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-Nov Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port
29 #include <linux/kprobes.h>
30 #include <linux/ptrace.h>
31 #include <linux/preempt.h>
32 #include <linux/extable.h>
33 #include <linux/kdebug.h>
34 #include <linux/slab.h>
35 #include <asm/code-patching.h>
36 #include <asm/cacheflush.h>
37 #include <asm/sstep.h>
38 #include <asm/sections.h>
39 #include <linux/uaccess.h>
41 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
42 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
44 struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}};
46 int is_current_kprobe_addr(unsigned long addr)
48 struct kprobe *p = kprobe_running();
49 return (p && (unsigned long)p->addr == addr) ? 1 : 0;
52 bool arch_within_kprobe_blacklist(unsigned long addr)
54 return (addr >= (unsigned long)__kprobes_text_start &&
55 addr < (unsigned long)__kprobes_text_end) ||
56 (addr >= (unsigned long)_stext &&
57 addr < (unsigned long)__head_end);
60 kprobe_opcode_t *kprobe_lookup_name(const char *name, unsigned int offset)
62 kprobe_opcode_t *addr;
64 #ifdef PPC64_ELF_ABI_v2
65 /* PPC64 ABIv2 needs local entry point */
66 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
67 if (addr && !offset) {
68 #ifdef CONFIG_KPROBES_ON_FTRACE
71 * Per livepatch.h, ftrace location is always within the first
72 * 16 bytes of a function on powerpc with -mprofile-kernel.
74 faddr = ftrace_location_range((unsigned long)addr,
75 (unsigned long)addr + 16);
77 addr = (kprobe_opcode_t *)faddr;
80 addr = (kprobe_opcode_t *)ppc_function_entry(addr);
82 #elif defined(PPC64_ELF_ABI_v1)
84 * 64bit powerpc ABIv1 uses function descriptors:
85 * - Check for the dot variant of the symbol first.
86 * - If that fails, try looking up the symbol provided.
88 * This ensures we always get to the actual symbol and not
91 * Also handle <module:symbol> format.
93 char dot_name[MODULE_NAME_LEN + 1 + KSYM_NAME_LEN];
95 bool dot_appended = false;
96 if ((modsym = strchr(name, ':')) != NULL) {
98 if (*modsym != '\0' && *modsym != '.') {
99 /* Convert to <module:.symbol> */
100 strncpy(dot_name, name, modsym - name);
101 dot_name[modsym - name] = '.';
102 dot_name[modsym - name + 1] = '\0';
103 strncat(dot_name, modsym,
104 sizeof(dot_name) - (modsym - name) - 2);
108 strncat(dot_name, name, sizeof(dot_name) - 1);
110 } else if (name[0] != '.') {
113 strncat(dot_name, name, KSYM_NAME_LEN - 2);
117 strncat(dot_name, name, KSYM_NAME_LEN - 1);
119 addr = (kprobe_opcode_t *)kallsyms_lookup_name(dot_name);
120 if (!addr && dot_appended) {
121 /* Let's try the original non-dot symbol lookup */
122 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
125 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name);
131 int arch_prepare_kprobe(struct kprobe *p)
134 kprobe_opcode_t insn = *p->addr;
136 if ((unsigned long)p->addr & 0x03) {
137 printk("Attempt to register kprobe at an unaligned address\n");
139 } else if (IS_MTMSRD(insn) || IS_RFID(insn) || IS_RFI(insn)) {
140 printk("Cannot register a kprobe on rfi/rfid or mtmsr[d]\n");
144 /* insn must be on a special executable page on ppc64. This is
145 * not explicitly required on ppc32 (right now), but it doesn't hurt */
147 p->ainsn.insn = get_insn_slot();
153 memcpy(p->ainsn.insn, p->addr,
154 MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
155 p->opcode = *p->addr;
156 flush_icache_range((unsigned long)p->ainsn.insn,
157 (unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t));
160 p->ainsn.boostable = 0;
163 NOKPROBE_SYMBOL(arch_prepare_kprobe);
165 void arch_arm_kprobe(struct kprobe *p)
167 *p->addr = BREAKPOINT_INSTRUCTION;
168 flush_icache_range((unsigned long) p->addr,
169 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
171 NOKPROBE_SYMBOL(arch_arm_kprobe);
173 void arch_disarm_kprobe(struct kprobe *p)
175 *p->addr = p->opcode;
176 flush_icache_range((unsigned long) p->addr,
177 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
179 NOKPROBE_SYMBOL(arch_disarm_kprobe);
181 void arch_remove_kprobe(struct kprobe *p)
184 free_insn_slot(p->ainsn.insn, 0);
185 p->ainsn.insn = NULL;
188 NOKPROBE_SYMBOL(arch_remove_kprobe);
