Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Kernel Probes (KProbes)
4 *
5 * Copyright (C) IBM Corporation, 2002, 2004
6 *
7 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
8 * Probes initial implementation (includes suggestions from
9 * Rusty Russell).
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.
19 */
20
21#define pr_fmt(fmt) "kprobes: " fmt
22
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/kallsyms.h>
30#include <linux/freezer.h>
31#include <linux/seq_file.h>
32#include <linux/debugfs.h>
33#include <linux/sysctl.h>
34#include <linux/kdebug.h>
35#include <linux/memory.h>
36#include <linux/ftrace.h>
37#include <linux/cpu.h>
38#include <linux/jump_label.h>
39#include <linux/static_call.h>
40#include <linux/perf_event.h>
41#include <linux/execmem.h>
42
43#include <asm/sections.h>
44#include <asm/cacheflush.h>
45#include <asm/errno.h>
46#include <linux/uaccess.h>
47
48#define KPROBE_HASH_BITS 6
49#define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
50
51#if !defined(CONFIG_OPTPROBES) || !defined(CONFIG_SYSCTL)
52#define kprobe_sysctls_init() do { } while (0)
53#endif
54
55static int kprobes_initialized;
56/* kprobe_table can be accessed by
57 * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
58 * Or
59 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
60 */
61static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
62
63/* NOTE: change this value only with 'kprobe_mutex' held */
64static bool kprobes_all_disarmed;
65
66/* This protects 'kprobe_table' and 'optimizing_list' */
67static DEFINE_MUTEX(kprobe_mutex);
68static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
69
70kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
71 unsigned int __unused)
72{
73 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
74}
75
76/*
77 * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
78 * kprobes can not probe.
79 */
80static LIST_HEAD(kprobe_blacklist);
81
82#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
83/*
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
88 */
89struct kprobe_insn_page {
90 struct list_head list;
91 kprobe_opcode_t *insns; /* Page of instruction slots */
92 struct kprobe_insn_cache *cache;
93 int nused;
94 int ngarbage;
95 char slot_used[];
96};
97
98static int slots_per_page(struct kprobe_insn_cache *c)
99{
100 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
101}
102
103enum kprobe_slot_state {
104 SLOT_CLEAN = 0,
105 SLOT_DIRTY = 1,
106 SLOT_USED = 2,
107};
108
109void __weak *alloc_insn_page(void)
110{
111 /*
112 * Use execmem_alloc() so this page is within +/- 2GB of where the
113 * kernel image and loaded module images reside. This is required
114 * for most of the architectures.
115 * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
116 */
117 return execmem_alloc(EXECMEM_KPROBES, PAGE_SIZE);
118}
119
120static void free_insn_page(void *page)
121{
122 execmem_free(page);
123}
124
125struct kprobe_insn_cache kprobe_insn_slots = {
126 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
127 .alloc = alloc_insn_page,
128 .free = free_insn_page,
129 .sym = KPROBE_INSN_PAGE_SYM,
130 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
131 .insn_size = MAX_INSN_SIZE,
132 .nr_garbage = 0,
133};
134static int collect_garbage_slots(struct kprobe_insn_cache *c);
135
136/**
137 * __get_insn_slot() - Find a slot on an executable page for an instruction.
138 * We allocate an executable page if there's no room on existing ones.
139 */
140kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
141{
142 struct kprobe_insn_page *kip;
143 kprobe_opcode_t *slot = NULL;
144
145 /* Since the slot array is not protected by rcu, we need a mutex */
146 mutex_lock(&c->mutex);
147 retry:
148 rcu_read_lock();
149 list_for_each_entry_rcu(kip, &c->pages, list) {
150 if (kip->nused < slots_per_page(c)) {
151 int i;
152
153 for (i = 0; i < slots_per_page(c); i++) {
154 if (kip->slot_used[i] == SLOT_CLEAN) {
155 kip->slot_used[i] = SLOT_USED;
156 kip->nused++;
157 slot = kip->insns + (i * c->insn_size);
158 rcu_read_unlock();
159 goto out;
160 }
161 }
162 /* kip->nused is broken. Fix it. */
163 kip->nused = slots_per_page(c);
164 WARN_ON(1);
165 }
166 }
167 rcu_read_unlock();
168
169 /* If there are any garbage slots, collect it and try again. */
170 if (c->nr_garbage && collect_garbage_slots(c) == 0)
171 goto retry;
172
173 /* All out of space. Need to allocate a new page. */
174 kip = kmalloc(struct_size(kip, slot_used, slots_per_page(c)), GFP_KERNEL);
175 if (!kip)
176 goto out;
177
178 kip->insns = c->alloc();
179 if (!kip->insns) {
180 kfree(kip);
181 goto out;
182 }
183 INIT_LIST_HEAD(&kip->list);
184 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
185 kip->slot_used[0] = SLOT_USED;
186 kip->nused = 1;
187 kip->ngarbage = 0;
188 kip->cache = c;
189 list_add_rcu(&kip->list, &c->pages);
190 slot = kip->insns;
191
192 /* Record the perf ksymbol register event after adding the page */
193 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
194 PAGE_SIZE, false, c->sym);
195out:
196 mutex_unlock(&c->mutex);
197 return slot;
198}
199
200/* Return true if all garbages are collected, otherwise false. */
201static bool collect_one_slot(struct kprobe_insn_page *kip, int idx)
202{
203 kip->slot_used[idx] = SLOT_CLEAN;
204 kip->nused--;
205 if (kip->nused != 0)
206 return false;
207
208 /*
209 * Page is no longer in use. Free it unless
210 * it's the last one. We keep the last one
211 * so as not to have to set it up again the
212 * next time somebody inserts a probe.
