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>
53 #include <asm/cacheflush.h>
55 #include <asm/pgtable.h>
56 #include <asm/uaccess.h>
57 #include <asm/alternative.h>
59 #include <asm/debugreg.h>
61 void jprobe_return_end(void);
63 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
64 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
67 #define stack_addr(regs) ((unsigned long *)regs->sp)
70 * "®s->sp" looks wrong, but it's correct for x86_32. x86_32 CPUs
71 * don't save the ss and esp registers if the CPU is already in kernel
72 * mode when it traps. So for kprobes, regs->sp and regs->ss are not
73 * the [nonexistent] saved stack pointer and ss register, but rather
74 * the top 8 bytes of the pre-int3 stack. So ®s->sp happens to
75 * point to the top of the pre-int3 stack.
77 #define stack_addr(regs) ((unsigned long *)®s->sp)
80 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
81 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
82 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
83 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
84 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
87 * Undefined/reserved opcodes, conditional jump, Opcode Extension
88 * Groups, and some special opcodes can not boost.
90 static const u32 twobyte_is_boostable
[256 / 32] = {
91 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
92 /* ---------------------------------------------- */
93 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */
94 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 10 */
95 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */
96 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */
97 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */
98 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
99 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */
100 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */
101 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */
102 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */
103 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */
104 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */
105 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */
106 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */
107 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */
108 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */
109 /* ----------------------------------------------- */
110 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
114 struct kretprobe_blackpoint kretprobe_blacklist
[] = {
115 {"__switch_to", }, /* This function switches only current task, but
116 doesn't switch kernel stack.*/
117 {NULL
, NULL
} /* Terminator */
119 const int kretprobe_blacklist_size
= ARRAY_SIZE(kretprobe_blacklist
);
121 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
122 static void __kprobes
set_jmp_op(void *from
, void *to
)
124 struct __arch_jmp_op
{
127 } __attribute__((packed
)) * jop
;
128 jop
= (struct __arch_jmp_op
*)from
;
129 jop
->raddr
= (s32
)((long)(to
) - ((long)(from
) + 5));
130 jop
->op
= RELATIVEJUMP_INSTRUCTION
;
134 * Check for the REX prefix which can only exist on X86_64
135 * X86_32 always returns 0
137 static int __kprobes
is_REX_prefix(kprobe_opcode_t
*insn
)
140 if ((*insn
& 0xf0) == 0x40)
147 * Returns non-zero if opcode is boostable.
148 * RIP relative instructions are adjusted at copying time in 64 bits mode
150 static int __kprobes
can_boost(kprobe_opcode_t
*opcodes
)
152 kprobe_opcode_t opcode
;
153 kprobe_opcode_t
*orig_opcodes
= opcodes
;
155 if (search_exception_tables((unsigned long)opcodes
))
156 return 0; /* Page fault may occur on this address. */
159 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
161 opcode
= *(opcodes
++);
163 /* 2nd-byte opcode */
164 if (opcode
== 0x0f) {
165 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
167 return test_bit(*opcodes
,
168 (unsigned long *)twobyte_is_boostable
);
171 switch (opcode
& 0xf0) {
174 goto retry
; /* REX prefix is boostable */
177 if (0x63 < opcode
&& opcode
< 0x67)
178 goto retry
; /* prefixes */
179 /* can't boost Address-size override and bound */
180 return (opcode
!= 0x62 && opcode
!= 0x67);
182 return 0; /* can't boost conditional jump */
184 /* can't boost software-interruptions */
185 return (0xc1 < opcode
&& opcode
< 0xcc) || opcode
== 0xcf;
187 /* can boost AA* and XLAT */
188 return (opcode
== 0xd4 || opcode
== 0xd5 || opcode
== 0xd7);
190 /* can boost in/out and absolute jmps */
191 return ((opcode
& 0x04) || opcode
== 0xea);
193 if ((opcode
& 0x0c) == 0 && opcode
!= 0xf1)
194 goto retry
; /* lock/rep(ne) prefix */
195 /* clear and set flags are boostable */
196 return (opcode
== 0xf5 || (0xf7 < opcode
&& opcode
< 0xfe));
198 /* segment override prefixes are boostable */
199 if (opcode
== 0x26 || opcode
== 0x36 || opcode
== 0x3e)
200 goto retry
; /* prefixes */
201 /* CS override prefix and call are not boostable */
202 return (opcode
!= 0x2e && opcode
!= 0x9a);
206 /* Recover the probed instruction at addr for further analysis. */
207 static int recover_probed_instruction(kprobe_opcode_t
*buf
, unsigned long addr
)
210 kp
= get_kprobe((void *)addr
);
215 * Basically, kp->ainsn.insn has an original instruction.
