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 <asm-generic/sections.h>
41 #include <asm/cacheflush.h>
42 #include <asm/errno.h>
43 #include <asm/kdebug.h>
45 #define KPROBE_HASH_BITS 6
46 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
48 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
49 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
50 static atomic_t kprobe_count;
52 DEFINE_MUTEX(kprobe_mutex); /* Protects kprobe_table */
53 DEFINE_SPINLOCK(kretprobe_lock); /* Protects kretprobe_inst_table */
54 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
56 static struct notifier_block kprobe_page_fault_nb = {
57 .notifier_call = kprobe_exceptions_notify,
58 .priority = 0x7fffffff /* we need to notified first */
61 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
63 * kprobe->ainsn.insn points to the copy of the instruction to be
64 * single-stepped. x86_64, POWER4 and above have no-exec support and
65 * stepping on the instruction on a vmalloced/kmalloced/data page
66 * is a recipe for disaster
68 #define INSNS_PER_PAGE (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
70 struct kprobe_insn_page {
71 struct hlist_node hlist;
72 kprobe_opcode_t *insns; /* Page of instruction slots */
73 char slot_used[INSNS_PER_PAGE];
77 static struct hlist_head kprobe_insn_pages;
80 * get_insn_slot() - Find a slot on an executable page for an instruction.
81 * We allocate an executable page if there's no room on existing ones.
83 kprobe_opcode_t __kprobes *get_insn_slot(void)
85 struct kprobe_insn_page *kip;
86 struct hlist_node *pos;
88 hlist_for_each(pos, &kprobe_insn_pages) {
89 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
90 if (kip->nused < INSNS_PER_PAGE) {
92 for (i = 0; i < INSNS_PER_PAGE; i++) {
93 if (!kip->slot_used[i]) {
94 kip->slot_used[i] = 1;
96 return kip->insns + (i * MAX_INSN_SIZE);
99 /* Surprise! No unused slots. Fix kip->nused. */
100 kip->nused = INSNS_PER_PAGE;
104 /* All out of space. Need to allocate a new page. Use slot 0.*/
105 kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
111 * Use module_alloc so this page is within +/- 2GB of where the
112 * kernel image and loaded module images reside. This is required
113 * so x86_64 can correctly handle the %rip-relative fixups.
115 kip->insns = module_alloc(PAGE_SIZE);
120 INIT_HLIST_NODE(&kip->hlist);
121 hlist_add_head(&kip->hlist, &kprobe_insn_pages);
122 memset(kip->slot_used, 0, INSNS_PER_PAGE);
123 kip->slot_used[0] = 1;
128 void __kprobes free_insn_slot(kprobe_opcode_t *slot)
130 struct kprobe_insn_page *kip;
131 struct hlist_node *pos;
133 hlist_for_each(pos, &kprobe_insn_pages) {
134 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
135 if (kip->insns <= slot &&
136 slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
137 int i = (slot - kip->insns) / MAX_INSN_SIZE;
138 kip->slot_used[i] = 0;
140 if (kip->nused == 0) {
142 * Page is no longer in use. Free it unless
143 * it's the last one. We keep the last one
144 * so as not to have to set it up again the
145 * next time somebody inserts a probe.
