projects/linux·blob v5.13 · refs

2834 lines · 67.6 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Kernel Probes (KProbes)
4 * kernel/kprobes.c
5 *
6 * Copyright (C) IBM Corporation, 2002, 2004
7 *
8 * 2002-Oct Created by Vamsi Krishna S <[email protected]> Kernel
9 * Probes initial implementation (includes suggestions from
10 * Rusty Russell).
11 * 2004-Aug Updated by Prasanna S Panchamukhi <[email protected]> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <[email protected]> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <[email protected]> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <[email protected]>, Jim Keniston
18 * <[email protected]> and Prasanna S Panchamukhi
19 * <[email protected]> added function-return probes.
20 */
21#include <linux/kprobes.h>
22#include <linux/hash.h>
23#include <linux/init.h>
24#include <linux/slab.h>
25#include <linux/stddef.h>
26#include <linux/export.h>
27#include <linux/moduleloader.h>
28#include <linux/kallsyms.h>
29#include <linux/freezer.h>
30#include <linux/seq_file.h>
31#include <linux/debugfs.h>
32#include <linux/sysctl.h>
33#include <linux/kdebug.h>
34#include <linux/memory.h>
35#include <linux/ftrace.h>
36#include <linux/cpu.h>
37#include <linux/jump_label.h>
38#include <linux/perf_event.h>
39
40#include <asm/sections.h>
41#include <asm/cacheflush.h>
42#include <asm/errno.h>
43#include <linux/uaccess.h>
44
45#define KPROBE_HASH_BITS 6
46#define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
47
48
49static int kprobes_initialized;
50/* kprobe_table can be accessed by
51 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
52 * Or
53 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
54 */
55static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
56
57/* NOTE: change this value only with kprobe_mutex held */
58static bool kprobes_all_disarmed;
59
60/* This protects kprobe_table and optimizing_list */
61static DEFINE_MUTEX(kprobe_mutex);
62static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
63
64kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
65 unsigned int __unused)
66{
67 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
68}
69
70/* Blacklist -- list of struct kprobe_blacklist_entry */
71static LIST_HEAD(kprobe_blacklist);
72
73#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
74/*
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
79 */
80struct kprobe_insn_page {
81 struct list_head list;
82 kprobe_opcode_t *insns; /* Page of instruction slots */
83 struct kprobe_insn_cache *cache;
84 int nused;
85 int ngarbage;
86 char slot_used[];
87};
88
89#define KPROBE_INSN_PAGE_SIZE(slots) \
90 (offsetof(struct kprobe_insn_page, slot_used) + \
91 (sizeof(char) * (slots)))
92
93static int slots_per_page(struct kprobe_insn_cache *c)
94{
95 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
96}
97
98enum kprobe_slot_state {
99 SLOT_CLEAN = 0,
100 SLOT_DIRTY = 1,
101 SLOT_USED = 2,
102};
103
104void __weak *alloc_insn_page(void)
105{
106 return module_alloc(PAGE_SIZE);
107}
108
109void __weak free_insn_page(void *page)
110{
111 module_memfree(page);
112}
113
114struct kprobe_insn_cache kprobe_insn_slots = {
115 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
116 .alloc = alloc_insn_page,
117 .free = free_insn_page,
118 .sym = KPROBE_INSN_PAGE_SYM,
119 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
120 .insn_size = MAX_INSN_SIZE,
121 .nr_garbage = 0,
122};
123static int collect_garbage_slots(struct kprobe_insn_cache *c);
124
125/**
126 * __get_insn_slot() - Find a slot on an executable page for an instruction.
127 * We allocate an executable page if there's no room on existing ones.
128 */
129kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
130{
131 struct kprobe_insn_page *kip;
132 kprobe_opcode_t *slot = NULL;
133
134 /* Since the slot array is not protected by rcu, we need a mutex */
135 mutex_lock(&c->mutex);
136 retry:
137 rcu_read_lock();
138 list_for_each_entry_rcu(kip, &c->pages, list) {
139 if (kip->nused < slots_per_page(c)) {
140 int i;
141 for (i = 0; i < slots_per_page(c); i++) {
142 if (kip->slot_used[i] == SLOT_CLEAN) {
143 kip->slot_used[i] = SLOT_USED;
144 kip->nused++;
145 slot = kip->insns + (i * c->insn_size);
146 rcu_read_unlock();
147 goto out;
148 }
149 }
150 /* kip->nused is broken. Fix it. */
151 kip->nused = slots_per_page(c);
152 WARN_ON(1);
153 }
154 }
155 rcu_read_unlock();
156
157 /* If there are any garbage slots, collect it and try again. */
158 if (c->nr_garbage && collect_garbage_slots(c) == 0)
159 goto retry;
160
161 /* All out of space. Need to allocate a new page. */
162 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
163 if (!kip)
164 goto out;
165
166 /*
167 * Use module_alloc so this page is within +/- 2GB of where the
168 * kernel image and loaded module images reside. This is required
169 * so x86_64 can correctly handle the %rip-relative fixups.
170 */
171 kip->insns = c->alloc();
172 if (!kip->insns) {
173 kfree(kip);
174 goto out;
175 }
176 INIT_LIST_HEAD(&kip->list);
177 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
178 kip->slot_used[0] = SLOT_USED;
179 kip->nused = 1;
180 kip->ngarbage = 0;
181 kip->cache = c;
182 list_add_rcu(&kip->list, &c->pages);
183 slot = kip->insns;
184
185 /* Record the perf ksymbol register event after adding the page */
186 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
187 PAGE_SIZE, false, c->sym);
188out:
189 mutex_unlock(&c->mutex);
190 return slot;
191}
192
193/* Return 1 if all garbages are collected, otherwise 0. */
194static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
195{
196 kip->slot_used[idx] = SLOT_CLEAN;
197 kip->nused--;
198 if (kip->nused == 0) {
199 /*
200 * Page is no longer in use. Free it unless
201 * it's the last one. We keep the last one
202 * so as not to have to set it up again the
203 * next time somebody inserts a probe.
204 */
205 if (!list_is_singular(&kip->list)) {
206 /*
207 * Record perf ksymbol unregister event before removing
208 * the page.
209 */
210 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
211 (unsigned long)kip->insns, PAGE_SIZE, true,
212 kip->cache->sym);
213 list_del_rcu(&kip->list);
214 synchronize_rcu();
215 kip->cache->free(kip->insns);
216 kfree(kip);
217 }
218 return 1;
219 }
220 return 0;
221}
222
223static int collect_garbage_slots(struct kprobe_insn_cache *c)
224{
225 struct kprobe_insn_page *kip, *next;
226
227 /* Ensure no-one is interrupted on the garbages */
228 synchronize_rcu();
229
230 list_for_each_entry_safe(kip, next, &c->pages, list) {
231 int i;
232 if (kip->ngarbage == 0)
233 continue;
234 kip->ngarbage = 0; /* we will collect all garbages */
235 for (i = 0; i < slots_per_page(c); i++) {
236 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
237 break;
238 }
239 }
240 c->nr_garbage = 0;
241 return 0;
242}
243
244void __free_insn_slot(struct kprobe_insn_cache *c,
245 kprobe_opcode_t *slot, int dirty)
246{
247 struct kprobe_insn_page *kip;
248 long idx;
249
250 mutex_lock(&c->mutex);
251 rcu_read_lock();
252 list_for_each_entry_rcu(kip, &c->pages, list) {
253 idx = ((long)slot - (long)kip->insns) /
254 (c->insn_size * sizeof(kprobe_opcode_t));
255 if (idx >= 0 && idx < slots_per_page(c))
256 goto out;
257 }
258 /* Could not find this slot. */
259 WARN_ON(1);
260 kip = NULL;
261out:
262 rcu_read_unlock();
263 /* Mark and sweep: this may sleep */
264 if (kip) {
265 /* Check double free */
266 WARN_ON(kip->slot_used[idx] != SLOT_USED);
267 if (dirty) {
268 kip->slot_used[idx] = SLOT_DIRTY;
269 kip->ngarbage++;
270 if (++c->nr_garbage > slots_per_page(c))
271 collect_garbage_slots(c);
272 } else {
273 collect_one_slot(kip, idx);
274 }
275 }
276 mutex_unlock(&c->mutex);
277}
278
279/*
280 * Check given address is on the page of kprobe instruction slots.
281 * This will be used for checking whether the address on a stack
282 * is on a text area or not.
283 */
284bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
285{
286 struct kprobe_insn_page *kip;
287 bool ret = false;
288
289 rcu_read_lock();
290 list_for_each_entry_rcu(kip, &c->pages, list) {
291 if (addr >= (unsigned long)kip->insns &&
292 addr < (unsigned long)kip->insns + PAGE_SIZE) {
293 ret = true;
294 break;
295 }
296 }
297 rcu_read_unlock();
298
299 return ret;
300}
301
302int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
303 unsigned long *value, char *type, char *sym)
304{
305 struct kprobe_insn_page *kip;
306 int ret = -ERANGE;
307
308 rcu_read_lock();
309 list_for_each_entry_rcu(kip, &c->pages, list) {
310 if ((*symnum)--)
311 continue;
312 strlcpy(sym, c->sym, KSYM_NAME_LEN);
313 *type = 't';
314 *value = (unsigned long)kip->insns;
315 ret = 0;
316 break;
317 }
318 rcu_read_unlock();
319
320 return ret;
321}
322
323#ifdef CONFIG_OPTPROBES
324/* For optimized_kprobe buffer */
325struct kprobe_insn_cache kprobe_optinsn_slots = {
326 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
327 .alloc = alloc_insn_page,
328 .free = free_insn_page,
329 .sym = KPROBE_OPTINSN_PAGE_SYM,
330 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
331 /* .insn_size is initialized later */
332 .nr_garbage = 0,
333};
334#endif
335#endif
336
337/* We have preemption disabled.. so it is safe to use __ versions */
338static inline void set_kprobe_instance(struct kprobe *kp)
339{
340 __this_cpu_write(kprobe_instance, kp);
341}
342
343static inline void reset_kprobe_instance(void)
344{
345 __this_cpu_write(kprobe_instance, NULL);
346}
347
348/*
349 * This routine is called either:
350 * - under the kprobe_mutex - during kprobe_[un]register()
351 * OR
352 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
353 */
354struct kprobe *get_kprobe(void *addr)
355{
356 struct hlist_head *head;
357 struct kprobe *p;
358
359 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
360 hlist_for_each_entry_rcu(p, head, hlist,
361 lockdep_is_held(&kprobe_mutex)) {
362 if (p->addr == addr)
363 return p;
364 }
365
366 return NULL;
367}
368NOKPROBE_SYMBOL(get_kprobe);
369
370static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
371
372/* Return true if the kprobe is an aggregator */
373static inline int kprobe_aggrprobe(struct kprobe *p)
374{
375 return p->pre_handler == aggr_pre_handler;
376}
377
378/* Return true(!0) if the kprobe is unused */
379static inline int kprobe_unused(struct kprobe *p)
380{
381 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
382 list_empty(&p->list);
383}
384
385/*
386 * Keep all fields in the kprobe consistent
387 */
388static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
389{
390 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
391 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
392}
393
394#ifdef CONFIG_OPTPROBES
395/* NOTE: change this value only with kprobe_mutex held */
396static bool kprobes_allow_optimization;
397
398/*
399 * Call all pre_handler on the list, but ignores its return value.
