2 * Kernel Probes (KProbes)
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 suggestions from
24 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25 * hlists and exceptions notifier as suggested by Andi Kleen.
26 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27 * interface to access function arguments.
28 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29 * exceptions notifier to be first on the priority list.
30 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32 * <prasanna@in.ibm.com> added function-return probes.
34 #include <linux/kprobes.h>
35 #include <linux/hash.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/module.h>
39 #include <linux/moduleloader.h>
40 #include <linux/kallsyms.h>
41 #include <linux/freezer.h>
42 #include <asm-generic/sections.h>
43 #include <asm/cacheflush.h>
44 #include <asm/errno.h>
45 #include <asm/kdebug.h>
47 #define KPROBE_HASH_BITS 6
48 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
52 * Some oddball architectures like 64bit powerpc have function descriptors
53 * so this must be overridable.
55 #ifndef kprobe_lookup_name
56 #define kprobe_lookup_name(name, addr) \
57 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
60 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
61 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
62 static atomic_t kprobe_count;
64 DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */
65 DEFINE_SPINLOCK(kretprobe_lock); /* Protects kretprobe_inst_table */
66 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
68 static struct notifier_block kprobe_page_fault_nb = {
69 .notifier_call = kprobe_exceptions_notify,
70 .priority = 0x7fffffff /* we need to notified first */
73 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
75 * kprobe->ainsn.insn points to the copy of the instruction to be
76 * single-stepped. x86_64, POWER4 and above have no-exec support and
77 * stepping on the instruction on a vmalloced/kmalloced/data page
78 * is a recipe for disaster
80 #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
82 struct kprobe_insn_page {
83 struct hlist_node hlist;
84 kprobe_opcode_t *insns; /* Page of instruction slots */
85 char slot_used[INSNS_PER_PAGE];
90 static struct hlist_head kprobe_insn_pages;
91 static int kprobe_garbage_slots;
92 static int collect_garbage_slots(void);
94 static int __kprobes check_safety(void)
97 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
98 ret = freeze_processes();
100 struct task_struct *p, *q;
101 do_each_thread(p, q) {
102 if (p != current && p->state == TASK_RUNNING &&
104 printk("Check failed: %s is running\n",p->comm);
108 } while_each_thread(p, q);
119 * get_insn_slot() - Find a slot on an executable page for an instruction.
120 * We allocate an executable page if there's no room on existing ones.
122 kprobe_opcode_t __kprobes *get_insn_slot(void)
124 struct kprobe_insn_page *kip;
125 struct hlist_node *pos;
128 hlist_for_each(pos, &kprobe_insn_pages) {
129 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
130 if (kip->nused < INSNS_PER_PAGE) {
132 for (i = 0; i < INSNS_PER_PAGE; i++) {
133 if (!kip->slot_used[i]) {
134 kip->slot_used[i] = 1;
136 return kip->insns + (i * MAX_INSN_SIZE);
139 /* Surprise! No unused slots. Fix kip->nused. */
140 kip->nused = INSNS_PER_PAGE;
144 /* If there are any garbage slots, collect it and try again. */
145 if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
148 /* All out of space. Need to allocate a new page. Use slot 0. */
149 kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
155 * Use module_alloc so this page is within +/- 2GB of where the
156 * kernel image and loaded module images reside. This is required
157 * so x86_64 can correctly handle the %rip-relative fixups.
159 kip->insns = module_alloc(PAGE_SIZE);
164 INIT_HLIST_NODE(&kip->hlist);
165 hlist_add_head(&kip->hlist, &kprobe_insn_pages);
166 memset(kip->slot_used, 0, INSNS_PER_PAGE);
167 kip->slot_used[0] = 1;
173 /* Return 1 if all garbages are collected, otherwise 0. */
174 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
176 kip->slot_used[idx] = 0;
178 if (kip->nused == 0) {
180 * Page is no longer in use. Free it unless
181 * it's the last one. We keep the last one
182 * so as not to have to set it up again the
183 * next time somebody inserts a probe.