216 * However, RIP-relative instruction can not do single-stepping
217 * at different place, fix_riprel() tweaks the displacement of
218 * that instruction. In that case, we can't recover the instruction
219 * from the kp->ainsn.insn.
221 * On the other hand, kp->opcode has a copy of the first byte of
222 * the probed instruction, which is overwritten by int3. And
223 * the instruction at kp->addr is not modified by kprobes except
224 * for the first byte, we can recover the original instruction
225 * from it and kp->opcode.
227 memcpy(buf
, kp
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
232 /* Dummy buffers for kallsyms_lookup */
233 static char __dummy_buf
[KSYM_NAME_LEN
];
235 /* Check if paddr is at an instruction boundary */
236 static int __kprobes
can_probe(unsigned long paddr
)
239 unsigned long addr
, offset
= 0;
241 kprobe_opcode_t buf
[MAX_INSN_SIZE
];
243 if (!kallsyms_lookup(paddr
, NULL
, &offset
, NULL
, __dummy_buf
))
246 /* Decode instructions */
247 addr
= paddr
- offset
;
248 while (addr
< paddr
) {
249 kernel_insn_init(&insn
, (void *)addr
);
250 insn_get_opcode(&insn
);
253 * Check if the instruction has been modified by another
254 * kprobe, in which case we replace the breakpoint by the
255 * original instruction in our buffer.
257 if (insn
.opcode
.bytes
[0] == BREAKPOINT_INSTRUCTION
) {
258 ret
= recover_probed_instruction(buf
, addr
);
261 * Another debugging subsystem might insert
262 * this breakpoint. In that case, we can't
266 kernel_insn_init(&insn
, buf
);
268 insn_get_length(&insn
);
272 return (addr
== paddr
);
276 * Returns non-zero if opcode modifies the interrupt flag.
278 static int __kprobes
is_IF_modifier(kprobe_opcode_t
*insn
)
283 case 0xcf: /* iret/iretd */
284 case 0x9d: /* popf/popfd */
289 * on X86_64, 0x40-0x4f are REX prefixes so we need to look
290 * at the next byte instead.. but of course not recurse infinitely
292 if (is_REX_prefix(insn
))
293 return is_IF_modifier(++insn
);
299 * Adjust the displacement if the instruction uses the %rip-relative
301 * If it does, Return the address of the 32-bit displacement word.
302 * If not, return null.
303 * Only applicable to 64-bit x86.
305 static void __kprobes
fix_riprel(struct kprobe
*p
)
309 kernel_insn_init(&insn
, p
->ainsn
.insn
);
311 if (insn_rip_relative(&insn
)) {
314 insn_get_displacement(&insn
);
316 * The copied instruction uses the %rip-relative addressing
317 * mode. Adjust the displacement for the difference between
318 * the original location of this instruction and the location
319 * of the copy that will actually be run. The tricky bit here
320 * is making sure that the sign extension happens correctly in
321 * this calculation, since we need a signed 32-bit result to
322 * be sign-extended to 64 bits when it's added to the %rip
323 * value and yield the same 64-bit result that the sign-
324 * extension of the original signed 32-bit displacement would
327 newdisp
= (u8
*) p
->addr
+ (s64
) insn
.displacement
.value
-
328 (u8
*) p
->ainsn
.insn
;
329 BUG_ON((s64
) (s32
) newdisp
!= newdisp
); /* Sanity check. */
330 disp
= (u8
*) p
->ainsn
.insn
+ insn_offset_displacement(&insn
);
331 *(s32
*) disp
= (s32
) newdisp
;
336 static void __kprobes
arch_copy_kprobe(struct kprobe
*p
)
338 memcpy(p
->ainsn
.insn
, p
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
342 if (can_boost(p
->addr
))
343 p
->ainsn
.boostable
= 0;
345 p
->ainsn
.boostable
= -1;
347 p
->opcode
= *p
->addr
;
350 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
352 if (!can_probe((unsigned long)p
->addr
))
354 /* insn: must be on special executable page on x86. */
355 p
->ainsn
.insn
= get_insn_slot();
362 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
364 text_poke(p
->addr
, ((unsigned char []){BREAKPOINT_INSTRUCTION
}), 1);
367 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
369 text_poke(p
->addr
, &p
->opcode
, 1);
372 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
375 free_insn_slot(p
->ainsn
.insn
, (p
->ainsn
.boostable
== 1));
376 p
->ainsn
.insn
= NULL
;
380 static void __kprobes
