2 * Kernel Probes (KProbes)
3 * arch/i386/kernel/kprobes.c
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation ( includes contributions from
24 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
25 * interface to access function arguments.
26 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
27 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
28 * <prasanna@in.ibm.com> added function-return probes.
31 #include <linux/config.h>
32 #include <linux/kprobes.h>
33 #include <linux/ptrace.h>
34 #include <linux/preempt.h>
35 #include <asm/cacheflush.h>
36 #include <asm/kdebug.h>
39 void jprobe_return_end(void);
41 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
) = NULL
;
42 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
45 * returns non-zero if opcode modifies the interrupt flag.
47 static inline int is_IF_modifier(kprobe_opcode_t opcode
)
52 case 0xcf: /* iret/iretd */
53 case 0x9d: /* popf/popfd */
59 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
64 void __kprobes
arch_copy_kprobe(struct kprobe
*p
)
66 memcpy(p
->ainsn
.insn
, p
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
70 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
72 *p
->addr
= BREAKPOINT_INSTRUCTION
;
73 flush_icache_range((unsigned long) p
->addr
,
74 (unsigned long) p
->addr
+ sizeof(kprobe_opcode_t
));
77 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
80 flush_icache_range((unsigned long) p
->addr
,
81 (unsigned long) p
->addr
+ sizeof(kprobe_opcode_t
));
84 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
88 static inline void save_previous_kprobe(struct kprobe_ctlblk
*kcb
)
90 kcb
->prev_kprobe
.kp
= kprobe_running();
91 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
92 kcb
->prev_kprobe
.old_eflags
= kcb
->kprobe_old_eflags
;
93 kcb
->prev_kprobe
.saved_eflags
= kcb
->kprobe_saved_eflags
;
96 static inline void restore_previous_kprobe(struct kprobe_ctlblk
*kcb
)
98 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
99 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
100 kcb
->kprobe_old_eflags
= kcb
->prev_kprobe
.old_eflags
;
101 kcb
->kprobe_saved_eflags
= kcb
->prev_kprobe
.saved_eflags
;
104 static inline void set_current_kprobe(struct kprobe
*p
, struct pt_regs
*regs
,
105 struct kprobe_ctlblk
*kcb
)
107 __get_cpu_var(current_kprobe
) = p
;
108 kcb
->kprobe_saved_eflags
= kcb
->kprobe_old_eflags
109 = (regs
->eflags
& (TF_MASK
| IF_MASK
));
110 if (is_IF_modifier(p
->opcode
))
111 kcb
->kprobe_saved_eflags
&= ~IF_MASK
;
114 static inline void prepare_singlestep(struct kprobe
*p
, struct pt_regs
*regs
)
116 regs
->eflags
|= TF_MASK
;
117 regs
->eflags
&= ~IF_MASK
;
118 /*single step inline if the instruction is an int3*/
119 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
120 regs
->eip
= (unsigned long)p
->addr
;
122 regs
->eip
= (unsigned long)&p
->ainsn
.insn
;
125 /* Called with kretprobe_lock held */
126 void __kprobes
arch_prepare_kretprobe(struct kretprobe
*rp
,
127 struct pt_regs
*regs
)
129 unsigned long *sara
= (unsigned long *)®s
->esp
;
130 struct kretprobe_instance
*ri
;
132 if ((ri
= get_free_rp_inst(rp
)) != NULL
) {
135 ri
->ret_addr
= (kprobe_opcode_t
*) *sara
;
137 /* Replace the return addr with trampoline addr */
138 *sara
= (unsigned long) &kretprobe_trampoline
;
147 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
148 * remain disabled thorough out this function.
150 static int __kprobes
kprobe_handler(struct pt_regs
*regs
)
154 kprobe_opcode_t
*addr
= NULL
;
156 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
158 /* Check if the application is using LDT entry for its code segment and
159 * calculate the address by reading the base address from the LDT entry.