190 static nokprobe_inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
192 enable_single_step(regs);
195 * On powerpc we should single step on the original
196 * instruction even if the probed insn is a trap
197 * variant as values in regs could play a part in
198 * if the trap is taken or not
200 regs->nip = (unsigned long)p->ainsn.insn;
203 static nokprobe_inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
205 kcb->prev_kprobe.kp = kprobe_running();
206 kcb->prev_kprobe.status = kcb->kprobe_status;
207 kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr;
210 static nokprobe_inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
212 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp);
213 kcb->kprobe_status = kcb->prev_kprobe.status;
214 kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr;
217 static nokprobe_inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
218 struct kprobe_ctlblk *kcb)
220 __this_cpu_write(current_kprobe, p);
221 kcb->kprobe_saved_msr = regs->msr;
224 bool arch_function_offset_within_entry(unsigned long offset)
226 #ifdef PPC64_ELF_ABI_v2
227 #ifdef CONFIG_KPROBES_ON_FTRACE
237 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs)
239 ri->ret_addr = (kprobe_opcode_t *)regs->link;
241 /* Replace the return addr with trampoline addr */
242 regs->link = (unsigned long)kretprobe_trampoline;
244 NOKPROBE_SYMBOL(arch_prepare_kretprobe);
246 int try_to_emulate(struct kprobe *p, struct pt_regs *regs)
249 unsigned int insn = *p->ainsn.insn;
251 /* regs->nip is also adjusted if emulate_step returns 1 */
252 ret = emulate_step(regs, insn);
255 * Once this instruction has been boosted
256 * successfully, set the boostable flag
258 if (unlikely(p->ainsn.boostable == 0))
259 p->ainsn.boostable = 1;
260 } else if (ret < 0) {
262 * We don't allow kprobes on mtmsr(d)/rfi(d), etc.
263 * So, we should never get here... but, its still
264 * good to catch them, just in case...
266 printk("Can't step on instruction %x\n", insn);
269 /* This instruction can't be boosted */
270 p->ainsn.boostable = -1;
274 NOKPROBE_SYMBOL(try_to_emulate);
276 int kprobe_handler(struct pt_regs *regs)
280 unsigned int *addr = (unsigned int *)regs->nip;
281 struct kprobe_ctlblk *kcb;
287 * We don't want to be preempted for the entire
288 * duration of kprobe processing
291 kcb = get_kprobe_ctlblk();
293 /* Check we're not actually recursing */
294 if (kprobe_running()) {
295 p = get_kprobe(addr);
297 kprobe_opcode_t insn = *p->ainsn.insn;
298 if (kcb->kprobe_status == KPROBE_HIT_SS &&
300 /* Turn off 'trace' bits */
301 regs->msr &= ~MSR_SINGLESTEP;
302 regs->msr |= kcb->kprobe_saved_msr;
305 /* We have reentered the kprobe_handler(), since
306 * another probe was hit while within the handler.
307 * We here save the original kprobes variables and
308 * just single step on the instruction of the new probe
309 * without calling any user handlers.
311 save_previous_kprobe(kcb);
312 set_current_kprobe(p, regs, kcb);
313 kprobes_inc_nmissed_count(p);
314 kcb->kprobe_status = KPROBE_REENTER;
315 if (p->ainsn.boostable >= 0) {
316 ret = try_to_emulate(p, regs);
319 restore_previous_kprobe(kcb);
320 preempt_enable_no_resched();
324 prepare_singlestep(p, regs);
327 if (*addr != BREAKPOINT_INSTRUCTION) {
328 /* If trap variant, then it belongs not to us */
329 kprobe_opcode_t cur_insn = *addr;
330 if (is_trap(cur_insn))
332 /* The breakpoint instruction was removed by
333 * another cpu right after we hit, no further
334 * handling of this interrupt is appropriate
339 p = __this_cpu_read(current_kprobe);
340 if (p->break_handler && p->break_handler(p, regs)) {
341 if (!skip_singlestep(p, regs, kcb))
349 p = get_kprobe(addr);
351 if (*addr != BREAKPOINT_INSTRUCTION) {
353 * PowerPC has multiple variants of the "trap"
354 * instruction. If the current instruction is a
355 * trap variant, it could belong to someone else
357 kprobe_opcode_t cur_insn = *addr;
358 if (is_trap(cur_insn))
361 * The breakpoint instruction was removed right
362 * after we hit it. Another cpu has removed
363 * either a probepoint or a debugger breakpoint
364 * at this address. In either case, no further
365 * handling of this interrupt is appropriate.