save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
382 kcb
->prev_kprobe
.kp
= kprobe_running();
383 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
384 kcb
->prev_kprobe
.old_flags
= kcb
->kprobe_old_flags
;
385 kcb
->prev_kprobe
.saved_flags
= kcb
->kprobe_saved_flags
;
388 static void __kprobes
restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
390 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
391 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
392 kcb
->kprobe_old_flags
= kcb
->prev_kprobe
.old_flags
;
393 kcb
->kprobe_saved_flags
= kcb
->prev_kprobe
.saved_flags
;
396 static void __kprobes
set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
397 struct kprobe_ctlblk
*kcb
)
399 __get_cpu_var(current_kprobe
) = p
;
400 kcb
->kprobe_saved_flags
= kcb
->kprobe_old_flags
401 = (regs
->flags
& (X86_EFLAGS_TF
| X86_EFLAGS_IF
));
402 if (is_IF_modifier(p
->ainsn
.insn
))
403 kcb
->kprobe_saved_flags
&= ~X86_EFLAGS_IF
;
406 static void __kprobes
clear_btf(void)
408 if (test_thread_flag(TIF_DEBUGCTLMSR
))
409 update_debugctlmsr(0);
412 static void __kprobes
restore_btf(void)
414 if (test_thread_flag(TIF_DEBUGCTLMSR
))
415 update_debugctlmsr(current
->thread
.debugctlmsr
);
418 static void __kprobes
prepare_singlestep(struct kprobe
*p
, struct pt_regs
*regs
)
421 regs
->flags
|= X86_EFLAGS_TF
;
422 regs
->flags
&= ~X86_EFLAGS_IF
;
423 /* single step inline if the instruction is an int3 */
424 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
425 regs
->ip
= (unsigned long)p
->addr
;
427 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
430 void __kprobes
arch_prepare_kretprobe(struct kretprobe_instance
*ri
,
431 struct pt_regs
*regs
)
433 unsigned long *sara
= stack_addr(regs
);
435 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
437 /* Replace the return addr with trampoline addr */
438 *sara
= (unsigned long) &kretprobe_trampoline
;
441 static void __kprobes
setup_singlestep(struct kprobe
*p
, struct pt_regs
*regs
,
442 struct kprobe_ctlblk
*kcb
)
444 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_FREEZER)
445 if (p
->ainsn
.boostable
== 1 && !p
->post_handler
) {
446 /* Boost up -- we can execute copied instructions directly */
447 reset_current_kprobe();
448 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
449 preempt_enable_no_resched();
453 prepare_singlestep(p
, regs
);
454 kcb
->kprobe_status
= KPROBE_HIT_SS
;
458 * We have reentered the kprobe_handler(), since another probe was hit while
459 * within the handler. We save the original kprobes variables and just single
460 * step on the instruction of the new probe without calling any user handlers.
462 static int __kprobes
reenter_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
463 struct kprobe_ctlblk
*kcb
)
465 switch (kcb
->kprobe_status
) {
466 case KPROBE_HIT_SSDONE
:
467 case KPROBE_HIT_ACTIVE
:
468 save_previous_kprobe(kcb
);
469 set_current_kprobe(p
, regs
, kcb
);
470 kprobes_inc_nmissed_count(p
);
471 prepare_singlestep(p
, regs
);
472 kcb
->kprobe_status
= KPROBE_REENTER
;
475 /* A probe has been hit in the codepath leading up to, or just
476 * after, single-stepping of a probed instruction. This entire
477 * codepath should strictly reside in .kprobes.text section.
478 * Raise a BUG or we'll continue in an endless reentering loop
479 * and eventually a stack overflow.
481 printk(KERN_WARNING
"Unrecoverable kprobe detected at %p.\n",
486 /* impossible cases */
495 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
496 * remain disabled thorough out this function.
498 static int __kprobes
kprobe_handler(struct pt_regs
*regs
)
500 kprobe_opcode_t
*addr
;
502 struct kprobe_ctlblk
*kcb
;
504 addr
= (kprobe_opcode_t
*)(regs
->ip
- sizeof(kprobe_opcode_t
));
505 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
507 * The breakpoint instruction was removed right
508 * after we hit it. Another cpu has removed
509 * either a probepoint or a debugger breakpoint
510 * at this address. In either case, no further
511 * handling of this interrupt is appropriate.