213 */
214 if (!list_is_singular(&kip->list)) {
215 /*
216 * Record perf ksymbol unregister event before removing
217 * the page.
218 */
219 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
220 (unsigned long)kip->insns, PAGE_SIZE, true,
221 kip->cache->sym);
222 list_del_rcu(&kip->list);
223 synchronize_rcu();
224 kip->cache->free(kip->insns);
225 kfree(kip);
226 }
227 return true;
228}
229
230static int collect_garbage_slots(struct kprobe_insn_cache *c)
231{
232 struct kprobe_insn_page *kip, *next;
233
234 /* Ensure no-one is interrupted on the garbages */
235 synchronize_rcu();
236
237 list_for_each_entry_safe(kip, next, &c->pages, list) {
238 int i;
239
240 if (kip->ngarbage == 0)
241 continue;
242 kip->ngarbage = 0; /* we will collect all garbages */
243 for (i = 0; i < slots_per_page(c); i++) {
244 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
245 break;
246 }
247 }
248 c->nr_garbage = 0;
249 return 0;
250}
251
252void __free_insn_slot(struct kprobe_insn_cache *c,
253 kprobe_opcode_t *slot, int dirty)
254{
255 struct kprobe_insn_page *kip;
256 long idx;
257
258 mutex_lock(&c->mutex);
259 rcu_read_lock();
260 list_for_each_entry_rcu(kip, &c->pages, list) {
261 idx = ((long)slot - (long)kip->insns) /
262 (c->insn_size * sizeof(kprobe_opcode_t));
263 if (idx >= 0 && idx < slots_per_page(c))
264 goto out;
265 }
266 /* Could not find this slot. */
267 WARN_ON(1);
268 kip = NULL;
269out:
270 rcu_read_unlock();
271 /* Mark and sweep: this may sleep */
272 if (kip) {
273 /* Check double free */
274 WARN_ON(kip->slot_used[idx] != SLOT_USED);
275 if (dirty) {
276 kip->slot_used[idx] = SLOT_DIRTY;
277 kip->ngarbage++;
278 if (++c->nr_garbage > slots_per_page(c))
279 collect_garbage_slots(c);
280 } else {
281 collect_one_slot(kip, idx);
282 }
283 }
284 mutex_unlock(&c->mutex);
285}
286
287/*
288 * Check given address is on the page of kprobe instruction slots.
289 * This will be used for checking whether the address on a stack
290 * is on a text area or not.
291 */
292bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
293{
294 struct kprobe_insn_page *kip;
295 bool ret = false;
296
297 rcu_read_lock();
298 list_for_each_entry_rcu(kip, &c->pages, list) {
299 if (addr >= (unsigned long)kip->insns &&
300 addr < (unsigned long)kip->insns + PAGE_SIZE) {
301 ret = true;
302 break;
303 }
304 }
305 rcu_read_unlock();
306
307 return ret;
308}
309
310int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
311 unsigned long *value, char *type, char *sym)
312{
313 struct kprobe_insn_page *kip;
314 int ret = -ERANGE;
315
316 rcu_read_lock();
317 list_for_each_entry_rcu(kip, &c->pages, list) {
318 if ((*symnum)--)
319 continue;
320 strscpy(sym, c->sym, KSYM_NAME_LEN);
321 *type = 't';
322 *value = (unsigned long)kip->insns;
323 ret = 0;
324 break;
325 }
326 rcu_read_unlock();
327
328 return ret;
329}
330
331#ifdef CONFIG_OPTPROBES
332void __weak *alloc_optinsn_page(void)
333{
334 return alloc_insn_page();
335}
336
337void __weak free_optinsn_page(void *page)
338{
339 free_insn_page(page);
340}
341
342/* For optimized_kprobe buffer */
343struct kprobe_insn_cache kprobe_optinsn_slots = {
344 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
345 .alloc = alloc_optinsn_page,
346 .free = free_optinsn_page,
347 .sym = KPROBE_OPTINSN_PAGE_SYM,
348 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
349 /* .insn_size is initialized later */
350 .nr_garbage = 0,
351};
352#endif /* CONFIG_OPTPROBES */
353#endif /* __ARCH_WANT_KPROBES_INSN_SLOT */
354
355/* We have preemption disabled.. so it is safe to use __ versions */
356static inline void set_kprobe_instance(struct kprobe *kp)
357{
358 __this_cpu_write(kprobe_instance, kp);
359}
360
361static inline void reset_kprobe_instance(void)
362{
363 __this_cpu_write(kprobe_instance, NULL);
364}
365
366/*
367 * This routine is called either:
368 * - under the 'kprobe_mutex' - during kprobe_[un]register().