400 * This must be called from arch-dep optimized caller.
401 */
402void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
403{
404 struct kprobe *kp;
405
406 list_for_each_entry_rcu(kp, &p->list, list) {
407 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
408 set_kprobe_instance(kp);
409 kp->pre_handler(kp, regs);
410 }
411 reset_kprobe_instance();
412 }
413}
414NOKPROBE_SYMBOL(opt_pre_handler);
415
416/* Free optimized instructions and optimized_kprobe */
417static void free_aggr_kprobe(struct kprobe *p)
418{
419 struct optimized_kprobe *op;
420
421 op = container_of(p, struct optimized_kprobe, kp);
422 arch_remove_optimized_kprobe(op);
423 arch_remove_kprobe(p);
424 kfree(op);
425}
426
427/* Return true(!0) if the kprobe is ready for optimization. */
428static inline int kprobe_optready(struct kprobe *p)
429{
430 struct optimized_kprobe *op;
431
432 if (kprobe_aggrprobe(p)) {
433 op = container_of(p, struct optimized_kprobe, kp);
434 return arch_prepared_optinsn(&op->optinsn);
435 }
436
437 return 0;
438}
439
440/* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
441static inline int kprobe_disarmed(struct kprobe *p)
442{
443 struct optimized_kprobe *op;
444
445 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
446 if (!kprobe_aggrprobe(p))
447 return kprobe_disabled(p);
448
449 op = container_of(p, struct optimized_kprobe, kp);
450
451 return kprobe_disabled(p) && list_empty(&op->list);
452}
453
454/* Return true(!0) if the probe is queued on (un)optimizing lists */
455static int kprobe_queued(struct kprobe *p)
456{
457 struct optimized_kprobe *op;
458
459 if (kprobe_aggrprobe(p)) {
460 op = container_of(p, struct optimized_kprobe, kp);
461 if (!list_empty(&op->list))
462 return 1;
463 }
464 return 0;
465}
466
467/*
468 * Return an optimized kprobe whose optimizing code replaces
469 * instructions including addr (exclude breakpoint).
470 */
471static struct kprobe *get_optimized_kprobe(unsigned long addr)
472{
473 int i;
474 struct kprobe *p = NULL;
475 struct optimized_kprobe *op;
476
477 /* Don't check i == 0, since that is a breakpoint case. */
478 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
479 p = get_kprobe((void *)(addr - i));
480
481 if (p && kprobe_optready(p)) {
482 op = container_of(p, struct optimized_kprobe, kp);
483 if (arch_within_optimized_kprobe(op, addr))
484 return p;
485 }
486
487 return NULL;
488}
489
490/* Optimization staging list, protected by kprobe_mutex */
491static LIST_HEAD(optimizing_list);
492static LIST_HEAD(unoptimizing_list);
493static LIST_HEAD(freeing_list);
494
495static void kprobe_optimizer(struct work_struct *work);
496static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
497#define OPTIMIZE_DELAY 5
498
499/*
500 * Optimize (replace a breakpoint with a jump) kprobes listed on
501 * optimizing_list.
502 */
503static void do_optimize_kprobes(void)
504{
505 lockdep_assert_held(&text_mutex);
506 /*
507 * The optimization/unoptimization refers online_cpus via
508 * stop_machine() and cpu-hotplug modifies online_cpus.
509 * And same time, text_mutex will be held in cpu-hotplug and here.
510 * This combination can cause a deadlock (cpu-hotplug try to lock
511 * text_mutex but stop_machine can not be done because online_cpus
512 * has been changed)
513 * To avoid this deadlock, caller must have locked cpu hotplug
514 * for preventing cpu-hotplug outside of text_mutex locking.
515 */
516 lockdep_assert_cpus_held();
517
518 /* Optimization never be done when disarmed */
519 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
520 list_empty(&optimizing_list))
521 return;
522
523 arch_optimize_kprobes(&optimizing_list);
524}
525
526/*
527 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
528 * if need) kprobes listed on unoptimizing_list.
529 */
530static void do_unoptimize_kprobes(void)
531{
532 struct optimized_kprobe *op, *tmp;
533
534 lockdep_assert_held(&text_mutex);
535 /* See comment in do_optimize_kprobes() */
536 lockdep_assert_cpus_held();
537
538 /* Unoptimization must be done anytime */
539 if (list_empty(&unoptimizing_list))
540 return;
541
542 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
543 /* Loop free_list for disarming */
544 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
545 /* Switching from detour code to origin */
546 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
547 /* Disarm probes if marked disabled */
548 if (kprobe_disabled(&op->kp))
549 arch_disarm_kprobe(&op->kp);
550 if (kprobe_unused(&op->kp)) {
551 /*
552 * Remove unused probes from hash list. After waiting
553 * for synchronization, these probes are reclaimed.
554 * (reclaiming is done by do_free_cleaned_kprobes.)
555 */
556 hlist_del_rcu(&op->kp.hlist);
557 } else
558 list_del_init(&op->list);
559 }
560}
561
562/* Reclaim all kprobes on the free_list */
563static void do_free_cleaned_kprobes(void)
564{
565 struct optimized_kprobe *op, *tmp;
566
567 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
568 list_del_init(&op->list);
569 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
570 /*
571 * This must not happen, but if there is a kprobe
572 * still in use, keep it on kprobes hash list.
573 */
574 continue;
575 }
576 free_aggr_kprobe(&op->kp);
577 }
578}
579
580/* Start optimizer after OPTIMIZE_DELAY passed */
581static void kick_kprobe_optimizer(void)
582{
583 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
584}
585
586/* Kprobe jump optimizer */
587static void kprobe_optimizer(struct work_struct *work)
588{
589 mutex_lock(&kprobe_mutex);
590 cpus_read_lock();
591 mutex_lock(&text_mutex);
592
593 /*
594 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
595 * kprobes before waiting for quiesence period.
596 */
597 do_unoptimize_kprobes();
598
599 /*
600 * Step 2: Wait for quiesence period to ensure all potentially
601 * preempted tasks to have normally scheduled. Because optprobe
602 * may modify multiple instructions, there is a chance that Nth
603 * instruction is preempted. In that case, such tasks can return
604 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
605 * Note that on non-preemptive kernel, this is transparently converted
606 * to synchronoze_sched() to wait for all interrupts to have completed.
607 */
608 synchronize_rcu_tasks();
609
610 /* Step 3: Optimize kprobes after quiesence period */
611 do_optimize_kprobes();
612
613 /* Step 4: Free cleaned kprobes after quiesence period */
614 do_free_cleaned_kprobes();
615
616 mutex_unlock(&text_mutex);
617 cpus_read_unlock();
618
619 /* Step 5: Kick optimizer again if needed */
620 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
621 kick_kprobe_optimizer();
622
623 mutex_unlock(&kprobe_mutex);
624}
625
626/* Wait for completing optimization and unoptimization */
627void wait_for_kprobe_optimizer(void)
628{
629 mutex_lock(&kprobe_mutex);
630
631 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
632 mutex_unlock(&kprobe_mutex);
633
634 /* this will also make optimizing_work execute immmediately */
635 flush_delayed_work(&optimizing_work);
636 /* @optimizing_work might not have been queued yet, relax */
637 cpu_relax();
638
639 mutex_lock(&kprobe_mutex);
640 }
641
642 mutex_unlock(&kprobe_mutex);
643}
644
645static bool optprobe_queued_unopt(struct optimized_kprobe *op)
646{
647 struct optimized_kprobe *_op;
648
649 list_for_each_entry(_op, &unoptimizing_list, list) {
650 if (op == _op)
651 return true;
652 }
653
654 return false;
655}
656
657/* Optimize kprobe if p is ready to be optimized */
658static void optimize_kprobe(struct kprobe *p)
659{
660 struct optimized_kprobe *op;
661
662 /* Check if the kprobe is disabled or not ready for optimization. */
663 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
664 (kprobe_disabled(p) || kprobes_all_disarmed))
665 return;
666
667 /* kprobes with post_handler can not be optimized */
668 if (p->post_handler)
669 return;
670
671 op = container_of(p, struct optimized_kprobe, kp);
672
673 /* Check there is no other kprobes at the optimized instructions */
674 if (arch_check_optimized_kprobe(op) < 0)
675 return;
676
677 /* Check if it is already optimized. */
678 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
679 if (optprobe_queued_unopt(op)) {
680 /* This is under unoptimizing. Just dequeue the probe */
681 list_del_init(&op->list);
682 }
683 return;
684 }
685 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
686
687 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
688 if (WARN_ON_ONCE(!list_empty(&op->list)))
689 return;
690
691 list_add(&op->list, &optimizing_list);
692 kick_kprobe_optimizer();
693}
694
695/* Short cut to direct unoptimizing */
696static void force_unoptimize_kprobe(struct optimized_kprobe *op)
697{
698 lockdep_assert_cpus_held();
699 arch_unoptimize_kprobe(op);
700 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
701}
702
703/* Unoptimize a kprobe if p is optimized */
704static void unoptimize_kprobe(struct kprobe *p, bool force)
705{
706 struct optimized_kprobe *op;
707
708 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
709 return; /* This is not an optprobe nor optimized */
710
711 op = container_of(p, struct optimized_kprobe, kp);
712 if (!kprobe_optimized(p))
713 return;
714
715 if (!list_empty(&op->list)) {
716 if (optprobe_queued_unopt(op)) {
717 /* Queued in unoptimizing queue */
718 if (force) {
719 /*
720 * Forcibly unoptimize the kprobe here, and queue it
721 * in the freeing list for release afterwards.