512 * Back up over the (now missing) int3 and run
513 * the original instruction.
515 regs
->ip
= (unsigned long)addr
;
520 * We don't want to be preempted for the entire
521 * duration of kprobe processing. We conditionally
522 * re-enable preemption at the end of this function,
523 * and also in reenter_kprobe() and setup_singlestep().
527 kcb
= get_kprobe_ctlblk();
528 p
= get_kprobe(addr
);
531 if (kprobe_running()) {
532 if (reenter_kprobe(p
, regs
, kcb
))
535 set_current_kprobe(p
, regs
, kcb
);
536 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
539 * If we have no pre-handler or it returned 0, we
540 * continue with normal processing. If we have a
541 * pre-handler and it returned non-zero, it prepped
542 * for calling the break_handler below on re-entry
543 * for jprobe processing, so get out doing nothing
546 if (!p
->pre_handler
|| !p
->pre_handler(p
, regs
))
547 setup_singlestep(p
, regs
, kcb
);
550 } else if (kprobe_running()) {
551 p
= __get_cpu_var(current_kprobe
);
552 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
553 setup_singlestep(p
, regs
, kcb
);
556 } /* else: not a kprobe fault; let the kernel handle it */
558 preempt_enable_no_resched();
563 * When a retprobed function returns, this code saves registers and
564 * calls trampoline_handler() runs, which calls the kretprobe's handler.
566 static void __used __kprobes
kretprobe_trampoline_holder(void)
569 ".global kretprobe_trampoline\n"
570 "kretprobe_trampoline: \n"
572 /* We don't bother saving the ss register */
576 * Skip cs, ip, orig_ax.
577 * trampoline_handler() will plug in these values
596 " call trampoline_handler\n"
597 /* Replace saved sp with true return address. */
598 " movq %rax, 152(%rsp)\n"
614 /* Skip orig_ax, ip, cs */
620 * Skip cs, ip, orig_ax and gs.
621 * trampoline_handler() will plug in these values
635 " call trampoline_handler\n"
636 /* Move flags to cs */
637 " movl 56(%esp), %edx\n"
638 " movl %edx, 52(%esp)\n"
639 /* Replace saved flags with true return address. */
640 " movl %eax, 56(%esp)\n"
648 /* Skip ds, es, fs, gs, orig_ax and ip */
656 * Called from kretprobe_trampoline
658 static __used __kprobes
void *trampoline_handler(struct pt_regs
*regs
)
660 struct kretprobe_instance
*ri
= NULL
;
661 struct hlist_head
*head
, empty_rp
;
662 struct hlist_node
*node
, *tmp
;
663 unsigned long flags
, orig_ret_address
= 0;
664 unsigned long trampoline_address
= (unsigned long)&kretprobe_trampoline
;
666 INIT_HLIST_HEAD(&empty_rp
);
667 kretprobe_hash_lock(current
, &head
, &flags
);
668 /* fixup registers */
670 regs
->cs
= __KERNEL_CS
;
672 regs
->cs
= __KERNEL_CS
| get_kernel_rpl();
675 regs
->ip
= trampoline_address
;
676 regs
->orig_ax
= ~0UL;
679 * It is possible to have multiple instances associated with a given
680 * task either because multiple functions in the call path have
681 * return probes installed on them, and/or more than one
682 * return probe was registered for a target function.
684 * We can handle this because:
685 * - instances are always pushed into the head of the list
686 * - when multiple return probes are registered for the same
687 * function, the (chronologically) first instance's ret_addr
688 * will be the real return address, and all the rest will
689 * point to kretprobe_trampoline.
691 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
692 if (ri
->task
!= current
)
693 /* another task is sharing our hash bucket */
696 if (ri
->rp
&& ri
->rp
->handler
) {
697 __get_cpu_var(current_kprobe
) = &ri
->rp
->kp
;
698 get_kprobe_ctlblk()->kprobe_status
= KPROBE_HIT_ACTIVE
;
699 ri
->rp
->handler(ri
, regs
);
700 __get_cpu_var(current_kprobe
) = NULL
;
703 orig_ret_address
= (unsigned long)ri
->ret_addr
;
704 recycle_rp_inst(ri
, &empty_rp
);
706 if (orig_ret_address
!= trampoline_address
)
708 * This is the real return address. Any other
709 * instances associated with this task are for
710 * other calls deeper on the call stack
715 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
717 kretprobe_hash_unlock(current
, &flags
);
719 hlist_for_each_entry_safe(ri
, node
, tmp
, &empty_rp
, hlist
) {
720 hlist_del(&ri
->hlist
);
723 return (void *)orig_ret_address
;
727 * Called after single-stepping. p->addr is the address of the
728 * instruction whose first byte has been replaced by the "int 3"
729 * instruction. To avoid the SMP problems that can occur when we
730 * temporarily put back the original opcode to single-step, we
731 * single-stepped a copy of the instruction. The address of this
732 * copy is p->ainsn.insn.