147 hlist_del(&kip->hlist);
148 if (hlist_empty(&kprobe_insn_pages)) {
149 INIT_HLIST_NODE(&kip->hlist);
150 hlist_add_head(&kip->hlist,
153 module_free(NULL, kip->insns);
163 /* We have preemption disabled.. so it is safe to use __ versions */
164 static inline void set_kprobe_instance(struct kprobe *kp)
166 __get_cpu_var(kprobe_instance) = kp;
169 static inline void reset_kprobe_instance(void)
171 __get_cpu_var(kprobe_instance) = NULL;
175 * This routine is called either:
176 * - under the kprobe_mutex - during kprobe_[un]register()
178 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
180 struct kprobe __kprobes *get_kprobe(void *addr)
182 struct hlist_head *head;
183 struct hlist_node *node;
186 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
187 hlist_for_each_entry_rcu(p, node, head, hlist) {
195 * Aggregate handlers for multiple kprobes support - these handlers
196 * take care of invoking the individual kprobe handlers on p->list
198 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
202 list_for_each_entry_rcu(kp, &p->list, list) {
203 if (kp->pre_handler) {
204 set_kprobe_instance(kp);
205 if (kp->pre_handler(kp, regs))
208 reset_kprobe_instance();
213 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
218 list_for_each_entry_rcu(kp, &p->list, list) {
219 if (kp->post_handler) {
220 set_kprobe_instance(kp);
221 kp->post_handler(kp, regs, flags);
222 reset_kprobe_instance();
228 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
231 struct kprobe *cur = __get_cpu_var(kprobe_instance);
234 * if we faulted "during" the execution of a user specified
235 * probe handler, invoke just that probe's fault handler
237 if (cur && cur->fault_handler) {
238 if (cur->fault_handler(cur, regs, trapnr))
244 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
246 struct kprobe *cur = __get_cpu_var(kprobe_instance);
249 if (cur && cur->break_handler) {
250 if (cur->break_handler(cur, regs))
253 reset_kprobe_instance();
257 /* Walks the list and increments nmissed count for multiprobe case */
258 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
261 if (p->pre_handler != aggr_pre_handler) {
264 list_for_each_entry_rcu(kp, &p->list, list)
270 /* Called with kretprobe_lock held */
271 struct kretprobe_instance __kprobes *get_free_rp_inst(struct kretprobe *rp)
273 struct hlist_node *node;
274 struct kretprobe_instance *ri;
275 hlist_for_each_entry(ri, node, &rp->free_instances, uflist)
280 /* Called with kretprobe_lock held */
281 static struct kretprobe_instance __kprobes *get_used_rp_inst(struct kretprobe
284 struct hlist_node *node;
285 struct kretprobe_instance *ri;
286 hlist_for_each_entry(ri, node, &rp->used_instances, uflist)
291 /* Called with kretprobe_lock held */
292 void __kprobes add_rp_inst(struct kretprobe_instance *ri)
295 * Remove rp inst off the free list -
296 * Add it back when probed function returns
298 hlist_del(&ri->uflist);
300 /* Add rp inst onto table */
301 INIT_HLIST_NODE(&ri->hlist);
302 hlist_add_head(&ri->hlist,
303 &kretprobe_inst_table[hash_ptr(ri->task, KPROBE_HASH_BITS)]);
305 /* Also add this rp inst to the used list. */
306 INIT_HLIST_NODE(&ri->uflist);
307 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
310 /* Called with kretprobe_lock held */
311 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri)
313 /* remove rp inst off the rprobe_inst_table */
314 hlist_del(&ri->hlist);
316 /* remove rp inst off the used list */
317 hlist_del(&ri->uflist);
318 /* put rp inst back onto the free list */
319 INIT_HLIST_NODE(&ri->uflist);
320 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
326 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
328 return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
332 * This function is called from finish_task_switch when task tk becomes dead,
333 * so that we can recycle any function-return probe instances associated
334 * with this task. These left over instances represent probed functions
335 * that have been called but will never return.