722 */
723 force_unoptimize_kprobe(op);
724 list_move(&op->list, &freeing_list);
725 }
726 } else {
727 /* Dequeue from the optimizing queue */
728 list_del_init(&op->list);
729 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
730 }
731 return;
732 }
733
734 /* Optimized kprobe case */
735 if (force) {
736 /* Forcibly update the code: this is a special case */
737 force_unoptimize_kprobe(op);
738 } else {
739 list_add(&op->list, &unoptimizing_list);
740 kick_kprobe_optimizer();
741 }
742}
743
744/* Cancel unoptimizing for reusing */
745static int reuse_unused_kprobe(struct kprobe *ap)
746{
747 struct optimized_kprobe *op;
748
749 /*
750 * Unused kprobe MUST be on the way of delayed unoptimizing (means
751 * there is still a relative jump) and disabled.
752 */
753 op = container_of(ap, struct optimized_kprobe, kp);
754 WARN_ON_ONCE(list_empty(&op->list));
755 /* Enable the probe again */
756 ap->flags &= ~KPROBE_FLAG_DISABLED;
757 /* Optimize it again (remove from op->list) */
758 if (!kprobe_optready(ap))
759 return -EINVAL;
760
761 optimize_kprobe(ap);
762 return 0;
763}
764
765/* Remove optimized instructions */
766static void kill_optimized_kprobe(struct kprobe *p)
767{
768 struct optimized_kprobe *op;
769
770 op = container_of(p, struct optimized_kprobe, kp);
771 if (!list_empty(&op->list))
772 /* Dequeue from the (un)optimization queue */
773 list_del_init(&op->list);
774 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
775
776 if (kprobe_unused(p)) {
777 /* Enqueue if it is unused */
778 list_add(&op->list, &freeing_list);
779 /*
780 * Remove unused probes from the hash list. After waiting
781 * for synchronization, this probe is reclaimed.
782 * (reclaiming is done by do_free_cleaned_kprobes().)
783 */
784 hlist_del_rcu(&op->kp.hlist);
785 }
786
787 /* Don't touch the code, because it is already freed. */
788 arch_remove_optimized_kprobe(op);
789}
790
791static inline
792void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
793{
794 if (!kprobe_ftrace(p))
795 arch_prepare_optimized_kprobe(op, p);
796}
797
798/* Try to prepare optimized instructions */
799static void prepare_optimized_kprobe(struct kprobe *p)
800{
801 struct optimized_kprobe *op;
802
803 op = container_of(p, struct optimized_kprobe, kp);
804 __prepare_optimized_kprobe(op, p);
805}
806
807/* Allocate new optimized_kprobe and try to prepare optimized instructions */
808static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
809{
810 struct optimized_kprobe *op;
811
812 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
813 if (!op)
814 return NULL;
815
816 INIT_LIST_HEAD(&op->list);
817 op->kp.addr = p->addr;
818 __prepare_optimized_kprobe(op, p);
819
820 return &op->kp;
821}
822
823static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
824
825/*
826 * Prepare an optimized_kprobe and optimize it
827 * NOTE: p must be a normal registered kprobe
828 */
829static void try_to_optimize_kprobe(struct kprobe *p)
830{
831 struct kprobe *ap;
832 struct optimized_kprobe *op;
833
834 /* Impossible to optimize ftrace-based kprobe */
835 if (kprobe_ftrace(p))
836 return;
837
838 /* For preparing optimization, jump_label_text_reserved() is called */
839 cpus_read_lock();
840 jump_label_lock();
841 mutex_lock(&text_mutex);
842
843 ap = alloc_aggr_kprobe(p);
844 if (!ap)
845 goto out;
846
847 op = container_of(ap, struct optimized_kprobe, kp);
848 if (!arch_prepared_optinsn(&op->optinsn)) {
849 /* If failed to setup optimizing, fallback to kprobe */
850 arch_remove_optimized_kprobe(op);
851 kfree(op);
852 goto out;
853 }
854
855 init_aggr_kprobe(ap, p);
856 optimize_kprobe(ap); /* This just kicks optimizer thread */
857
858out:
859 mutex_unlock(&text_mutex);
860 jump_label_unlock();
861 cpus_read_unlock();
862}
863
864static void optimize_all_kprobes(void)
865{
866 struct hlist_head *head;
867 struct kprobe *p;
868 unsigned int i;
869
870 mutex_lock(&kprobe_mutex);
871 /* If optimization is already allowed, just return */
872 if (kprobes_allow_optimization)
873 goto out;
874
875 cpus_read_lock();
876 kprobes_allow_optimization = true;
877 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
878 head = &kprobe_table[i];
879 hlist_for_each_entry(p, head, hlist)
880 if (!kprobe_disabled(p))
881 optimize_kprobe(p);
882 }
883 cpus_read_unlock();
884 printk(KERN_INFO "Kprobes globally optimized\n");
885out:
886 mutex_unlock(&kprobe_mutex);
887}
888
889#ifdef CONFIG_SYSCTL
890static void unoptimize_all_kprobes(void)
891{
892 struct hlist_head *head;
893 struct kprobe *p;
894 unsigned int i;
895
896 mutex_lock(&kprobe_mutex);
897 /* If optimization is already prohibited, just return */
898 if (!kprobes_allow_optimization) {
899 mutex_unlock(&kprobe_mutex);
900 return;
901 }
902
903 cpus_read_lock();
904 kprobes_allow_optimization = false;
905 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
906 head = &kprobe_table[i];
907 hlist_for_each_entry(p, head, hlist) {
908 if (!kprobe_disabled(p))
909 unoptimize_kprobe(p, false);
910 }
911 }
912 cpus_read_unlock();
913 mutex_unlock(&kprobe_mutex);
914
915 /* Wait for unoptimizing completion */
916 wait_for_kprobe_optimizer();
917 printk(KERN_INFO "Kprobes globally unoptimized\n");
918}
919
920static DEFINE_MUTEX(kprobe_sysctl_mutex);
921int sysctl_kprobes_optimization;
922int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
923 void *buffer, size_t *length,
924 loff_t *ppos)
925{
926 int ret;
927
928 mutex_lock(&kprobe_sysctl_mutex);
929 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
930 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
931
932 if (sysctl_kprobes_optimization)
933 optimize_all_kprobes();
934 else
935 unoptimize_all_kprobes();
936 mutex_unlock(&kprobe_sysctl_mutex);
937
938 return ret;
939}
940#endif /* CONFIG_SYSCTL */
941
942/* Put a breakpoint for a probe. Must be called with text_mutex locked */
943static void __arm_kprobe(struct kprobe *p)
944{
945 struct kprobe *_p;
946
947 /* Check collision with other optimized kprobes */
948 _p = get_optimized_kprobe((unsigned long)p->addr);
949 if (unlikely(_p))
950 /* Fallback to unoptimized kprobe */
951 unoptimize_kprobe(_p, true);
952
953 arch_arm_kprobe(p);
954 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
955}
956
957/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
958static void __disarm_kprobe(struct kprobe *p, bool reopt)
959{
960 struct kprobe *_p;
961
962 /* Try to unoptimize */
963 unoptimize_kprobe(p, kprobes_all_disarmed);
964
965 if (!kprobe_queued(p)) {
966 arch_disarm_kprobe(p);
967 /* If another kprobe was blocked, optimize it. */
968 _p = get_optimized_kprobe((unsigned long)p->addr);
969 if (unlikely(_p) && reopt)
970 optimize_kprobe(_p);
971 }
972 /* TODO: reoptimize others after unoptimized this probe */
973}
974
975#else /* !CONFIG_OPTPROBES */
976
977#define optimize_kprobe(p) do {} while (0)
978#define unoptimize_kprobe(p, f) do {} while (0)
979#define kill_optimized_kprobe(p) do {} while (0)
980#define prepare_optimized_kprobe(p) do {} while (0)
981#define try_to_optimize_kprobe(p) do {} while (0)
982#define __arm_kprobe(p) arch_arm_kprobe(p)
983#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
984#define kprobe_disarmed(p) kprobe_disabled(p)
985#define wait_for_kprobe_optimizer() do {} while (0)
986
987static int reuse_unused_kprobe(struct kprobe *ap)
988{
989 /*
990 * If the optimized kprobe is NOT supported, the aggr kprobe is
991 * released at the same time that the last aggregated kprobe is
992 * unregistered.
993 * Thus there should be no chance to reuse unused kprobe.
994 */
995 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
996 return -EINVAL;
997}
998
999static void free_aggr_kprobe(struct kprobe *p)
1000{
1001 arch_remove_kprobe(p);
1002 kfree(p);
1003}
1004
1005static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1006{
1007 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1008}
1009#endif /* CONFIG_OPTPROBES */
1010
1011#ifdef CONFIG_KPROBES_ON_FTRACE
1012static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1013 .func = kprobe_ftrace_handler,
1014 .flags = FTRACE_OPS_FL_SAVE_REGS,
1015};
1016
1017static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1018 .func = kprobe_ftrace_handler,
1019 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1020};
1021
1022static int kprobe_ipmodify_enabled;
1023static int kprobe_ftrace_enabled;
1024
1025/* Must ensure p->addr is really on ftrace */
1026static int prepare_kprobe(struct kprobe *p)
1027{
1028 if (!kprobe_ftrace(p))
1029 return arch_prepare_kprobe(p);
1030
1031 return arch_prepare_kprobe_ftrace(p);
1032}
1033
1034/* Caller must lock kprobe_mutex */
1035static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1036 int *cnt)
1037{
1038 int ret = 0;
1039
1040 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1041 if (ret) {
1042 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1043 p->addr, ret);
1044 return ret;
1045 }
1046
1047 if (*cnt == 0) {
1048 ret = register_ftrace_function(ops);
1049 if (ret) {
1050 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1051 goto err_ftrace;
1052 }
1053 }
1054
1055 (*cnt)++;
1056 return ret;
1057
1058err_ftrace:
1059 /*
1060 * At this point, sinec ops is not registered, we should be sefe from
1061 * registering empty filter.
