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, 2006
20 * s390 port, used ppc64 as template. Mike Grundy <grundym@us.ibm.com>
23 #include <linux/kprobes.h>
24 #include <linux/ptrace.h>
25 #include <linux/preempt.h>
26 #include <linux/stop_machine.h>
27 #include <linux/kdebug.h>
28 #include <linux/uaccess.h>
29 #include <asm/cacheflush.h>
30 #include <asm/sections.h>
31 #include <linux/module.h>
32 #include <linux/slab.h>
33 #include <linux/hardirq.h>
35 DEFINE_PER_CPU(struct kprobe
*, current_kprobe
);
36 DEFINE_PER_CPU(struct kprobe_ctlblk
, kprobe_ctlblk
);
38 struct kretprobe_blackpoint kretprobe_blacklist
[] = { };
40 static int __kprobes
is_prohibited_opcode(kprobe_opcode_t
*insn
)
42 switch (insn
[0] >> 8) {
43 case 0x0c: /* bassm */
47 case 0xac: /* stnsm */
48 case 0xad: /* stosm */
53 case 0xb25a: /* bsa */
54 case 0xb240: /* bakr */
55 case 0xb258: /* bsg */
58 case 0xb98d: /* epsw */
64 static int __kprobes
get_fixup_type(kprobe_opcode_t
*insn
)
66 /* default fixup method */
67 int fixup
= FIXUP_PSW_NORMAL
;
69 switch (insn
[0] >> 8) {
72 fixup
= FIXUP_RETURN_REGISTER
;
73 /* if r2 = 0, no branch will be taken */
74 if ((insn
[0] & 0x0f) == 0)
75 fixup
|= FIXUP_BRANCH_NOT_TAKEN
;
79 fixup
= FIXUP_BRANCH_NOT_TAKEN
;
83 fixup
= FIXUP_RETURN_REGISTER
;
89 fixup
= FIXUP_BRANCH_NOT_TAKEN
;
92 fixup
= FIXUP_NOT_REQUIRED
;
94 case 0xb2: /* lpswe */
95 if ((insn
[0] & 0xff) == 0xb2)
96 fixup
= FIXUP_NOT_REQUIRED
;
99 if ((insn
[0] & 0x0f) == 0x05)
100 fixup
|= FIXUP_RETURN_REGISTER
;
103 if ((insn
[0] & 0x0f) == 0x00 || /* larl */
104 (insn
[0] & 0x0f) == 0x05) /* brasl */
105 fixup
|= FIXUP_RETURN_REGISTER
;
108 if ((insn
[2] & 0xff) == 0x44 || /* bxhg */
109 (insn
[2] & 0xff) == 0x45) /* bxleg */
110 fixup
= FIXUP_BRANCH_NOT_TAKEN
;
112 case 0xe3: /* bctg */
113 if ((insn
[2] & 0xff) == 0x46)
114 fixup
= FIXUP_BRANCH_NOT_TAKEN
;
120 int __kprobes
arch_prepare_kprobe(struct kprobe
*p
)
122 if ((unsigned long) p
->addr
& 0x01)
125 /* Make sure the probe isn't going on a difficult instruction */
126 if (is_prohibited_opcode(p
->addr
))
129 p
->opcode
= *p
->addr
;
130 memcpy(p
->ainsn
.insn
, p
->addr
, ((p
->opcode
>> 14) + 3) & -2);
135 struct ins_replace_args
{
136 kprobe_opcode_t
*ptr
;
137 kprobe_opcode_t opcode
;
140 static int __kprobes
swap_instruction(void *aref
)
142 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
143 unsigned long status
= kcb
->kprobe_status
;
144 struct ins_replace_args
*args
= aref
;
146 kcb
->kprobe_status
= KPROBE_SWAP_INST
;
147 probe_kernel_write(args
->ptr
, &args
->opcode
, sizeof(args
->opcode
));
148 kcb
->kprobe_status
= status
;
152 void __kprobes
arch_arm_kprobe(struct kprobe
*p
)
154 struct ins_replace_args args
;
157 args
.opcode
= BREAKPOINT_INSTRUCTION
;
158 stop_machine(swap_instruction
, &args
, NULL
);
161 void __kprobes
arch_disarm_kprobe(struct kprobe
*p
)
163 struct ins_replace_args args
;
166 args
.opcode
= p
->opcode
;
