1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Kernel Probes (KProbes)
5 * Copyright (C) IBM Corporation, 2002, 2004
7 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
8 * Probes initial implementation (includes suggestions from
10 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
11 * hlists and exceptions notifier as suggested by Andi Kleen.
12 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
13 * interface to access function arguments.
14 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
15 * exceptions notifier to be first on the priority list.
16 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
17 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
18 * <prasanna@in.ibm.com> added function-return probes.
21 #define pr_fmt(fmt) "kprobes: " fmt
23 #include <linux/kprobes.h>
24 #include <linux/hash.h>
25 #include <linux/init.h>
26 #include <linux/slab.h>
27 #include <linux/stddef.h>
28 #include <linux/export.h>
29 #include <linux/moduleloader.h>
30 #include <linux/kallsyms.h>
31 #include <linux/freezer.h>
32 #include <linux/seq_file.h>
33 #include <linux/debugfs.h>
34 #include <linux/sysctl.h>
35 #include <linux/kdebug.h>
36 #include <linux/memory.h>
37 #include <linux/ftrace.h>
38 #include <linux/cpu.h>
39 #include <linux/jump_label.h>
40 #include <linux/static_call.h>
41 #include <linux/perf_event.h>
43 #include <asm/sections.h>
44 #include <asm/cacheflush.h>
45 #include <asm/errno.h>
46 #include <linux/uaccess.h>
48 #define KPROBE_HASH_BITS 6
49 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
51 #if !defined(CONFIG_OPTPROBES) || !defined(CONFIG_SYSCTL)
52 #define kprobe_sysctls_init() do { } while (0)
55 static int kprobes_initialized;
56 /* kprobe_table can be accessed by
57 * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
59 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
61 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
63 /* NOTE: change this value only with 'kprobe_mutex' held */
64 static bool kprobes_all_disarmed;
66 /* This protects 'kprobe_table' and 'optimizing_list' */
67 static DEFINE_MUTEX(kprobe_mutex);
68 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
70 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
71 unsigned int __unused)
73 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
77 * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
78 * kprobes can not probe.
80 static LIST_HEAD(kprobe_blacklist);
82 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
84 * 'kprobe::ainsn.insn' points to the copy of the instruction to be
85 * single-stepped. x86_64, POWER4 and above have no-exec support and
86 * stepping on the instruction on a vmalloced/kmalloced/data page
87 * is a recipe for disaster
89 struct kprobe_insn_page {
90 struct list_head list;
91 kprobe_opcode_t *insns; /* Page of instruction slots */
92 struct kprobe_insn_cache *cache;
98 #define KPROBE_INSN_PAGE_SIZE(slots) \
99 (offsetof(struct kprobe_insn_page, slot_used) + \
100 (sizeof(char) * (slots)))
102 static int slots_per_page(struct kprobe_insn_cache *c)
104 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
107 enum kprobe_slot_state {
113 void __weak *alloc_insn_page(void)
116 * Use module_alloc() so this page is within +/- 2GB of where the
117 * kernel image and loaded module images reside. This is required
118 * for most of the architectures.
119 * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
121 return module_alloc(PAGE_SIZE);
124 static void free_insn_page(void *page)
126 module_memfree(page);
129 struct kprobe_insn_cache kprobe_insn_slots = {
130 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
131 .alloc = alloc_insn_page,
132 .free = free_insn_page,
133 .sym = KPROBE_INSN_PAGE_SYM,
134 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
135 .insn_size = MAX_INSN_SIZE,
138 static int collect_garbage_slots(struct kprobe_insn_cache *c);
141 * __get_insn_slot() - Find a slot on an executable page for an instruction.
142 * We allocate an executable page if there's no room on existing ones.
144 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
146 struct kprobe_insn_page *kip;
147 kprobe_opcode_t *slot = NULL;
149 /* Since the slot array is not protected by rcu, we need a mutex */
150 mutex_lock(&c->mutex);
153 list_for_each_entry_rcu(kip, &c->pages, list) {
154 if (kip->nused < slots_per_page(c)) {
157 for (i = 0; i < slots_per_page(c); i++) {
158 if (kip->slot_used[i] == SLOT_CLEAN) {
159 kip->slot_used[i] = SLOT_USED;
161 slot = kip->insns + (i * c->insn_size);
166 /* kip->nused is broken. Fix it. */
167 kip->nused = slots_per_page(c);
173 /* If there are any garbage slots, collect it and try again. */
174 if (c->nr_garbage && collect_garbage_slots(c) == 0)
177 /* All out of space. Need to allocate a new page. */
178 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
182 kip->insns = c->alloc();
187 INIT_LIST_HEAD(&kip->list);
188 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
189 kip->slot_used[0] = SLOT_USED;
193 list_add_rcu(&kip->list, &c->pages);
196 /* Record the perf ksymbol register event after adding the page */
197 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
198 PAGE_SIZE, false, c->sym);
200 mutex_unlock(&c->mutex);
204 /* Return true if all garbages are collected, otherwise false. */
205 static bool collect_one_slot(struct kprobe_insn_page *kip, int idx)
207 kip->slot_used[idx] = SLOT_CLEAN;
209 if (kip->nused == 0) {
211 * Page is no longer in use. Free it unless
212 * it's the last one. We keep the last one
213 * so as not to have to set it up again the
214 * next time somebody inserts a probe.
216 if (!list_is_singular(&kip->list)) {
218 * Record perf ksymbol unregister event before removing
221 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
222 (unsigned long)kip->insns, PAGE_SIZE, true,
224 list_del_rcu(&kip->list);
226 kip->cache->free(kip->insns);
234 static int collect_garbage_slots(struct kprobe_insn_cache *c)
236 struct kprobe_insn_page *kip, *next;
238 /* Ensure no-one is interrupted on the garbages */
241 list_for_each_entry_safe(kip, next, &c->pages, list) {
244 if (kip->ngarbage == 0)
246 kip->ngarbage = 0; /* we will collect all garbages */
247 for (i = 0; i < slots_per_page(c); i++) {
248 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
256 void __free_insn_slot(struct kprobe_insn_cache *c,
257 kprobe_opcode_t *slot, int dirty)
259 struct kprobe_insn_page *kip;
262 mutex_lock(&c->mutex);
264 list_for_each_entry_rcu(kip, &c->pages, list) {
265 idx = ((long)slot - (long)kip->insns) /
266 (c->insn_size * sizeof(kprobe_opcode_t));
267 if (idx >= 0 && idx < slots_per_page(c))
270 /* Could not find this slot. */
275 /* Mark and sweep: this may sleep */
277 /* Check double free */
278 WARN_ON(kip->slot_used[idx] != SLOT_USED);
280 kip->slot_used[idx] = SLOT_DIRTY;
282 if (++c->nr_garbage > slots_per_page(c))
283 collect_garbage_slots(c);
285 collect_one_slot(kip, idx);
288 mutex_unlock(&c->mutex);
292 * Check given address is on the page of kprobe instruction slots.
293 * This will be used for checking whether the address on a stack
294 * is on a text area or not.
296 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
298 struct kprobe_insn_page *kip;
302 list_for_each_entry_rcu(kip, &c->pages, list) {
303 if (addr >= (unsigned long)kip->insns &&
304 addr < (unsigned long)kip->insns + PAGE_SIZE) {
314 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
315 unsigned long *value, char *type, char *sym)
317 struct kprobe_insn_page *kip;
321 list_for_each_entry_rcu(kip, &c->pages, list) {
324 strscpy(sym, c->sym, KSYM_NAME_LEN);
326 *value = (unsigned long)kip->insns;
335 #ifdef CONFIG_OPTPROBES
336 void __weak *alloc_optinsn_page(void)
338 return alloc_insn_page();
341 void __weak free_optinsn_page(void *page)
343 free_insn_page(page);
346 /* For optimized_kprobe buffer */
347 struct kprobe_insn_cache kprobe_optinsn_slots = {
348 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
349 .alloc = alloc_optinsn_page,
350 .free = free_optinsn_page,
351 .sym = KPROBE_OPTINSN_PAGE_SYM,
352 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
353 /* .insn_size is initialized later */
359 /* We have preemption disabled.. so it is safe to use __ versions */
360 static inline void set_kprobe_instance(struct kprobe *kp)
362 __this_cpu_write(kprobe_instance, kp);
365 static inline void reset_kprobe_instance(void)
367 __this_cpu_write(kprobe_instance, NULL);
371 * This routine is called either:
372 * - under the 'kprobe_mutex' - during kprobe_[un]register().
374 * - with preemption disabled - from architecture specific code.