1062 */
1063 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1064 return ret;
1065}
1066
1067static int arm_kprobe_ftrace(struct kprobe *p)
1068{
1069 bool ipmodify = (p->post_handler != NULL);
1070
1071 return __arm_kprobe_ftrace(p,
1072 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1073 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1074}
1075
1076/* Caller must lock kprobe_mutex */
1077static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1078 int *cnt)
1079{
1080 int ret = 0;
1081
1082 if (*cnt == 1) {
1083 ret = unregister_ftrace_function(ops);
1084 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1085 return ret;
1086 }
1087
1088 (*cnt)--;
1089
1090 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1091 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1092 p->addr, ret);
1093 return ret;
1094}
1095
1096static int disarm_kprobe_ftrace(struct kprobe *p)
1097{
1098 bool ipmodify = (p->post_handler != NULL);
1099
1100 return __disarm_kprobe_ftrace(p,
1101 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1102 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1103}
1104#else /* !CONFIG_KPROBES_ON_FTRACE */
1105static inline int prepare_kprobe(struct kprobe *p)
1106{
1107 return arch_prepare_kprobe(p);
1108}
1109
1110static inline int arm_kprobe_ftrace(struct kprobe *p)
1111{
1112 return -ENODEV;
1113}
1114
1115static inline int disarm_kprobe_ftrace(struct kprobe *p)
1116{
1117 return -ENODEV;
1118}
1119#endif
1120
1121/* Arm a kprobe with text_mutex */
1122static int arm_kprobe(struct kprobe *kp)
1123{
1124 if (unlikely(kprobe_ftrace(kp)))
1125 return arm_kprobe_ftrace(kp);
1126
1127 cpus_read_lock();
1128 mutex_lock(&text_mutex);
1129 __arm_kprobe(kp);
1130 mutex_unlock(&text_mutex);
1131 cpus_read_unlock();
1132
1133 return 0;
1134}
1135
1136/* Disarm a kprobe with text_mutex */
1137static int disarm_kprobe(struct kprobe *kp, bool reopt)
1138{
1139 if (unlikely(kprobe_ftrace(kp)))
1140 return disarm_kprobe_ftrace(kp);
1141
1142 cpus_read_lock();
1143 mutex_lock(&text_mutex);
1144 __disarm_kprobe(kp, reopt);
1145 mutex_unlock(&text_mutex);
1146 cpus_read_unlock();
1147
1148 return 0;
1149}
1150
1151/*
1152 * Aggregate handlers for multiple kprobes support - these handlers
1153 * take care of invoking the individual kprobe handlers on p->list
1154 */
1155static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1156{
1157 struct kprobe *kp;
1158
1159 list_for_each_entry_rcu(kp, &p->list, list) {
1160 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1161 set_kprobe_instance(kp);
1162 if (kp->pre_handler(kp, regs))
1163 return 1;
1164 }
1165 reset_kprobe_instance();
1166 }
1167 return 0;
1168}
1169NOKPROBE_SYMBOL(aggr_pre_handler);
1170
1171static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1172 unsigned long flags)
1173{
1174 struct kprobe *kp;
1175
1176 list_for_each_entry_rcu(kp, &p->list, list) {
1177 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1178 set_kprobe_instance(kp);
1179 kp->post_handler(kp, regs, flags);
1180 reset_kprobe_instance();
1181 }
1182 }
1183}
1184NOKPROBE_SYMBOL(aggr_post_handler);
1185
1186static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1187 int trapnr)
1188{
1189 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1190
1191 /*
1192 * if we faulted "during" the execution of a user specified
1193 * probe handler, invoke just that probe's fault handler
1194 */
1195 if (cur && cur->fault_handler) {
1196 if (cur->fault_handler(cur, regs, trapnr))
1197 return 1;
1198 }
1199 return 0;
1200}
1201NOKPROBE_SYMBOL(aggr_fault_handler);
1202
1203/* Walks the list and increments nmissed count for multiprobe case */
1204void kprobes_inc_nmissed_count(struct kprobe *p)
1205{
1206 struct kprobe *kp;
1207 if (!kprobe_aggrprobe(p)) {
1208 p->nmissed++;
1209 } else {
1210 list_for_each_entry_rcu(kp, &p->list, list)
1211 kp->nmissed++;
1212 }
1213 return;
1214}
1215NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1216
1217static void free_rp_inst_rcu(struct rcu_head *head)
1218{
1219 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1220
1221 if (refcount_dec_and_test(&ri->rph->ref))
1222 kfree(ri->rph);
1223 kfree(ri);
1224}
1225NOKPROBE_SYMBOL(free_rp_inst_rcu);
1226
1227static void recycle_rp_inst(struct kretprobe_instance *ri)
1228{
1229 struct kretprobe *rp = get_kretprobe(ri);
1230
1231 if (likely(rp)) {
1232 freelist_add(&ri->freelist, &rp->freelist);
1233 } else
1234 call_rcu(&ri->rcu, free_rp_inst_rcu);
1235}
1236NOKPROBE_SYMBOL(recycle_rp_inst);
1237
1238static struct kprobe kprobe_busy = {
1239 .addr = (void *) get_kprobe,
1240};
1241
1242void kprobe_busy_begin(void)
1243{
1244 struct kprobe_ctlblk *kcb;
1245
1246 preempt_disable();
1247 __this_cpu_write(current_kprobe, &kprobe_busy);
1248 kcb = get_kprobe_ctlblk();
1249 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1250}
1251
1252void kprobe_busy_end(void)
1253{
1254 __this_cpu_write(current_kprobe, NULL);
1255 preempt_enable();
1256}
1257
1258/*
1259 * This function is called from finish_task_switch when task tk becomes dead,
1260 * so that we can recycle any function-return probe instances associated
1261 * with this task. These left over instances represent probed functions
1262 * that have been called but will never return.
1263 */
1264void kprobe_flush_task(struct task_struct *tk)
1265{
1266 struct kretprobe_instance *ri;
1267 struct llist_node *node;
1268
1269 /* Early boot, not yet initialized. */
1270 if (unlikely(!kprobes_initialized))
1271 return;
1272
1273 kprobe_busy_begin();
1274
1275 node = __llist_del_all(&tk->kretprobe_instances);
1276 while (node) {
1277 ri = container_of(node, struct kretprobe_instance, llist);
1278 node = node->next;
1279
1280 recycle_rp_inst(ri);
1281 }
1282
1283 kprobe_busy_end();
1284}
1285NOKPROBE_SYMBOL(kprobe_flush_task);
1286
1287static inline void free_rp_inst(struct kretprobe *rp)
1288{
1289 struct kretprobe_instance *ri;
1290 struct freelist_node *node;
1291 int count = 0;
1292
1293 node = rp->freelist.head;
1294 while (node) {
1295 ri = container_of(node, struct kretprobe_instance, freelist);
1296 node = node->next;
1297
1298 kfree(ri);
1299 count++;
1300 }
1301
1302 if (refcount_sub_and_test(count, &rp->rph->ref)) {
1303 kfree(rp->rph);
1304 rp->rph = NULL;
1305 }
1306}
1307
1308/* Add the new probe to ap->list */
1309static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1310{
1311 if (p->post_handler)
1312 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1313
1314 list_add_rcu(&p->list, &ap->list);
1315 if (p->post_handler && !ap->post_handler)
1316 ap->post_handler = aggr_post_handler;
1317
1318 return 0;
1319}
1320
1321/*
1322 * Fill in the required fields of the "manager kprobe". Replace the
1323 * earlier kprobe in the hlist with the manager kprobe
1324 */
1325static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1326{
1327 /* Copy p's insn slot to ap */
1328 copy_kprobe(p, ap);
1329 flush_insn_slot(ap);
1330 ap->addr = p->addr;
1331 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1332 ap->pre_handler = aggr_pre_handler;
1333 ap->fault_handler = aggr_fault_handler;
1334 /* We don't care the kprobe which has gone. */
1335 if (p->post_handler && !kprobe_gone(p))
1336 ap->post_handler = aggr_post_handler;
1337
1338 INIT_LIST_HEAD(&ap->list);
1339 INIT_HLIST_NODE(&ap->hlist);
1340
1341 list_add_rcu(&p->list, &ap->list);
1342 hlist_replace_rcu(&p->hlist, &ap->hlist);
1343}
1344
1345/*
1346 * This is the second or subsequent kprobe at the address - handle
1347 * the intricacies
1348 */
1349static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1350{
1351 int ret = 0;
1352 struct kprobe *ap = orig_p;
1353
1354 cpus_read_lock();
1355
1356 /* For preparing optimization, jump_label_text_reserved() is called */
1357 jump_label_lock();
1358 mutex_lock(&text_mutex);
1359
1360 if (!kprobe_aggrprobe(orig_p)) {
1361 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1362 ap = alloc_aggr_kprobe(orig_p);
1363 if (!ap) {
1364 ret = -ENOMEM;
1365 goto out;
1366 }
1367 init_aggr_kprobe(ap, orig_p);
1368 } else if (kprobe_unused(ap)) {
1369 /* This probe is going to die. Rescue it */
1370 ret = reuse_unused_kprobe(ap);
1371 if (ret)
1372 goto out;
1373 }
1374
1375 if (kprobe_gone(ap)) {
1376 /*
1377 * Attempting to insert new probe at the same location that
1378 * had a probe in the module vaddr area which already
1379 * freed. So, the instruction slot has already been
1380 * released. We need a new slot for the new probe.
1381 */
1382 ret = arch_prepare_kprobe(ap);
1383 if (ret)
1384 /*
1385 * Even if fail to allocate new slot, don't need to
1386 * free aggr_probe. It will be used next time, or
1387 * freed by unregister_kprobe.
1388 */
1389 goto out;
1390
1391 /* Prepare optimized instructions if possible. */
1392 prepare_optimized_kprobe(ap);
1393
1394 /*
1395 * Clear gone flag to prevent allocating new slot again, and
1396 * set disabled flag because it is not armed yet.
1397 */
1398 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1399 | KPROBE_FLAG_DISABLED;
1400 }
1401
1402 /* Copy ap's insn slot to p */
1403 copy_kprobe(ap, p);
1404 ret = add_new_kprobe(ap, p);
1405
1406out:
1407 mutex_unlock(&text_mutex);
1408 jump_label_unlock();
1409 cpus_read_unlock();
1410
1411 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1412 ap->flags &= ~KPROBE_FLAG_DISABLED;
1413 if (!kprobes_all_disarmed) {
1414 /* Arm the breakpoint again. */
1415 ret = arm_kprobe(ap);
1416 if (ret) {
1417 ap->flags |= KPROBE_FLAG_DISABLED;
1418 list_del_rcu(&p->list);
1419 synchronize_rcu();
1420 }
1421 }
1422 }
1423 return ret;
1424}
1425
1426bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1427{
1428 /* The __kprobes marked functions and entry code must not be probed */
1429 return addr >= (unsigned long)__kprobes_text_start &&
1430 addr < (unsigned long)__kprobes_text_end;
1431}
1432
1433static bool __within_kprobe_blacklist(unsigned long addr)
1434{
1435 struct kprobe_blacklist_entry *ent;
1436
1437 if (arch_within_kprobe_blacklist(addr))
1438 return true;
1439 /*
1440 * If there exists a kprobe_blacklist, verify and
1441 * fail any probe registration in the prohibited area
1442 */
1443 list_for_each_entry(ent, &kprobe_blacklist, list) {
1444 if (addr >= ent->start_addr && addr < ent->end_addr)
1445 return true;
1446 }
1447 return false;
1448}
1449
1450bool within_kprobe_blacklist(unsigned long addr)
1451{
1452 char symname[KSYM_NAME_LEN], *p;
1453
1454 if (__within_kprobe_blacklist(addr))
1455 return true;
1456
1457 /* Check if the address is on a suffixed-symbol */
1458 if (!lookup_symbol_name(addr, symname)) {
1459 p = strchr(symname, '.');
1460 if (!p)
1461 return false;
1462 *p = '\0';
1463 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1464 if (addr)
1465 return __within_kprobe_blacklist(addr);
1466 }
1467 return false;
1468}
1469
1470/*
1471 * If we have a symbol_name argument, look it up and add the offset field
1472 * to it. This way, we can specify a relative address to a symbol.