369 * OR
370 * - with preemption disabled - from architecture specific code.
371 */
372struct kprobe *get_kprobe(void *addr)
373{
374 struct hlist_head *head;
375 struct kprobe *p;
376
377 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
378 hlist_for_each_entry_rcu(p, head, hlist,
379 lockdep_is_held(&kprobe_mutex)) {
380 if (p->addr == addr)
381 return p;
382 }
383
384 return NULL;
385}
386NOKPROBE_SYMBOL(get_kprobe);
387
388static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
389
390/* Return true if 'p' is an aggregator */
391static inline bool kprobe_aggrprobe(struct kprobe *p)
392{
393 return p->pre_handler == aggr_pre_handler;
394}
395
396/* Return true if 'p' is unused */
397static inline bool kprobe_unused(struct kprobe *p)
398{
399 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
400 list_empty(&p->list);
401}
402
403/* Keep all fields in the kprobe consistent. */
404static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
405{
406 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
407 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
408}
409
410#ifdef CONFIG_OPTPROBES
411/* NOTE: This is protected by 'kprobe_mutex'. */
412static bool kprobes_allow_optimization;
413
414/*
415 * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
416 * This must be called from arch-dep optimized caller.
417 */
418void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
419{
420 struct kprobe *kp;
421
422 list_for_each_entry_rcu(kp, &p->list, list) {
423 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
424 set_kprobe_instance(kp);
425 kp->pre_handler(kp, regs);
426 }
427 reset_kprobe_instance();
428 }
429}
430NOKPROBE_SYMBOL(opt_pre_handler);
431
432/* Free optimized instructions and optimized_kprobe */
433static void free_aggr_kprobe(struct kprobe *p)
434{
435 struct optimized_kprobe *op;
436
437 op = container_of(p, struct optimized_kprobe, kp);
438 arch_remove_optimized_kprobe(op);
439 arch_remove_kprobe(p);
440 kfree(op);
441}
442
443/* Return true if the kprobe is ready for optimization. */
444static inline int kprobe_optready(struct kprobe *p)
445{
446 struct optimized_kprobe *op;
447
448 if (kprobe_aggrprobe(p)) {
449 op = container_of(p, struct optimized_kprobe, kp);
450 return arch_prepared_optinsn(&op->optinsn);
451 }
452
453 return 0;
454}
455
456/* Return true if the kprobe is disarmed. Note: p must be on hash list */
457bool kprobe_disarmed(struct kprobe *p)
458{
459 struct optimized_kprobe *op;
460
461 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
462 if (!kprobe_aggrprobe(p))
463 return kprobe_disabled(p);
464
465 op = container_of(p, struct optimized_kprobe, kp);
466
467 return kprobe_disabled(p) && list_empty(&op->list);
468}
469
470/* Return true if the probe is queued on (un)optimizing lists */
471static bool kprobe_queued(struct kprobe *p)
472{
473 struct optimized_kprobe *op;
474
475 if (kprobe_aggrprobe(p)) {
476 op = container_of(p, struct optimized_kprobe, kp);
477 if (!list_empty(&op->list))
478 return true;
479 }
480 return false;
481}
482
483/*
484 * Return an optimized kprobe whose optimizing code replaces
485 * instructions including 'addr' (exclude breakpoint).
486 */
487static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
488{
489 int i;
490 struct kprobe *p = NULL;
491 struct optimized_kprobe *op;
492
493 /* Don't check i == 0, since that is a breakpoint case. */
494 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
495 p = get_kprobe(addr - i);
496
497 if (p && kprobe_optready(p)) {
498 op = container_of(p, struct optimized_kprobe, kp);
499 if (arch_within_optimized_kprobe(op, addr))
500 return p;
501 }
502
503 return NULL;
504}
505
506/* Optimization staging list, protected by 'kprobe_mutex' */
507static LIST_HEAD(optimizing_list);
508static LIST_HEAD(unoptimizing_list);
509static LIST_HEAD(freeing_list);
510
511static void kprobe_optimizer(struct work_struct *work);
512static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
513#define OPTIMIZE_DELAY 5
514
515/*
516 * Optimize (replace a breakpoint with a jump) kprobes listed on
517 * 'optimizing_list'.
518 */
519static void do_optimize_kprobes(void)
520{
521 lockdep_assert_held(&text_mutex);
522 /*
523 * The optimization/unoptimization refers 'online_cpus' via
524 * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
525 * And same time, 'text_mutex' will be held in cpu-hotplug and here.
526 * This combination can cause a deadlock (cpu-hotplug tries to lock
527 * 'text_mutex' but stop_machine() can not be done because
528 * the 'online_cpus' has been changed)
529 * To avoid this deadlock, caller must have locked cpu-hotplug
530 * for preventing cpu-hotplug outside of 'text_mutex' locking.
