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) = NULL;
36 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
38 struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}};
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((kprobe_opcode_t *) 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 per_cr_bits kprobe_per_regs[1];
180 /* Set up the per control reg info, will pass to lctl */
181 memset(kprobe_per_regs, 0, sizeof(per_cr_bits));
182 kprobe_per_regs[0].em_instruction_fetch = 1;
183 kprobe_per_regs[0].starting_addr = ip;
184 kprobe_per_regs[0].ending_addr = ip;
186 /* Save control regs and psw mask */
187 __ctl_store(kcb->kprobe_saved_ctl, 9, 11);
188 kcb->kprobe_saved_imask = regs->psw.mask &
189 (PSW_MASK_PER | PSW_MASK_IO | PSW_MASK_EXT);
191 /* Set PER control regs, turns on single step for the given address */
192 __ctl_load(kprobe_per_regs, 9, 11);
193 regs->psw.mask |= PSW_MASK_PER;
194 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
195 regs->psw.addr = ip | PSW_ADDR_AMODE;
198 static void __kprobes disable_singlestep(struct kprobe_ctlblk *kcb,
199 struct pt_regs *regs,
202 /* Restore control regs and psw mask, set new psw address */
203 __ctl_load(kcb->kprobe_saved_ctl, 9, 11);
204 regs->psw.mask &= ~PSW_MASK_PER;
205 regs->psw.mask |= kcb->kprobe_saved_imask;
206 regs->psw.addr = ip | PSW_ADDR_AMODE;
210 * Activate a kprobe by storing its pointer to current_kprobe. The
211 * previous kprobe is stored in kcb->prev_kprobe. A stack of up to
212 * two kprobes can be active, see KPROBE_REENTER.
214 static void __kprobes push_kprobe(struct kprobe_ctlblk *kcb, struct kprobe *p)
216 kcb->prev_kprobe.kp = __get_cpu_var(current_kprobe);
217 kcb->prev_kprobe.status = kcb->kprobe_status;
218 __get_cpu_var(current_kprobe) = p;
222 * Deactivate a kprobe by backing up to the previous state. If the
223 * current state is KPROBE_REENTER prev_kprobe.kp will be non-NULL,
224 * for any other state prev_kprobe.kp will be NULL.
226 static void __kprobes pop_kprobe(struct kprobe_ctlblk *kcb)
228 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
229 kcb->kprobe_status = kcb->prev_kprobe.status;
232 void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
233 struct pt_regs *regs)
235 ri->ret_addr = (kprobe_opcode_t *) regs->gprs[14];
237 /* Replace the return addr with trampoline addr */
238 regs->gprs[14] = (unsigned long)&kretprobe_trampoline;
241 static void __kprobes kprobe_reenter_check(struct kprobe_ctlblk *kcb,
244 switch (kcb->kprobe_status) {
245 case KPROBE_HIT_SSDONE:
246 case KPROBE_HIT_ACTIVE:
247 kprobes_inc_nmissed_count(p);
253 * A kprobe on the code path to single step an instruction
254 * is a BUG. The code path resides in the .kprobes.text
255 * section and is executed with interrupts disabled.
257 printk(KERN_EMERG "Invalid kprobe detected at %p.\n", p->addr);
263 static int __kprobes kprobe_handler(struct pt_regs *regs)
265 struct kprobe_ctlblk *kcb;
269 * We want to disable preemption for the entire duration of kprobe
270 * processing. That includes the calls to the pre/post handlers
271 * and single stepping the kprobe instruction.
274 kcb = get_kprobe_ctlblk();
275 p = get_kprobe((void *)((regs->psw.addr & PSW_ADDR_INSN) - 2));
278 if (kprobe_running()) {
280 * We have hit a kprobe while another is still
281 * active. This can happen in the pre and post
282 * handler. Single step the instruction of the
283 * new probe but do not call any handler function
284 * of this secondary kprobe.
285 * push_kprobe and pop_kprobe saves and restores
286 * the currently active kprobe.
288 kprobe_reenter_check(kcb, p);
290 kcb->kprobe_status = KPROBE_REENTER;
293 * If we have no pre-handler or it returned 0, we
294 * continue with single stepping. If we have a
295 * pre-handler and it returned non-zero, it prepped
296 * for calling the break_handler below on re-entry
297 * for jprobe processing, so get out doing nothing
301 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
302 if (p->pre_handler && p->pre_handler(p, regs))
304 kcb->kprobe_status = KPROBE_HIT_SS;
306 enable_singlestep(kcb, regs, (unsigned long) p->ainsn.insn);
308 } else if (kprobe_running()) {
309 p = __get_cpu_var(current_kprobe);
310 if (p->break_handler && p->break_handler(p, regs)) {
312 * Continuation after the jprobe completed and
313 * caused the jprobe_return trap. The jprobe
314 * break_handler "returns" to the original
315 * function that still has the kprobe breakpoint
316 * installed. We continue with single stepping.