376 struct kprobe *get_kprobe(void *addr)
378 struct hlist_head *head;
381 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
382 hlist_for_each_entry_rcu(p, head, hlist,
383 lockdep_is_held(&kprobe_mutex)) {
390 NOKPROBE_SYMBOL(get_kprobe);
392 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
394 /* Return true if 'p' is an aggregator */
395 static inline bool kprobe_aggrprobe(struct kprobe *p)
397 return p->pre_handler == aggr_pre_handler;
400 /* Return true if 'p' is unused */
401 static inline bool kprobe_unused(struct kprobe *p)
403 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
404 list_empty(&p->list);
407 /* Keep all fields in the kprobe consistent. */
408 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
410 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
411 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
414 #ifdef CONFIG_OPTPROBES
415 /* NOTE: This is protected by 'kprobe_mutex'. */
416 static bool kprobes_allow_optimization;
419 * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
420 * This must be called from arch-dep optimized caller.
422 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
426 list_for_each_entry_rcu(kp, &p->list, list) {
427 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
428 set_kprobe_instance(kp);
429 kp->pre_handler(kp, regs);
431 reset_kprobe_instance();
434 NOKPROBE_SYMBOL(opt_pre_handler);
436 /* Free optimized instructions and optimized_kprobe */
437 static void free_aggr_kprobe(struct kprobe *p)
439 struct optimized_kprobe *op;
441 op = container_of(p, struct optimized_kprobe, kp);
442 arch_remove_optimized_kprobe(op);
443 arch_remove_kprobe(p);
447 /* Return true if the kprobe is ready for optimization. */
448 static inline int kprobe_optready(struct kprobe *p)
450 struct optimized_kprobe *op;
452 if (kprobe_aggrprobe(p)) {
453 op = container_of(p, struct optimized_kprobe, kp);
454 return arch_prepared_optinsn(&op->optinsn);
460 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
461 bool kprobe_disarmed(struct kprobe *p)
463 struct optimized_kprobe *op;
465 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
466 if (!kprobe_aggrprobe(p))
467 return kprobe_disabled(p);
469 op = container_of(p, struct optimized_kprobe, kp);
471 return kprobe_disabled(p) && list_empty(&op->list);
474 /* Return true if the probe is queued on (un)optimizing lists */
475 static bool kprobe_queued(struct kprobe *p)
477 struct optimized_kprobe *op;
479 if (kprobe_aggrprobe(p)) {
480 op = container_of(p, struct optimized_kprobe, kp);
481 if (!list_empty(&op->list))
488 * Return an optimized kprobe whose optimizing code replaces
489 * instructions including 'addr' (exclude breakpoint).
491 static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
494 struct kprobe *p = NULL;
495 struct optimized_kprobe *op;
497 /* Don't check i == 0, since that is a breakpoint case. */
498 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
499 p = get_kprobe(addr - i);
501 if (p && kprobe_optready(p)) {
502 op = container_of(p, struct optimized_kprobe, kp);
503 if (arch_within_optimized_kprobe(op, addr))
510 /* Optimization staging list, protected by 'kprobe_mutex' */
511 static LIST_HEAD(optimizing_list);
512 static LIST_HEAD(unoptimizing_list);
513 static LIST_HEAD(freeing_list);
515 static void kprobe_optimizer(struct work_struct *work);
516 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
517 #define OPTIMIZE_DELAY 5
520 * Optimize (replace a breakpoint with a jump) kprobes listed on
523 static void do_optimize_kprobes(void)
525 lockdep_assert_held(&text_mutex);
527 * The optimization/unoptimization refers 'online_cpus' via
528 * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
529 * And same time, 'text_mutex' will be held in cpu-hotplug and here.
530 * This combination can cause a deadlock (cpu-hotplug tries to lock
531 * 'text_mutex' but stop_machine() can not be done because
532 * the 'online_cpus' has been changed)
533 * To avoid this deadlock, caller must have locked cpu-hotplug
534 * for preventing cpu-hotplug outside of 'text_mutex' locking.
536 lockdep_assert_cpus_held();
538 /* Optimization never be done when disarmed */
539 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
540 list_empty(&optimizing_list))
543 arch_optimize_kprobes(&optimizing_list);
547 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
548 * if need) kprobes listed on 'unoptimizing_list'.
550 static void do_unoptimize_kprobes(void)
552 struct optimized_kprobe *op, *tmp;
554 lockdep_assert_held(&text_mutex);
555 /* See comment in do_optimize_kprobes() */
556 lockdep_assert_cpus_held();
558 if (!list_empty(&unoptimizing_list))
559 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
561 /* Loop on 'freeing_list' for disarming and removing from kprobe hash list */
562 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
563 /* Switching from detour code to origin */
564 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
565 /* Disarm probes if marked disabled and not gone */
566 if (kprobe_disabled(&op->kp) && !kprobe_gone(&op->kp))
567 arch_disarm_kprobe(&op->kp);
568 if (kprobe_unused(&op->kp)) {
570 * Remove unused probes from hash list. After waiting
571 * for synchronization, these probes are reclaimed.
572 * (reclaiming is done by do_free_cleaned_kprobes().)
574 hlist_del_rcu(&op->kp.hlist);
576 list_del_init(&op->list);
580 /* Reclaim all kprobes on the 'freeing_list' */
581 static void do_free_cleaned_kprobes(void)
583 struct optimized_kprobe *op, *tmp;
585 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
586 list_del_init(&op->list);
587 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
589 * This must not happen, but if there is a kprobe
590 * still in use, keep it on kprobes hash list.
594 free_aggr_kprobe(&op->kp);
598 /* Start optimizer after OPTIMIZE_DELAY passed */
599 static void kick_kprobe_optimizer(void)
601 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
604 /* Kprobe jump optimizer */
605 static void kprobe_optimizer(struct work_struct *work)
607 mutex_lock(&kprobe_mutex);
609 mutex_lock(&text_mutex);
612 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
613 * kprobes before waiting for quiesence period.
615 do_unoptimize_kprobes();
618 * Step 2: Wait for quiesence period to ensure all potentially
619 * preempted tasks to have normally scheduled. Because optprobe
620 * may modify multiple instructions, there is a chance that Nth
621 * instruction is preempted. In that case, such tasks can return
622 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
623 * Note that on non-preemptive kernel, this is transparently converted
624 * to synchronoze_sched() to wait for all interrupts to have completed.
626 synchronize_rcu_tasks();
628 /* Step 3: Optimize kprobes after quiesence period */
629 do_optimize_kprobes();
631 /* Step 4: Free cleaned kprobes after quiesence period */
632 do_free_cleaned_kprobes();
634 mutex_unlock(&text_mutex);
637 /* Step 5: Kick optimizer again if needed */
638 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
639 kick_kprobe_optimizer();
641 mutex_unlock(&kprobe_mutex);
644 /* Wait for completing optimization and unoptimization */
645 void wait_for_kprobe_optimizer(void)
647 mutex_lock(&kprobe_mutex);
649 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
650 mutex_unlock(&kprobe_mutex);
652 /* This will also make 'optimizing_work' execute immmediately */
653 flush_delayed_work(&optimizing_work);
654 /* 'optimizing_work' might not have been queued yet, relax */
657 mutex_lock(&kprobe_mutex);
660 mutex_unlock(&kprobe_mutex);
663 bool optprobe_queued_unopt(struct optimized_kprobe *op)
665 struct optimized_kprobe *_op;
667 list_for_each_entry(_op, &unoptimizing_list, list) {
675 /* Optimize kprobe if p is ready to be optimized */
676 static void optimize_kprobe(struct kprobe *p)
678 struct optimized_kprobe *op;
680 /* Check if the kprobe is disabled or not ready for optimization. */
681 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
682 (kprobe_disabled(p) || kprobes_all_disarmed))
685 /* kprobes with 'post_handler' can not be optimized */
689 op = container_of(p, struct optimized_kprobe, kp);
691 /* Check there is no other kprobes at the optimized instructions */
692 if (arch_check_optimized_kprobe(op) < 0)
695 /* Check if it is already optimized. */
696 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
697 if (optprobe_queued_unopt(op)) {
698 /* This is under unoptimizing. Just dequeue the probe */
699 list_del_init(&op->list);
703 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
706 * On the 'unoptimizing_list' and 'optimizing_list',
707 * 'op' must have OPTIMIZED flag
709 if (WARN_ON_ONCE(!list_empty(&op->list)))
712 list_add(&op->list, &optimizing_list);
713 kick_kprobe_optimizer();
716 /* Short cut to direct unoptimizing */
717 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
719 lockdep_assert_cpus_held();
720 arch_unoptimize_kprobe(op);
721 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
724 /* Unoptimize a kprobe if p is optimized */
725 static void unoptimize_kprobe(struct kprobe *p, bool force)
727 struct optimized_kprobe *op;
729 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
730 return; /* This is not an optprobe nor optimized */
732 op = container_of(p, struct optimized_kprobe, kp);
733 if (!kprobe_optimized(p))
736 if (!list_empty(&op->list)) {
737 if (optprobe_queued_unopt(op)) {
738 /* Queued in unoptimizing queue */
741 * Forcibly unoptimize the kprobe here, and queue it
742 * in the freeing list for release afterwards.