337 void __kprobes kprobe_flush_task(struct task_struct *tk)
339 struct kretprobe_instance *ri;
340 struct hlist_head *head;
341 struct hlist_node *node, *tmp;
342 unsigned long flags = 0;
344 spin_lock_irqsave(&kretprobe_lock, flags);
345 head = kretprobe_inst_table_head(tk);
346 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
350 spin_unlock_irqrestore(&kretprobe_lock, flags);
353 static inline void free_rp_inst(struct kretprobe *rp)
355 struct kretprobe_instance *ri;
356 while ((ri = get_free_rp_inst(rp)) != NULL) {
357 hlist_del(&ri->uflist);
363 * Keep all fields in the kprobe consistent
365 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
367 memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
368 memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
372 * Add the new probe to old_p->list. Fail if this is the
373 * second jprobe at the address - two jprobes can't coexist
375 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
377 if (p->break_handler) {
378 if (old_p->break_handler)
380 list_add_tail_rcu(&p->list, &old_p->list);
381 old_p->break_handler = aggr_break_handler;
383 list_add_rcu(&p->list, &old_p->list);
384 if (p->post_handler && !old_p->post_handler)
385 old_p->post_handler = aggr_post_handler;
390 * Fill in the required fields of the "manager kprobe". Replace the
391 * earlier kprobe in the hlist with the manager kprobe
393 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
398 ap->pre_handler = aggr_pre_handler;
399 ap->fault_handler = aggr_fault_handler;
401 ap->post_handler = aggr_post_handler;
402 if (p->break_handler)
403 ap->break_handler = aggr_break_handler;
405 INIT_LIST_HEAD(&ap->list);
406 list_add_rcu(&p->list, &ap->list);
408 hlist_replace_rcu(&p->hlist, &ap->hlist);
412 * This is the second or subsequent kprobe at the address - handle
415 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
421 if (old_p->pre_handler == aggr_pre_handler) {
422 copy_kprobe(old_p, p);
423 ret = add_new_kprobe(old_p, p);
425 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
428 add_aggr_kprobe(ap, old_p);
430 ret = add_new_kprobe(ap, p);
435 static int __kprobes in_kprobes_functions(unsigned long addr)
437 if (addr >= (unsigned long)__kprobes_text_start
438 && addr < (unsigned long)__kprobes_text_end)
443 static int __kprobes __register_kprobe(struct kprobe *p,
444 unsigned long called_from)
447 struct kprobe *old_p;
448 struct module *probed_mod;
450 if ((!kernel_text_address((unsigned long) p->addr)) ||
451 in_kprobes_functions((unsigned long) p->addr))
454 p->mod_refcounted = 0;
455 /* Check are we probing a module */
456 if ((probed_mod = module_text_address((unsigned long) p->addr))) {
457 struct module *calling_mod = module_text_address(called_from);
458 /* We must allow modules to probe themself and
459 * in this case avoid incrementing the module refcount,
460 * so as to allow unloading of self probing modules.
462 if (calling_mod && (calling_mod != probed_mod)) {
463 if (unlikely(!try_module_get(probed_mod)))
465 p->mod_refcounted = 1;
471 mutex_lock(&kprobe_mutex);
472 old_p = get_kprobe(p->addr);
474 ret = register_aggr_kprobe(old_p, p);
476 atomic_inc(&kprobe_count);
480 if ((ret = arch_prepare_kprobe(p)) != 0)
483 INIT_HLIST_NODE(&p->hlist);
484 hlist_add_head_rcu(&p->hlist,
485 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
487 if (atomic_add_return(1, &kprobe_count) == \
488 (ARCH_INACTIVE_KPROBE_COUNT + 1))
489 register_page_fault_notifier(&kprobe_page_fault_nb);
494 mutex_unlock(&kprobe_mutex);
496 if (ret && probed_mod)
497 module_put(probed_mod);
501 int __kprobes register_kprobe(struct kprobe *p)
503 return __register_kprobe(p,
504 (unsigned long)__builtin_return_address(0));
507 void __kprobes unregister_kprobe(struct kprobe *p)
510 struct kprobe *old_p, *list_p;
513 mutex_lock(&kprobe_mutex);
514 old_p = get_kprobe(p->addr);
515 if (unlikely(!old_p)) {
516 mutex_unlock(&kprobe_mutex);
520 list_for_each_entry_rcu(list_p, &old_p->list, list)
522 /* kprobe p is a valid probe */
524 mutex_unlock(&kprobe_mutex);
528 if ((old_p == p) || ((old_p->pre_handler == aggr_pre_handler) &&
529 (p->list.next == &old_p->list) &&
530 (p->list.prev == &old_p->list))) {
531 /* Only probe on the hash list */
532 arch_disarm_kprobe(p);
533 hlist_del_rcu(&old_p->hlist);
536 list_del_rcu(&p->list);
540 mutex_unlock(&kprobe_mutex);
543 if (p->mod_refcounted &&
544 (mod = module_text_address((unsigned long)p->addr)))
549 list_del_rcu(&p->list);
552 arch_remove_kprobe(p);
554 mutex_lock(&kprobe_mutex);
555 if (p->break_handler)
556 old_p->break_handler = NULL;
557 if (p->post_handler){
558 list_for_each_entry_rcu(list_p, &old_p->list, list){
559 if (list_p->post_handler){
565 old_p->post_handler = NULL;
567 mutex_unlock(&kprobe_mutex);
570 /* Call unregister_page_fault_notifier()
571 * if no probes are active
573 mutex_lock(&kprobe_mutex);
574 if (atomic_add_return(-1, &kprobe_count) == \
575 ARCH_INACTIVE_KPROBE_COUNT)
576 unregister_page_fault_notifier(&kprobe_page_fault_nb);
577 mutex_unlock(&kprobe_mutex);
581 static struct notifier_block kprobe_exceptions_nb = {
582 .notifier_call = kprobe_exceptions_notify,
583 .priority = 0x7fffffff /* we need to be notified first */
587 int __kprobes register_jprobe(struct jprobe *jp)
589 /* Todo: Verify probepoint is a function entry point */
590 jp->kp.pre_handler = setjmp_pre_handler;
591 jp->kp.break_handler = longjmp_break_handler;
593 return __register_kprobe(&jp->kp,
594 (unsigned long)__builtin_return_address(0));
597 void __kprobes unregister_jprobe(struct jprobe *jp)
599 unregister_kprobe(&jp->kp);
602 #ifdef ARCH_SUPPORTS_KRETPROBES
605 * This kprobe pre_handler is registered with every kretprobe. When probe
606 * hits it will set up the return probe.