167 stop_machine(swap_instruction
, &args
, NULL
);
170 void __kprobes
arch_remove_kprobe(struct kprobe
*p
)
174 static void __kprobes
enable_singlestep(struct kprobe_ctlblk
*kcb
,
175 struct pt_regs
*regs
,
178 struct per_regs per_kprobe
;
180 /* Set up the PER control registers %cr9-%cr11 */
181 per_kprobe
.control
= PER_EVENT_IFETCH
;
182 per_kprobe
.start
= ip
;
185 /* Save control regs and psw mask */
186 __ctl_store(kcb
->kprobe_saved_ctl
, 9, 11);
187 kcb
->kprobe_saved_imask
= regs
->psw
.mask
&
188 (PSW_MASK_PER
| PSW_MASK_IO
| PSW_MASK_EXT
);
190 /* Set PER control regs, turns on single step for the given address */
191 __ctl_load(per_kprobe
, 9, 11);
192 regs
->psw
.mask
|= PSW_MASK_PER
;
193 regs
->psw
.mask
&= ~(PSW_MASK_IO
| PSW_MASK_EXT
);
194 regs
->psw
.addr
= ip
| PSW_ADDR_AMODE
;
197 static void __kprobes
disable_singlestep(struct kprobe_ctlblk
*kcb
,
198 struct pt_regs
*regs
,
201 /* Restore control regs and psw mask, set new psw address */
202 __ctl_load(kcb
->kprobe_saved_ctl
, 9, 11);
203 regs
->psw
.mask
&= ~PSW_MASK_PER
;
204 regs
->psw
.mask
|= kcb
->kprobe_saved_imask
;
205 regs
->psw
.addr
= ip
| PSW_ADDR_AMODE
;
209 * Activate a kprobe by storing its pointer to current_kprobe. The
210 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
211 * two kprobes can be active, see KPROBE_REENTER.
213 static void __kprobes
push_kprobe(struct kprobe_ctlblk
*kcb
, struct kprobe
*p
)
215 kcb
->prev_kprobe
.kp
= __get_cpu_var(current_kprobe
);
216 kcb
->prev_kprobe
.status
= kcb
->kprobe_status
;
217 __get_cpu_var(current_kprobe
) = p
;
221 * Deactivate a kprobe by backing up to the previous state. If the
222 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
223 * for any other state prev_kprobe.kp will be NULL.
225 static void __kprobes
pop_kprobe(struct kprobe_ctlblk
*kcb
)
227 __get_cpu_var(current_kprobe
) = kcb
->prev_kprobe
.kp
;
228 kcb
->kprobe_status
= kcb
->prev_kprobe
.status
;
231 void __kprobes
arch_prepare_kretprobe(struct kretprobe_instance
*ri
,
232 struct pt_regs
*regs
)
234 ri
->ret_addr
= (kprobe_opcode_t
*) regs
->gprs
[14];
236 /* Replace the return addr with trampoline addr */
237 regs
->gprs
[14] = (unsigned long) &kretprobe_trampoline
;
240 static void __kprobes
kprobe_reenter_check(struct kprobe_ctlblk
*kcb
,
243 switch (kcb
->kprobe_status
) {
244 case KPROBE_HIT_SSDONE
:
245 case KPROBE_HIT_ACTIVE
:
246 kprobes_inc_nmissed_count(p
);
252 * A kprobe on the code path to single step an instruction
253 * is a BUG. The code path resides in the .kprobes.text
254 * section and is executed with interrupts disabled.
256 printk(KERN_EMERG
"Invalid kprobe detected at %p.\n", p
->addr
);
262 static int __kprobes
kprobe_handler(struct pt_regs
*regs
)
264 struct kprobe_ctlblk
*kcb
;
268 * We want to disable preemption for the entire duration of kprobe
269 * processing. That includes the calls to the pre/post handlers
270 * and single stepping the kprobe instruction.