531 */
532 lockdep_assert_cpus_held();
533
534 /* Optimization never be done when disarmed */
535 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
536 list_empty(&optimizing_list))
537 return;
538
539 arch_optimize_kprobes(&optimizing_list);
540}
541
542/*
543 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
544 * if need) kprobes listed on 'unoptimizing_list'.
545 */
546static void do_unoptimize_kprobes(void)
547{
548 struct optimized_kprobe *op, *tmp;
549
550 lockdep_assert_held(&text_mutex);
551 /* See comment in do_optimize_kprobes() */
552 lockdep_assert_cpus_held();
553
554 if (!list_empty(&unoptimizing_list))
555 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
556
557 /* Loop on 'freeing_list' for disarming and removing from kprobe hash list */
558 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
559 /* Switching from detour code to origin */
560 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
561 /* Disarm probes if marked disabled and not gone */
562 if (kprobe_disabled(&op->kp) && !kprobe_gone(&op->kp))
563 arch_disarm_kprobe(&op->kp);
564 if (kprobe_unused(&op->kp)) {
565 /*
566 * Remove unused probes from hash list. After waiting
567 * for synchronization, these probes are reclaimed.
568 * (reclaiming is done by do_free_cleaned_kprobes().)
569 */
570 hlist_del_rcu(&op->kp.hlist);
571 } else
572 list_del_init(&op->list);
573 }
574}
575
576/* Reclaim all kprobes on the 'freeing_list' */
577static void do_free_cleaned_kprobes(void)
578{
579 struct optimized_kprobe *op, *tmp;
580
581 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
582 list_del_init(&op->list);
583 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
584 /*
585 * This must not happen, but if there is a kprobe
586 * still in use, keep it on kprobes hash list.
587 */
588 continue;
589 }
590 free_aggr_kprobe(&op->kp);
591 }
592}
593
594/* Start optimizer after OPTIMIZE_DELAY passed */
595static void kick_kprobe_optimizer(void)
596{
597 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
598}
599
600/* Kprobe jump optimizer */
601static void kprobe_optimizer(struct work_struct *work)
602{
603 mutex_lock(&kprobe_mutex);
604 cpus_read_lock();
605 mutex_lock(&text_mutex);
606
607 /*
608 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
609 * kprobes before waiting for quiesence period.
610 */
611 do_unoptimize_kprobes();
612
613 /*
614 * Step 2: Wait for quiesence period to ensure all potentially
615 * preempted tasks to have normally scheduled. Because optprobe
616 * may modify multiple instructions, there is a chance that Nth
617 * instruction is preempted. In that case, such tasks can return
618 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
619 * Note that on non-preemptive kernel, this is transparently converted
620 * to synchronoze_sched() to wait for all interrupts to have completed.
621 */
622 synchronize_rcu_tasks();
623
624 /* Step 3: Optimize kprobes after quiesence period */
625 do_optimize_kprobes();
626
627 /* Step 4: Free cleaned kprobes after quiesence period */
628 do_free_cleaned_kprobes();
629
630 mutex_unlock(&text_mutex);
631 cpus_read_unlock();
632
633 /* Step 5: Kick optimizer again if needed */
634 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
635 kick_kprobe_optimizer();
636
637 mutex_unlock(&kprobe_mutex);
638}
639
640/* Wait for completing optimization and unoptimization */
641void wait_for_kprobe_optimizer(void)
642{
643 mutex_lock(&kprobe_mutex);
644
645 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
646 mutex_unlock(&kprobe_mutex);
647
648 /* This will also make 'optimizing_work' execute immmediately */
649 flush_delayed_work(&optimizing_work);
650 /* 'optimizing_work' might not have been queued yet, relax */
651 cpu_relax();
652
653 mutex_lock(&kprobe_mutex);
654 }
655
656 mutex_unlock(&kprobe_mutex);
657}
658
659bool optprobe_queued_unopt(struct optimized_kprobe *op)
660{
661 struct optimized_kprobe *_op;
662
663 list_for_each_entry(_op, &unoptimizing_list, list) {
664 if (op == _op)
665 return true;
666 }
667
668 return false;
669}
670
671/* Optimize kprobe if p is ready to be optimized */
672static void optimize_kprobe(struct kprobe *p)
673{
674 struct optimized_kprobe *op;
675
676 /* Check if the kprobe is disabled or not ready for optimization. */
677 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
678 (kprobe_disabled(p) || kprobes_all_disarmed))
679 return;
680
681 /* kprobes with 'post_handler' can not be optimized */
682 if (p->post_handler)
683 return;
684
685 op = container_of(p, struct optimized_kprobe, kp);
686
687 /* Check there is no other kprobes at the optimized instructions */
688 if (arch_check_optimized_kprobe(op) < 0)
689 return;
690
691 /* Check if it is already optimized. */
692 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
693 if (optprobe_queued_unopt(op)) {
694 /* This is under unoptimizing. Just dequeue the probe */
695 list_del_init(&op->list);
696 }
697 return;
698 }
699 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
700
701 /*
702 * On the 'unoptimizing_list' and 'optimizing_list',
703 * 'op' must have OPTIMIZED flag
704 */
705 if (WARN_ON_ONCE(!list_empty(&op->list)))
706 return;
707
708 list_add(&op->list, &optimizing_list);
709 kick_kprobe_optimizer();
710}
711
712/* Short cut to direct unoptimizing */
713static void force_unoptimize_kprobe(struct optimized_kprobe *op)
714{
715 lockdep_assert_cpus_held();
716 arch_unoptimize_kprobe(op);
717 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
718}
719
720/* Unoptimize a kprobe if p is optimized */
721static void unoptimize_kprobe(struct kprobe *p, bool force)
722{
723 struct optimized_kprobe *op;
724
725 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
726 return; /* This is not an optprobe nor optimized */
727
728 op = container_of(p, struct optimized_kprobe, kp);
729 if (!kprobe_optimized(p))
730 return;
731
732 if (!list_empty(&op->list)) {
733 if (optprobe_queued_unopt(op)) {
734 /* Queued in unoptimizing queue */
735 if (force) {
736 /*
737 * Forcibly unoptimize the kprobe here, and queue it
738 * in the freeing list for release afterwards.