608 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
609 struct pt_regs *regs)
611 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
612 unsigned long flags = 0;
614 /*TODO: consider to only swap the RA after the last pre_handler fired */
615 spin_lock_irqsave(&kretprobe_lock, flags);
616 arch_prepare_kretprobe(rp, regs);
617 spin_unlock_irqrestore(&kretprobe_lock, flags);
621 int __kprobes register_kretprobe(struct kretprobe *rp)
624 struct kretprobe_instance *inst;
627 rp->kp.pre_handler = pre_handler_kretprobe;
628 rp->kp.post_handler = NULL;
629 rp->kp.fault_handler = NULL;
630 rp->kp.break_handler = NULL;
632 /* Pre-allocate memory for max kretprobe instances */
633 if (rp->maxactive <= 0) {
634 #ifdef CONFIG_PREEMPT
635 rp->maxactive = max(10, 2 * NR_CPUS);
637 rp->maxactive = NR_CPUS;
640 INIT_HLIST_HEAD(&rp->used_instances);
641 INIT_HLIST_HEAD(&rp->free_instances);
642 for (i = 0; i < rp->maxactive; i++) {
643 inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL);
648 INIT_HLIST_NODE(&inst->uflist);
649 hlist_add_head(&inst->uflist, &rp->free_instances);
653 /* Establish function entry probe point */
654 if ((ret = __register_kprobe(&rp->kp,
655 (unsigned long)__builtin_return_address(0))) != 0)
660 #else /* ARCH_SUPPORTS_KRETPROBES */
662 int __kprobes register_kretprobe(struct kretprobe *rp)
667 #endif /* ARCH_SUPPORTS_KRETPROBES */
669 void __kprobes unregister_kretprobe(struct kretprobe *rp)
672 struct kretprobe_instance *ri;
674 unregister_kprobe(&rp->kp);
676 spin_lock_irqsave(&kretprobe_lock, flags);
677 while ((ri = get_used_rp_inst(rp)) != NULL) {
679 hlist_del(&ri->uflist);
681 spin_unlock_irqrestore(&kretprobe_lock, flags);
685 static int __init init_kprobes(void)
689 /* FIXME allocate the probe table, currently defined statically */
690 /* initialize all list heads */
691 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
692 INIT_HLIST_HEAD(&kprobe_table[i]);
693 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
695 atomic_set(&kprobe_count, 0);
697 err = arch_init_kprobes();
699 err = register_die_notifier(&kprobe_exceptions_nb);
704 __initcall(init_kprobes);
706 EXPORT_SYMBOL_GPL(register_kprobe);
707 EXPORT_SYMBOL_GPL(unregister_kprobe);
708 EXPORT_SYMBOL_GPL(register_jprobe);
709 EXPORT_SYMBOL_GPL(unregister_jprobe);
710 EXPORT_SYMBOL_GPL(jprobe_return);
711 EXPORT_SYMBOL_GPL(register_kretprobe);
712 EXPORT_SYMBOL_GPL(unregister_kretprobe);