161 if ((regs
->xcs
& 4) && (current
->mm
)) {
162 lp
= (unsigned long *) ((unsigned long)((regs
->xcs
>> 3) * 8)
163 + (char *) current
->mm
->context
.ldt
);
164 addr
= (kprobe_opcode_t
*) (get_desc_base(lp
) + regs
->eip
-
165 sizeof(kprobe_opcode_t
));
167 addr
= (kprobe_opcode_t
*)(regs
->eip
- sizeof(kprobe_opcode_t
));
169 /* Check we're not actually recursing */
170 if (kprobe_running()) {
171 p
= get_kprobe(addr
);
173 if (kcb
->kprobe_status
== KPROBE_HIT_SS
&&
174 *p
->ainsn
.insn
== BREAKPOINT_INSTRUCTION
) {
175 regs
->eflags
&= ~TF_MASK
;
176 regs
->eflags
|= kcb
->kprobe_saved_eflags
;
179 /* We have reentered the kprobe_handler(), since
180 * another probe was hit while within the handler.
181 * We here save the original kprobes variables and
182 * just single step on the instruction of the new probe
183 * without calling any user handlers.
185 save_previous_kprobe(kcb
);
186 set_current_kprobe(p
, regs
, kcb
);
188 prepare_singlestep(p
, regs
);
189 kcb
->kprobe_status
= KPROBE_REENTER
;
192 p
= __get_cpu_var(current_kprobe
);
193 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
200 p
= get_kprobe(addr
);
202 if (regs
->eflags
& VM_MASK
) {
203 /* We are in virtual-8086 mode. Return 0 */
207 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
209 * The breakpoint instruction was removed right
210 * after we hit it. Another cpu has removed
211 * either a probepoint or a debugger breakpoint
212 * at this address. In either case, no further
213 * handling of this interrupt is appropriate.
214 * Back up over the (now missing) int3 and run
215 * the original instruction.
217 regs
->eip
-= sizeof(kprobe_opcode_t
);
220 /* Not one of ours: let kernel handle it */
225 * This preempt_disable() matches the preempt_enable_no_resched()
226 * in post_kprobe_handler()
229 set_current_kprobe(p
, regs
, kcb
);
230 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
232 if (p
->pre_handler
&& p
->pre_handler(p
, regs
))
233 /* handler has already set things up, so skip ss setup */
237 prepare_singlestep(p
, regs
);
238 kcb
->kprobe_status
= KPROBE_HIT_SS
;
246 * For function-return probes, init_kprobes() establishes a probepoint
247 * here. When a retprobed function returns, this probe is hit and
248 * trampoline_probe_handler() runs, calling the kretprobe's handler.
250 void kretprobe_trampoline_holder(void)
252 asm volatile ( ".global kretprobe_trampoline\n"
253 "kretprobe_trampoline: \n"
258 * Called when we hit the probe point at kretprobe_trampoline
260 int __kprobes
trampoline_probe_handler(struct kprobe
*p
, struct pt_regs
*regs
)
262 struct kretprobe_instance
*ri
= NULL
;
263 struct hlist_head
*head
;
264 struct hlist_node
*node
, *tmp
;
265 unsigned long flags
, orig_ret_address
= 0;
266 unsigned long trampoline_address
=(unsigned long)&kretprobe_trampoline
;
268 spin_lock_irqsave(&kretprobe_lock
, flags
);
269 head
= kretprobe_inst_table_head(current
);
272 * It is possible to have multiple instances associated with a given
273 * task either because an multiple functions in the call path
274 * have a return probe installed on them, and/or more then one return
275 * return probe was registered for a target function.