744 force_unoptimize_kprobe(op);
745 list_move(&op->list, &freeing_list);
748 /* Dequeue from the optimizing queue */
749 list_del_init(&op->list);
750 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
755 /* Optimized kprobe case */
757 /* Forcibly update the code: this is a special case */
758 force_unoptimize_kprobe(op);
760 list_add(&op->list, &unoptimizing_list);
761 kick_kprobe_optimizer();
765 /* Cancel unoptimizing for reusing */
766 static int reuse_unused_kprobe(struct kprobe *ap)
768 struct optimized_kprobe *op;
771 * Unused kprobe MUST be on the way of delayed unoptimizing (means
772 * there is still a relative jump) and disabled.
774 op = container_of(ap, struct optimized_kprobe, kp);
775 WARN_ON_ONCE(list_empty(&op->list));
776 /* Enable the probe again */
777 ap->flags &= ~KPROBE_FLAG_DISABLED;
778 /* Optimize it again. (remove from 'op->list') */
779 if (!kprobe_optready(ap))
786 /* Remove optimized instructions */
787 static void kill_optimized_kprobe(struct kprobe *p)
789 struct optimized_kprobe *op;
791 op = container_of(p, struct optimized_kprobe, kp);
792 if (!list_empty(&op->list))
793 /* Dequeue from the (un)optimization queue */
794 list_del_init(&op->list);
795 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
797 if (kprobe_unused(p)) {
799 * Unused kprobe is on unoptimizing or freeing list. We move it
800 * to freeing_list and let the kprobe_optimizer() remove it from
801 * the kprobe hash list and free it.
803 if (optprobe_queued_unopt(op))
804 list_move(&op->list, &freeing_list);
807 /* Don't touch the code, because it is already freed. */
808 arch_remove_optimized_kprobe(op);
812 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
814 if (!kprobe_ftrace(p))
815 arch_prepare_optimized_kprobe(op, p);
818 /* Try to prepare optimized instructions */
819 static void prepare_optimized_kprobe(struct kprobe *p)
821 struct optimized_kprobe *op;
823 op = container_of(p, struct optimized_kprobe, kp);
824 __prepare_optimized_kprobe(op, p);
827 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
828 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
830 struct optimized_kprobe *op;
832 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
836 INIT_LIST_HEAD(&op->list);
837 op->kp.addr = p->addr;
838 __prepare_optimized_kprobe(op, p);
843 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
846 * Prepare an optimized_kprobe and optimize it.
847 * NOTE: 'p' must be a normal registered kprobe.
849 static void try_to_optimize_kprobe(struct kprobe *p)
852 struct optimized_kprobe *op;
854 /* Impossible to optimize ftrace-based kprobe. */
855 if (kprobe_ftrace(p))
858 /* For preparing optimization, jump_label_text_reserved() is called. */
861 mutex_lock(&text_mutex);
863 ap = alloc_aggr_kprobe(p);
867 op = container_of(ap, struct optimized_kprobe, kp);
868 if (!arch_prepared_optinsn(&op->optinsn)) {
869 /* If failed to setup optimizing, fallback to kprobe. */
870 arch_remove_optimized_kprobe(op);
875 init_aggr_kprobe(ap, p);
876 optimize_kprobe(ap); /* This just kicks optimizer thread. */
879 mutex_unlock(&text_mutex);
884 static void optimize_all_kprobes(void)
886 struct hlist_head *head;
890 mutex_lock(&kprobe_mutex);
891 /* If optimization is already allowed, just return. */
892 if (kprobes_allow_optimization)
896 kprobes_allow_optimization = true;
897 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
898 head = &kprobe_table[i];
899 hlist_for_each_entry(p, head, hlist)
900 if (!kprobe_disabled(p))
904 pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
906 mutex_unlock(&kprobe_mutex);
910 static void unoptimize_all_kprobes(void)
912 struct hlist_head *head;
916 mutex_lock(&kprobe_mutex);
917 /* If optimization is already prohibited, just return. */
918 if (!kprobes_allow_optimization) {
919 mutex_unlock(&kprobe_mutex);
924 kprobes_allow_optimization = false;
925 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
926 head = &kprobe_table[i];
927 hlist_for_each_entry(p, head, hlist) {
928 if (!kprobe_disabled(p))
929 unoptimize_kprobe(p, false);
933 mutex_unlock(&kprobe_mutex);
935 /* Wait for unoptimizing completion. */
936 wait_for_kprobe_optimizer();
937 pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
940 static DEFINE_MUTEX(kprobe_sysctl_mutex);
941 static int sysctl_kprobes_optimization;
942 static int proc_kprobes_optimization_handler(struct ctl_table *table,
943 int write, void *buffer,
944 size_t *length, loff_t *ppos)
948 mutex_lock(&kprobe_sysctl_mutex);
949 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
950 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
952 if (sysctl_kprobes_optimization)
953 optimize_all_kprobes();
955 unoptimize_all_kprobes();
956 mutex_unlock(&kprobe_sysctl_mutex);
961 static struct ctl_table kprobe_sysctls[] = {
963 .procname = "kprobes-optimization",
964 .data = &sysctl_kprobes_optimization,
965 .maxlen = sizeof(int),
967 .proc_handler = proc_kprobes_optimization_handler,
968 .extra1 = SYSCTL_ZERO,
969 .extra2 = SYSCTL_ONE,
974 static void __init kprobe_sysctls_init(void)
976 register_sysctl_init("debug", kprobe_sysctls);
978 #endif /* CONFIG_SYSCTL */
980 /* Put a breakpoint for a probe. */
981 static void __arm_kprobe(struct kprobe *p)
985 lockdep_assert_held(&text_mutex);
987 /* Find the overlapping optimized kprobes. */
988 _p = get_optimized_kprobe(p->addr);
990 /* Fallback to unoptimized kprobe */
991 unoptimize_kprobe(_p, true);
994 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
997 /* Remove the breakpoint of a probe. */
998 static void __disarm_kprobe(struct kprobe *p, bool reopt)
1002 lockdep_assert_held(&text_mutex);
1004 /* Try to unoptimize */
1005 unoptimize_kprobe(p, kprobes_all_disarmed);
1007 if (!kprobe_queued(p)) {
1008 arch_disarm_kprobe(p);
1009 /* If another kprobe was blocked, re-optimize it. */
1010 _p = get_optimized_kprobe(p->addr);
1011 if (unlikely(_p) && reopt)
1012 optimize_kprobe(_p);
1015 * TODO: Since unoptimization and real disarming will be done by
1016 * the worker thread, we can not check whether another probe are
1017 * unoptimized because of this probe here. It should be re-optimized
1018 * by the worker thread.
1022 #else /* !CONFIG_OPTPROBES */
1024 #define optimize_kprobe(p) do {} while (0)
1025 #define unoptimize_kprobe(p, f) do {} while (0)
1026 #define kill_optimized_kprobe(p) do {} while (0)
1027 #define prepare_optimized_kprobe(p) do {} while (0)
1028 #define try_to_optimize_kprobe(p) do {} while (0)
1029 #define __arm_kprobe(p) arch_arm_kprobe(p)
1030 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
1031 #define kprobe_disarmed(p) kprobe_disabled(p)
1032 #define wait_for_kprobe_optimizer() do {} while (0)
1034 static int reuse_unused_kprobe(struct kprobe *ap)
1037 * If the optimized kprobe is NOT supported, the aggr kprobe is
1038 * released at the same time that the last aggregated kprobe is
1040 * Thus there should be no chance to reuse unused kprobe.
1046 static void free_aggr_kprobe(struct kprobe *p)
1048 arch_remove_kprobe(p);
1052 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1054 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1056 #endif /* CONFIG_OPTPROBES */
1058 #ifdef CONFIG_KPROBES_ON_FTRACE
1059 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1060 .func = kprobe_ftrace_handler,
1061 .flags = FTRACE_OPS_FL_SAVE_REGS,
1064 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1065 .func = kprobe_ftrace_handler,
1066 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1069 static int kprobe_ipmodify_enabled;
1070 static int kprobe_ftrace_enabled;
1072 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1077 lockdep_assert_held(&kprobe_mutex);
1079 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1080 if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1084 ret = register_ftrace_function(ops);
1085 if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1094 * At this point, sinec ops is not registered, we should be sefe from
1095 * registering empty filter.