739 */
740 force_unoptimize_kprobe(op);
741 list_move(&op->list, &freeing_list);
742 }
743 } else {
744 /* Dequeue from the optimizing queue */
745 list_del_init(&op->list);
746 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
747 }
748 return;
749 }
750
751 /* Optimized kprobe case */
752 if (force) {
753 /* Forcibly update the code: this is a special case */
754 force_unoptimize_kprobe(op);
755 } else {
756 list_add(&op->list, &unoptimizing_list);
757 kick_kprobe_optimizer();
758 }
759}
760
761/* Cancel unoptimizing for reusing */
762static int reuse_unused_kprobe(struct kprobe *ap)
763{
764 struct optimized_kprobe *op;
765
766 /*
767 * Unused kprobe MUST be on the way of delayed unoptimizing (means
768 * there is still a relative jump) and disabled.
769 */
770 op = container_of(ap, struct optimized_kprobe, kp);
771 WARN_ON_ONCE(list_empty(&op->list));
772 /* Enable the probe again */
773 ap->flags &= ~KPROBE_FLAG_DISABLED;
774 /* Optimize it again. (remove from 'op->list') */
775 if (!kprobe_optready(ap))
776 return -EINVAL;
777
778 optimize_kprobe(ap);
779 return 0;
780}
781
782/* Remove optimized instructions */
783static void kill_optimized_kprobe(struct kprobe *p)
784{
785 struct optimized_kprobe *op;
786
787 op = container_of(p, struct optimized_kprobe, kp);
788 if (!list_empty(&op->list))
789 /* Dequeue from the (un)optimization queue */
790 list_del_init(&op->list);
791 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
792
793 if (kprobe_unused(p)) {
794 /*
795 * Unused kprobe is on unoptimizing or freeing list. We move it
796 * to freeing_list and let the kprobe_optimizer() remove it from
797 * the kprobe hash list and free it.
798 */
799 if (optprobe_queued_unopt(op))
800 list_move(&op->list, &freeing_list);
801 }
802
803 /* Don't touch the code, because it is already freed. */
804 arch_remove_optimized_kprobe(op);
805}
806
807static inline
808void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
809{
810 if (!kprobe_ftrace(p))
811 arch_prepare_optimized_kprobe(op, p);
812}
813
814/* Try to prepare optimized instructions */
815static void prepare_optimized_kprobe(struct kprobe *p)
816{
817 struct optimized_kprobe *op;
818
819 op = container_of(p, struct optimized_kprobe, kp);
820 __prepare_optimized_kprobe(op, p);
821}
822
823/* Allocate new optimized_kprobe and try to prepare optimized instructions. */
824static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
825{
826 struct optimized_kprobe *op;
827
828 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
829 if (!op)
830 return NULL;
831
832 INIT_LIST_HEAD(&op->list);
833 op->kp.addr = p->addr;
834 __prepare_optimized_kprobe(op, p);
835
836 return &op->kp;
837}
838
839static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
840
841/*
842 * Prepare an optimized_kprobe and optimize it.
843 * NOTE: 'p' must be a normal registered kprobe.