277 * We can handle this because:
278 * - instances are always inserted at the head of the list
279 * - when multiple return probes are registered for the same
280 * function, the first instance's ret_addr will point to the
281 * real return address, and all the rest will point to
282 * kretprobe_trampoline
284 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
285 if (ri
->task
!= current
)
286 /* another task is sharing our hash bucket */
289 if (ri
->rp
&& ri
->rp
->handler
)
290 ri
->rp
->handler(ri
, regs
);
292 orig_ret_address
= (unsigned long)ri
->ret_addr
;
295 if (orig_ret_address
!= trampoline_address
)
297 * This is the real return address. Any other
298 * instances associated with this task are for
299 * other calls deeper on the call stack
304 BUG_ON(!orig_ret_address
|| (orig_ret_address
== trampoline_address
));
305 regs
->eip
= orig_ret_address
;
307 reset_current_kprobe();
308 spin_unlock_irqrestore(&kretprobe_lock
, flags
);
309 preempt_enable_no_resched();
312 * By returning a non-zero value, we are telling
313 * kprobe_handler() that we have handled unlocking
314 * and re-enabling preemption
320 * Called after single-stepping. p->addr is the address of the
321 * instruction whose first byte has been replaced by the "int 3"
322 * instruction. To avoid the SMP problems that can occur when we
323 * temporarily put back the original opcode to single-step, we
324 * single-stepped a copy of the instruction. The address of this
325 * copy is p->ainsn.insn.
327 * This function prepares to return from the post-single-step
328 * interrupt. We have to fix up the stack as follows:
330 * 0) Except in the case of absolute or indirect jump or call instructions,
331 * the new eip is relative to the copied instruction. We need to make
332 * it relative to the original instruction.
334 * 1) If the single-stepped instruction was pushfl, then the TF and IF
335 * flags are set in the just-pushed eflags, and may need to be cleared.
337 * 2) If the single-stepped instruction was a call, the return address
338 * that is atop the stack is the address following the copied instruction.
339 * We need to make it the address following the original instruction.
341 static void __kprobes
resume_execution(struct kprobe
*p
,
342 struct pt_regs
*regs
, struct kprobe_ctlblk
*kcb
)
344 unsigned long *tos
= (unsigned long *)®s
->esp
;
345 unsigned long next_eip
= 0;
346 unsigned long copy_eip
= (unsigned long)&p
->ainsn
.insn
;
347 unsigned long orig_eip
= (unsigned long)p
->addr
;
349 switch (p
->ainsn
.insn
[0]) {
350 case 0x9c: /* pushfl */
351 *tos
&= ~(TF_MASK
| IF_MASK
);
352 *tos
|= kcb
->kprobe_old_eflags
;
354 case 0xc3: /* ret/lret */
358 regs
->eflags
&= ~TF_MASK
;
359 /* eip is already adjusted, no more changes required*/
361 case 0xe8: /* call relative - Fix return addr */
362 *tos
= orig_eip
+ (*tos
- copy_eip
);
365 if ((p
->ainsn
.insn
[1] & 0x30) == 0x10) {
366 /* call absolute, indirect */
367 /* Fix return addr; eip is correct. */
368 next_eip
= regs
->eip
;
369 *tos
= orig_eip
+ (*tos
- copy_eip
);
370 } else if (((p
->ainsn
.insn
[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
371 ((p
->ainsn
.insn
[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
372 /* eip is correct. */
373 next_eip
= regs
->eip
;
376 case 0xea: /* jmp absolute -- eip is correct */
377 next_eip
= regs
->eip
;
383 regs
->eflags
&= ~TF_MASK
;
385 regs
->eip
= next_eip
;
387 regs
->eip
= orig_eip
+ (regs
->eip
- copy_eip
);
392 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
393 * remain disabled thoroughout this function.