369 /* Not one of ours: let kernel handle it */
373 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
374 set_current_kprobe(p, regs, kcb);
375 if (p->pre_handler && p->pre_handler(p, regs))
376 /* handler has already set things up, so skip ss setup */
380 if (p->ainsn.boostable >= 0) {
381 ret = try_to_emulate(p, regs);
385 p->post_handler(p, regs, 0);
387 kcb->kprobe_status = KPROBE_HIT_SSDONE;
388 reset_current_kprobe();
389 preempt_enable_no_resched();
393 prepare_singlestep(p, regs);
394 kcb->kprobe_status = KPROBE_HIT_SS;
398 preempt_enable_no_resched();
401 NOKPROBE_SYMBOL(kprobe_handler);
404 * Function return probe trampoline:
405 * - init_kprobes() establishes a probepoint here
406 * - When the probed function returns, this probe
407 * causes the handlers to fire
409 asm(".global kretprobe_trampoline\n"
410 ".type kretprobe_trampoline, @function\n"
411 "kretprobe_trampoline:\n"
414 ".size kretprobe_trampoline, .-kretprobe_trampoline\n");
417 * Called when the probe at kretprobe trampoline is hit
419 static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
421 struct kretprobe_instance *ri = NULL;
422 struct hlist_head *head, empty_rp;
423 struct hlist_node *tmp;
424 unsigned long flags, orig_ret_address = 0;
425 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
427 INIT_HLIST_HEAD(&empty_rp);
428 kretprobe_hash_lock(current, &head, &flags);
431 * It is possible to have multiple instances associated with a given
432 * task either because an multiple functions in the call path
433 * have a return probe installed on them, and/or more than one return
434 * return probe was registered for a target function.
436 * We can handle this because:
437 * - instances are always inserted at the head of the list
438 * - when multiple return probes are registered for the same
439 * function, the first instance's ret_addr will point to the
440 * real return address, and all the rest will point to
441 * kretprobe_trampoline
443 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
444 if (ri->task != current)
445 /* another task is sharing our hash bucket */
448 if (ri->rp && ri->rp->handler)
449 ri->rp->handler(ri, regs);
451 orig_ret_address = (unsigned long)ri->ret_addr;
452 recycle_rp_inst(ri, &empty_rp);
454 if (orig_ret_address != trampoline_address)
456 * This is the real return address. Any other
457 * instances associated with this task are for
458 * other calls deeper on the call stack
463 kretprobe_assert(ri, orig_ret_address, trampoline_address);
464 regs->nip = orig_ret_address;
466 * Make LR point to the orig_ret_address.
467 * When the 'nop' inside the kretprobe_trampoline
468 * is optimized, we can do a 'blr' after executing the
469 * detour buffer code.
471 regs->link = orig_ret_address;
473 reset_current_kprobe();
474 kretprobe_hash_unlock(current, &flags);
475 preempt_enable_no_resched();
477 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
478 hlist_del(&ri->hlist);
482 * By returning a non-zero value, we are telling
483 * kprobe_handler() that we don't want the post_handler
484 * to run (and have re-enabled preemption)
488 NOKPROBE_SYMBOL(trampoline_probe_handler);
491 * Called after single-stepping. p->addr is the address of the
492 * instruction whose first byte has been replaced by the "breakpoint"
493 * instruction. To avoid the SMP problems that can occur when we
494 * temporarily put back the original opcode to single-step, we
495 * single-stepped a copy of the instruction. The address of this
496 * copy is p->ainsn.insn.