318 kcb->kprobe_status = KPROBE_HIT_SS;
319 enable_singlestep(kcb, regs,
320 (unsigned long) p->ainsn.insn);
323 * No kprobe at this address and the current kprobe
324 * has no break handler (no jprobe!). The kernel just
325 * exploded, let the standard trap handler pick up the
329 * No kprobe at this address and no active kprobe. The trap has
330 * not been caused by a kprobe breakpoint. The race of breakpoint
331 * vs. kprobe remove does not exist because on s390 as we use
332 * stop_machine to arm/disarm the breakpoints.
334 preempt_enable_no_resched();
339 * Function return probe trampoline:
340 * - init_kprobes() establishes a probepoint here
341 * - When the probed function returns, this probe
342 * causes the handlers to fire
344 static void __used kretprobe_trampoline_holder(void)
346 asm volatile(".global kretprobe_trampoline\n"
347 "kretprobe_trampoline: bcr 0,0\n");
351 * Called when the probe at kretprobe trampoline is hit
353 static int __kprobes trampoline_probe_handler(struct kprobe *p,
354 struct pt_regs *regs)
356 struct kretprobe_instance *ri = NULL;
357 struct hlist_head *head, empty_rp;
358 struct hlist_node *node, *tmp;
359 unsigned long flags, orig_ret_address = 0;
360 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline;
361 kprobe_opcode_t *correct_ret_addr = NULL;
363 INIT_HLIST_HEAD(&empty_rp);
364 kretprobe_hash_lock(current, &head, &flags);
367 * It is possible to have multiple instances associated with a given
368 * task either because an multiple functions in the call path
369 * have a return probe installed on them, and/or more than one return
370 * return probe was registered for a target function.
372 * We can handle this because:
373 * - instances are always inserted at the head of the list
374 * - when multiple return probes are registered for the same
375 * function, the first instance's ret_addr will point to the
376 * real return address, and all the rest will point to
377 * kretprobe_trampoline
379 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
380 if (ri->task != current)
381 /* another task is sharing our hash bucket */
384 orig_ret_address = (unsigned long)ri->ret_addr;
386 if (orig_ret_address != trampoline_address)
388 * This is the real return address. Any other
389 * instances associated with this task are for
390 * other calls deeper on the call stack
395 kretprobe_assert(ri, orig_ret_address, trampoline_address);
397 correct_ret_addr = ri->ret_addr;
398 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
399 if (ri->task != current)
400 /* another task is sharing our hash bucket */
403 orig_ret_address = (unsigned long)ri->ret_addr;
405 if (ri->rp && ri->rp->handler) {
406 ri->ret_addr = correct_ret_addr;
407 ri->rp->handler(ri, regs);
410 recycle_rp_inst(ri, &empty_rp);
412 if (orig_ret_address != trampoline_address) {
414 * This is the real return address. Any other
415 * instances associated with this task are for
416 * other calls deeper on the call stack
422 regs->psw.addr = orig_ret_address | PSW_ADDR_AMODE;
424 pop_kprobe(get_kprobe_ctlblk());
425 kretprobe_hash_unlock(current, &flags);
426 preempt_enable_no_resched();
428 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
429 hlist_del(&ri->hlist);
433 * By returning a non-zero value, we are telling
434 * kprobe_handler() that we don't want the post_handler
435 * to run (and have re-enabled preemption)
441 * Called after single-stepping. p->addr is the address of the
442 * instruction whose first byte has been replaced by the "breakpoint"
443 * instruction. To avoid the SMP problems that can occur when we
444 * temporarily put back the original opcode to single-step, we
445 * single-stepped a copy of the instruction. The address of this
446 * copy is p->ainsn.insn.