395 static inline int post_kprobe_handler(struct pt_regs
*regs
)
397 struct kprobe
*cur
= kprobe_running();
398 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
403 if ((kcb
->kprobe_status
!= KPROBE_REENTER
) && cur
->post_handler
) {
404 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
405 cur
->post_handler(cur
, regs
, 0);
408 resume_execution(cur
, regs
, kcb
);
409 regs
->eflags
|= kcb
->kprobe_saved_eflags
;
411 /*Restore back the original saved kprobes variables and continue. */
412 if (kcb
->kprobe_status
== KPROBE_REENTER
) {
413 restore_previous_kprobe(kcb
);
416 reset_current_kprobe();
418 preempt_enable_no_resched();
421 * if somebody else is singlestepping across a probe point, eflags
422 * will have TF set, in which case, continue the remaining processing
423 * of do_debug, as if this is not a probe hit.
425 if (regs
->eflags
& TF_MASK
)
431 static inline int kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
433 struct kprobe
*cur
= kprobe_running();
434 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
436 if (cur
->fault_handler
&& cur
->fault_handler(cur
, regs
, trapnr
))
439 if (kcb
->kprobe_status
& KPROBE_HIT_SS
) {
440 resume_execution(cur
, regs
, kcb
);
441 regs
->eflags
|= kcb
->kprobe_old_eflags
;
443 reset_current_kprobe();
444 preempt_enable_no_resched();
450 * Wrapper routine to for handling exceptions.
452 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
453 unsigned long val
, void *data
)
455 struct die_args
*args
= (struct die_args
*)data
;
456 int ret
= NOTIFY_DONE
;
461 if (kprobe_handler(args
->regs
))
465 if (post_kprobe_handler(args
->regs
))
470 if (kprobe_running() &&
471 kprobe_fault_handler(args
->regs
, args
->trapnr
))
481 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
483 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
485 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
487 kcb
->jprobe_saved_regs
= *regs
;
488 kcb
->jprobe_saved_esp
= ®s
->esp
;
489 addr
= (unsigned long)(kcb
->jprobe_saved_esp
);
492 * TBD: As Linus pointed out, gcc assumes that the callee
493 * owns the argument space and could overwrite it, e.g.
494 * tailcall optimization. So, to be absolutely safe
495 * we also save and restore enough stack bytes to cover
498 memcpy(kcb
->jprobes_stack
, (kprobe_opcode_t
*)addr
,
499 MIN_STACK_SIZE(addr
));
500 regs
->eflags
&= ~IF_MASK
;
501 regs
->eip
= (unsigned long)(jp
->entry
);
505 void __kprobes
jprobe_return(void)
507 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
509 asm volatile (" xchgl %%ebx,%%esp \n"
511 " .globl jprobe_return_end \n"
512 " jprobe_return_end: \n"
514 (kcb
->jprobe_saved_esp
):"memory");
517 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
519 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
520 u8
*addr
= (u8
*) (regs
->eip
- 1);
521 unsigned long stack_addr
= (unsigned long)(kcb
->jprobe_saved_esp
);
522 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
524 if ((addr
> (u8
*) jprobe_return
) && (addr
< (u8
*) jprobe_return_end
)) {
525 if (®s
->esp
!= kcb
->jprobe_saved_esp
) {
526 struct pt_regs
*saved_regs
=
527 container_of(kcb
->jprobe_saved_esp
,
528 struct pt_regs
, esp
);
529 printk("current esp %p does not match saved esp %p\n",
530 ®s
->esp
, kcb
->jprobe_saved_esp
);
531 printk("Saved registers for jprobe %p\n", jp
);
532 show_registers(saved_regs
);
533 printk("Current registers\n");
534 show_registers(regs
);
537 *regs
= kcb
->jprobe_saved_regs
;
538 memcpy((kprobe_opcode_t
*) stack_addr
, kcb
->jprobes_stack
,
539 MIN_STACK_SIZE(stack_addr
));
545 static struct kprobe trampoline_p
= {
546 .addr
= (kprobe_opcode_t
*) &kretprobe_trampoline
,
547 .pre_handler
= trampoline_probe_handler
550 int __init
arch_init_kprobes(void)
552 return register_kprobe(&trampoline_p
);