1097 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1101 static int arm_kprobe_ftrace(struct kprobe *p)
1103 bool ipmodify = (p->post_handler != NULL);
1105 return __arm_kprobe_ftrace(p,
1106 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1107 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1110 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1115 lockdep_assert_held(&kprobe_mutex);
1118 ret = unregister_ftrace_function(ops);
1119 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1125 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1126 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1131 static int disarm_kprobe_ftrace(struct kprobe *p)
1133 bool ipmodify = (p->post_handler != NULL);
1135 return __disarm_kprobe_ftrace(p,
1136 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1137 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1139 #else /* !CONFIG_KPROBES_ON_FTRACE */
1140 static inline int arm_kprobe_ftrace(struct kprobe *p)
1145 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1151 static int prepare_kprobe(struct kprobe *p)
1153 /* Must ensure p->addr is really on ftrace */
1154 if (kprobe_ftrace(p))
1155 return arch_prepare_kprobe_ftrace(p);
1157 return arch_prepare_kprobe(p);
1160 static int arm_kprobe(struct kprobe *kp)
1162 if (unlikely(kprobe_ftrace(kp)))
1163 return arm_kprobe_ftrace(kp);
1166 mutex_lock(&text_mutex);
1168 mutex_unlock(&text_mutex);
1174 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1176 if (unlikely(kprobe_ftrace(kp)))
1177 return disarm_kprobe_ftrace(kp);
1180 mutex_lock(&text_mutex);
1181 __disarm_kprobe(kp, reopt);
1182 mutex_unlock(&text_mutex);
1189 * Aggregate handlers for multiple kprobes support - these handlers
1190 * take care of invoking the individual kprobe handlers on p->list
1192 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1196 list_for_each_entry_rcu(kp, &p->list, list) {
1197 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1198 set_kprobe_instance(kp);
1199 if (kp->pre_handler(kp, regs))
1202 reset_kprobe_instance();
1206 NOKPROBE_SYMBOL(aggr_pre_handler);
1208 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1209 unsigned long flags)
1213 list_for_each_entry_rcu(kp, &p->list, list) {
1214 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1215 set_kprobe_instance(kp);
1216 kp->post_handler(kp, regs, flags);
1217 reset_kprobe_instance();
1221 NOKPROBE_SYMBOL(aggr_post_handler);
1223 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
1224 void kprobes_inc_nmissed_count(struct kprobe *p)
1228 if (!kprobe_aggrprobe(p)) {
1231 list_for_each_entry_rcu(kp, &p->list, list)
1235 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1237 static struct kprobe kprobe_busy = {
1238 .addr = (void *) get_kprobe,
1241 void kprobe_busy_begin(void)
1243 struct kprobe_ctlblk *kcb;
1246 __this_cpu_write(current_kprobe, &kprobe_busy);
1247 kcb = get_kprobe_ctlblk();
1248 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1251 void kprobe_busy_end(void)
1253 __this_cpu_write(current_kprobe, NULL);
1257 /* Add the new probe to 'ap->list'. */
1258 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1260 if (p->post_handler)
1261 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1263 list_add_rcu(&p->list, &ap->list);
1264 if (p->post_handler && !ap->post_handler)
1265 ap->post_handler = aggr_post_handler;
1271 * Fill in the required fields of the aggregator kprobe. Replace the
1272 * earlier kprobe in the hlist with the aggregator kprobe.
1274 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1276 /* Copy the insn slot of 'p' to 'ap'. */
1278 flush_insn_slot(ap);
1280 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1281 ap->pre_handler = aggr_pre_handler;
1282 /* We don't care the kprobe which has gone. */
1283 if (p->post_handler && !kprobe_gone(p))
1284 ap->post_handler = aggr_post_handler;
1286 INIT_LIST_HEAD(&ap->list);
1287 INIT_HLIST_NODE(&ap->hlist);
1289 list_add_rcu(&p->list, &ap->list);
1290 hlist_replace_rcu(&p->hlist, &ap->hlist);
1294 * This registers the second or subsequent kprobe at the same address.
1296 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1299 struct kprobe *ap = orig_p;
1303 /* For preparing optimization, jump_label_text_reserved() is called */
1305 mutex_lock(&text_mutex);
1307 if (!kprobe_aggrprobe(orig_p)) {
1308 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1309 ap = alloc_aggr_kprobe(orig_p);
1314 init_aggr_kprobe(ap, orig_p);
1315 } else if (kprobe_unused(ap)) {
1316 /* This probe is going to die. Rescue it */
1317 ret = reuse_unused_kprobe(ap);
1322 if (kprobe_gone(ap)) {
1324 * Attempting to insert new probe at the same location that
1325 * had a probe in the module vaddr area which already
1326 * freed. So, the instruction slot has already been
1327 * released. We need a new slot for the new probe.
1329 ret = arch_prepare_kprobe(ap);
1332 * Even if fail to allocate new slot, don't need to
1333 * free the 'ap'. It will be used next time, or
1334 * freed by unregister_kprobe().
1338 /* Prepare optimized instructions if possible. */
1339 prepare_optimized_kprobe(ap);
1342 * Clear gone flag to prevent allocating new slot again, and
1343 * set disabled flag because it is not armed yet.
1345 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1346 | KPROBE_FLAG_DISABLED;
1349 /* Copy the insn slot of 'p' to 'ap'. */
1351 ret = add_new_kprobe(ap, p);
1354 mutex_unlock(&text_mutex);
1355 jump_label_unlock();
1358 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1359 ap->flags &= ~KPROBE_FLAG_DISABLED;
1360 if (!kprobes_all_disarmed) {
1361 /* Arm the breakpoint again. */
1362 ret = arm_kprobe(ap);
1364 ap->flags |= KPROBE_FLAG_DISABLED;
1365 list_del_rcu(&p->list);
1373 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1375 /* The '__kprobes' functions and entry code must not be probed. */
1376 return addr >= (unsigned long)__kprobes_text_start &&
1377 addr < (unsigned long)__kprobes_text_end;
1380 static bool __within_kprobe_blacklist(unsigned long addr)
1382 struct kprobe_blacklist_entry *ent;
1384 if (arch_within_kprobe_blacklist(addr))
1387 * If 'kprobe_blacklist' is defined, check the address and
1388 * reject any probe registration in the prohibited area.
1390 list_for_each_entry(ent, &kprobe_blacklist, list) {
1391 if (addr >= ent->start_addr && addr < ent->end_addr)
1397 bool within_kprobe_blacklist(unsigned long addr)
1399 char symname[KSYM_NAME_LEN], *p;
1401 if (__within_kprobe_blacklist(addr))
1404 /* Check if the address is on a suffixed-symbol */
1405 if (!lookup_symbol_name(addr, symname)) {
1406 p = strchr(symname, '.');
1410 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1412 return __within_kprobe_blacklist(addr);
1418 * arch_adjust_kprobe_addr - adjust the address
1419 * @addr: symbol base address
1420 * @offset: offset within the symbol
1421 * @on_func_entry: was this @addr+@offset on the function entry
1423 * Typically returns @addr + @offset, except for special cases where the
1424 * function might be prefixed by a CFI landing pad, in that case any offset
1425 * inside the landing pad is mapped to the first 'real' instruction of the
1428 * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1429 * instruction at +0.
1431 kprobe_opcode_t *__weak arch_adjust_kprobe_addr(unsigned long addr,
1432 unsigned long offset,
1433 bool *on_func_entry)
1435 *on_func_entry = !offset;
1436 return (kprobe_opcode_t *)(addr + offset);
1440 * If 'symbol_name' is specified, look it up and add the 'offset'
1441 * to it. This way, we can specify a relative address to a symbol.
1442 * This returns encoded errors if it fails to look up symbol or invalid
1443 * combination of parameters.
1445 static kprobe_opcode_t *
1446 _kprobe_addr(kprobe_opcode_t *addr, const char *symbol_name,
1447 unsigned long offset, bool *on_func_entry)
1449 if ((symbol_name && addr) || (!symbol_name && !addr))
1454 * Input: @sym + @offset
1455 * Output: @addr + @offset
1457 * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1458 * argument into it's output!
1460 addr = kprobe_lookup_name(symbol_name, offset);
1462 return ERR_PTR(-ENOENT);
1466 * So here we have @addr + @offset, displace it into a new
1467 * @addr' + @offset' where @addr' is the symbol start address.
1469 addr = (void *)addr + offset;
1470 if (!kallsyms_lookup_size_offset((unsigned long)addr, NULL, &offset))
1471 return ERR_PTR(-ENOENT);
1472 addr = (void *)addr - offset;
1475 * Then ask the architecture to re-combine them, taking care of
1476 * magical function entry details while telling us if this was indeed
1477 * at the start of the function.