498 int kprobe_post_handler(struct pt_regs *regs)
500 struct kprobe *cur = kprobe_running();
501 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
503 if (!cur || user_mode(regs))
506 /* make sure we got here for instruction we have a kprobe on */
507 if (((unsigned long)cur->ainsn.insn + 4) != regs->nip)
510 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
511 kcb->kprobe_status = KPROBE_HIT_SSDONE;
512 cur->post_handler(cur, regs, 0);
515 /* Adjust nip to after the single-stepped instruction */
516 regs->nip = (unsigned long)cur->addr + 4;
517 regs->msr |= kcb->kprobe_saved_msr;
519 /*Restore back the original saved kprobes variables and continue. */
520 if (kcb->kprobe_status == KPROBE_REENTER) {
521 restore_previous_kprobe(kcb);
524 reset_current_kprobe();
526 preempt_enable_no_resched();
529 * if somebody else is singlestepping across a probe point, msr
530 * will have DE/SE set, in which case, continue the remaining processing
531 * of do_debug, as if this is not a probe hit.
533 if (regs->msr & MSR_SINGLESTEP)
538 NOKPROBE_SYMBOL(kprobe_post_handler);
540 int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
542 struct kprobe *cur = kprobe_running();
543 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
544 const struct exception_table_entry *entry;
546 switch(kcb->kprobe_status) {
550 * We are here because the instruction being single
551 * stepped caused a page fault. We reset the current
552 * kprobe and the nip points back to the probe address
553 * and allow the page fault handler to continue as a
556 regs->nip = (unsigned long)cur->addr;
557 regs->msr &= ~MSR_SINGLESTEP; /* Turn off 'trace' bits */
558 regs->msr |= kcb->kprobe_saved_msr;
559 if (kcb->kprobe_status == KPROBE_REENTER)
560 restore_previous_kprobe(kcb);
562 reset_current_kprobe();
563 preempt_enable_no_resched();
565 case KPROBE_HIT_ACTIVE:
566 case KPROBE_HIT_SSDONE:
568 * We increment the nmissed count for accounting,
569 * we can also use npre/npostfault count for accounting
570 * these specific fault cases.
572 kprobes_inc_nmissed_count(cur);
575 * We come here because instructions in the pre/post
576 * handler caused the page_fault, this could happen
577 * if handler tries to access user space by
578 * copy_from_user(), get_user() etc. Let the
579 * user-specified handler try to fix it first.
581 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
585 * In case the user-specified fault handler returned
586 * zero, try to fix up.
588 if ((entry = search_exception_tables(regs->nip)) != NULL) {
589 regs->nip = extable_fixup(entry);
594 * fixup_exception() could not handle it,
595 * Let do_page_fault() fix it.
603 NOKPROBE_SYMBOL(kprobe_fault_handler);
605 unsigned long arch_deref_entry_point(void *entry)
607 return ppc_global_function_entry(entry);
609 NOKPROBE_SYMBOL(arch_deref_entry_point);
611 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
613 struct jprobe *jp = container_of(p, struct jprobe, kp);
614 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
616 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs));
618 /* setup return addr to the jprobe handler routine */
619 regs->nip = arch_deref_entry_point(jp->entry);
620 #ifdef PPC64_ELF_ABI_v2
621 regs->gpr[12] = (unsigned long)jp->entry;
622 #elif defined(PPC64_ELF_ABI_v1)
623 regs->gpr[2] = (unsigned long)(((func_descr_t *)jp->entry)->toc);
627 * jprobes use jprobe_return() which skips the normal return
628 * path of the function, and this messes up the accounting of the
629 * function graph tracer.
631 * Pause function graph tracing while performing the jprobe function.
633 pause_graph_tracing();
637 NOKPROBE_SYMBOL(setjmp_pre_handler);
639 void __used jprobe_return(void)
641 asm volatile("trap" ::: "memory");
643 NOKPROBE_SYMBOL(jprobe_return);
645 static void __used jprobe_return_end(void)
648 NOKPROBE_SYMBOL(jprobe_return_end);
650 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
652 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
655 * FIXME - we should ideally be validating that we got here 'cos
656 * of the "trap" in jprobe_return() above, before restoring the
659 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs));
660 /* It's OK to start function graph tracing again */
661 unpause_graph_tracing();
662 preempt_enable_no_resched();
665 NOKPROBE_SYMBOL(longjmp_break_handler);
667 static struct kprobe trampoline_p = {
668 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
669 .pre_handler = trampoline_probe_handler
672 int __init arch_init_kprobes(void)
674 return register_kprobe(&trampoline_p);
677 int arch_trampoline_kprobe(struct kprobe *p)
679 if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
684 NOKPROBE_SYMBOL(arch_trampoline_kprobe);