185 hlist_del(&kip->hlist);
186 if (hlist_empty(&kprobe_insn_pages)) {
187 INIT_HLIST_NODE(&kip->hlist);
188 hlist_add_head(&kip->hlist,
191 module_free(NULL, kip->insns);
199 static int __kprobes collect_garbage_slots(void)
201 struct kprobe_insn_page *kip;
202 struct hlist_node *pos, *next;
204 /* Ensure no-one is preepmted on the garbages */
205 if (check_safety() != 0)
208 hlist_for_each_safe(pos, next, &kprobe_insn_pages) {
210 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
211 if (kip->ngarbage == 0)
213 kip->ngarbage = 0; /* we will collect all garbages */
214 for (i = 0; i < INSNS_PER_PAGE; i++) {
215 if (kip->slot_used[i] == -1 &&
216 collect_one_slot(kip, i))
220 kprobe_garbage_slots = 0;
224 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
226 struct kprobe_insn_page *kip;
227 struct hlist_node *pos;
229 hlist_for_each(pos, &kprobe_insn_pages) {
230 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
231 if (kip->insns <= slot &&
232 slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
233 int i = (slot - kip->insns) / MAX_INSN_SIZE;
235 kip->slot_used[i] = -1;
238 collect_one_slot(kip, i);
243 if (dirty && (++kprobe_garbage_slots > INSNS_PER_PAGE)) {
244 collect_garbage_slots();
249 /* We have preemption disabled.. so it is safe to use __ versions */
250 static inline void set_kprobe_instance(struct kprobe *kp)
252 __get_cpu_var(kprobe_instance) = kp;
255 static inline void reset_kprobe_instance(void)
257 __get_cpu_var(kprobe_instance) = NULL;
261 * This routine is called either:
262 * - under the kprobe_mutex - during kprobe_[un]register()
264 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
266 struct kprobe __kprobes *get_kprobe(void *addr)
268 struct hlist_head *head;
269 struct hlist_node *node;
272 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
273 hlist_for_each_entry_rcu(p, node, head, hlist) {
281 * Aggregate handlers for multiple kprobes support - these handlers
282 * take care of invoking the individual kprobe handlers on p->list
284 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
288 list_for_each_entry_rcu(kp, &p->list, list) {
289 if (kp->pre_handler) {
290 set_kprobe_instance(kp);
291 if (kp->pre_handler(kp, regs))
294 reset_kprobe_instance();
299 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
304 list_for_each_entry_rcu(kp, &p->list, list) {
305 if (kp->post_handler) {
306 set_kprobe_instance(kp);
307 kp->post_handler(kp, regs, flags);
308 reset_kprobe_instance();
314 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
317 struct kprobe *cur = __get_cpu_var(kprobe_instance);
320 * if we faulted "during" the execution of a user specified
321 * probe handler, invoke just that probe's fault handler
323 if (cur && cur->fault_handler) {
324 if (cur->fault_handler(cur, regs, trapnr))
330 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
332 struct kprobe *cur = __get_cpu_var(kprobe_instance);
335 if (cur && cur->break_handler) {
336 if (cur->break_handler(cur, regs))
339 reset_kprobe_instance();
343 /* Walks the list and increments nmissed count for multiprobe case */
344 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
347 if (p->pre_handler != aggr_pre_handler) {
350 list_for_each_entry_rcu(kp, &p->list, list)
356 /* Called with kretprobe_lock held */
357 struct kretprobe_instance __kprobes *get_free_rp_inst(struct kretprobe *rp)
359 struct hlist_node *node;
360 struct kretprobe_instance *ri;
361 hlist_for_each_entry(ri, node, &rp->free_instances, uflist)
366 /* Called with kretprobe_lock held */
367 static struct kretprobe_instance __kprobes *get_used_rp_inst(struct kretprobe
370 struct hlist_node *node;
371 struct kretprobe_instance *ri;
372 hlist_for_each_entry(ri, node, &rp->used_instances, uflist)
377 /* Called with kretprobe_lock held */
378 void __kprobes add_rp_inst(struct kretprobe_instance *ri)
381 * Remove rp inst off the free list -
382 * Add it back when probed function returns