273 kcb
= get_kprobe_ctlblk();
274 p
= get_kprobe((void *)((regs
->psw
.addr
& PSW_ADDR_INSN
) - 2));
277 if (kprobe_running()) {
279 * We have hit a kprobe while another is still
280 * active. This can happen in the pre and post
281 * handler. Single step the instruction of the
282 * new probe but do not call any handler function
283 * of this secondary kprobe.
284 * push_kprobe and pop_kprobe saves and restores
285 * the currently active kprobe.
287 kprobe_reenter_check(kcb
, p
);
289 kcb
->kprobe_status
= KPROBE_REENTER
;
292 * If we have no pre-handler or it returned 0, we
293 * continue with single stepping. If we have a
294 * pre-handler and it returned non-zero, it prepped
295 * for calling the break_handler below on re-entry
296 * for jprobe processing, so get out doing nothing
300 kcb
->kprobe_status
= KPROBE_HIT_ACTIVE
;
301 if (p
->pre_handler
&& p
->pre_handler(p
, regs
))
303 kcb
->kprobe_status
= KPROBE_HIT_SS
;
305 enable_singlestep(kcb
, regs
, (unsigned long) p
->ainsn
.insn
);
307 } else if (kprobe_running()) {
308 p
= __get_cpu_var(current_kprobe
);
309 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
311 * Continuation after the jprobe completed and
312 * caused the jprobe_return trap. The jprobe
313 * break_handler "returns" to the original
314 * function that still has the kprobe breakpoint
315 * installed. We continue with single stepping.
317 kcb
->kprobe_status
= KPROBE_HIT_SS
;
318 enable_singlestep(kcb
, regs
,
319 (unsigned long) p
->ainsn
.insn
);
322 * No kprobe at this address and the current kprobe
323 * has no break handler (no jprobe!). The kernel just
324 * exploded, let the standard trap handler pick up the
328 * No kprobe at this address and no active kprobe. The trap has
329 * not been caused by a kprobe breakpoint. The race of breakpoint
330 * vs. kprobe remove does not exist because on s390 as we use
331 * stop_machine to arm/disarm the breakpoints.
333 preempt_enable_no_resched();
338 * Function return probe trampoline:
339 * - init_kprobes() establishes a probepoint here
340 * - When the probed function returns, this probe
341 * causes the handlers to fire
343 static void __used
kretprobe_trampoline_holder(void)
345 asm volatile(".global kretprobe_trampoline\n"
346 "kretprobe_trampoline: bcr 0,0\n");
350 * Called when the probe at kretprobe trampoline is hit
352 static int __kprobes
trampoline_probe_handler(struct kprobe
*p
,
353 struct pt_regs
*regs
)
355 struct kretprobe_instance
*ri
;
356 struct hlist_head
*head
, empty_rp
;
357 struct hlist_node
*node
, *tmp
;
358 unsigned long flags
, orig_ret_address
;
359 unsigned long trampoline_address
;
360 kprobe_opcode_t
*correct_ret_addr
;
362 INIT_HLIST_HEAD(&empty_rp
);
363 kretprobe_hash_lock(current
, &head
, &flags
);
366 * It is possible to have multiple instances associated with a given
367 * task either because an multiple functions in the call path
368 * have a return probe installed on them, and/or more than one return
369 * return probe was registered for a target function.