1473 * This returns encoded errors if it fails to look up symbol or invalid
1474 * combination of parameters.
1475 */
1476static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1477 const char *symbol_name, unsigned int offset)
1478{
1479 if ((symbol_name && addr) || (!symbol_name && !addr))
1480 goto invalid;
1481
1482 if (symbol_name) {
1483 addr = kprobe_lookup_name(symbol_name, offset);
1484 if (!addr)
1485 return ERR_PTR(-ENOENT);
1486 }
1487
1488 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1489 if (addr)
1490 return addr;
1491
1492invalid:
1493 return ERR_PTR(-EINVAL);
1494}
1495
1496static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1497{
1498 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1499}
1500
1501/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1502static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1503{
1504 struct kprobe *ap, *list_p;
1505
1506 lockdep_assert_held(&kprobe_mutex);
1507
1508 ap = get_kprobe(p->addr);
1509 if (unlikely(!ap))
1510 return NULL;
1511
1512 if (p != ap) {
1513 list_for_each_entry(list_p, &ap->list, list)
1514 if (list_p == p)
1515 /* kprobe p is a valid probe */
1516 goto valid;
1517 return NULL;
1518 }
1519valid:
1520 return ap;
1521}
1522
1523/*
1524 * Warn and return error if the kprobe is being re-registered since
1525 * there must be a software bug.
1526 */
1527static inline int warn_kprobe_rereg(struct kprobe *p)
1528{
1529 int ret = 0;
1530
1531 mutex_lock(&kprobe_mutex);
1532 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1533 ret = -EINVAL;
1534 mutex_unlock(&kprobe_mutex);
1535
1536 return ret;
1537}
1538
1539int __weak arch_check_ftrace_location(struct kprobe *p)
1540{
1541 unsigned long ftrace_addr;
1542
1543 ftrace_addr = ftrace_location((unsigned long)p->addr);
1544 if (ftrace_addr) {
1545#ifdef CONFIG_KPROBES_ON_FTRACE
1546 /* Given address is not on the instruction boundary */
1547 if ((unsigned long)p->addr != ftrace_addr)
1548 return -EILSEQ;
1549 p->flags |= KPROBE_FLAG_FTRACE;
1550#else /* !CONFIG_KPROBES_ON_FTRACE */
1551 return -EINVAL;
1552#endif
1553 }
1554 return 0;
1555}
1556
1557static int check_kprobe_address_safe(struct kprobe *p,
1558 struct module **probed_mod)
1559{
1560 int ret;
1561
1562 ret = arch_check_ftrace_location(p);
1563 if (ret)
1564 return ret;
1565 jump_label_lock();
1566 preempt_disable();
1567
1568 /* Ensure it is not in reserved area nor out of text */
1569 if (!kernel_text_address((unsigned long) p->addr) ||
1570 within_kprobe_blacklist((unsigned long) p->addr) ||
1571 jump_label_text_reserved(p->addr, p->addr) ||
1572 find_bug((unsigned long)p->addr)) {
1573 ret = -EINVAL;
1574 goto out;
1575 }
1576
1577 /* Check if are we probing a module */
1578 *probed_mod = __module_text_address((unsigned long) p->addr);
1579 if (*probed_mod) {
1580 /*
1581 * We must hold a refcount of the probed module while updating
1582 * its code to prohibit unexpected unloading.
1583 */
1584 if (unlikely(!try_module_get(*probed_mod))) {
1585 ret = -ENOENT;
1586 goto out;
1587 }
1588
1589 /*
1590 * If the module freed .init.text, we couldn't insert
1591 * kprobes in there.
1592 */
1593 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1594 (*probed_mod)->state != MODULE_STATE_COMING) {
1595 module_put(*probed_mod);
1596 *probed_mod = NULL;
1597 ret = -ENOENT;
1598 }
1599 }
1600out:
1601 preempt_enable();
1602 jump_label_unlock();
1603
1604 return ret;
1605}
1606
1607int register_kprobe(struct kprobe *p)
1608{
1609 int ret;
1610 struct kprobe *old_p;
1611 struct module *probed_mod;
1612 kprobe_opcode_t *addr;
1613
1614 /* Adjust probe address from symbol */
1615 addr = kprobe_addr(p);
1616 if (IS_ERR(addr))
1617 return PTR_ERR(addr);
1618 p->addr = addr;
1619
1620 ret = warn_kprobe_rereg(p);
1621 if (ret)
1622 return ret;
1623
1624 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1625 p->flags &= KPROBE_FLAG_DISABLED;
1626 p->nmissed = 0;
1627 INIT_LIST_HEAD(&p->list);
1628
1629 ret = check_kprobe_address_safe(p, &probed_mod);
1630 if (ret)
1631 return ret;
1632
1633 mutex_lock(&kprobe_mutex);
1634
1635 old_p = get_kprobe(p->addr);
1636 if (old_p) {
1637 /* Since this may unoptimize old_p, locking text_mutex. */
1638 ret = register_aggr_kprobe(old_p, p);
1639 goto out;
1640 }
1641
1642 cpus_read_lock();
1643 /* Prevent text modification */
1644 mutex_lock(&text_mutex);
1645 ret = prepare_kprobe(p);
1646 mutex_unlock(&text_mutex);
1647 cpus_read_unlock();
1648 if (ret)
1649 goto out;
1650
1651 INIT_HLIST_NODE(&p->hlist);
1652 hlist_add_head_rcu(&p->hlist,
1653 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1654
1655 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1656 ret = arm_kprobe(p);
1657 if (ret) {
1658 hlist_del_rcu(&p->hlist);
1659 synchronize_rcu();
1660 goto out;
1661 }
1662 }
1663
1664 /* Try to optimize kprobe */
1665 try_to_optimize_kprobe(p);
1666out:
1667 mutex_unlock(&kprobe_mutex);
1668
1669 if (probed_mod)
1670 module_put(probed_mod);
1671
1672 return ret;
1673}
1674EXPORT_SYMBOL_GPL(register_kprobe);
1675
1676/* Check if all probes on the aggrprobe are disabled */
1677static int aggr_kprobe_disabled(struct kprobe *ap)
1678{
1679 struct kprobe *kp;
1680
1681 lockdep_assert_held(&kprobe_mutex);
1682
1683 list_for_each_entry(kp, &ap->list, list)
1684 if (!kprobe_disabled(kp))
1685 /*
1686 * There is an active probe on the list.
1687 * We can't disable this ap.
1688 */
1689 return 0;
1690
1691 return 1;
1692}
1693
1694/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1695static struct kprobe *__disable_kprobe(struct kprobe *p)
1696{
1697 struct kprobe *orig_p;
1698 int ret;
1699
1700 /* Get an original kprobe for return */
1701 orig_p = __get_valid_kprobe(p);
1702 if (unlikely(orig_p == NULL))
1703 return ERR_PTR(-EINVAL);
1704
1705 if (!kprobe_disabled(p)) {
1706 /* Disable probe if it is a child probe */
1707 if (p != orig_p)
1708 p->flags |= KPROBE_FLAG_DISABLED;
1709
1710 /* Try to disarm and disable this/parent probe */
1711 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1712 /*
1713 * If kprobes_all_disarmed is set, orig_p
1714 * should have already been disarmed, so
1715 * skip unneed disarming process.
1716 */
1717 if (!kprobes_all_disarmed) {
1718 ret = disarm_kprobe(orig_p, true);
1719 if (ret) {
1720 p->flags &= ~KPROBE_FLAG_DISABLED;
1721 return ERR_PTR(ret);
1722 }
1723 }
1724 orig_p->flags |= KPROBE_FLAG_DISABLED;
1725 }
1726 }
1727
1728 return orig_p;
1729}
1730
1731/*
1732 * Unregister a kprobe without a scheduler synchronization.
1733 */
1734static int __unregister_kprobe_top(struct kprobe *p)
1735{
1736 struct kprobe *ap, *list_p;
1737
1738 /* Disable kprobe. This will disarm it if needed. */
1739 ap = __disable_kprobe(p);
1740 if (IS_ERR(ap))
1741 return PTR_ERR(ap);
1742
1743 if (ap == p)
1744 /*
1745 * This probe is an independent(and non-optimized) kprobe
1746 * (not an aggrprobe). Remove from the hash list.
1747 */
1748 goto disarmed;
1749
1750 /* Following process expects this probe is an aggrprobe */
1751 WARN_ON(!kprobe_aggrprobe(ap));
1752
1753 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1754 /*
1755 * !disarmed could be happen if the probe is under delayed
1756 * unoptimizing.
1757 */
1758 goto disarmed;
1759 else {
1760 /* If disabling probe has special handlers, update aggrprobe */
1761 if (p->post_handler && !kprobe_gone(p)) {
1762 list_for_each_entry(list_p, &ap->list, list) {
1763 if ((list_p != p) && (list_p->post_handler))
1764 goto noclean;
1765 }
1766 ap->post_handler = NULL;
1767 }
1768noclean:
1769 /*
1770 * Remove from the aggrprobe: this path will do nothing in
1771 * __unregister_kprobe_bottom().
1772 */
1773 list_del_rcu(&p->list);
1774 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1775 /*
1776 * Try to optimize this probe again, because post
1777 * handler may have been changed.