1479 addr = arch_adjust_kprobe_addr((unsigned long)addr, offset, on_func_entry);
1484 return ERR_PTR(-EINVAL);
1487 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1490 return _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1494 * Check the 'p' is valid and return the aggregator kprobe
1495 * at the same address.
1497 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1499 struct kprobe *ap, *list_p;
1501 lockdep_assert_held(&kprobe_mutex);
1503 ap = get_kprobe(p->addr);
1508 list_for_each_entry(list_p, &ap->list, list)
1510 /* kprobe p is a valid probe */
1519 * Warn and return error if the kprobe is being re-registered since
1520 * there must be a software bug.
1522 static inline int warn_kprobe_rereg(struct kprobe *p)
1526 mutex_lock(&kprobe_mutex);
1527 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1529 mutex_unlock(&kprobe_mutex);
1534 static int check_ftrace_location(struct kprobe *p)
1536 unsigned long addr = (unsigned long)p->addr;
1538 if (ftrace_location(addr) == addr) {
1539 #ifdef CONFIG_KPROBES_ON_FTRACE
1540 p->flags |= KPROBE_FLAG_FTRACE;
1541 #else /* !CONFIG_KPROBES_ON_FTRACE */
1548 static bool is_cfi_preamble_symbol(unsigned long addr)
1550 char symbuf[KSYM_NAME_LEN];
1552 if (lookup_symbol_name(addr, symbuf))
1555 return str_has_prefix("__cfi_", symbuf) ||
1556 str_has_prefix("__pfx_", symbuf);
1559 static int check_kprobe_address_safe(struct kprobe *p,
1560 struct module **probed_mod)
1564 ret = check_ftrace_location(p);
1570 /* Ensure it is not in reserved area nor out of text */
1571 if (!(core_kernel_text((unsigned long) p->addr) ||
1572 is_module_text_address((unsigned long) p->addr)) ||
1573 in_gate_area_no_mm((unsigned long) p->addr) ||
1574 within_kprobe_blacklist((unsigned long) p->addr) ||
1575 jump_label_text_reserved(p->addr, p->addr) ||
1576 static_call_text_reserved(p->addr, p->addr) ||
1577 find_bug((unsigned long)p->addr) ||
1578 is_cfi_preamble_symbol((unsigned long)p->addr)) {
1583 /* Check if 'p' is probing a module. */
1584 *probed_mod = __module_text_address((unsigned long) p->addr);
1587 * We must hold a refcount of the probed module while updating
1588 * its code to prohibit unexpected unloading.
1590 if (unlikely(!try_module_get(*probed_mod))) {
1596 * If the module freed '.init.text', we couldn't insert
1599 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1600 (*probed_mod)->state != MODULE_STATE_COMING) {
1601 module_put(*probed_mod);
1608 jump_label_unlock();
1613 int register_kprobe(struct kprobe *p)
1616 struct kprobe *old_p;
1617 struct module *probed_mod;
1618 kprobe_opcode_t *addr;
1621 /* Adjust probe address from symbol */
1622 addr = _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1624 return PTR_ERR(addr);
1627 ret = warn_kprobe_rereg(p);
1631 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1632 p->flags &= KPROBE_FLAG_DISABLED;
1634 INIT_LIST_HEAD(&p->list);
1636 ret = check_kprobe_address_safe(p, &probed_mod);
1640 mutex_lock(&kprobe_mutex);
1643 p->flags |= KPROBE_FLAG_ON_FUNC_ENTRY;
1645 old_p = get_kprobe(p->addr);
1647 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1648 ret = register_aggr_kprobe(old_p, p);
1653 /* Prevent text modification */
1654 mutex_lock(&text_mutex);
1655 ret = prepare_kprobe(p);
1656 mutex_unlock(&text_mutex);
1661 INIT_HLIST_NODE(&p->hlist);
1662 hlist_add_head_rcu(&p->hlist,
1663 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1665 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1666 ret = arm_kprobe(p);
1668 hlist_del_rcu(&p->hlist);
1674 /* Try to optimize kprobe */
1675 try_to_optimize_kprobe(p);
1677 mutex_unlock(&kprobe_mutex);
1680 module_put(probed_mod);
1684 EXPORT_SYMBOL_GPL(register_kprobe);
1686 /* Check if all probes on the 'ap' are disabled. */
1687 static bool aggr_kprobe_disabled(struct kprobe *ap)
1691 lockdep_assert_held(&kprobe_mutex);
1693 list_for_each_entry(kp, &ap->list, list)
1694 if (!kprobe_disabled(kp))
1696 * Since there is an active probe on the list,
1697 * we can't disable this 'ap'.
1704 static struct kprobe *__disable_kprobe(struct kprobe *p)
1706 struct kprobe *orig_p;
1709 lockdep_assert_held(&kprobe_mutex);
1711 /* Get an original kprobe for return */
1712 orig_p = __get_valid_kprobe(p);
1713 if (unlikely(orig_p == NULL))
1714 return ERR_PTR(-EINVAL);
1716 if (!kprobe_disabled(p)) {
1717 /* Disable probe if it is a child probe */
1719 p->flags |= KPROBE_FLAG_DISABLED;
1721 /* Try to disarm and disable this/parent probe */
1722 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1724 * Don't be lazy here. Even if 'kprobes_all_disarmed'
1725 * is false, 'orig_p' might not have been armed yet.
1726 * Note arm_all_kprobes() __tries__ to arm all kprobes
1727 * on the best effort basis.
1729 if (!kprobes_all_disarmed && !kprobe_disabled(orig_p)) {
1730 ret = disarm_kprobe(orig_p, true);
1732 p->flags &= ~KPROBE_FLAG_DISABLED;
1733 return ERR_PTR(ret);
1736 orig_p->flags |= KPROBE_FLAG_DISABLED;
1744 * Unregister a kprobe without a scheduler synchronization.
1746 static int __unregister_kprobe_top(struct kprobe *p)
1748 struct kprobe *ap, *list_p;
1750 /* Disable kprobe. This will disarm it if needed. */
1751 ap = __disable_kprobe(p);
1757 * This probe is an independent(and non-optimized) kprobe
1758 * (not an aggrprobe). Remove from the hash list.
1762 /* Following process expects this probe is an aggrprobe */
1763 WARN_ON(!kprobe_aggrprobe(ap));
1765 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1767 * !disarmed could be happen if the probe is under delayed
1772 /* If disabling probe has special handlers, update aggrprobe */
1773 if (p->post_handler && !kprobe_gone(p)) {
1774 list_for_each_entry(list_p, &ap->list, list) {
1775 if ((list_p != p) && (list_p->post_handler))
1779 * For the kprobe-on-ftrace case, we keep the
1780 * post_handler setting to identify this aggrprobe
1781 * armed with kprobe_ipmodify_ops.
1783 if (!kprobe_ftrace(ap))
1784 ap->post_handler = NULL;
1788 * Remove from the aggrprobe: this path will do nothing in
1789 * __unregister_kprobe_bottom().
1791 list_del_rcu(&p->list);
1792 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1794 * Try to optimize this probe again, because post
1795 * handler may have been changed.
1797 optimize_kprobe(ap);
1802 hlist_del_rcu(&ap->hlist);
1806 static void __unregister_kprobe_bottom(struct kprobe *p)
1810 if (list_empty(&p->list))
1811 /* This is an independent kprobe */
1812 arch_remove_kprobe(p);
1813 else if (list_is_singular(&p->list)) {
1814 /* This is the last child of an aggrprobe */
1815 ap = list_entry(p->list.next, struct kprobe, list);
1817 free_aggr_kprobe(ap);
1819 /* Otherwise, do nothing. */
1822 int register_kprobes(struct kprobe **kps, int num)
1828 for (i = 0; i < num; i++) {
1829 ret = register_kprobe(kps[i]);
1832 unregister_kprobes(kps, i);
1838 EXPORT_SYMBOL_GPL(register_kprobes);
1840 void unregister_kprobe(struct kprobe *p)
1842 unregister_kprobes(&p, 1);
1844 EXPORT_SYMBOL_GPL(unregister_kprobe);
1846 void unregister_kprobes(struct kprobe **kps, int num)
1852 mutex_lock(&kprobe_mutex);
1853 for (i = 0; i < num; i++)
1854 if (__unregister_kprobe_top(kps[i]) < 0)
1855 kps[i]->addr = NULL;
1856 mutex_unlock(&kprobe_mutex);
1859 for (i = 0; i < num; i++)
1861 __unregister_kprobe_bottom(kps[i]);
1863 EXPORT_SYMBOL_GPL(unregister_kprobes);
1865 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1866 unsigned long val, void *data)
1870 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1872 static struct notifier_block kprobe_exceptions_nb = {
1873 .notifier_call = kprobe_exceptions_notify,
1874 .priority = 0x7fffffff /* we need to be notified first */
1877 #ifdef CONFIG_KRETPROBES
1879 #if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
1880 static void free_rp_inst_rcu(struct rcu_head *head)
1882 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1884 if (refcount_dec_and_test(&ri->rph->ref))
1888 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1890 static void recycle_rp_inst(struct kretprobe_instance *ri)
1892 struct kretprobe *rp = get_kretprobe(ri);
1895 freelist_add(&ri->freelist, &rp->freelist);
1897 call_rcu(&ri->rcu, free_rp_inst_rcu);
1899 NOKPROBE_SYMBOL(recycle_rp_inst);
1902 * This function is called from delayed_put_task_struct() when a task is
1903 * dead and cleaned up to recycle any kretprobe instances associated with
1904 * this task. These left over instances represent probed functions that
1905 * have been called but will never return.
