *from
, void *to
)
1113 __synthesize_relative_insn(from
, to
, RELATIVECALL_OPCODE
);
1116 /* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
1117 static void __kprobes
synthesize_set_arg1(kprobe_opcode_t
*addr
,
1120 #ifdef CONFIG_X86_64
1126 *(unsigned long *)addr
= val
;
1129 static void __used __kprobes
kprobes_optinsn_template_holder(void)
1132 ".global optprobe_template_entry\n"
1133 "optprobe_template_entry: \n"
1134 #ifdef CONFIG_X86_64
1135 /* We don't bother saving the ss register */
1139 " movq %rsp, %rsi\n"
1140 ".global optprobe_template_val\n"
1141 "optprobe_template_val: \n"
1144 ".global optprobe_template_call\n"
1145 "optprobe_template_call: \n"
1147 /* Move flags to rsp */
1148 " movq 144(%rsp), %rdx\n"
1149 " movq %rdx, 152(%rsp)\n"
1151 /* Skip flags entry */
1154 #else /* CONFIG_X86_32 */
1157 " movl %esp, %edx\n"
1158 ".global optprobe_template_val\n"
1159 "optprobe_template_val: \n"
1161 ".global optprobe_template_call\n"
1162 "optprobe_template_call: \n"
1165 " addl $4, %esp\n" /* skip cs */
1168 ".global optprobe_template_end\n"
1169 "optprobe_template_end: \n");
1172 #define TMPL_MOVE_IDX \
1173 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
1174 #define TMPL_CALL_IDX \
1175 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
1176 #define TMPL_END_IDX \
1177 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
1179 #define INT3_SIZE sizeof(kprobe_opcode_t)
1181 /* Optimized kprobe call back function: called from optinsn */
1182 static void __kprobes
optimized_callback(struct optimized_kprobe
*op
,
1183 struct pt_regs
*regs
)
1185 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1187 /* This is possible if op is under delayed unoptimizing */
1188 if (kprobe_disabled(&op
->kp
))
1192 if (kprobe_running()) {
1193 kprobes_inc_nmissed_count(&op
->kp
);
1195 /* Save skipped registers */
1196 #ifdef CONFIG_X86_64
1197 regs
->cs
= __KERNEL_CS
;
1199 regs
->cs
= __KERNEL_CS
| get_kernel_rpl();
1202 regs
->ip
= (unsigned long)op
->kp
.addr
+ INT3_SIZE
;
1203 regs
->orig_ax
= ~0UL;
1205 __get_cpu_var(current_kprobe
) = &op
->kp
;
1206 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
1207 opt_pre_handler(&op
->kp
, regs
);
1208 __get_cpu_var(current_kprobe
) = NULL
;
1210 preempt_enable_no_resched();
1213 static int __kprobes
copy_optimized_instructions(u8
*dest
, u8
*src
)
1217 while (len
< RELATIVEJUMP_SIZE
) {
1218 ret
= __copy_instruction(dest
+ len
, src
+ len
, 1);
1219 if (!ret
|| !can_boost(dest
+ len
))
1223 /* Check whether the address range is reserved */
1224 if (ftrace_text_reserved(src
, src
+ len
- 1) ||
1225 alternatives_text_reserved(src
, src
+ len
- 1) ||
1226 jump_label_text_reserved(src
, src
+ len
- 1))
1232 /* Check whether insn is indirect jump */
1233 static int __kprobes
insn_is_indirect_jump(struct insn
*insn
)
1235 return ((insn
->opcode
.bytes
[0] == 0xff &&
1236 (X86_MODRM_REG(insn
->modrm
.value
) & 6) == 4) || /* Jump */
1237 insn
->opcode
.bytes
[0] == 0xea); /* Segment based jump */
1240 /* Check whether insn jumps into specified address range */
1241 static int insn_jump_into_range(struct insn
*insn
, unsigned long start
, int len
)
1243 unsigned long target
= 0;
1245 switch (insn
->opcode
.bytes
[0]) {
1246 case 0xe0: /* loopne */
1247 case 0xe1: /* loope */
1248 case 0xe2: /* loop */
1249 case 0xe3: /* jcxz */
1250 case 0xe9: /* near relative jump */
1251 case 0xeb: /* short relative jump */
1254 if ((insn
->opcode
.bytes
[1] & 0xf0) == 0x80) /* jcc near */
1258 if ((insn
->opcode
.bytes
[0] & 0xf0) == 0x70) /* jcc short */
1262 target
= (unsigned long)insn
->next_byte
+ insn
->immediate
.value
;
1264 return (start
<= target
&& target
<= start
+ len
);
1267 /* Decode whole function to ensure any instructions don't jump into target */
1268 static int __kprobes
can_optimize(unsigned long paddr
)
1271 unsigned long addr
, size
= 0, offset
= 0;
1273 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
1275 /* Lookup symbol including addr */
1276 if (!kallsyms_lookup_size_offset(paddr
, &size
, &offset
))
1279 /* Check there is enough space for a relative jump. */
1280 if (size
- offset
< RELATIVEJUMP_SIZE
)
1283 /* Decode instructions */
1284 addr
= paddr
- offset
;
1285 while (addr
< paddr
- offset
+ size
) { /* Decode until function end */
1286 if (search_exception_tables(addr
))
1288 * Since some fixup code will jumps into this function,
1289 * we can't optimize kprobe in this function.