1778 */
1779 optimize_kprobe(ap);
1780 }
1781 return 0;
1782
1783disarmed:
1784 hlist_del_rcu(&ap->hlist);
1785 return 0;
1786}
1787
1788static void __unregister_kprobe_bottom(struct kprobe *p)
1789{
1790 struct kprobe *ap;
1791
1792 if (list_empty(&p->list))
1793 /* This is an independent kprobe */
1794 arch_remove_kprobe(p);
1795 else if (list_is_singular(&p->list)) {
1796 /* This is the last child of an aggrprobe */
1797 ap = list_entry(p->list.next, struct kprobe, list);
1798 list_del(&p->list);
1799 free_aggr_kprobe(ap);
1800 }
1801 /* Otherwise, do nothing. */
1802}
1803
1804int register_kprobes(struct kprobe **kps, int num)
1805{
1806 int i, ret = 0;
1807
1808 if (num <= 0)
1809 return -EINVAL;
1810 for (i = 0; i < num; i++) {
1811 ret = register_kprobe(kps[i]);
1812 if (ret < 0) {
1813 if (i > 0)
1814 unregister_kprobes(kps, i);
1815 break;
1816 }
1817 }
1818 return ret;
1819}
1820EXPORT_SYMBOL_GPL(register_kprobes);
1821
1822void unregister_kprobe(struct kprobe *p)
1823{
1824 unregister_kprobes(&p, 1);
1825}
1826EXPORT_SYMBOL_GPL(unregister_kprobe);
1827
1828void unregister_kprobes(struct kprobe **kps, int num)
1829{
1830 int i;
1831
1832 if (num <= 0)
1833 return;
1834 mutex_lock(&kprobe_mutex);
1835 for (i = 0; i < num; i++)
1836 if (__unregister_kprobe_top(kps[i]) < 0)
1837 kps[i]->addr = NULL;
1838 mutex_unlock(&kprobe_mutex);
1839
1840 synchronize_rcu();
1841 for (i = 0; i < num; i++)
1842 if (kps[i]->addr)
1843 __unregister_kprobe_bottom(kps[i]);
1844}
1845EXPORT_SYMBOL_GPL(unregister_kprobes);
1846
1847int __weak kprobe_exceptions_notify(struct notifier_block *self,
1848 unsigned long val, void *data)
1849{
1850 return NOTIFY_DONE;
1851}
1852NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1853
1854static struct notifier_block kprobe_exceptions_nb = {
1855 .notifier_call = kprobe_exceptions_notify,
1856 .priority = 0x7fffffff /* we need to be notified first */
1857};
1858
1859unsigned long __weak arch_deref_entry_point(void *entry)
1860{
1861 return (unsigned long)entry;
1862}
1863
1864#ifdef CONFIG_KRETPROBES
1865
1866unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1867 void *trampoline_address,
1868 void *frame_pointer)
1869{
1870 kprobe_opcode_t *correct_ret_addr = NULL;
1871 struct kretprobe_instance *ri = NULL;
1872 struct llist_node *first, *node;
1873 struct kretprobe *rp;
1874
1875 /* Find all nodes for this frame. */
1876 first = node = current->kretprobe_instances.first;
1877 while (node) {
1878 ri = container_of(node, struct kretprobe_instance, llist);
1879
1880 BUG_ON(ri->fp != frame_pointer);
1881
1882 if (ri->ret_addr != trampoline_address) {
1883 correct_ret_addr = ri->ret_addr;
1884 /*
1885 * This is the real return address. Any other
1886 * instances associated with this task are for
1887 * other calls deeper on the call stack
1888 */
1889 goto found;
1890 }
1891
1892 node = node->next;
1893 }
1894 pr_err("Oops! Kretprobe fails to find correct return address.\n");
1895 BUG_ON(1);
1896
1897found:
1898 /* Unlink all nodes for this frame. */
1899 current->kretprobe_instances.first = node->next;
1900 node->next = NULL;
1901
1902 /* Run them.. */
1903 while (first) {
1904 ri = container_of(first, struct kretprobe_instance, llist);
1905 first = first->next;
1906
1907 rp = get_kretprobe(ri);
1908 if (rp && rp->handler) {
1909 struct kprobe *prev = kprobe_running();
1910
1911 __this_cpu_write(current_kprobe, &rp->kp);
1912 ri->ret_addr = correct_ret_addr;
1913 rp->handler(ri, regs);
1914 __this_cpu_write(current_kprobe, prev);
1915 }
1916
1917 recycle_rp_inst(ri);
1918 }
1919
1920 return (unsigned long)correct_ret_addr;
1921}
1922NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1923
1924/*
1925 * This kprobe pre_handler is registered with every kretprobe. When probe
1926 * hits it will set up the return probe.
1927 */
1928static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1929{
1930 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1931 struct kretprobe_instance *ri;
1932 struct freelist_node *fn;
1933
1934 fn = freelist_try_get(&rp->freelist);
1935 if (!fn) {
1936 rp->nmissed++;
1937 return 0;
1938 }
1939
1940 ri = container_of(fn, struct kretprobe_instance, freelist);
1941
1942 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1943 freelist_add(&ri->freelist, &rp->freelist);
1944 return 0;
1945 }
1946
1947 arch_prepare_kretprobe(ri, regs);
1948
1949 __llist_add(&ri->llist, &current->kretprobe_instances);
1950
1951 return 0;
1952}
1953NOKPROBE_SYMBOL(pre_handler_kretprobe);
1954
1955bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1956{
1957 return !offset;
1958}
1959
1960/**
1961 * kprobe_on_func_entry() -- check whether given address is function entry
1962 * @addr: Target address
1963 * @sym: Target symbol name
1964 * @offset: The offset from the symbol or the address
1965 *
1966 * This checks whether the given @addr+@offset or @sym+@offset is on the
1967 * function entry address or not.
1968 * This returns 0 if it is the function entry, or -EINVAL if it is not.
1969 * And also it returns -ENOENT if it fails the symbol or address lookup.
1970 * Caller must pass @addr or @sym (either one must be NULL), or this
1971 * returns -EINVAL.
1972 */
1973int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1974{
1975 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1976
1977 if (IS_ERR(kp_addr))
1978 return PTR_ERR(kp_addr);
1979
1980 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
1981 return -ENOENT;
1982
1983 if (!arch_kprobe_on_func_entry(offset))
1984 return -EINVAL;
1985
1986 return 0;
1987}
1988
1989int register_kretprobe(struct kretprobe *rp)
1990{
1991 int ret;
1992 struct kretprobe_instance *inst;
1993 int i;
1994 void *addr;
1995
1996 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
1997 if (ret)
1998 return ret;
1999
2000 /* If only rp->kp.addr is specified, check reregistering kprobes */
2001 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2002 return -EINVAL;
2003
2004 if (kretprobe_blacklist_size) {
2005 addr = kprobe_addr(&rp->kp);
2006 if (IS_ERR(addr))
2007 return PTR_ERR(addr);
2008
2009 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2010 if (kretprobe_blacklist[i].addr == addr)
2011 return -EINVAL;
2012 }
2013 }
2014
2015 rp->kp.pre_handler = pre_handler_kretprobe;
2016 rp->kp.post_handler = NULL;
2017 rp->kp.fault_handler = NULL;
2018
2019 /* Pre-allocate memory for max kretprobe instances */
2020 if (rp->maxactive <= 0) {
2021#ifdef CONFIG_PREEMPTION
2022 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2023#else
2024 rp->maxactive = num_possible_cpus();
2025#endif
2026 }
2027 rp->freelist.head = NULL;
2028 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2029 if (!rp->rph)
2030 return -ENOMEM;
2031
2032 rp->rph->rp = rp;
2033 for (i = 0; i < rp->maxactive; i++) {
2034 inst = kzalloc(sizeof(struct kretprobe_instance) +
2035 rp->data_size, GFP_KERNEL);
2036 if (inst == NULL) {
2037 refcount_set(&rp->rph->ref, i);
2038 free_rp_inst(rp);
2039 return -ENOMEM;
2040 }
2041 inst->rph = rp->rph;
2042 freelist_add(&inst->freelist, &rp->freelist);
2043 }
2044 refcount_set(&rp->rph->ref, i);
2045
2046 rp->nmissed = 0;
2047 /* Establish function entry probe point */
2048 ret = register_kprobe(&rp->kp);
2049 if (ret != 0)
2050 free_rp_inst(rp);
2051 return ret;
2052}
2053EXPORT_SYMBOL_GPL(register_kretprobe);
2054
2055int register_kretprobes(struct kretprobe **rps, int num)
2056{
2057 int ret = 0, i;
2058
2059 if (num <= 0)
2060 return -EINVAL;
2061 for (i = 0; i < num; i++) {
2062 ret = register_kretprobe(rps[i]);
2063 if (ret < 0) {
2064 if (i > 0)
2065 unregister_kretprobes(rps, i);
2066 break;
2067 }
2068 }
2069 return ret;
2070}
2071EXPORT_SYMBOL_GPL(register_kretprobes);
2072
2073void unregister_kretprobe(struct kretprobe *rp)
2074{
2075 unregister_kretprobes(&rp, 1);
2076}
2077EXPORT_SYMBOL_GPL(unregister_kretprobe);
2078
2079void unregister_kretprobes(struct kretprobe **rps, int num)
2080{
2081 int i;
2082
2083 if (num <= 0)
2084 return;
2085 mutex_lock(&kprobe_mutex);
2086 for (i = 0; i < num; i++) {
2087 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2088 rps[i]->kp.addr = NULL;
2089 rps[i]->rph->rp = NULL;
2090 }
2091 mutex_unlock(&kprobe_mutex);
2092
2093 synchronize_rcu();
2094 for (i = 0; i < num; i++) {
2095 if (rps[i]->kp.addr) {
2096 __unregister_kprobe_bottom(&rps[i]->kp);
2097 free_rp_inst(rps[i]);
2098 }
2099 }
2100}
2101EXPORT_SYMBOL_GPL(unregister_kretprobes);
2102
2103#else /* CONFIG_KRETPROBES */
2104int register_kretprobe(struct kretprobe *rp)
2105{
2106 return -ENOSYS;
2107}
2108EXPORT_SYMBOL_GPL(register_kretprobe);
2109
2110int register_kretprobes(struct kretprobe **rps, int num)
2111{
2112 return -ENOSYS;
2113}
2114EXPORT_SYMBOL_GPL(register_kretprobes);
2115
2116void unregister_kretprobe(struct kretprobe *rp)
2117{
2118}
2119EXPORT_SYMBOL_GPL(unregister_kretprobe);
2120
2121void unregister_kretprobes(struct kretprobe **rps, int num)
2122{
2123}
2124EXPORT_SYMBOL_GPL(unregister_kretprobes);
2125
2126static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2127{
2128 return 0;
2129}
2130NOKPROBE_SYMBOL(pre_handler_kretprobe);
2131
2132#endif /* CONFIG_KRETPROBES */
2133
2134/* Set the kprobe gone and remove its instruction buffer. */
2135static void kill_kprobe(struct kprobe *p)
2136{
2137 struct kprobe *kp;
2138
2139 lockdep_assert_held(&kprobe_mutex);
2140
2141 p->flags |= KPROBE_FLAG_GONE;
2142 if (kprobe_aggrprobe(p)) {
2143 /*
2144 * If this is an aggr_kprobe, we have to list all the
2145 * chained probes and mark them GONE.
