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/preempt.h>
48 #include <linux/module.h>
49 #include <linux/kdebug.h>
51 #include <asm/cacheflush.h>
53 #include <asm/pgtable.h>
54 #include <asm/uaccess.h>
55 #include <asm/alternative.h>
57 void jprobe_return_end(void);
59 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
60 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
63 #define stack_addr(regs) ((unsigned long *)regs->sp)
66 * "®s->sp" looks wrong, but it's correct for x86_32. x86_32 CPUs
67 * don't save the ss and esp registers if the CPU is already in kernel
68 * mode when it traps. So for kprobes, regs->sp and regs->ss are not
69 * the [nonexistent] saved stack pointer and ss register, but rather
70 * the top 8 bytes of the pre-int3 stack. So ®s->sp happens to
71 * point to the top of the pre-int3 stack.
73 #define stack_addr(regs) ((unsigned long *)®s->sp)
76 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\
77 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \
78 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \
79 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \
80 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \
83 * Undefined/reserved opcodes, conditional jump, Opcode Extension
84 * Groups, and some special opcodes can not boost.
86 static const 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, 0) , /* 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 */
108 static const u32 onebyte_has_modrm
[256 / 32] = {
109 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
110 /* ----------------------------------------------- */
111 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 00 */
112 W(0x10, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 10 */
113 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) | /* 20 */
114 W(0x30, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0) , /* 30 */
115 W(0x40, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 40 */
116 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */
117 W(0x60, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0, 0) | /* 60 */
118 W(0x70, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 70 */
119 W(0x80, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 80 */
120 W(0x90, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 90 */
121 W(0xa0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* a0 */
122 W(0xb0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* b0 */
123 W(0xc0, 1, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* c0 */
124 W(0xd0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) , /* d0 */
125 W(0xe0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* e0 */
126 W(0xf0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) /* f0 */
127 /* ----------------------------------------------- */
128 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
130 static const u32 twobyte_has_modrm
[256 / 32] = {
131 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
132 /* ----------------------------------------------- */
133 W(0x00, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1) | /* 0f */
134 W(0x10, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0) , /* 1f */
135 W(0x20, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* 2f */
136 W(0x30, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 3f */
137 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 4f */
138 W(0x50, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 5f */
139 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 6f */
140 W(0x70, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1) , /* 7f */
141 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 8f */
142 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 9f */
143 W(0xa0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) | /* af */
144 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 1, 1, 1) , /* bf */
145 W(0xc0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0) | /* cf */
146 W(0xd0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* df */
147 W(0xe0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* ef */
148 W(0xf0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0) /* ff */
149 /* ----------------------------------------------- */
150 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
154 struct kretprobe_blackpoint kretprobe_blacklist
[] = {
155 {"__switch_to", }, /* This function switches only current task, but
156 doesn't switch kernel stack.*/
157 {NULL
, NULL
} /* Terminator */
159 const int kretprobe_blacklist_size
= ARRAY_SIZE(kretprobe_blacklist
);
161 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
162 static void __kprobes
set_jmp_op(void *from
, void *to
)
164 struct __arch_jmp_op
{
167 } __attribute__((packed
)) * jop
;
168 jop
= (struct __arch_jmp_op
*)from
;
169 jop
->raddr
= (s32
)((long)(to
) - ((long)(from
) + 5));
170 jop
->op
= RELATIVEJUMP_INSTRUCTION
;
174 * Returns non-zero if opcode is boostable.