1907 void kprobe_flush_task(struct task_struct *tk)
1909 struct kretprobe_instance *ri;
1910 struct llist_node *node;
1912 /* Early boot, not yet initialized. */
1913 if (unlikely(!kprobes_initialized))
1916 kprobe_busy_begin();
1918 node = __llist_del_all(&tk->kretprobe_instances);
1920 ri = container_of(node, struct kretprobe_instance, llist);
1923 recycle_rp_inst(ri);
1928 NOKPROBE_SYMBOL(kprobe_flush_task);
1930 static inline void free_rp_inst(struct kretprobe *rp)
1932 struct kretprobe_instance *ri;
1933 struct freelist_node *node;
1936 node = rp->freelist.head;
1938 ri = container_of(node, struct kretprobe_instance, freelist);
1945 if (refcount_sub_and_test(count, &rp->rph->ref)) {
1951 /* This assumes the 'tsk' is the current task or the is not running. */
1952 static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1953 struct llist_node **cur)
1955 struct kretprobe_instance *ri = NULL;
1956 struct llist_node *node = *cur;
1959 node = tsk->kretprobe_instances.first;
1964 ri = container_of(node, struct kretprobe_instance, llist);
1965 if (ri->ret_addr != kretprobe_trampoline_addr()) {
1967 return ri->ret_addr;
1973 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1976 * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1978 * @fp: A frame pointer
1979 * @cur: a storage of the loop cursor llist_node pointer for next call
1981 * Find the correct return address modified by a kretprobe on @tsk in unsigned
1982 * long type. If it finds the return address, this returns that address value,
1983 * or this returns 0.
1984 * The @tsk must be 'current' or a task which is not running. @fp is a hint
1985 * to get the currect return address - which is compared with the
1986 * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1987 * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
1988 * first call, but '@cur' itself must NOT NULL.
1990 unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
1991 struct llist_node **cur)
1993 struct kretprobe_instance *ri = NULL;
1994 kprobe_opcode_t *ret;
1996 if (WARN_ON_ONCE(!cur))
2000 ret = __kretprobe_find_ret_addr(tsk, cur);
2003 ri = container_of(*cur, struct kretprobe_instance, llist);
2004 } while (ri->fp != fp);
2006 return (unsigned long)ret;
2008 NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
2010 void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
2011 kprobe_opcode_t *correct_ret_addr)
2014 * Do nothing by default. Please fill this to update the fake return
2015 * address on the stack with the correct one on each arch if possible.
2019 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
2020 void *frame_pointer)
2022 kprobe_opcode_t *correct_ret_addr = NULL;
2023 struct kretprobe_instance *ri = NULL;
2024 struct llist_node *first, *node = NULL;
2025 struct kretprobe *rp;
2027 /* Find correct address and all nodes for this frame. */
2028 correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
2029 if (!correct_ret_addr) {
2030 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2035 * Set the return address as the instruction pointer, because if the
2036 * user handler calls stack_trace_save_regs() with this 'regs',
2037 * the stack trace will start from the instruction pointer.
2039 instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
2041 /* Run the user handler of the nodes. */
2042 first = current->kretprobe_instances.first;
2044 ri = container_of(first, struct kretprobe_instance, llist);
2046 if (WARN_ON_ONCE(ri->fp != frame_pointer))
2049 rp = get_kretprobe(ri);
2050 if (rp && rp->handler) {
2051 struct kprobe *prev = kprobe_running();
2053 __this_cpu_write(current_kprobe, &rp->kp);
2054 ri->ret_addr = correct_ret_addr;
2055 rp->handler(ri, regs);
2056 __this_cpu_write(current_kprobe, prev);
2061 first = first->next;
2064 arch_kretprobe_fixup_return(regs, correct_ret_addr);
2066 /* Unlink all nodes for this frame. */
2067 first = current->kretprobe_instances.first;
2068 current->kretprobe_instances.first = node->next;
2071 /* Recycle free instances. */
2073 ri = container_of(first, struct kretprobe_instance, llist);
2074 first = first->next;
2076 recycle_rp_inst(ri);
2079 return (unsigned long)correct_ret_addr;
2081 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2084 * This kprobe pre_handler is registered with every kretprobe. When probe
2085 * hits it will set up the return probe.
2087 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2089 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2090 struct kretprobe_instance *ri;
2091 struct freelist_node *fn;
2093 fn = freelist_try_get(&rp->freelist);
2099 ri = container_of(fn, struct kretprobe_instance, freelist);
2101 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2102 freelist_add(&ri->freelist, &rp->freelist);
2106 arch_prepare_kretprobe(ri, regs);
2108 __llist_add(&ri->llist, ¤t->kretprobe_instances);
2112 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2113 #else /* CONFIG_KRETPROBE_ON_RETHOOK */
2115 * This kprobe pre_handler is registered with every kretprobe. When probe
2116 * hits it will set up the return probe.
2118 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2120 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2121 struct kretprobe_instance *ri;
2122 struct rethook_node *rhn;
2124 rhn = rethook_try_get(rp->rh);
2130 ri = container_of(rhn, struct kretprobe_instance, node);
2132 if (rp->entry_handler && rp->entry_handler(ri, regs))
2133 rethook_recycle(rhn);
2135 rethook_hook(rhn, regs, kprobe_ftrace(p));
2139 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2141 static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
2142 struct pt_regs *regs)
2144 struct kretprobe *rp = (struct kretprobe *)data;
2145 struct kretprobe_instance *ri;
2146 struct kprobe_ctlblk *kcb;
2148 /* The data must NOT be null. This means rethook data structure is broken. */
2149 if (WARN_ON_ONCE(!data) || !rp->handler)
2152 __this_cpu_write(current_kprobe, &rp->kp);
2153 kcb = get_kprobe_ctlblk();
2154 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
2156 ri = container_of(rh, struct kretprobe_instance, node);
2157 rp->handler(ri, regs);
2159 __this_cpu_write(current_kprobe, NULL);
2161 NOKPROBE_SYMBOL(kretprobe_rethook_handler);
2163 #endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2166 * kprobe_on_func_entry() -- check whether given address is function entry
2167 * @addr: Target address
2168 * @sym: Target symbol name
2169 * @offset: The offset from the symbol or the address
2171 * This checks whether the given @addr+@offset or @sym+@offset is on the
2172 * function entry address or not.
2173 * This returns 0 if it is the function entry, or -EINVAL if it is not.
2174 * And also it returns -ENOENT if it fails the symbol or address lookup.