2146 */
2147 list_for_each_entry(kp, &p->list, list)
2148 kp->flags |= KPROBE_FLAG_GONE;
2149 p->post_handler = NULL;
2150 kill_optimized_kprobe(p);
2151 }
2152 /*
2153 * Here, we can remove insn_slot safely, because no thread calls
2154 * the original probed function (which will be freed soon) any more.
2155 */
2156 arch_remove_kprobe(p);
2157
2158 /*
2159 * The module is going away. We should disarm the kprobe which
2160 * is using ftrace, because ftrace framework is still available at
2161 * MODULE_STATE_GOING notification.
2162 */
2163 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2164 disarm_kprobe_ftrace(p);
2165}
2166
2167/* Disable one kprobe */
2168int disable_kprobe(struct kprobe *kp)
2169{
2170 int ret = 0;
2171 struct kprobe *p;
2172
2173 mutex_lock(&kprobe_mutex);
2174
2175 /* Disable this kprobe */
2176 p = __disable_kprobe(kp);
2177 if (IS_ERR(p))
2178 ret = PTR_ERR(p);
2179
2180 mutex_unlock(&kprobe_mutex);
2181 return ret;
2182}
2183EXPORT_SYMBOL_GPL(disable_kprobe);
2184
2185/* Enable one kprobe */
2186int enable_kprobe(struct kprobe *kp)
2187{
2188 int ret = 0;
2189 struct kprobe *p;
2190
2191 mutex_lock(&kprobe_mutex);
2192
2193 /* Check whether specified probe is valid. */
2194 p = __get_valid_kprobe(kp);
2195 if (unlikely(p == NULL)) {
2196 ret = -EINVAL;
2197 goto out;
2198 }
2199
2200 if (kprobe_gone(kp)) {
2201 /* This kprobe has gone, we couldn't enable it. */
2202 ret = -EINVAL;
2203 goto out;
2204 }
2205
2206 if (p != kp)
2207 kp->flags &= ~KPROBE_FLAG_DISABLED;
2208
2209 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2210 p->flags &= ~KPROBE_FLAG_DISABLED;
2211 ret = arm_kprobe(p);
2212 if (ret)
2213 p->flags |= KPROBE_FLAG_DISABLED;
2214 }
2215out:
2216 mutex_unlock(&kprobe_mutex);
2217 return ret;
2218}
2219EXPORT_SYMBOL_GPL(enable_kprobe);
2220
2221/* Caller must NOT call this in usual path. This is only for critical case */
2222void dump_kprobe(struct kprobe *kp)
2223{
2224 pr_err("Dumping kprobe:\n");
2225 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2226 kp->symbol_name, kp->offset, kp->addr);
2227}
2228NOKPROBE_SYMBOL(dump_kprobe);
2229
2230int kprobe_add_ksym_blacklist(unsigned long entry)
2231{
2232 struct kprobe_blacklist_entry *ent;
2233 unsigned long offset = 0, size = 0;
2234
2235 if (!kernel_text_address(entry) ||
2236 !kallsyms_lookup_size_offset(entry, &size, &offset))
2237 return -EINVAL;
2238
2239 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2240 if (!ent)
2241 return -ENOMEM;
2242 ent->start_addr = entry;
2243 ent->end_addr = entry + size;
2244 INIT_LIST_HEAD(&ent->list);
2245 list_add_tail(&ent->list, &kprobe_blacklist);
2246
2247 return (int)size;
2248}
2249
2250/* Add all symbols in given area into kprobe blacklist */
2251int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2252{
2253 unsigned long entry;
2254 int ret = 0;
2255
2256 for (entry = start; entry < end; entry += ret) {
2257 ret = kprobe_add_ksym_blacklist(entry);
2258 if (ret < 0)
2259 return ret;
2260 if (ret == 0) /* In case of alias symbol */
2261 ret = 1;
2262 }
2263 return 0;
2264}
2265
2266/* Remove all symbols in given area from kprobe blacklist */
2267static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2268{
2269 struct kprobe_blacklist_entry *ent, *n;
2270
2271 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2272 if (ent->start_addr < start || ent->start_addr >= end)
2273 continue;
2274 list_del(&ent->list);
2275 kfree(ent);
2276 }
2277}
2278
2279static void kprobe_remove_ksym_blacklist(unsigned long entry)
2280{
2281 kprobe_remove_area_blacklist(entry, entry + 1);
2282}
2283
2284int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2285 char *type, char *sym)
2286{
2287 return -ERANGE;
2288}
2289
2290int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2291 char *sym)
2292{
2293#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2294 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2295 return 0;
2296#ifdef CONFIG_OPTPROBES
2297 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2298 return 0;
2299#endif
2300#endif
2301 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2302 return 0;
2303 return -ERANGE;
2304}
2305
2306int __init __weak arch_populate_kprobe_blacklist(void)
2307{
2308 return 0;
2309}
2310
2311/*
2312 * Lookup and populate the kprobe_blacklist.
2313 *
2314 * Unlike the kretprobe blacklist, we'll need to determine
2315 * the range of addresses that belong to the said functions,
2316 * since a kprobe need not necessarily be at the beginning
2317 * of a function.
2318 */
2319static int __init populate_kprobe_blacklist(unsigned long *start,
2320 unsigned long *end)
2321{
2322 unsigned long entry;
2323 unsigned long *iter;
2324 int ret;
2325
2326 for (iter = start; iter < end; iter++) {
2327 entry = arch_deref_entry_point((void *)*iter);
2328 ret = kprobe_add_ksym_blacklist(entry);
2329 if (ret == -EINVAL)
2330 continue;
2331 if (ret < 0)
2332 return ret;
2333 }
2334
2335 /* Symbols in __kprobes_text are blacklisted */
2336 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2337 (unsigned long)__kprobes_text_end);
2338 if (ret)
2339 return ret;
2340
2341 /* Symbols in noinstr section are blacklisted */
2342 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2343 (unsigned long)__noinstr_text_end);
2344
2345 return ret ? : arch_populate_kprobe_blacklist();
2346}
2347
2348static void add_module_kprobe_blacklist(struct module *mod)
2349{
2350 unsigned long start, end;
2351 int i;
2352
2353 if (mod->kprobe_blacklist) {
2354 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2355 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2356 }
2357
2358 start = (unsigned long)mod->kprobes_text_start;
2359 if (start) {
2360 end = start + mod->kprobes_text_size;
2361 kprobe_add_area_blacklist(start, end);
2362 }
2363
2364 start = (unsigned long)mod->noinstr_text_start;
2365 if (start) {
2366 end = start + mod->noinstr_text_size;
2367 kprobe_add_area_blacklist(start, end);
2368 }
2369}
2370
2371static void remove_module_kprobe_blacklist(struct module *mod)
2372{
2373 unsigned long start, end;
2374 int i;
2375
2376 if (mod->kprobe_blacklist) {
2377 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2378 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2379 }
2380
2381 start = (unsigned long)mod->kprobes_text_start;
2382 if (start) {
2383 end = start + mod->kprobes_text_size;
2384 kprobe_remove_area_blacklist(start, end);
2385 }
2386
2387 start = (unsigned long)mod->noinstr_text_start;
2388 if (start) {
2389 end = start + mod->noinstr_text_size;
2390 kprobe_remove_area_blacklist(start, end);
2391 }
2392}
2393
2394/* Module notifier call back, checking kprobes on the module */
2395static int kprobes_module_callback(struct notifier_block *nb,
2396 unsigned long val, void *data)
2397{
2398 struct module *mod = data;
2399 struct hlist_head *head;
2400 struct kprobe *p;
2401 unsigned int i;
2402 int checkcore = (val == MODULE_STATE_GOING);
2403
2404 if (val == MODULE_STATE_COMING) {
2405 mutex_lock(&kprobe_mutex);
2406 add_module_kprobe_blacklist(mod);
2407 mutex_unlock(&kprobe_mutex);
2408 }
2409 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2410 return NOTIFY_DONE;
2411
2412 /*
2413 * When MODULE_STATE_GOING was notified, both of module .text and
2414 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2415 * notified, only .init.text section would be freed. We need to
2416 * disable kprobes which have been inserted in the sections.
2417 */
2418 mutex_lock(&kprobe_mutex);
2419 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2420 head = &kprobe_table[i];
2421 hlist_for_each_entry(p, head, hlist)
2422 if (within_module_init((unsigned long)p->addr, mod) ||
2423 (checkcore &&
2424 within_module_core((unsigned long)p->addr, mod))) {
2425 /*
2426 * The vaddr this probe is installed will soon
2427 * be vfreed buy not synced to disk. Hence,
2428 * disarming the breakpoint isn't needed.
2429 *
2430 * Note, this will also move any optimized probes
2431 * that are pending to be removed from their
2432 * corresponding lists to the freeing_list and
2433 * will not be touched by the delayed
2434 * kprobe_optimizer work handler.