371 * We can handle this because:
372 * - instances are always inserted at the head of the list
373 * - when multiple return probes are registered for the same
374 * function, the first instance's ret_addr will point to the
375 * real return address, and all the rest will point to
376 * kretprobe_trampoline
379 orig_ret_address
= 0;
380 correct_ret_addr
= NULL
;
381 trampoline_address
= (unsigned long) &kretprobe_trampoline
;
382 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
383 if (ri
->task
!= current
)
384 /* another task is sharing our hash bucket */
387 orig_ret_address
= (unsigned long) ri
->ret_addr
;
389 if (orig_ret_address
!= trampoline_address
)
391 * This is the real return address. Any other
392 * instances associated with this task are for
393 * other calls deeper on the call stack
398 kretprobe_assert(ri
, orig_ret_address
, trampoline_address
);
400 correct_ret_addr
= ri
->ret_addr
;
401 hlist_for_each_entry_safe(ri
, node
, tmp
, head
, hlist
) {
402 if (ri
->task
!= current
)
403 /* another task is sharing our hash bucket */
406 orig_ret_address
= (unsigned long) ri
->ret_addr
;
408 if (ri
->rp
&& ri
->rp
->handler
) {
409 ri
->ret_addr
= correct_ret_addr
;
410 ri
->rp
->handler(ri
, regs
);
413 recycle_rp_inst(ri
, &empty_rp
);
415 if (orig_ret_address
!= trampoline_address
)
417 * This is the real return address. Any other
418 * instances associated with this task are for
419 * other calls deeper on the call stack
424 regs
->psw
.addr
= orig_ret_address
| PSW_ADDR_AMODE
;
426 pop_kprobe(get_kprobe_ctlblk());
427 kretprobe_hash_unlock(current
, &flags
);
428 preempt_enable_no_resched();
430 hlist_for_each_entry_safe(ri
, node
, tmp
, &empty_rp
, hlist
) {
431 hlist_del(&ri
->hlist
);
435 * By returning a non-zero value, we are telling
436 * kprobe_handler() that we don't want the post_handler
437 * to run (and have re-enabled preemption)
443 * Called after single-stepping. p->addr is the address of the
444 * instruction whose first byte has been replaced by the "breakpoint"
445 * instruction. To avoid the SMP problems that can occur when we
446 * temporarily put back the original opcode to single-step, we
447 * single-stepped a copy of the instruction. The address of this
448 * copy is p->ainsn.insn.
450 static void __kprobes
resume_execution(struct kprobe
*p
, struct pt_regs
*regs
)
452 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
453 unsigned long ip
= regs
->psw
.addr
& PSW_ADDR_INSN
;
454 int fixup
= get_fixup_type(p
->ainsn
.insn
);
456 if (fixup
& FIXUP_PSW_NORMAL
)
457 ip
+= (unsigned long) p
->addr
- (unsigned long) p
->ainsn
.insn
;
459 if (fixup
& FIXUP_BRANCH_NOT_TAKEN
) {
460 int ilen
= ((p
->ainsn
.insn
[0] >> 14) + 3) & -2;
461 if (ip
- (unsigned long) p
->ainsn
.insn
== ilen
)
462 ip
= (unsigned long) p
->addr
+ ilen
;
465 if (fixup
& FIXUP_RETURN_REGISTER
) {
466 int reg
= (p
->ainsn
.insn
[0] & 0xf0) >> 4;
467 regs
->gprs
[reg
] += (unsigned long) p
->addr
-
468 (unsigned long) p
->ainsn
.insn
;
471 disable_singlestep(kcb
, regs
, ip
);
474 static int __kprobes
post_kprobe_handler(struct pt_regs
*regs
)
476 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
477 struct kprobe
*p
= kprobe_running();
482 if (kcb
->kprobe_status
!= KPROBE_REENTER
&& p
->post_handler
) {
483 kcb
->kprobe_status
= KPROBE_HIT_SSDONE
;
484 p
->post_handler(p
, regs
, 0);
487 resume_execution(p
, regs
);
489 preempt_enable_no_resched();
492 * if somebody else is singlestepping across a probe point, psw mask
493 * will have PER set, in which case, continue the remaining processing
494 * of do_single_step, as if this is not a probe hit.
496 if (regs
->psw
.mask
& PSW_MASK_PER
)
502 static int __kprobes
kprobe_trap_handler(struct pt_regs
*regs
, int trapnr
)
504 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
505 struct kprobe
*p
= kprobe_running();
506 const struct exception_table_entry
*entry
;
508 switch(kcb
->kprobe_status
) {
509 case KPROBE_SWAP_INST
:
510 /* We are here because the instruction replacement failed */
515 * We are here because the instruction being single
516 * stepped caused a page fault. We reset the current
517 * kprobe and the nip points back to the probe address
518 * and allow the page fault handler to continue as a
521 disable_singlestep(kcb
, regs
, (unsigned long) p
->addr
);
523 preempt_enable_no_resched();
525 case KPROBE_HIT_ACTIVE
:
526 case KPROBE_HIT_SSDONE
:
528 * We increment the nmissed count for accounting,
529 * we can also use npre/npostfault count for accouting
530 * these specific fault cases.