384 hlist_del(&ri->uflist);
386 /* Add rp inst onto table */
387 INIT_HLIST_NODE(&ri->hlist);
388 hlist_add_head(&ri->hlist,
389 &kretprobe_inst_table[hash_ptr(ri->task, KPROBE_HASH_BITS)]);
391 /* Also add this rp inst to the used list. */
392 INIT_HLIST_NODE(&ri->uflist);
393 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
396 /* Called with kretprobe_lock held */
397 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
398 struct hlist_head *head)
400 /* remove rp inst off the rprobe_inst_table */
401 hlist_del(&ri->hlist);
403 /* remove rp inst off the used list */
404 hlist_del(&ri->uflist);
405 /* put rp inst back onto the free list */
406 INIT_HLIST_NODE(&ri->uflist);
407 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
410 hlist_add_head(&ri->hlist, head);
413 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
415 return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
419 * This function is called from finish_task_switch when task tk becomes dead,
420 * so that we can recycle any function-return probe instances associated
421 * with this task. These left over instances represent probed functions
422 * that have been called but will never return.
424 void __kprobes kprobe_flush_task(struct task_struct *tk)
426 struct kretprobe_instance *ri;
427 struct hlist_head *head, empty_rp;
428 struct hlist_node *node, *tmp;
429 unsigned long flags = 0;
431 INIT_HLIST_HEAD(&empty_rp);
432 spin_lock_irqsave(&kretprobe_lock, flags);
433 head = kretprobe_inst_table_head(tk);
434 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
436 recycle_rp_inst(ri, &empty_rp);
438 spin_unlock_irqrestore(&kretprobe_lock, flags);
440 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
441 hlist_del(&ri->hlist);
446 static inline void free_rp_inst(struct kretprobe *rp)
448 struct kretprobe_instance *ri;
449 while ((ri = get_free_rp_inst(rp)) != NULL) {
450 hlist_del(&ri->uflist);
456 * Keep all fields in the kprobe consistent
458 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
460 memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
461 memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
465 * Add the new probe to old_p->list. Fail if this is the
466 * second jprobe at the address - two jprobes can't coexist
468 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
470 if (p->break_handler) {
471 if (old_p->break_handler)
473 list_add_tail_rcu(&p->list, &old_p->list);
474 old_p->break_handler = aggr_break_handler;
476 list_add_rcu(&p->list, &old_p->list);
477 if (p->post_handler && !old_p->post_handler)
478 old_p->post_handler = aggr_post_handler;
483 * Fill in the required fields of the "manager kprobe". Replace the
484 * earlier kprobe in the hlist with the manager kprobe
486 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
491 ap->pre_handler = aggr_pre_handler;
492 ap->fault_handler = aggr_fault_handler;
494 ap->post_handler = aggr_post_handler;
495 if (p->break_handler)
496 ap->break_handler = aggr_break_handler;
498 INIT_LIST_HEAD(&ap->list);
499 list_add_rcu(&p->list, &ap->list);
501 hlist_replace_rcu(&p->hlist, &ap->hlist);
505 * This is the second or subsequent kprobe at the address - handle
508 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
514 if (old_p->pre_handler == aggr_pre_handler) {
515 copy_kprobe(old_p, p);
516 ret = add_new_kprobe(old_p, p);
518 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
521 add_aggr_kprobe(ap, old_p);
523 ret = add_new_kprobe(ap, p);
528 static int __kprobes in_kprobes_functions(unsigned long addr)
530 if (addr >= (unsigned long)__kprobes_text_start
531 && addr < (unsigned long)__kprobes_text_end)
536 static int __kprobes __register_kprobe(struct kprobe *p,
537 unsigned long called_from)
540 struct kprobe *old_p;
541 struct module *probed_mod;
544 * If we have a symbol_name argument look it up,
545 * and add it to the address. That way the addr
546 * field can either be global or relative to a symbol.