2175 * Caller must pass @addr or @sym (either one must be NULL), or this
2178 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2181 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset, &on_func_entry);
2183 if (IS_ERR(kp_addr))
2184 return PTR_ERR(kp_addr);
2192 int register_kretprobe(struct kretprobe *rp)
2195 struct kretprobe_instance *inst;
2199 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2203 /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2204 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2207 if (kretprobe_blacklist_size) {
2208 addr = kprobe_addr(&rp->kp);
2210 return PTR_ERR(addr);
2212 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2213 if (kretprobe_blacklist[i].addr == addr)
2218 if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2221 rp->kp.pre_handler = pre_handler_kretprobe;
2222 rp->kp.post_handler = NULL;
2224 /* Pre-allocate memory for max kretprobe instances */
2225 if (rp->maxactive <= 0)
2226 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2228 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2229 rp->rh = rethook_alloc((void *)rp, kretprobe_rethook_handler);
2233 for (i = 0; i < rp->maxactive; i++) {
2234 inst = kzalloc(sizeof(struct kretprobe_instance) +
2235 rp->data_size, GFP_KERNEL);
2237 rethook_free(rp->rh);
2241 rethook_add_node(rp->rh, &inst->node);
2244 /* Establish function entry probe point */
2245 ret = register_kprobe(&rp->kp);
2247 rethook_free(rp->rh);
2250 #else /* !CONFIG_KRETPROBE_ON_RETHOOK */
2251 rp->freelist.head = NULL;
2252 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2256 rcu_assign_pointer(rp->rph->rp, rp);
2257 for (i = 0; i < rp->maxactive; i++) {
2258 inst = kzalloc(sizeof(struct kretprobe_instance) +
2259 rp->data_size, GFP_KERNEL);
2261 refcount_set(&rp->rph->ref, i);
2265 inst->rph = rp->rph;
2266 freelist_add(&inst->freelist, &rp->freelist);
2268 refcount_set(&rp->rph->ref, i);
2271 /* Establish function entry probe point */
2272 ret = register_kprobe(&rp->kp);
2278 EXPORT_SYMBOL_GPL(register_kretprobe);
2280 int register_kretprobes(struct kretprobe **rps, int num)
2286 for (i = 0; i < num; i++) {
2287 ret = register_kretprobe(rps[i]);
2290 unregister_kretprobes(rps, i);
2296 EXPORT_SYMBOL_GPL(register_kretprobes);
2298 void unregister_kretprobe(struct kretprobe *rp)
2300 unregister_kretprobes(&rp, 1);
2302 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2304 void unregister_kretprobes(struct kretprobe **rps, int num)
2310 mutex_lock(&kprobe_mutex);
2311 for (i = 0; i < num; i++) {
2312 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2313 rps[i]->kp.addr = NULL;
2314 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2315 rethook_free(rps[i]->rh);
2317 rcu_assign_pointer(rps[i]->rph->rp, NULL);
2320 mutex_unlock(&kprobe_mutex);
2323 for (i = 0; i < num; i++) {
2324 if (rps[i]->kp.addr) {
2325 __unregister_kprobe_bottom(&rps[i]->kp);
2326 #ifndef CONFIG_KRETPROBE_ON_RETHOOK
2327 free_rp_inst(rps[i]);
2332 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2334 #else /* CONFIG_KRETPROBES */
2335 int register_kretprobe(struct kretprobe *rp)
2339 EXPORT_SYMBOL_GPL(register_kretprobe);
2341 int register_kretprobes(struct kretprobe **rps, int num)
2345 EXPORT_SYMBOL_GPL(register_kretprobes);
2347 void unregister_kretprobe(struct kretprobe *rp)
2350 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2352 void unregister_kretprobes(struct kretprobe **rps, int num)
2355 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2357 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2361 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2363 #endif /* CONFIG_KRETPROBES */
2365 /* Set the kprobe gone and remove its instruction buffer. */
2366 static void kill_kprobe(struct kprobe *p)
2370 lockdep_assert_held(&kprobe_mutex);
2373 * The module is going away. We should disarm the kprobe which
2374 * is using ftrace, because ftrace framework is still available at
2375 * 'MODULE_STATE_GOING' notification.
2377 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2378 disarm_kprobe_ftrace(p);
2380 p->flags |= KPROBE_FLAG_GONE;
2381 if (kprobe_aggrprobe(p)) {
2383 * If this is an aggr_kprobe, we have to list all the
2384 * chained probes and mark them GONE.
2386 list_for_each_entry(kp, &p->list, list)
2387 kp->flags |= KPROBE_FLAG_GONE;
2388 p->post_handler = NULL;
2389 kill_optimized_kprobe(p);
2392 * Here, we can remove insn_slot safely, because no thread calls
2393 * the original probed function (which will be freed soon) any more.
2395 arch_remove_kprobe(p);
2398 /* Disable one kprobe */
2399 int disable_kprobe(struct kprobe *kp)
2404 mutex_lock(&kprobe_mutex);
2406 /* Disable this kprobe */
2407 p = __disable_kprobe(kp);
2411 mutex_unlock(&kprobe_mutex);
2414 EXPORT_SYMBOL_GPL(disable_kprobe);
2416 /* Enable one kprobe */
2417 int enable_kprobe(struct kprobe *kp)
2422 mutex_lock(&kprobe_mutex);
2424 /* Check whether specified probe is valid. */
2425 p = __get_valid_kprobe(kp);
2426 if (unlikely(p == NULL)) {
2431 if (kprobe_gone(kp)) {
2432 /* This kprobe has gone, we couldn't enable it. */
2438 kp->flags &= ~KPROBE_FLAG_DISABLED;
2440 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2441 p->flags &= ~KPROBE_FLAG_DISABLED;
2442 ret = arm_kprobe(p);
2444 p->flags |= KPROBE_FLAG_DISABLED;
2446 kp->flags |= KPROBE_FLAG_DISABLED;
2450 mutex_unlock(&kprobe_mutex);
2453 EXPORT_SYMBOL_GPL(enable_kprobe);
2455 /* Caller must NOT call this in usual path. This is only for critical case */
2456 void dump_kprobe(struct kprobe *kp)
2458 pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2459 kp->symbol_name, kp->offset, kp->addr);
2461 NOKPROBE_SYMBOL(dump_kprobe);
2463 int kprobe_add_ksym_blacklist(unsigned long entry)
2465 struct kprobe_blacklist_entry *ent;
2466 unsigned long offset = 0, size = 0;
2468 if (!kernel_text_address(entry) ||
2469 !kallsyms_lookup_size_offset(entry, &size, &offset))
2472 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2475 ent->start_addr = entry;
2476 ent->end_addr = entry + size;
2477 INIT_LIST_HEAD(&ent->list);
2478 list_add_tail(&ent->list, &kprobe_blacklist);
2483 /* Add all symbols in given area into kprobe blacklist */
2484 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2486 unsigned long entry;
2489 for (entry = start; entry < end; entry += ret) {
2490 ret = kprobe_add_ksym_blacklist(entry);
2493 if (ret == 0) /* In case of alias symbol */
2499 /* Remove all symbols in given area from kprobe blacklist */
2500 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2502 struct kprobe_blacklist_entry *ent, *n;
2504 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2505 if (ent->start_addr < start || ent->start_addr >= end)
2507 list_del(&ent->list);
2512 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2514 kprobe_remove_area_blacklist(entry, entry + 1);
2517 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2518 char *type, char *sym)
2523 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2526 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2527 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2529 #ifdef CONFIG_OPTPROBES
2530 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2534 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2539 int __init __weak arch_populate_kprobe_blacklist(void)
2545 * Lookup and populate the kprobe_blacklist.
2547 * Unlike the kretprobe blacklist, we'll need to determine
2548 * the range of addresses that belong to the said functions,
2549 * since a kprobe need not necessarily be at the beginning
2552 static int __init populate_kprobe_blacklist(unsigned long *start,
2555 unsigned long entry;
2556 unsigned long *iter;
2559 for (iter = start; iter < end; iter++) {
2560 entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2561 ret = kprobe_add_ksym_blacklist(entry);
2568 /* Symbols in '__kprobes_text' are blacklisted */
2569 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2570 (unsigned long)__kprobes_text_end);
2574 /* Symbols in 'noinstr' section are blacklisted */
2575 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2576 (unsigned long)__noinstr_text_end);
2578 return ret ? : arch_populate_kprobe_blacklist();
2581 static void add_module_kprobe_blacklist(struct module *mod)
2583 unsigned long start, end;
2586 if (mod->kprobe_blacklist) {
2587 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2588 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2591 start = (unsigned long)mod->kprobes_text_start;
2593 end = start + mod->kprobes_text_size;
2594 kprobe_add_area_blacklist(start, end);
2597 start = (unsigned long)mod->noinstr_text_start;
2599 end = start + mod->noinstr_text_size;
2600 kprobe_add_area_blacklist(start, end);
2604 static void remove_module_kprobe_blacklist(struct module *mod)
2606 unsigned long start, end;
2609 if (mod->kprobe_blacklist) {
2610 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2611 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2614 start = (unsigned long)mod->kprobes_text_start;
2616 end = start + mod->kprobes_text_size;
2617 kprobe_remove_area_blacklist(start, end);
2620 start = (unsigned long)mod->noinstr_text_start;
2622 end = start + mod->noinstr_text_size;
2623 kprobe_remove_area_blacklist(start, end);
2627 /* Module notifier call back, checking kprobes on the module */
2628 static int kprobes_module_callback(struct notifier_block *nb,
2629 unsigned long val, void *data)
2631 struct module *mod = data;
2632 struct hlist_head *head;
2635 int checkcore = (val == MODULE_STATE_GOING);
2637 if (val == MODULE_STATE_COMING) {
2638 mutex_lock(&kprobe_mutex);
2639 add_module_kprobe_blacklist(mod);
2640 mutex_unlock(&kprobe_mutex);
2642 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2646 * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2647 * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2648 * notified, only '.init.text' section would be freed. We need to
2649 * disable kprobes which have been inserted in the sections.