734 * This function prepares to return from the post-single-step
735 * interrupt. We have to fix up the stack as follows:
737 * 0) Except in the case of absolute or indirect jump or call instructions,
738 * the new ip is relative to the copied instruction. We need to make
739 * it relative to the original instruction.
741 * 1) If the single-stepped instruction was pushfl, then the TF and IF
742 * flags are set in the just-pushed flags, and may need to be cleared.
744 * 2) If the single-stepped instruction was a call, the return address
745 * that is atop the stack is the address following the copied instruction.
746 * We need to make it the address following the original instruction.
748 * If this is the first time we've single-stepped the instruction at
749 * this probepoint, and the instruction is boostable, boost it: add a
750 * jump instruction after the copied instruction, that jumps to the next
751 * instruction after the probepoint.
753 static void __kprobes
resume_execution(struct kprobe
*p
,
754 struct pt_regs
*regs
, struct kprobe_ctlblk
*kcb
)
756 unsigned long *tos
= stack_addr(regs
);
757 unsigned long copy_ip
= (unsigned long)p
->ainsn
.insn
;
758 unsigned long orig_ip
= (unsigned long)p
->addr
;
759 kprobe_opcode_t
*insn
= p
->ainsn
.insn
;
761 /*skip the REX prefix*/
762 if (is_REX_prefix(insn
))
765 regs
->flags
&= ~X86_EFLAGS_TF
;
767 case 0x9c: /* pushfl */
768 *tos
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_IF
);
769 *tos
|= kcb
->kprobe_old_flags
;
771 case 0xc2: /* iret/ret/lret */
776 case 0xea: /* jmp absolute -- ip is correct */
777 /* ip is already adjusted, no more changes required */
778 p
->ainsn
.boostable
= 1;
780 case 0xe8: /* call relative - Fix return addr */
781 *tos
= orig_ip
+ (*tos
- copy_ip
);
784 case 0x9a: /* call absolute -- same as call absolute, indirect */
785 *tos
= orig_ip
+ (*tos
- copy_ip
);
789 if ((insn
[1] & 0x30) == 0x10) {
791 * call absolute, indirect
792 * Fix return addr; ip is correct.
793 * But this is not boostable
795 *tos
= orig_ip
+ (*tos
- copy_ip
);
797 } else if (((insn
[1] & 0x31) == 0x20) ||
798 ((insn
[1] & 0x31) == 0x21)) {
800 * jmp near and far, absolute indirect
801 * ip is correct. And this is boostable
803 p
->ainsn
.boostable
= 1;
810 if (p
->ainsn
.boostable
== 0) {
811 if ((regs
->ip
> copy_ip
) &&
812 (regs
->ip
- copy_ip
) + 5 < MAX_INSN_SIZE
) {
814 * These instructions can be executed directly if it
815 * jumps back to correct address.
817 set_jmp_op((void *)regs
->ip
,
818 (void *)orig_ip
+ (regs
->ip
- copy_ip
));
819 p
->ainsn
.boostable
= 1;
821 p
->ainsn
.boostable
= -1;
825 regs
->ip
+= orig_ip
- copy_ip
;
832 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
833 * remain disabled thoroughout this function.
835 static int __kprobes
post_kprobe_handler(struct pt_regs
*regs
)
837 struct kprobe
*cur
= kprobe_running();
838 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
843 resume_execution(cur
, regs
, kcb
);
844 regs
->flags
|= kcb
->kprobe_saved_flags
;
846 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
847 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
848 cur
->post_handler(cur
, regs
, 0);
851 /* Restore back the original saved kprobes variables and continue. */
852 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
853 restore_previous_kprobe(kcb
);
856 reset_current_kprobe();
858 preempt_enable_no_resched();
861 * if somebody else is singlestepping across a probe point, flags
862 * will have TF set, in which case, continue the remaining processing
863 * of do_debug, as if this is not a probe hit.