175 * RIP relative instructions are adjusted at copying time in 64 bits mode
177 static int __kprobes
can_boost(kprobe_opcode_t
*opcodes
)
179 kprobe_opcode_t opcode
;
180 kprobe_opcode_t
*orig_opcodes
= opcodes
;
183 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
185 opcode
= *(opcodes
++);
187 /* 2nd-byte opcode */
188 if (opcode
== 0x0f) {
189 if (opcodes
- orig_opcodes
> MAX_INSN_SIZE
- 1)
191 return test_bit(*opcodes
,
192 (unsigned long *)twobyte_is_boostable
);
195 switch (opcode
& 0xf0) {
198 goto retry
; /* REX prefix is boostable */
201 if (0x63 < opcode
&& opcode
< 0x67)
202 goto retry
; /* prefixes */
203 /* can't boost Address-size override and bound */
204 return (opcode
!= 0x62 && opcode
!= 0x67);
206 return 0; /* can't boost conditional jump */
208 /* can't boost software-interruptions */
209 return (0xc1 < opcode
&& opcode
< 0xcc) || opcode
== 0xcf;
211 /* can boost AA* and XLAT */
212 return (opcode
== 0xd4 || opcode
== 0xd5 || opcode
== 0xd7);
214 /* can boost in/out and absolute jmps */
215 return ((opcode
& 0x04) || opcode
== 0xea);
217 if ((opcode
& 0x0c) == 0 && opcode
!= 0xf1)
218 goto retry
; /* lock/rep(ne) prefix */
219 /* clear and set flags are boostable */
220 return (opcode
== 0xf5 || (0xf7 < opcode
&& opcode
< 0xfe));
222 /* segment override prefixes are boostable */
223 if (opcode
== 0x26 || opcode
== 0x36 || opcode
== 0x3e)
224 goto retry
; /* prefixes */
225 /* CS override prefix and call are not boostable */
226 return (opcode
!= 0x2e && opcode
!= 0x9a);
231 * Returns non-zero if opcode modifies the interrupt flag.
233 static int __kprobes
is_IF_modifier(kprobe_opcode_t
*insn
)
238 case 0xcf: /* iret/iretd */
239 case 0x9d: /* popf/popfd */
244 * on 64 bit x86, 0x40-0x4f are prefixes so we need to look
245 * at the next byte instead.. but of course not recurse infinitely
247 if (*insn
>= 0x40 && *insn
<= 0x4f)
248 return is_IF_modifier(++insn
);
255 * Adjust the displacement if the instruction uses the %rip-relative
257 * If it does, Return the address of the 32-bit displacement word.
258 * If not, return null.
260 static void __kprobes
fix_riprel(struct kprobe
*p
)
262 u8
*insn
= p
->ainsn
.insn
;
266 /* Skip legacy instruction prefixes. */
286 /* Skip REX instruction prefix. */
287 if ((*insn
& 0xf0) == 0x40)
291 /* Two-byte opcode. */
293 need_modrm
= test_bit(*insn
,
294 (unsigned long *)twobyte_has_modrm
);
296 /* One-byte opcode. */
297 need_modrm
= test_bit(*insn
,
298 (unsigned long *)onebyte_has_modrm
);
302 if ((modrm
& 0xc7) == 0x05) {
303 /* %rip+disp32 addressing mode */
304 /* Displacement follows ModRM byte. */
307 * The copied instruction uses the %rip-relative
308 * addressing mode. Adjust the displacement for the
309 * difference between the original location of this
310 * instruction and the location of the copy that will
311 * actually be run. The tricky bit here is making sure
312 * that the sign extension happens correctly in this
313 * calculation, since we need a signed 32-bit result to
314 * be sign-extended to 64 bits when it's added to the
315 * %rip value and yield the same 64-bit result that the
316 * sign-extension of the original signed 32-bit
317 * displacement would have given.
319 disp
= (u8
*) p
->addr
+ *((s32
*) insn
) -
320 (u8
*) p
->ainsn
.insn
;
321 BUG_ON((s64
) (s32
) disp
!= disp
); /* Sanity check. */
322 *(s32
*)insn
= (s32
) disp
;
328 static void __kprobes
arch_copy_kprobe(struct kprobe
*p
)
330 memcpy(p
->ainsn
.insn
, p
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
334 if (can_boost(p
->addr
))
335 p
->ainsn
.boostable
= 0;
337 p
->ainsn
.boostable
= -1;
339 p
->opcode
= *p
->addr
;
342 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
344 /* insn: must be on special executable page on x86. */
345 p
->ainsn
.insn
= get_insn_slot();
352 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
354 text_poke(p
->addr
, ((unsigned char []){BREAKPOINT_INSTRUCTION
}), 1);
357 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
359 text_poke(p
->addr
, &p
->opcode
, 1);
362 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
364 mutex_lock(&kprobe_mutex
);
365 free_insn_slot(p
->ainsn
.insn
, (p
->ainsn
.boostable
== 1));