844 */
845static void try_to_optimize_kprobe(struct kprobe *p)
846{
847 struct kprobe *ap;
848 struct optimized_kprobe *op;
849
850 /* Impossible to optimize ftrace-based kprobe. */
851 if (kprobe_ftrace(p))
852 return;
853
854 /* For preparing optimization, jump_label_text_reserved() is called. */
855 cpus_read_lock();
856 jump_label_lock();
857 mutex_lock(&text_mutex);
858
859 ap = alloc_aggr_kprobe(p);
860 if (!ap)
861 goto out;
862
863 op = container_of(ap, struct optimized_kprobe, kp);
864 if (!arch_prepared_optinsn(&op->optinsn)) {
865 /* If failed to setup optimizing, fallback to kprobe. */
866 arch_remove_optimized_kprobe(op);
867 kfree(op);
868 goto out;
869 }
870
871 init_aggr_kprobe(ap, p);
872 optimize_kprobe(ap); /* This just kicks optimizer thread. */
873
874out:
875 mutex_unlock(&text_mutex);
876 jump_label_unlock();
877 cpus_read_unlock();
878}
879
880static void optimize_all_kprobes(void)
881{
882 struct hlist_head *head;
883 struct kprobe *p;
884 unsigned int i;
885
886 mutex_lock(&kprobe_mutex);
887 /* If optimization is already allowed, just return. */
888 if (kprobes_allow_optimization)
889 goto out;
890
891 cpus_read_lock();
892 kprobes_allow_optimization = true;
893 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
894 head = &kprobe_table[i];
895 hlist_for_each_entry(p, head, hlist)
896 if (!kprobe_disabled(p))
897 optimize_kprobe(p);
898 }
899 cpus_read_unlock();
900 pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
901out:
902 mutex_unlock(&kprobe_mutex);
903}
904
905#ifdef CONFIG_SYSCTL
906static void unoptimize_all_kprobes(void)
907{
908 struct hlist_head *head;
909 struct kprobe *p;
910 unsigned int i;
911
912 mutex_lock(&kprobe_mutex);
913 /* If optimization is already prohibited, just return. */
914 if (!kprobes_allow_optimization) {
915 mutex_unlock(&kprobe_mutex);
916 return;
917 }
918
919 cpus_read_lock();
920 kprobes_allow_optimization = false;
921 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
922 head = &kprobe_table[i];
923 hlist_for_each_entry(p, head, hlist) {
924 if (!kprobe_disabled(p))
925 unoptimize_kprobe(p, false);
926 }
927 }
928 cpus_read_unlock();
929 mutex_unlock(&kprobe_mutex);
930
931 /* Wait for unoptimizing completion. */
932 wait_for_kprobe_optimizer();
933 pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
934}
935
936static DEFINE_MUTEX(kprobe_sysctl_mutex);
937static int sysctl_kprobes_optimization;
938static int proc_kprobes_optimization_handler(const struct ctl_table *table,
939 int write, void *buffer,
940 size_t *length, loff_t *ppos)
941{
942 int ret;
943
944 mutex_lock(&kprobe_sysctl_mutex);
945 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
946 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
947
948 if (sysctl_kprobes_optimization)
949 optimize_all_kprobes();
950 else
951 unoptimize_all_kprobes();
952 mutex_unlock(&kprobe_sysctl_mutex);
953
954 return ret;
955}
956
957static struct ctl_table kprobe_sysctls[] = {
958 {
959 .procname = "kprobes-optimization",
960 .data = &sysctl_kprobes_optimization,
961 .maxlen = sizeof(int),
962 .mode = 0644,
963 .proc_handler = proc_kprobes_optimization_handler,
964 .extra1 = SYSCTL_ZERO,
965 .extra2 = SYSCTL_ONE,
966 },
967};
968
969static void __init kprobe_sysctls_init(void)
970{
971 register_sysctl_init("debug", kprobe_sysctls);
972}
973#endif /* CONFIG_SYSCTL */
974
975/* Put a breakpoint for a probe. */
976static void __arm_kprobe(struct kprobe *p)
977{
978 struct kprobe *_p;
979
980 lockdep_assert_held(&text_mutex);
981
982 /* Find the overlapping optimized kprobes. */
983 _p = get_optimized_kprobe(p->addr);
984 if (unlikely(_p))
985 /* Fallback to unoptimized kprobe */
986 unoptimize_kprobe(_p, true);
987
988 arch_arm_kprobe(p);
989 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
990}
991
992/* Remove the breakpoint of a probe. */
993static void __disarm_kprobe(struct kprobe *p, bool reopt)
994{
995 struct kprobe *_p;
996
997 lockdep_assert_held(&text_mutex);
998
999 /* Try to unoptimize */
1000 unoptimize_kprobe(p, kprobes_all_disarmed);
1001
1002 if (!kprobe_queued(p)) {
1003 arch_disarm_kprobe(p);
1004 /* If another kprobe was blocked, re-optimize it. */
1005 _p = get_optimized_kprobe(p->addr);
1006 if (unlikely(_p) && reopt)
1007 optimize_kprobe(_p);
1008 }
1009 /*
1010 * TODO: Since unoptimization and real disarming will be done by
1011 * the worker thread, we can not check whether another probe are
1012 * unoptimized because of this probe here. It should be re-optimized
1013 * by the worker thread.