2651 mutex_lock(&kprobe_mutex);
2652 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2653 head = &kprobe_table[i];
2654 hlist_for_each_entry(p, head, hlist)
2655 if (within_module_init((unsigned long)p->addr, mod) ||
2657 within_module_core((unsigned long)p->addr, mod))) {
2659 * The vaddr this probe is installed will soon
2660 * be vfreed buy not synced to disk. Hence,
2661 * disarming the breakpoint isn't needed.
2663 * Note, this will also move any optimized probes
2664 * that are pending to be removed from their
2665 * corresponding lists to the 'freeing_list' and
2666 * will not be touched by the delayed
2667 * kprobe_optimizer() work handler.
2672 if (val == MODULE_STATE_GOING)
2673 remove_module_kprobe_blacklist(mod);
2674 mutex_unlock(&kprobe_mutex);
2678 static struct notifier_block kprobe_module_nb = {
2679 .notifier_call = kprobes_module_callback,
2683 void kprobe_free_init_mem(void)
2685 void *start = (void *)(&__init_begin);
2686 void *end = (void *)(&__init_end);
2687 struct hlist_head *head;
2691 mutex_lock(&kprobe_mutex);
2693 /* Kill all kprobes on initmem because the target code has been freed. */
2694 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2695 head = &kprobe_table[i];
2696 hlist_for_each_entry(p, head, hlist) {
2697 if (start <= (void *)p->addr && (void *)p->addr < end)
2702 mutex_unlock(&kprobe_mutex);
2705 static int __init init_kprobes(void)
2709 /* FIXME allocate the probe table, currently defined statically */
2710 /* initialize all list heads */
2711 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2712 INIT_HLIST_HEAD(&kprobe_table[i]);
2714 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2715 __stop_kprobe_blacklist);
2717 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2719 if (kretprobe_blacklist_size) {
2720 /* lookup the function address from its name */
2721 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2722 kretprobe_blacklist[i].addr =
2723 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2724 if (!kretprobe_blacklist[i].addr)
2725 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2726 kretprobe_blacklist[i].name);
2730 /* By default, kprobes are armed */
2731 kprobes_all_disarmed = false;
2733 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2734 /* Init 'kprobe_optinsn_slots' for allocation */
2735 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2738 err = arch_init_kprobes();
2740 err = register_die_notifier(&kprobe_exceptions_nb);
2742 err = register_module_notifier(&kprobe_module_nb);
2744 kprobes_initialized = (err == 0);
2745 kprobe_sysctls_init();
2748 early_initcall(init_kprobes);
2750 #if defined(CONFIG_OPTPROBES)
2751 static int __init init_optprobes(void)
2754 * Enable kprobe optimization - this kicks the optimizer which
2755 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2756 * not spawned in early initcall. So delay the optimization.
2758 optimize_all_kprobes();
2762 subsys_initcall(init_optprobes);
2765 #ifdef CONFIG_DEBUG_FS
2766 static void report_probe(struct seq_file *pi, struct kprobe *p,
2767 const char *sym, int offset, char *modname, struct kprobe *pp)
2770 void *addr = p->addr;
2772 if (p->pre_handler == pre_handler_kretprobe)
2777 if (!kallsyms_show_value(pi->file->f_cred))
2781 seq_printf(pi, "%px %s %s+0x%x %s ",
2782 addr, kprobe_type, sym, offset,
2783 (modname ? modname : " "));
2784 else /* try to use %pS */
2785 seq_printf(pi, "%px %s %pS ",
2786 addr, kprobe_type, p->addr);
2790 seq_printf(pi, "%s%s%s%s\n",
2791 (kprobe_gone(p) ? "[GONE]" : ""),
2792 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2793 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2794 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2797 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2799 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2802 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2805 if (*pos >= KPROBE_TABLE_SIZE)
2810 static void kprobe_seq_stop(struct seq_file *f, void *v)
2815 static int show_kprobe_addr(struct seq_file *pi, void *v)
2817 struct hlist_head *head;
2818 struct kprobe *p, *kp;
2819 const char *sym = NULL;
2820 unsigned int i = *(loff_t *) v;
2821 unsigned long offset = 0;
2822 char *modname, namebuf[KSYM_NAME_LEN];
2824 head = &kprobe_table[i];
2826 hlist_for_each_entry_rcu(p, head, hlist) {
2827 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2828 &offset, &modname, namebuf);
2829 if (kprobe_aggrprobe(p)) {
2830 list_for_each_entry_rcu(kp, &p->list, list)
2831 report_probe(pi, kp, sym, offset, modname, p);
2833 report_probe(pi, p, sym, offset, modname, NULL);
2839 static const struct seq_operations kprobes_sops = {
2840 .start = kprobe_seq_start,
2841 .next = kprobe_seq_next,
2842 .stop = kprobe_seq_stop,
2843 .show = show_kprobe_addr
2846 DEFINE_SEQ_ATTRIBUTE(kprobes);
2848 /* kprobes/blacklist -- shows which functions can not be probed */
2849 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2851 mutex_lock(&kprobe_mutex);
2852 return seq_list_start(&kprobe_blacklist, *pos);
2855 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2857 return seq_list_next(v, &kprobe_blacklist, pos);
2860 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2862 struct kprobe_blacklist_entry *ent =
2863 list_entry(v, struct kprobe_blacklist_entry, list);
2866 * If '/proc/kallsyms' is not showing kernel address, we won't
2867 * show them here either.
2869 if (!kallsyms_show_value(m->file->f_cred))
2870 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2871 (void *)ent->start_addr);
2873 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2874 (void *)ent->end_addr, (void *)ent->start_addr);
2878 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2880 mutex_unlock(&kprobe_mutex);
2883 static const struct seq_operations kprobe_blacklist_sops = {
2884 .start = kprobe_blacklist_seq_start,
2885 .next = kprobe_blacklist_seq_next,
2886 .stop = kprobe_blacklist_seq_stop,
2887 .show = kprobe_blacklist_seq_show,
2889 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2891 static int arm_all_kprobes(void)
2893 struct hlist_head *head;
2895 unsigned int i, total = 0, errors = 0;
2898 mutex_lock(&kprobe_mutex);
2900 /* If kprobes are armed, just return */
2901 if (!kprobes_all_disarmed)
2902 goto already_enabled;
2905 * optimize_kprobe() called by arm_kprobe() checks
2906 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2909 kprobes_all_disarmed = false;
2910 /* Arming kprobes doesn't optimize kprobe itself */
2911 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2912 head = &kprobe_table[i];
2913 /* Arm all kprobes on a best-effort basis */
2914 hlist_for_each_entry(p, head, hlist) {
2915 if (!kprobe_disabled(p)) {
2916 err = arm_kprobe(p);
2927 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2930 pr_info("Kprobes globally enabled\n");
2933 mutex_unlock(&kprobe_mutex);
2937 static int disarm_all_kprobes(void)
2939 struct hlist_head *head;
2941 unsigned int i, total = 0, errors = 0;
2944 mutex_lock(&kprobe_mutex);
2946 /* If kprobes are already disarmed, just return */
2947 if (kprobes_all_disarmed) {
2948 mutex_unlock(&kprobe_mutex);
2952 kprobes_all_disarmed = true;
2954 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2955 head = &kprobe_table[i];
2956 /* Disarm all kprobes on a best-effort basis */
2957 hlist_for_each_entry(p, head, hlist) {
2958 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2959 err = disarm_kprobe(p, false);
2970 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2973 pr_info("Kprobes globally disabled\n");
2975 mutex_unlock(&kprobe_mutex);
2977 /* Wait for disarming all kprobes by optimizer */
2978 wait_for_kprobe_optimizer();
2984 * XXX: The debugfs bool file interface doesn't allow for callbacks
2985 * when the bool state is switched. We can reuse that facility when
2988 static ssize_t read_enabled_file_bool(struct file *file,
2989 char __user *user_buf, size_t count, loff_t *ppos)
2993 if (!kprobes_all_disarmed)
2999 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
3002 static ssize_t write_enabled_file_bool(struct file *file,
3003 const char __user *user_buf, size_t count, loff_t *ppos)
3008 ret = kstrtobool_from_user(user_buf, count, &enable);
3012 ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
3019 static const struct file_operations fops_kp = {
3020 .read = read_enabled_file_bool,
3021 .write = write_enabled_file_bool,
3022 .llseek = default_llseek,
3025 static int __init debugfs_kprobe_init(void)
3029 dir = debugfs_create_dir("kprobes", NULL);
3031 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
3033 debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
3035 debugfs_create_file("blacklist", 0400, dir, NULL,
3036 &kprobe_blacklist_fops);
3041 late_initcall(debugfs_kprobe_init);
3042 #endif /* CONFIG_DEBUG_FS */