366 mutex_unlock(&kprobe_mutex
);
369 static void __kprobes
save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
371 kcb
->prev_kprobe
.kp
= kprobe_running();
372 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
373 kcb
->prev_kprobe
.old_flags
= kcb
->kprobe_old_flags
;
374 kcb
->prev_kprobe
.saved_flags
= kcb
->kprobe_saved_flags
;
377 static void __kprobes
restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
379 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
380 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
381 kcb
->kprobe_old_flags
= kcb
->prev_kprobe
.old_flags
;
382 kcb
->kprobe_saved_flags
= kcb
->prev_kprobe
.saved_flags
;
385 static void __kprobes
set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
386 struct kprobe_ctlblk
*kcb
)
388 __get_cpu_var(current_kprobe
) = p
;
389 kcb
->kprobe_saved_flags
= kcb
->kprobe_old_flags
390 = (regs
->flags
& (X86_EFLAGS_TF
| X86_EFLAGS_IF
));
391 if (is_IF_modifier(p
->ainsn
.insn
))
392 kcb
->kprobe_saved_flags
&= ~X86_EFLAGS_IF
;
395 static void __kprobes
clear_btf(void)
397 if (test_thread_flag(TIF_DEBUGCTLMSR
))
398 wrmsr(MSR_IA32_DEBUGCTLMSR
, 0, 0);
401 static void __kprobes
restore_btf(void)
403 if (test_thread_flag(TIF_DEBUGCTLMSR
))
404 wrmsr(MSR_IA32_DEBUGCTLMSR
, current
->thread
.debugctlmsr
, 0);
407 static void __kprobes
prepare_singlestep(struct kprobe
*p
, struct pt_regs
*regs
)
410 regs
->flags
|= X86_EFLAGS_TF
;
411 regs
->flags
&= ~X86_EFLAGS_IF
;
412 /* single step inline if the instruction is an int3 */
413 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
414 regs
->ip
= (unsigned long)p
->addr
;
416 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
419 /* Called with kretprobe_lock held */
420 void __kprobes
arch_prepare_kretprobe(struct kretprobe_instance
*ri
,
421 struct pt_regs
*regs
)
423 unsigned long *sara
= stack_addr(regs
);
425 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
427 /* Replace the return addr with trampoline addr */
428 *sara
= (unsigned long) &kretprobe_trampoline
;
432 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
433 * remain disabled thorough out this function.
435 static int __kprobes
kprobe_handler(struct pt_regs
*regs
)
439 kprobe_opcode_t
*addr
;
440 struct kprobe_ctlblk
*kcb
;
442 addr
= (kprobe_opcode_t
*)(regs
->ip
- sizeof(kprobe_opcode_t
));
445 * We don't want to be preempted for the entire
446 * duration of kprobe processing
449 kcb
= get_kprobe_ctlblk();
451 /* Check we're not actually recursing */
452 if (kprobe_running()) {
453 p
= get_kprobe(addr
);
455 if (kcb
->kprobe_status
== KPROBE_HIT_SS
&&
456 *p
->ainsn
.insn
== BREAKPOINT_INSTRUCTION
) {
457 regs
->flags
&= ~X86_EFLAGS_TF
;
458 regs
->flags
|= kcb
->kprobe_saved_flags
;
461 } else if (kcb
->kprobe_status
== KPROBE_HIT_SSDONE
) {
462 /* TODO: Provide re-entrancy from
463 * post_kprobes_handler() and avoid exception
464 * stack corruption while single-stepping on
465 * the instruction of the new probe.
467 arch_disarm_kprobe(p
);
468 regs
->ip
= (unsigned long)p
->addr
;
469 reset_current_kprobe();
473 /* We have reentered the kprobe_handler(), since
474 * another probe was hit while within the handler.
475 * We here save the original kprobes variables and
476 * just single step on the instruction of the new probe
477 * without calling any user handlers.
479 save_previous_kprobe(kcb
);
480 set_current_kprobe(p
, regs
, kcb
);
481 kprobes_inc_nmissed_count(p
);
482 prepare_singlestep(p
, regs
);
483 kcb
->kprobe_status
= KPROBE_REENTER
;
486 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
487 /* The breakpoint instruction was removed by
488 * another cpu right after we hit, no further
489 * handling of this interrupt is appropriate
491 regs
->ip
= (unsigned long)addr
;
495 p
= __get_cpu_var(current_kprobe
);
496 if (p
->break_handler
&& p
->break_handler(p
, regs
))
502 p
= get_kprobe(addr
);
504 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
506 * The breakpoint instruction was removed right
507 * after we hit it. Another cpu has removed
508 * either a probepoint or a debugger breakpoint
509 * at this address. In either case, no further
510 * handling of this interrupt is appropriate.