548 if (p->symbol_name) {
551 kprobe_lookup_name(p->symbol_name, p->addr);
556 p->addr = (kprobe_opcode_t *)(((char *)p->addr)+ p->offset);
558 if ((!kernel_text_address((unsigned long) p->addr)) ||
559 in_kprobes_functions((unsigned long) p->addr))
562 p->mod_refcounted = 0;
563 /* Check are we probing a module */
564 if ((probed_mod = module_text_address((unsigned long) p->addr))) {
565 struct module *calling_mod = module_text_address(called_from);
566 /* We must allow modules to probe themself and
567 * in this case avoid incrementing the module refcount,
568 * so as to allow unloading of self probing modules.
570 if (calling_mod && (calling_mod != probed_mod)) {
571 if (unlikely(!try_module_get(probed_mod)))
573 p->mod_refcounted = 1;
579 mutex_lock(&kprobe_mutex);
580 old_p = get_kprobe(p->addr);
582 ret = register_aggr_kprobe(old_p, p);
584 atomic_inc(&kprobe_count);
588 if ((ret = arch_prepare_kprobe(p)) != 0)
591 INIT_HLIST_NODE(&p->hlist);
592 hlist_add_head_rcu(&p->hlist,
593 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
595 if (atomic_add_return(1, &kprobe_count) == \
596 (ARCH_INACTIVE_KPROBE_COUNT + 1))
597 register_page_fault_notifier(&kprobe_page_fault_nb);
602 mutex_unlock(&kprobe_mutex);
604 if (ret && probed_mod)
605 module_put(probed_mod);
609 int __kprobes register_kprobe(struct kprobe *p)
611 return __register_kprobe(p,
612 (unsigned long)__builtin_return_address(0));
615 void __kprobes unregister_kprobe(struct kprobe *p)
618 struct kprobe *old_p, *list_p;
621 mutex_lock(&kprobe_mutex);
622 old_p = get_kprobe(p->addr);
623 if (unlikely(!old_p)) {
624 mutex_unlock(&kprobe_mutex);
628 list_for_each_entry_rcu(list_p, &old_p->list, list)
630 /* kprobe p is a valid probe */
632 mutex_unlock(&kprobe_mutex);
636 if ((old_p == p) || ((old_p->pre_handler == aggr_pre_handler) &&
637 (p->list.next == &old_p->list) &&
638 (p->list.prev == &old_p->list))) {
639 /* Only probe on the hash list */
640 arch_disarm_kprobe(p);
641 hlist_del_rcu(&old_p->hlist);
644 list_del_rcu(&p->list);
648 mutex_unlock(&kprobe_mutex);
651 if (p->mod_refcounted &&
652 (mod = module_text_address((unsigned long)p->addr)))
657 list_del_rcu(&p->list);
660 arch_remove_kprobe(p);
662 mutex_lock(&kprobe_mutex);
663 if (p->break_handler)
664 old_p->break_handler = NULL;
665 if (p->post_handler){
666 list_for_each_entry_rcu(list_p, &old_p->list, list){
667 if (list_p->post_handler){
673 old_p->post_handler = NULL;
675 mutex_unlock(&kprobe_mutex);
678 /* Call unregister_page_fault_notifier()
679 * if no probes are active
681 mutex_lock(&kprobe_mutex);
682 if (atomic_add_return(-1, &kprobe_count) == \
683 ARCH_INACTIVE_KPROBE_COUNT)
684 unregister_page_fault_notifier(&kprobe_page_fault_nb);
685 mutex_unlock(&kprobe_mutex);
689 static struct notifier_block kprobe_exceptions_nb = {
690 .notifier_call = kprobe_exceptions_notify,
691 .priority = 0x7fffffff /* we need to be notified first */
695 int __kprobes register_jprobe(struct jprobe *jp)
697 /* Todo: Verify probepoint is a function entry point */
698 jp->kp.pre_handler = setjmp_pre_handler;
699 jp->kp.break_handler = longjmp_break_handler;
701 return __register_kprobe(&jp->kp,
702 (unsigned long)__builtin_return_address(0));
705 void __kprobes unregister_jprobe(struct jprobe *jp)
707 unregister_kprobe(&jp->kp);
710 #ifdef ARCH_SUPPORTS_KRETPROBES
713 * This kprobe pre_handler is registered with every kretprobe. When probe
714 * hits it will set up the return probe.