865 if (regs
->flags
& X86_EFLAGS_TF
)
871 int __kprobes
kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
873 struct kprobe
*cur
= kprobe_running();
874 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
876 switch (kcb
->kprobe_status
) {
880 * We are here because the instruction being single
881 * stepped caused a page fault. We reset the current
882 * kprobe and the ip points back to the probe address
883 * and allow the page fault handler to continue as a
886 regs
->ip
= (unsigned long)cur
->addr
;
887 regs
->flags
|= kcb
->kprobe_old_flags
;
888 if (kcb
->kprobe_status
== KPROBE_REENTER
)
889 restore_previous_kprobe(kcb
);
891 reset_current_kprobe();
892 preempt_enable_no_resched();
894 case KPROBE_HIT_ACTIVE
:
895 case KPROBE_HIT_SSDONE
:
897 * We increment the nmissed count for accounting,
898 * we can also use npre/npostfault count for accounting
899 * these specific fault cases.
901 kprobes_inc_nmissed_count(cur
);
904 * We come here because instructions in the pre/post
905 * handler caused the page_fault, this could happen
906 * if handler tries to access user space by
907 * copy_from_user(), get_user() etc. Let the
908 * user-specified handler try to fix it first.
910 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
914 * In case the user-specified fault handler returned
915 * zero, try to fix up.
917 if (fixup_exception(regs
))
921 * fixup routine could not handle it,
922 * Let do_page_fault() fix it.
932 * Wrapper routine for handling exceptions.
934 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
935 unsigned long val
, void *data
)
937 struct die_args
*args
= data
;
938 int ret
= NOTIFY_DONE
;
940 if (args
->regs
&& user_mode_vm(args
->regs
))
945 if (kprobe_handler(args
->regs
))
949 if (post_kprobe_handler(args
->regs
)) {
951 * Reset the BS bit in dr6 (pointed by args->err) to
952 * denote completion of processing
954 (*(unsigned long *)ERR_PTR(args
->err
)) &= ~DR_STEP
;
960 * To be potentially processing a kprobe fault and to
961 * trust the result from kprobe_running(), we have
962 * be non-preemptible.
964 if (!preemptible() && kprobe_running() &&
965 kprobe_fault_handler(args
->regs
, args
->trapnr
))
974 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
976 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
978 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
980 kcb
->jprobe_saved_regs
= *regs
;
981 kcb
->jprobe_saved_sp
= stack_addr(regs
);
982 addr
= (unsigned long)(kcb
->jprobe_saved_sp
);
985 * As Linus pointed out, gcc assumes that the callee
986 * owns the argument space and could overwrite it, e.g.
987 * tailcall optimization. So, to be absolutely safe
988 * we also save and restore enough stack bytes to cover
991 memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
,
992 MIN_STACK_SIZE(addr
));
993 regs
->flags
&= ~X86_EFLAGS_IF
;
994 trace_hardirqs_off();
995 regs
->ip
= (unsigned long)(jp
->entry
);
999 void __kprobes
jprobe_return(void)
1001 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1004 #ifdef CONFIG_X86_64
1005 " xchg %%rbx,%%rsp \n"
1007 " xchgl %%ebx,%%esp \n"
1010 " .globl jprobe_return_end\n"
1011 " jprobe_return_end: \n"
1013 (kcb
->jprobe_saved_sp
):"memory");
1016 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
1018 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
1019 u8
*addr
= (u8
*) (regs
->ip
- 1);
1020 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1022 if ((addr
> (u8
*) jprobe_return
) &&
1023 (addr
< (u8
*) jprobe_return_end
)) {
1024 if (stack_addr(regs
) != kcb
->jprobe_saved_sp
) {
1025 struct pt_regs
*saved_regs
= &kcb
->jprobe_saved_regs
;
1027 "current sp %p does not match saved sp %p\n",
1028 stack_addr(regs
), kcb
->jprobe_saved_sp
);
1029 printk(KERN_ERR
"Saved registers for jprobe %p\n", jp
);
1030 show_registers(saved_regs
);
1031 printk(KERN_ERR
"Current registers\n");
1032 show_registers(regs
);
1035 *regs
= kcb
->jprobe_saved_regs
;
1036 memcpy((kprobe_opcode_t
*)(kcb
->jprobe_saved_sp
),
1038 MIN_STACK_SIZE(kcb
->jprobe_saved_sp
));
1039 preempt_enable_no_resched();
1045 int __init
arch_init_kprobes(void)
1050 int __kprobes
arch_trampoline_kprobe(struct kprobe
*p
)