532 kprobes_inc_nmissed_count(p
);
535 * We come here because instructions in the pre/post
536 * handler caused the page_fault, this could happen
537 * if handler tries to access user space by
538 * copy_from_user(), get_user() etc. Let the
539 * user-specified handler try to fix it first.
541 if (p
->fault_handler
&& p
->fault_handler(p
, regs
, trapnr
))
545 * In case the user-specified fault handler returned
546 * zero, try to fix up.
548 entry
= search_exception_tables(regs
->psw
.addr
& PSW_ADDR_INSN
);
550 regs
->psw
.addr
= entry
->fixup
| PSW_ADDR_AMODE
;
555 * fixup_exception() could not handle it,
556 * Let do_page_fault() fix it.
565 int __kprobes
kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
569 if (regs
->psw
.mask
& (PSW_MASK_IO
| PSW_MASK_EXT
))
571 ret
= kprobe_trap_handler(regs
, trapnr
);
572 if (regs
->psw
.mask
& (PSW_MASK_IO
| PSW_MASK_EXT
))
573 local_irq_restore(regs
->psw
.mask
& ~PSW_MASK_PER
);
578 * Wrapper routine to for handling exceptions.
580 int __kprobes
kprobe_exceptions_notify(struct notifier_block
*self
,
581 unsigned long val
, void *data
)
583 struct die_args
*args
= (struct die_args
*) data
;
584 struct pt_regs
*regs
= args
->regs
;
585 int ret
= NOTIFY_DONE
;
587 if (regs
->psw
.mask
& (PSW_MASK_IO
| PSW_MASK_EXT
))
592 if (kprobe_handler(regs
))
596 if (post_kprobe_handler(regs
))
600 if (!preemptible() && kprobe_running() &&
601 kprobe_trap_handler(regs
, args
->trapnr
))
608 if (regs
->psw
.mask
& (PSW_MASK_IO
| PSW_MASK_EXT
))
609 local_irq_restore(regs
->psw
.mask
& ~PSW_MASK_PER
);
614 int __kprobes
setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
616 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
617 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
620 memcpy(&kcb
->jprobe_saved_regs
, regs
, sizeof(struct pt_regs
));
622 /* setup return addr to the jprobe handler routine */
623 regs
->psw
.addr
= (unsigned long) jp
->entry
| PSW_ADDR_AMODE
;
624 regs
->psw
.mask
&= ~(PSW_MASK_IO
| PSW_MASK_EXT
);
626 /* r15 is the stack pointer */
627 stack
= (unsigned long) regs
->gprs
[15];
629 memcpy(kcb
->jprobes_stack
, (void *) stack
, MIN_STACK_SIZE(stack
));
633 void __kprobes
jprobe_return(void)
635 asm volatile(".word 0x0002");
638 static void __used __kprobes
jprobe_return_end(void)
640 asm volatile("bcr 0,0");
643 int __kprobes
longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
645 struct kprobe_ctlblk
*kcb
= get_kprobe_ctlblk();
648 stack
= (unsigned long) kcb
->jprobe_saved_regs
.gprs
[15];
650 /* Put the regs back */
651 memcpy(regs
, &kcb
->jprobe_saved_regs
, sizeof(struct pt_regs
));
652 /* put the stack back */
653 memcpy((void *) stack
, kcb
->jprobes_stack
, MIN_STACK_SIZE(stack
));
654 preempt_enable_no_resched();
658 static struct kprobe trampoline
= {
659 .addr
= (kprobe_opcode_t
*) &kretprobe_trampoline
,
660 .pre_handler
= trampoline_probe_handler
663 int __init
arch_init_kprobes(void)
665 return register_kprobe(&trampoline
);
668 int __kprobes
arch_trampoline_kprobe(struct kprobe
*p
)
670 return p
->addr
== (kprobe_opcode_t
*) &kretprobe_trampoline
;