716 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
717 struct pt_regs *regs)
719 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
720 unsigned long flags = 0;
722 /*TODO: consider to only swap the RA after the last pre_handler fired */
723 spin_lock_irqsave(&kretprobe_lock, flags);
724 arch_prepare_kretprobe(rp, regs);
725 spin_unlock_irqrestore(&kretprobe_lock, flags);
729 int __kprobes register_kretprobe(struct kretprobe *rp)
732 struct kretprobe_instance *inst;
735 rp->kp.pre_handler = pre_handler_kretprobe;
736 rp->kp.post_handler = NULL;
737 rp->kp.fault_handler = NULL;
738 rp->kp.break_handler = NULL;
740 /* Pre-allocate memory for max kretprobe instances */
741 if (rp->maxactive <= 0) {
742 #ifdef CONFIG_PREEMPT
743 rp->maxactive = max(10, 2 * NR_CPUS);
745 rp->maxactive = NR_CPUS;
748 INIT_HLIST_HEAD(&rp->used_instances);
749 INIT_HLIST_HEAD(&rp->free_instances);
750 for (i = 0; i < rp->maxactive; i++) {
751 inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL);
756 INIT_HLIST_NODE(&inst->uflist);
757 hlist_add_head(&inst->uflist, &rp->free_instances);
761 /* Establish function entry probe point */
762 if ((ret = __register_kprobe(&rp->kp,
763 (unsigned long)__builtin_return_address(0))) != 0)
768 #else /* ARCH_SUPPORTS_KRETPROBES */
770 int __kprobes register_kretprobe(struct kretprobe *rp)
775 #endif /* ARCH_SUPPORTS_KRETPROBES */
777 void __kprobes unregister_kretprobe(struct kretprobe *rp)
780 struct kretprobe_instance *ri;
782 unregister_kprobe(&rp->kp);
784 spin_lock_irqsave(&kretprobe_lock, flags);
785 while ((ri = get_used_rp_inst(rp)) != NULL) {
787 hlist_del(&ri->uflist);
789 spin_unlock_irqrestore(&kretprobe_lock, flags);
793 static int __init init_kprobes(void)
797 /* FIXME allocate the probe table, currently defined statically */
798 /* initialize all list heads */
799 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
800 INIT_HLIST_HEAD(&kprobe_table[i]);
801 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
803 atomic_set(&kprobe_count, 0);
805 err = arch_init_kprobes();
807 err = register_die_notifier(&kprobe_exceptions_nb);
812 __initcall(init_kprobes);
814 EXPORT_SYMBOL_GPL(register_kprobe);
815 EXPORT_SYMBOL_GPL(unregister_kprobe);
816 EXPORT_SYMBOL_GPL(register_jprobe);
817 EXPORT_SYMBOL_GPL(unregister_jprobe);
818 EXPORT_SYMBOL_GPL(jprobe_return);
819 EXPORT_SYMBOL_GPL(register_kretprobe);
820 EXPORT_SYMBOL_GPL(unregister_kretprobe);