511 * Back up over the (now missing) int3 and run
512 * the original instruction.
514 regs
->ip
= (unsigned long)addr
;
517 /* Not one of ours: let kernel handle it */
521 set_current_kprobe(p
, regs
, kcb
);
522 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
524 if (p
->pre_handler
&& p
->pre_handler(p
, regs
))
525 /* handler has already set things up, so skip ss setup */
529 #if !defined(CONFIG_PREEMPT) || defined(CONFIG_PM)
530 if (p
->ainsn
.boostable
== 1 && !p
->post_handler
) {
531 /* Boost up -- we can execute copied instructions directly */
532 reset_current_kprobe();
533 regs
->ip
= (unsigned long)p
->ainsn
.insn
;
534 preempt_enable_no_resched();
538 prepare_singlestep(p
, regs
);
539 kcb
->kprobe_status
= KPROBE_HIT_SS
;
543 preempt_enable_no_resched();
548 * When a retprobed function returns, this code saves registers and
549 * calls trampoline_handler() runs, which calls the kretprobe's handler.
551 void __kprobes
kretprobe_trampoline_holder(void)
554 ".global kretprobe_trampoline\n"
555 "kretprobe_trampoline: \n"
557 /* We don't bother saving the ss register */
561 * Skip cs, ip, orig_ax.
562 * trampoline_handler() will plug in these values
581 " call trampoline_handler\n"
582 /* Replace saved sp with true return address. */
583 " movq %rax, 152(%rsp)\n"
599 /* Skip orig_ax, ip, cs */
605 * Skip cs, ip, orig_ax.
606 * trampoline_handler() will plug in these values
620 " call trampoline_handler\n"
621 /* Move flags to cs */
622 " movl 52(%esp), %edx\n"
623 " movl %edx, 48(%esp)\n"
624 /* Replace saved flags with true return address. */
625 " movl %eax, 52(%esp)\n"
633 /* Skip ip, orig_ax, es, ds, fs */
641 * Called from kretprobe_trampoline
643 void * __kprobes
trampoline_handler(struct pt_regs
*regs
)
645 struct kretprobe_instance
*ri
= NULL
;
646 struct hlist_head
*head
, empty_rp
;
647 struct hlist_node
*node
, *tmp
;
648 unsigned long flags
, orig_ret_address
= 0;
649 unsigned long trampoline_address
= (unsigned long)&kretprobe_trampoline
;
651 INIT_HLIST_HEAD(&empty_rp
);
652 spin_lock_irqsave(&kretprobe_lock
, flags
);
653 head
= kretprobe_inst_table_head(current
);
654 /* fixup registers */
656 regs
->cs
= __KERNEL_CS
;
658 regs
->cs
= __KERNEL_CS
| get_kernel_rpl();
660 regs
->ip
= trampoline_address
;
661 regs
->orig_ax
= ~0UL;
664 * It is possible to have multiple instances associated with a given
665 * task either because multiple functions in the call path have
666 * return probes installed on them, and/or more then one
667 * return probe was registered for a target function.
669 * We can handle this because:
670 * - instances are always pushed into the head of the list
671 * - when multiple return probes are registered for the same
672 * function, the (chronologically) first instance's ret_addr
673 * will be the real return address, and all the rest will
674 * point to kretprobe_trampoline.
676 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
677 if (ri
->task
!= current
)
678 /* another task is sharing our hash bucket */
681 if (ri
->rp
&& ri
->rp
->handler
) {
682 __get_cpu_var(current_kprobe
) = &ri
->rp
->kp
;
683 get_kprobe_ctlblk()->kprobe_status
= KPROBE_HIT_ACTIVE
;
684 ri
->rp
->handler(ri
, regs
);
685 __get_cpu_var(current_kprobe
) = NULL
;
688 orig_ret_address
= (unsigned long)ri
->ret_addr
;
689 recycle_rp_inst(ri
, &empty_rp
);
691 if (orig_ret_address
!= trampoline_address
)
693 * This is the real return address. Any other
694 * instances associated with this task are for
695 * other calls deeper on the call stack
700 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
702 spin_unlock_irqrestore(&kretprobe_lock
, flags
);
704 hlist_for_each_entry_safe(ri
, node
, tmp
, &empty_rp
, hlist
) {
705 hlist_del(&ri
->hlist
);
708 return (void *)orig_ret_address
;
712 * Called after single-stepping. p->addr is the address of the
713 * instruction whose first byte has been replaced by the "int 3"
714 * instruction. To avoid the SMP problems that can occur when we
715 * temporarily put back the original opcode to single-step, we
716 * single-stepped a copy of the instruction. The address of this
717 * copy is p->ainsn.insn.