1014 */
1015}
1016
1017#else /* !CONFIG_OPTPROBES */
1018
1019#define optimize_kprobe(p) do {} while (0)
1020#define unoptimize_kprobe(p, f) do {} while (0)
1021#define kill_optimized_kprobe(p) do {} while (0)
1022#define prepare_optimized_kprobe(p) do {} while (0)
1023#define try_to_optimize_kprobe(p) do {} while (0)
1024#define __arm_kprobe(p) arch_arm_kprobe(p)
1025#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
1026#define kprobe_disarmed(p) kprobe_disabled(p)
1027#define wait_for_kprobe_optimizer() do {} while (0)
1028
1029static int reuse_unused_kprobe(struct kprobe *ap)
1030{
1031 /*
1032 * If the optimized kprobe is NOT supported, the aggr kprobe is
1033 * released at the same time that the last aggregated kprobe is
1034 * unregistered.
1035 * Thus there should be no chance to reuse unused kprobe.
1036 */
1037 WARN_ON_ONCE(1);
1038 return -EINVAL;
1039}
1040
1041static void free_aggr_kprobe(struct kprobe *p)
1042{
1043 arch_remove_kprobe(p);
1044 kfree(p);
1045}
1046
1047static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1048{
1049 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1050}
1051#endif /* CONFIG_OPTPROBES */
1052
1053#ifdef CONFIG_KPROBES_ON_FTRACE
1054static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1055 .func = kprobe_ftrace_handler,
1056 .flags = FTRACE_OPS_FL_SAVE_REGS,
1057};
1058
1059static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1060 .func = kprobe_ftrace_handler,
1061 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1062};
1063
1064static int kprobe_ipmodify_enabled;
1065static int kprobe_ftrace_enabled;
1066bool kprobe_ftrace_disabled;
1067
1068static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1069 int *cnt)
1070{
1071 int ret;
1072
1073 lockdep_assert_held(&kprobe_mutex);
1074
1075 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1076 if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1077 return ret;
1078
1079 if (*cnt == 0) {
1080 ret = register_ftrace_function(ops);
1081 if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1082 goto err_ftrace;
1083 }
1084
1085 (*cnt)++;
1086 return ret;
1087
1088err_ftrace:
1089 /*
1090 * At this point, sinec ops is not registered, we should be sefe from
1091 * registering empty filter.
1092 */
1093 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1094 return ret;
1095}
1096
1097static int arm_kprobe_ftrace(struct kprobe *p)
1098{
1099 bool ipmodify = (p->post_handler != NULL);
1100
1101 return __arm_kprobe_ftrace(p,
1102 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1103 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1104}
1105
1106static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1107 int *cnt)
1108{
1109 int ret;
1110
1111 lockdep_assert_held(&kprobe_mutex);
1112
1113 if (*cnt == 1) {
1114 ret = unregister_ftrace_function(ops);
1115 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1116 return ret;
1117 }
1118
1119 (*cnt)--;
1120
1121 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1122 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1123 p->addr, ret);
1124 return ret;
1125}
1126
1127static int disarm_kprobe_ftrace(struct kprobe *p)
1128{
1129 bool ipmodify = (p->post_handler != NULL);
1130
1131 return __disarm_kprobe_ftrace(p,
1132 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1133 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1134}
1135
1136void kprobe_ftrace_kill(void)
1137{
1138 kprobe_ftrace_disabled = true;
1139}
1140#else /* !CONFIG_KPROBES_ON_FTRACE */
1141static inline int arm_kprobe_ftrace(struct kprobe *p)
1142{
1143 return -ENODEV;
1144}
1145
1146static inline int disarm_kprobe_ftrace(struct kprobe *p)
1147{
1148 return -ENODEV;
1149}
1150#endif
1151
1152static int prepare_kprobe(struct kprobe *p)
1153{
1154 /* Must ensure p->addr is really on ftrace */
1155 if (kprobe_ftrace(p))
1156 return arch_prepare_kprobe_ftrace(p);
1157
1158 return arch_prepare_kprobe(p);
1159}
1160
1161static int arm_kprobe(struct kprobe *kp)
1162{
1163 if (unlikely(kprobe_ftrace(kp)))
1164 return arm_kprobe_ftrace(kp);
1165
1166 cpus_read_lock();
1167 mutex_lock(&text_mutex);
1168 __arm_kprobe(kp);
1169 mutex_unlock(&text_mutex);
1170 cpus_read_unlock();
1171
1172 return 0;
1173}
1174
1175static int disarm_kprobe(struct kprobe *kp, bool reopt)
1176{
1177 if (unlikely(kprobe_ftrace(kp)))
1178 return disarm_kprobe_ftrace(kp);
1179
1180 cpus_read_lock();
1181 mutex_lock(&text_mutex);
1182 __disarm_kprobe(kp, reopt);
1183 mutex_unlock(&text_mutex);
1184 cpus_read_unlock();
1185
1186 return 0;
1187}
1188
1189/*
1190 * Aggregate handlers for multiple kprobes support - these handlers
1191 * take care of invoking the individual kprobe handlers on p->list
1192 */