1292 kernel_insn_init(&insn
, (void *)addr
);
1293 insn_get_opcode(&insn
);
1294 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
) {
1295 ret
= recover_probed_instruction(buf
, addr
);
1298 kernel_insn_init(&insn
, buf
);
1300 insn_get_length(&insn
);
1301 /* Recover address */
1302 insn
.kaddr
= (void *)addr
;
1303 insn
.next_byte
= (void *)(addr
+ insn
.length
);
1304 /* Check any instructions don't jump into target */
1305 if (insn_is_indirect_jump(&insn
) ||
1306 insn_jump_into_range(&insn
, paddr
+ INT3_SIZE
,
1307 RELATIVE_ADDR_SIZE
))
1309 addr
+= insn
.length
;
1315 /* Check optimized_kprobe can actually be optimized. */
1316 int __kprobes
arch_check_optimized_kprobe(struct optimized_kprobe
*op
)
1321 for (i
= 1; i
< op
->optinsn
.size
; i
++) {
1322 p
= get_kprobe(op
->kp
.addr
+ i
);
1323 if (p
&& !kprobe_disabled(p
))
1330 /* Check the addr is within the optimized instructions. */
1331 int __kprobes
arch_within_optimized_kprobe(struct optimized_kprobe
*op
,
1334 return ((unsigned long)op
->kp
.addr
<= addr
&&
1335 (unsigned long)op
->kp
.addr
+ op
->optinsn
.size
> addr
);
1338 /* Free optimized instruction slot */
1340 void __arch_remove_optimized_kprobe(struct optimized_kprobe
*op
, int dirty
)
1342 if (op
->optinsn
.insn
) {
1343 free_optinsn_slot(op
->optinsn
.insn
, dirty
);
1344 op
->optinsn
.insn
= NULL
;
1345 op
->optinsn
.size
= 0;
1349 void __kprobes
arch_remove_optimized_kprobe(struct optimized_kprobe
*op
)
1351 __arch_remove_optimized_kprobe(op
, 1);
1355 * Copy replacing target instructions
1356 * Target instructions MUST be relocatable (checked inside)
1358 int __kprobes
arch_prepare_optimized_kprobe(struct optimized_kprobe
*op
)
1364 if (!can_optimize((unsigned long)op
->kp
.addr
))
1367 op
->optinsn
.insn
= get_optinsn_slot();
1368 if (!op
->optinsn
.insn
)
1372 * Verify if the address gap is in 2GB range, because this uses
1375 rel
= (long)op
->optinsn
.insn
- (long)op
->kp
.addr
+ RELATIVEJUMP_SIZE
;
1376 if (abs(rel
) > 0x7fffffff)
1379 buf
= (u8
*)op
->optinsn
.insn
;
1381 /* Copy instructions into the out-of-line buffer */
1382 ret
= copy_optimized_instructions(buf
+ TMPL_END_IDX
, op
->kp
.addr
);
1384 __arch_remove_optimized_kprobe(op
, 0);
1387 op
->optinsn
.size
= ret
;
1389 /* Copy arch-dep-instance from template */
1390 memcpy(buf
, &optprobe_template_entry
, TMPL_END_IDX
);
1392 /* Set probe information */
1393 synthesize_set_arg1(buf
+ TMPL_MOVE_IDX
, (unsigned long)op
);
1395 /* Set probe function call */
1396 synthesize_relcall(buf
+ TMPL_CALL_IDX
, optimized_callback
);
1398 /* Set returning jmp instruction at the tail of out-of-line buffer */
1399 synthesize_reljump(buf
+ TMPL_END_IDX
+ op
->optinsn
.size
,
1400 (u8
*)op
->kp
.addr
+ op
->optinsn
.size
);
1402 flush_icache_range((unsigned long) buf
,
1403 (unsigned long) buf
+ TMPL_END_IDX
+
1404 op
->optinsn
.size
+ RELATIVEJUMP_SIZE
);
1408 #define MAX_OPTIMIZE_PROBES 256
1409 static struct text_poke_param
*jump_poke_params
;
1410 static struct jump_poke_buffer
{
1411 u8 buf
[RELATIVEJUMP_SIZE
];
1414 static void __kprobes
setup_optimize_kprobe(struct text_poke_param
*tprm
,
1416 struct optimized_kprobe
*op
)
1418 s32 rel
= (s32
)((long)op
->optinsn
.insn
-
1419 ((long)op
->kp
.addr
+ RELATIVEJUMP_SIZE
));
1421 /* Backup instructions which will be replaced by jump address */
1422 memcpy(op
->optinsn
.copied_insn
, op
->kp
.addr
+ INT3_SIZE
,
1423 RELATIVE_ADDR_SIZE
);
1425 insn_buf
[0] = RELATIVEJUMP_OPCODE
;
1426 *(s32
*)(&insn_buf
[1]) = rel
;
1428 tprm
->addr
= op
->kp
.addr
;
1429 tprm
->opcode
= insn_buf
;
1430 tprm
->len
= RELATIVEJUMP_SIZE
;
1434 * Replace breakpoints (int3) with relative jumps.