719 * This function prepares to return from the post-single-step
720 * interrupt. We have to fix up the stack as follows:
722 * 0) Except in the case of absolute or indirect jump or call instructions,
723 * the new ip is relative to the copied instruction. We need to make
724 * it relative to the original instruction.
726 * 1) If the single-stepped instruction was pushfl, then the TF and IF
727 * flags are set in the just-pushed flags, and may need to be cleared.
729 * 2) If the single-stepped instruction was a call, the return address
730 * that is atop the stack is the address following the copied instruction.
731 * We need to make it the address following the original instruction.
733 * If this is the first time we've single-stepped the instruction at
734 * this probepoint, and the instruction is boostable, boost it: add a
735 * jump instruction after the copied instruction, that jumps to the next
736 * instruction after the probepoint.
738 static void __kprobes
resume_execution(struct kprobe
*p
,
739 struct pt_regs
*regs
, struct kprobe_ctlblk
*kcb
)
741 unsigned long *tos
= stack_addr(regs
);
742 unsigned long copy_ip
= (unsigned long)p
->ainsn
.insn
;
743 unsigned long orig_ip
= (unsigned long)p
->addr
;
744 kprobe_opcode_t
*insn
= p
->ainsn
.insn
;
747 /*skip the REX prefix*/
748 if (*insn
>= 0x40 && *insn
<= 0x4f)
752 regs
->flags
&= ~X86_EFLAGS_TF
;
754 case 0x9c: /* pushfl */
755 *tos
&= ~(X86_EFLAGS_TF
| X86_EFLAGS_IF
);
756 *tos
|= kcb
->kprobe_old_flags
;
758 case 0xc2: /* iret/ret/lret */
763 case 0xea: /* jmp absolute -- ip is correct */
764 /* ip is already adjusted, no more changes required */
765 p
->ainsn
.boostable
= 1;
767 case 0xe8: /* call relative - Fix return addr */
768 *tos
= orig_ip
+ (*tos
- copy_ip
);
771 case 0x9a: /* call absolute -- same as call absolute, indirect */
772 *tos
= orig_ip
+ (*tos
- copy_ip
);
776 if ((insn
[1] & 0x30) == 0x10) {
778 * call absolute, indirect
779 * Fix return addr; ip is correct.
780 * But this is not boostable
782 *tos
= orig_ip
+ (*tos
- copy_ip
);
784 } else if (((insn
[1] & 0x31) == 0x20) ||
785 ((insn
[1] & 0x31) == 0x21)) {
787 * jmp near and far, absolute indirect
788 * ip is correct. And this is boostable
790 p
->ainsn
.boostable
= 1;
797 if (p
->ainsn
.boostable
== 0) {
798 if ((regs
->ip
> copy_ip
) &&
799 (regs
->ip
- copy_ip
) + 5 < MAX_INSN_SIZE
) {
801 * These instructions can be executed directly if it
802 * jumps back to correct address.
804 set_jmp_op((void *)regs
->ip
,
805 (void *)orig_ip
+ (regs
->ip
- copy_ip
));
806 p
->ainsn
.boostable
= 1;
808 p
->ainsn
.boostable
= -1;
812 regs
->ip
+= orig_ip
- copy_ip
;
819 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
820 * remain disabled thoroughout this function.
822 static int __kprobes
post_kprobe_handler(struct pt_regs
*regs
)
824 struct kprobe
*cur
= kprobe_running();
825 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
830 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
831 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
832 cur
->post_handler(cur
, regs
, 0);
835 resume_execution(cur
, regs
, kcb
);
836 regs
->flags
|= kcb
->kprobe_saved_flags
;
837 trace_hardirqs_fixup_flags(regs
->flags
);
839 /* Restore back the original saved kprobes variables and continue. */
840 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
841 restore_previous_kprobe(kcb
);
844 reset_current_kprobe();
846 preempt_enable_no_resched();
849 * if somebody else is singlestepping across a probe point, flags
850 * will have TF set, in which case, continue the remaining processing
851 * of do_debug, as if this is not a probe hit.