448 static void __kprobes resume_execution(struct kprobe *p, struct pt_regs *regs)
450 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
451 unsigned long ip = regs->psw.addr & PSW_ADDR_INSN;
452 int fixup = get_fixup_type(p->ainsn.insn);
454 if (fixup & FIXUP_PSW_NORMAL)
455 ip += (unsigned long) p->addr - (unsigned long) p->ainsn.insn;
457 if (fixup & FIXUP_BRANCH_NOT_TAKEN) {
458 int ilen = ((p->ainsn.insn[0] >> 14) + 3) & -2;
459 if (ip - (unsigned long) p->ainsn.insn == ilen)
460 ip = (unsigned long) p->addr + ilen;
463 if (fixup & FIXUP_RETURN_REGISTER) {
464 int reg = (p->ainsn.insn[0] & 0xf0) >> 4;
465 regs->gprs[reg] += (unsigned long) p->addr -
466 (unsigned long) p->ainsn.insn;
469 disable_singlestep(kcb, regs, ip);
472 static int __kprobes post_kprobe_handler(struct pt_regs *regs)
474 struct kprobe *cur = kprobe_running();
475 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
480 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
481 kcb->kprobe_status = KPROBE_HIT_SSDONE;
482 cur->post_handler(cur, regs, 0);
485 resume_execution(cur, regs);
487 preempt_enable_no_resched();
490 * if somebody else is singlestepping across a probe point, psw mask
491 * will have PER set, in which case, continue the remaining processing
492 * of do_single_step, as if this is not a probe hit.
494 if (regs->psw.mask & PSW_MASK_PER) {
501 static int __kprobes kprobe_trap_handler(struct pt_regs *regs, int trapnr)
503 struct kprobe *cur = kprobe_running();
504 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
505 const struct exception_table_entry *entry;
507 switch(kcb->kprobe_status) {
508 case KPROBE_SWAP_INST:
509 /* We are here because the instruction replacement failed */
514 * We are here because the instruction being single
515 * stepped caused a page fault. We reset the current
516 * kprobe and the nip points back to the probe address
517 * and allow the page fault handler to continue as a
520 disable_singlestep(kcb, regs, (unsigned long) cur->addr);
522 preempt_enable_no_resched();
524 case KPROBE_HIT_ACTIVE:
525 case KPROBE_HIT_SSDONE:
527 * We increment the nmissed count for accounting,
528 * we can also use npre/npostfault count for accouting
529 * these specific fault cases.
531 kprobes_inc_nmissed_count(cur);
534 * We come here because instructions in the pre/post
535 * handler caused the page_fault, this could happen
536 * if handler tries to access user space by
537 * copy_from_user(), get_user() etc. Let the
538 * user-specified handler try to fix it first.
540 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
544 * In case the user-specified fault handler returned
545 * zero, try to fix up.
547 entry = search_exception_tables(regs->psw.addr & PSW_ADDR_INSN);
549 regs->psw.addr = entry->fixup | PSW_ADDR_AMODE;
554 * fixup_exception() could not handle it,
555 * Let do_page_fault() fix it.
564 int __kprobes kprobe_fault_handler(struct pt_regs *regs, int trapnr)
568 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
570 ret = kprobe_trap_handler(regs, trapnr);
571 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
572 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
577 * Wrapper routine to for handling exceptions.
579 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
580 unsigned long val, void *data)
582 struct die_args *args = (struct die_args *)data;
583 struct pt_regs *regs = args->regs;
584 int ret = NOTIFY_DONE;
586 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
591 if (kprobe_handler(args->regs))
595 if (post_kprobe_handler(args->regs))
599 if (!preemptible() && kprobe_running() &&
600 kprobe_trap_handler(args->regs, args->trapnr))
607 if (regs->psw.mask & (PSW_MASK_IO | PSW_MASK_EXT))
608 local_irq_restore(regs->psw.mask & ~PSW_MASK_PER);
613 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
615 struct jprobe *jp = container_of(p, struct jprobe, kp);
616 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
619 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
621 /* setup return addr to the jprobe handler routine */
622 regs->psw.addr = (unsigned long)(jp->entry) | PSW_ADDR_AMODE;
623 regs->psw.mask &= ~(PSW_MASK_IO | PSW_MASK_EXT);
625 /* r15 is the stack pointer */
626 stack = (unsigned long) regs->gprs[15];
628 memcpy(kcb->jprobes_stack, (void *) stack, MIN_STACK_SIZE(stack));
632 void __kprobes jprobe_return(void)
634 asm volatile(".word 0x0002");
637 void __kprobes jprobe_return_end(void)
639 asm volatile("bcr 0,0");
642 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
644 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
647 stack = (unsigned long) kcb->jprobe_saved_regs.gprs[15];
649 /* Put the regs back */
650 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
651 /* put the stack back */
652 memcpy((void *) stack, kcb->jprobes_stack, MIN_STACK_SIZE(stack));
653 preempt_enable_no_resched();
657 static struct kprobe trampoline_p = {
658 .addr = (kprobe_opcode_t *) & kretprobe_trampoline,
659 .pre_handler = trampoline_probe_handler
662 int __init arch_init_kprobes(void)
664 return register_kprobe(&trampoline_p);
667 int __kprobes arch_trampoline_kprobe(struct kprobe *p)
669 if (p->addr == (kprobe_opcode_t *) & kretprobe_trampoline)