1435 * Caller must call with locking kprobe_mutex and text_mutex.
1437 void __kprobes
arch_optimize_kprobes(struct list_head
*oplist
)
1439 struct optimized_kprobe
*op
, *tmp
;
1442 list_for_each_entry_safe(op
, tmp
, oplist
, list
) {
1443 WARN_ON(kprobe_disabled(&op
->kp
));
1445 setup_optimize_kprobe(&jump_poke_params
[c
],
1446 jump_poke_bufs
[c
].buf
, op
);
1447 list_del_init(&op
->list
);
1448 if (++c
>= MAX_OPTIMIZE_PROBES
)
1453 * text_poke_smp doesn't support NMI/MCE code modifying.
1454 * However, since kprobes itself also doesn't support NMI/MCE
1455 * code probing, it's not a problem.
1457 text_poke_smp_batch(jump_poke_params
, c
);
1460 static void __kprobes
setup_unoptimize_kprobe(struct text_poke_param
*tprm
,
1462 struct optimized_kprobe
*op
)
1464 /* Set int3 to first byte for kprobes */
1465 insn_buf
[0] = BREAKPOINT_INSTRUCTION
;
1466 memcpy(insn_buf
+ 1, op
->optinsn
.copied_insn
, RELATIVE_ADDR_SIZE
);
1468 tprm
->addr
= op
->kp
.addr
;
1469 tprm
->opcode
= insn_buf
;
1470 tprm
->len
= RELATIVEJUMP_SIZE
;
1474 * Recover original instructions and breakpoints from relative jumps.
1475 * Caller must call with locking kprobe_mutex.
1477 extern void arch_unoptimize_kprobes(struct list_head
*oplist
,
1478 struct list_head
*done_list
)
1480 struct optimized_kprobe
*op
, *tmp
;
1483 list_for_each_entry_safe(op
, tmp
, oplist
, list
) {
1485 setup_unoptimize_kprobe(&jump_poke_params
[c
],
1486 jump_poke_bufs
[c
].buf
, op
);
1487 list_move(&op
->list
, done_list
);
1488 if (++c
>= MAX_OPTIMIZE_PROBES
)
1493 * text_poke_smp doesn't support NMI/MCE code modifying.
1494 * However, since kprobes itself also doesn't support NMI/MCE
1495 * code probing, it's not a problem.
1497 text_poke_smp_batch(jump_poke_params
, c
);
1500 /* Replace a relative jump with a breakpoint (int3). */
1501 void __kprobes
arch_unoptimize_kprobe(struct optimized_kprobe
*op
)
1503 u8 buf
[RELATIVEJUMP_SIZE
];
1505 /* Set int3 to first byte for kprobes */
1506 buf
[0] = BREAKPOINT_INSTRUCTION
;
1507 memcpy(buf
+ 1, op
->optinsn
.copied_insn
, RELATIVE_ADDR_SIZE
);
1508 text_poke_smp(op
->kp
.addr
, buf
, RELATIVEJUMP_SIZE
);
1511 static int __kprobes
setup_detour_execution(struct kprobe
*p
,
1512 struct pt_regs
*regs
,
1515 struct optimized_kprobe
*op
;
1517 if (p
->flags
& KPROBE_FLAG_OPTIMIZED
) {
1518 /* This kprobe is really able to run optimized path. */
1519 op
= container_of(p
, struct optimized_kprobe
, kp
);
1520 /* Detour through copied instructions */
1521 regs
->ip
= (unsigned long)op
->optinsn
.insn
+ TMPL_END_IDX
;
1523 reset_current_kprobe();
1524 preempt_enable_no_resched();
1530 static int __kprobes
init_poke_params(void)
1532 /* Allocate code buffer and parameter array */
1533 jump_poke_bufs
= kmalloc(sizeof(struct jump_poke_buffer
) *
1534 MAX_OPTIMIZE_PROBES
, GFP_KERNEL
);
1535 if (!jump_poke_bufs
)
1538 jump_poke_params
= kmalloc(sizeof(struct text_poke_param
) *
1539 MAX_OPTIMIZE_PROBES
, GFP_KERNEL
);
1540 if (!jump_poke_params
) {
1541 kfree(jump_poke_bufs
);
1542 jump_poke_bufs
= NULL
;
1548 #else /* !CONFIG_OPTPROBES */
1549 static int __kprobes
init_poke_params(void)
1555 int __init
arch_init_kprobes(void)
1557 return init_poke_params();
1560 int __kprobes
arch_trampoline_kprobe(struct kprobe
*p
)