853 if (regs
->flags
& X86_EFLAGS_TF
)
859 int __kprobes
kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
861 struct kprobe
*cur
= kprobe_running();
862 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
864 switch (kcb
->kprobe_status
) {
868 * We are here because the instruction being single
869 * stepped caused a page fault. We reset the current
870 * kprobe and the ip points back to the probe address
871 * and allow the page fault handler to continue as a
874 regs
->ip
= (unsigned long)cur
->addr
;
875 regs
->flags
|= kcb
->kprobe_old_flags
;
876 if (kcb
->kprobe_status
== KPROBE_REENTER
)
877 restore_previous_kprobe(kcb
);
879 reset_current_kprobe();
880 preempt_enable_no_resched();
882 case KPROBE_HIT_ACTIVE
:
883 case KPROBE_HIT_SSDONE
:
885 * We increment the nmissed count for accounting,
886 * we can also use npre/npostfault count for accounting
887 * these specific fault cases.
889 kprobes_inc_nmissed_count(cur
);
892 * We come here because instructions in the pre/post
893 * handler caused the page_fault, this could happen
894 * if handler tries to access user space by
895 * copy_from_user(), get_user() etc. Let the
896 * user-specified handler try to fix it first.
898 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
902 * In case the user-specified fault handler returned
903 * zero, try to fix up.
905 if (fixup_exception(regs
))
909 * fixup routine could not handle it,
910 * Let do_page_fault() fix it.
920 * Wrapper routine for handling exceptions.
922 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
923 unsigned long val
, void *data
)
925 struct die_args
*args
= (struct die_args
*)data
;
926 int ret
= NOTIFY_DONE
;
928 if (args
->regs
&& user_mode_vm(args
->regs
))
933 if (kprobe_handler(args
->regs
))
937 if (post_kprobe_handler(args
->regs
))
941 /* kprobe_running() needs smp_processor_id() */
943 if (kprobe_running() &&
944 kprobe_fault_handler(args
->regs
, args
->trapnr
))
954 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
956 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
958 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
960 kcb
->jprobe_saved_regs
= *regs
;
961 kcb
->jprobe_saved_sp
= stack_addr(regs
);
962 addr
= (unsigned long)(kcb
->jprobe_saved_sp
);
965 * As Linus pointed out, gcc assumes that the callee
966 * owns the argument space and could overwrite it, e.g.
967 * tailcall optimization. So, to be absolutely safe
968 * we also save and restore enough stack bytes to cover
971 memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
,
972 MIN_STACK_SIZE(addr
));
973 regs
->flags
&= ~X86_EFLAGS_IF
;
974 trace_hardirqs_off();
975 regs
->ip
= (unsigned long)(jp
->entry
);
979 void __kprobes
jprobe_return(void)
981 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
985 " xchg %%rbx,%%rsp \n"
987 " xchgl %%ebx,%%esp \n"
990 " .globl jprobe_return_end\n"
991 " jprobe_return_end: \n"
993 (kcb
->jprobe_saved_sp
):"memory");
996 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
998 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
999 u8
*addr
= (u8
*) (regs
->ip
- 1);
1000 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
1002 if ((addr
> (u8
*) jprobe_return
) &&
1003 (addr
< (u8
*) jprobe_return_end
)) {
1004 if (stack_addr(regs
) != kcb
->jprobe_saved_sp
) {
1005 struct pt_regs
*saved_regs
= &kcb
->jprobe_saved_regs
;
1007 "current sp %p does not match saved sp %p\n",
1008 stack_addr(regs
), kcb
->jprobe_saved_sp
);
1009 printk(KERN_ERR
"Saved registers for jprobe %p\n", jp
);
1010 show_registers(saved_regs
);
1011 printk(KERN_ERR
"Current registers\n");
1012 show_registers(regs
);
1015 *regs
= kcb
->jprobe_saved_regs
;
1016 memcpy((kprobe_opcode_t
*)(kcb
->jprobe_saved_sp
),
1018 MIN_STACK_SIZE(kcb
->jprobe_saved_sp
));
1019 preempt_enable_no_resched();
1025 int __init
arch_init_kprobes(void)
1030 int __kprobes
arch_trampoline_kprobe(struct kprobe
*p
)