2435 */
2436 kill_kprobe(p);
2437 }
2438 }
2439 if (val == MODULE_STATE_GOING)
2440 remove_module_kprobe_blacklist(mod);
2441 mutex_unlock(&kprobe_mutex);
2442 return NOTIFY_DONE;
2443}
2444
2445static struct notifier_block kprobe_module_nb = {
2446 .notifier_call = kprobes_module_callback,
2447 .priority = 0
2448};
2449
2450/* Markers of _kprobe_blacklist section */
2451extern unsigned long __start_kprobe_blacklist[];
2452extern unsigned long __stop_kprobe_blacklist[];
2453
2454void kprobe_free_init_mem(void)
2455{
2456 void *start = (void *)(&__init_begin);
2457 void *end = (void *)(&__init_end);
2458 struct hlist_head *head;
2459 struct kprobe *p;
2460 int i;
2461
2462 mutex_lock(&kprobe_mutex);
2463
2464 /* Kill all kprobes on initmem */
2465 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2466 head = &kprobe_table[i];
2467 hlist_for_each_entry(p, head, hlist) {
2468 if (start <= (void *)p->addr && (void *)p->addr < end)
2469 kill_kprobe(p);
2470 }
2471 }
2472
2473 mutex_unlock(&kprobe_mutex);
2474}
2475
2476static int __init init_kprobes(void)
2477{
2478 int i, err = 0;
2479
2480 /* FIXME allocate the probe table, currently defined statically */
2481 /* initialize all list heads */
2482 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2483 INIT_HLIST_HEAD(&kprobe_table[i]);
2484
2485 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2486 __stop_kprobe_blacklist);
2487 if (err) {
2488 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2489 pr_err("Please take care of using kprobes.\n");
2490 }
2491
2492 if (kretprobe_blacklist_size) {
2493 /* lookup the function address from its name */
2494 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2495 kretprobe_blacklist[i].addr =
2496 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2497 if (!kretprobe_blacklist[i].addr)
2498 printk("kretprobe: lookup failed: %s\n",
2499 kretprobe_blacklist[i].name);
2500 }
2501 }
2502
2503 /* By default, kprobes are armed */
2504 kprobes_all_disarmed = false;
2505
2506#if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2507 /* Init kprobe_optinsn_slots for allocation */
2508 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2509#endif
2510
2511 err = arch_init_kprobes();
2512 if (!err)
2513 err = register_die_notifier(&kprobe_exceptions_nb);
2514 if (!err)
2515 err = register_module_notifier(&kprobe_module_nb);
2516
2517 kprobes_initialized = (err == 0);
2518
2519 if (!err)
2520 init_test_probes();
2521 return err;
2522}
2523early_initcall(init_kprobes);
2524
2525#if defined(CONFIG_OPTPROBES)
2526static int __init init_optprobes(void)
2527{
2528 /*
2529 * Enable kprobe optimization - this kicks the optimizer which
2530 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2531 * not spawned in early initcall. So delay the optimization.
2532 */
2533 optimize_all_kprobes();
2534
2535 return 0;
2536}
2537subsys_initcall(init_optprobes);
2538#endif
2539
2540#ifdef CONFIG_DEBUG_FS
2541static void report_probe(struct seq_file *pi, struct kprobe *p,
2542 const char *sym, int offset, char *modname, struct kprobe *pp)
2543{
2544 char *kprobe_type;
2545 void *addr = p->addr;
2546
2547 if (p->pre_handler == pre_handler_kretprobe)
2548 kprobe_type = "r";
2549 else
2550 kprobe_type = "k";
2551
2552 if (!kallsyms_show_value(pi->file->f_cred))
2553 addr = NULL;
2554
2555 if (sym)
2556 seq_printf(pi, "%px %s %s+0x%x %s ",
2557 addr, kprobe_type, sym, offset,
2558 (modname ? modname : " "));
2559 else /* try to use %pS */
2560 seq_printf(pi, "%px %s %pS ",
2561 addr, kprobe_type, p->addr);
2562
2563 if (!pp)
2564 pp = p;
2565 seq_printf(pi, "%s%s%s%s\n",
2566 (kprobe_gone(p) ? "[GONE]" : ""),
2567 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2568 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2569 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2570}
2571
2572static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2573{
2574 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2575}
2576
2577static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2578{
2579 (*pos)++;
2580 if (*pos >= KPROBE_TABLE_SIZE)
2581 return NULL;
2582 return pos;
2583}
2584
2585static void kprobe_seq_stop(struct seq_file *f, void *v)
2586{
2587 /* Nothing to do */
2588}
2589
2590static int show_kprobe_addr(struct seq_file *pi, void *v)
2591{
2592 struct hlist_head *head;
2593 struct kprobe *p, *kp;
2594 const char *sym = NULL;
2595 unsigned int i = *(loff_t *) v;
2596 unsigned long offset = 0;
2597 char *modname, namebuf[KSYM_NAME_LEN];
2598
2599 head = &kprobe_table[i];
2600 preempt_disable();
2601 hlist_for_each_entry_rcu(p, head, hlist) {
2602 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2603 &offset, &modname, namebuf);
2604 if (kprobe_aggrprobe(p)) {
2605 list_for_each_entry_rcu(kp, &p->list, list)
2606 report_probe(pi, kp, sym, offset, modname, p);
2607 } else
2608 report_probe(pi, p, sym, offset, modname, NULL);
2609 }
2610 preempt_enable();
2611 return 0;
2612}
2613
2614static const struct seq_operations kprobes_sops = {
2615 .start = kprobe_seq_start,
2616 .next = kprobe_seq_next,
2617 .stop = kprobe_seq_stop,
2618 .show = show_kprobe_addr
2619};
2620
2621DEFINE_SEQ_ATTRIBUTE(kprobes);
2622
2623/* kprobes/blacklist -- shows which functions can not be probed */
2624static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2625{
2626 mutex_lock(&kprobe_mutex);
2627 return seq_list_start(&kprobe_blacklist, *pos);
2628}
2629
2630static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2631{
2632 return seq_list_next(v, &kprobe_blacklist, pos);
2633}
2634
2635static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2636{
2637 struct kprobe_blacklist_entry *ent =
2638 list_entry(v, struct kprobe_blacklist_entry, list);
2639
2640 /*
2641 * If /proc/kallsyms is not showing kernel address, we won't
2642 * show them here either.
2643 */
2644 if (!kallsyms_show_value(m->file->f_cred))
2645 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2646 (void *)ent->start_addr);
2647 else
2648 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2649 (void *)ent->end_addr, (void *)ent->start_addr);
2650 return 0;
2651}
2652
2653static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2654{
2655 mutex_unlock(&kprobe_mutex);
2656}
2657
2658static const struct seq_operations kprobe_blacklist_sops = {
2659 .start = kprobe_blacklist_seq_start,
2660 .next = kprobe_blacklist_seq_next,
2661 .stop = kprobe_blacklist_seq_stop,
2662 .show = kprobe_blacklist_seq_show,
2663};
2664DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2665
2666static int arm_all_kprobes(void)
2667{
2668 struct hlist_head *head;
2669 struct kprobe *p;
2670 unsigned int i, total = 0, errors = 0;
2671 int err, ret = 0;
2672
2673 mutex_lock(&kprobe_mutex);
2674
2675 /* If kprobes are armed, just return */
2676 if (!kprobes_all_disarmed)
2677 goto already_enabled;
2678
2679 /*
2680 * optimize_kprobe() called by arm_kprobe() checks
2681 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2682 * arm_kprobe.
2683 */
2684 kprobes_all_disarmed = false;
2685 /* Arming kprobes doesn't optimize kprobe itself */
2686 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2687 head = &kprobe_table[i];
2688 /* Arm all kprobes on a best-effort basis */
2689 hlist_for_each_entry(p, head, hlist) {
2690 if (!kprobe_disabled(p)) {
2691 err = arm_kprobe(p);
2692 if (err) {
2693 errors++;
2694 ret = err;
2695 }
2696 total++;
2697 }
2698 }
2699 }
2700
2701 if (errors)
2702 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2703 errors, total);
2704 else
2705 pr_info("Kprobes globally enabled\n");
2706
2707already_enabled:
2708 mutex_unlock(&kprobe_mutex);
2709 return ret;
2710}
2711
2712static int disarm_all_kprobes(void)
2713{
2714 struct hlist_head *head;
2715 struct kprobe *p;
2716 unsigned int i, total = 0, errors = 0;
2717 int err, ret = 0;
2718
2719 mutex_lock(&kprobe_mutex);
2720
2721 /* If kprobes are already disarmed, just return */
2722 if (kprobes_all_disarmed) {
2723 mutex_unlock(&kprobe_mutex);
2724 return 0;
2725 }
2726
2727 kprobes_all_disarmed = true;
2728
2729 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2730 head = &kprobe_table[i];
2731 /* Disarm all kprobes on a best-effort basis */
2732 hlist_for_each_entry(p, head, hlist) {
2733 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2734 err = disarm_kprobe(p, false);
2735 if (err) {
2736 errors++;
2737 ret = err;
2738 }
2739 total++;
2740 }
2741 }
2742 }
2743
2744 if (errors)
2745 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2746 errors, total);
2747 else
2748 pr_info("Kprobes globally disabled\n");
2749
2750 mutex_unlock(&kprobe_mutex);
2751
2752 /* Wait for disarming all kprobes by optimizer */
2753 wait_for_kprobe_optimizer();
2754
2755 return ret;
2756}
2757
2758/*
2759 * XXX: The debugfs bool file interface doesn't allow for callbacks
2760 * when the bool state is switched. We can reuse that facility when
2761 * available
2762 */
2763static ssize_t read_enabled_file_bool(struct file *file,
2764 char __user *user_buf, size_t count, loff_t *ppos)
2765{
2766 char buf[3];
2767
2768 if (!kprobes_all_disarmed)
2769 buf[0] = '1';
2770 else
2771 buf[0] = '0';
2772 buf[1] = '\n';
2773 buf[2] = 0x00;
2774 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2775}
2776
2777static ssize_t write_enabled_file_bool(struct file *file,
2778 const char __user *user_buf, size_t count, loff_t *ppos)
2779{
2780 char buf[32];
2781 size_t buf_size;
2782 int ret = 0;
2783
2784 buf_size = min(count, (sizeof(buf)-1));
2785 if (copy_from_user(buf, user_buf, buf_size))
2786 return -EFAULT;
2787
2788 buf[buf_size] = '\0';
2789 switch (buf[0]) {
2790 case 'y':
2791 case 'Y':
2792 case '1':
2793 ret = arm_all_kprobes();
2794 break;
2795 case 'n':
2796 case 'N':
2797 case '0':
2798 ret = disarm_all_kprobes();
2799 break;
2800 default:
2801 return -EINVAL;
2802 }
2803
2804 if (ret)
2805 return ret;
2806
2807 return count;
2808}
2809
2810static const struct file_operations fops_kp = {
2811 .read = read_enabled_file_bool,
2812 .write = write_enabled_file_bool,
2813 .llseek = default_llseek,
2814};
2815
2816static int __init debugfs_kprobe_init(void)
2817{
2818 struct dentry *dir;
2819 unsigned int value = 1;
2820
2821 dir = debugfs_create_dir("kprobes", NULL);
2822
2823 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2824
2825 debugfs_create_file("enabled", 0600, dir, &value, &fops_kp);
2826
2827 debugfs_create_file("blacklist", 0400, dir, NULL,
2828 &kprobe_blacklist_fops);
2829
2830 return 0;
2831}
2832
2833late_initcall(debugfs_kprobe_init);
2834#endif /* CONFIG_DEBUG_FS */