1193static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1194{
1195 struct kprobe *kp;
1196
1197 list_for_each_entry_rcu(kp, &p->list, list) {
1198 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1199 set_kprobe_instance(kp);
1200 if (kp->pre_handler(kp, regs))
1201 return 1;
1202 }
1203 reset_kprobe_instance();
1204 }
1205 return 0;
1206}
1207NOKPROBE_SYMBOL(aggr_pre_handler);
1208
1209static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1210 unsigned long flags)
1211{
1212 struct kprobe *kp;
1213
1214 list_for_each_entry_rcu(kp, &p->list, list) {
1215 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1216 set_kprobe_instance(kp);
1217 kp->post_handler(kp, regs, flags);
1218 reset_kprobe_instance();
1219 }
1220 }
1221}
1222NOKPROBE_SYMBOL(aggr_post_handler);
1223
1224/* Walks the list and increments 'nmissed' if 'p' has child probes. */
1225void kprobes_inc_nmissed_count(struct kprobe *p)
1226{
1227 struct kprobe *kp;
1228
1229 if (!kprobe_aggrprobe(p)) {
1230 p->nmissed++;
1231 } else {
1232 list_for_each_entry_rcu(kp, &p->list, list)
1233 kp->nmissed++;
1234 }
1235}
1236NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
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/* Add the new probe to 'ap->list'. */
1259static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1260{
1261 if (p->post_handler)
1262 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1263
1264 list_add_rcu(&p->list, &ap->list);
1265 if (p->post_handler && !ap->post_handler)
1266 ap->post_handler = aggr_post_handler;
1267
1268 return 0;
1269}
1270
1271/*
1272 * Fill in the required fields of the aggregator kprobe. Replace the
1273 * earlier kprobe in the hlist with the aggregator kprobe.
1274 */
1275static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1276{
1277 /* Copy the insn slot of 'p' to 'ap'. */
1278 copy_kprobe(p, ap);
1279 flush_insn_slot(ap);
1280 ap->addr = p->addr;
1281 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1282 ap->pre_handler = aggr_pre_handler;
1283 /* We don't care the kprobe which has gone. */
1284 if (p->post_handler && !kprobe_gone(p))
1285 ap->post_handler = aggr_post_handler;
1286
1287 INIT_LIST_HEAD(&ap->list);
1288 INIT_HLIST_NODE(&ap->hlist);
1289
1290 list_add_rcu(&p->list, &ap->list);
1291 hlist_replace_rcu(&p->hlist, &ap->hlist);
1292}
1293
1294/*
1295 * This registers the second or subsequent kprobe at the same address.
1296 */
1297static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1298{
1299 int ret = 0;
1300 struct kprobe *ap = orig_p;
1301
1302 cpus_read_lock();
1303
1304 /* For preparing optimization, jump_label_text_reserved() is called */
1305 jump_label_lock();
1306 mutex_lock(&text_mutex);
1307
1308 if (!kprobe_aggrprobe(orig_p)) {
1309 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1310 ap = alloc_aggr_kprobe(orig_p);
1311 if (!ap) {
1312 ret = -ENOMEM;
1313 goto out;
1314 }
1315 init_aggr_kprobe(ap, orig_p);
1316 } else if (kprobe_unused(ap)) {
1317 /* This probe is going to die. Rescue it */
1318 ret = reuse_unused_kprobe(ap);
1319 if (ret)
1320 goto out;
1321 }
1322
1323 if (kprobe_gone(ap)) {
1324 /*
1325 * Attempting to insert new probe at the same location that
1326 * had a probe in the module vaddr area which already
1327 * freed. So, the instruction slot has already been
1328 * released. We need a new slot for the new probe.
1329 */
1330 ret = arch_prepare_kprobe(ap);
1331 if (ret)
1332 /*
1333 * Even if fail to allocate new slot, don't need to
1334 * free the 'ap'. It will be used next time, or
1335 * freed by unregister_kprobe().
1336 */
1337 goto out;
1338
1339 /* Prepare optimized instructions if possible. */
1340 prepare_optimized_kprobe(ap);
1341
1342 /*
1343 * Clear gone flag to prevent allocating new slot again, and
1344 * set disabled flag because it is not armed yet.
1345 */
1346 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1347 | KPROBE_FLAG_DISABLED;
1348 }
1349
1350 /* Copy the insn slot of 'p' to 'ap'. */
1351 copy_kprobe(ap, p);
1352 ret = add_new_kprobe(ap, p);
1353
1354out:
1355 mutex_unlock(&text_mutex);
1356 jump_label_unlock();
1357 cpus_read_unlock();
1358
1359 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1360 ap->flags &= ~KPROBE_FLAG_DISABLED;
1361 if (!kprobes_all_disarmed) {
1362 /* Arm the breakpoint again. */
1363 ret = arm_kprobe(ap);
1364 if (ret) {
1365 ap->flags |= KPROBE_FLAG_DISABLED;
1366 list_del_rcu(&p->list);
1367 synchronize_rcu();
1368 }
1369 }
1370 }
1371 return ret;
1372}
1373
1374bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1375{
1376 /* The '__kprobes' functions and entry code must not be probed. */
1377 return addr >= (unsigned long)__kprobes_text_start &&
1378 addr < (unsigned long)__kprobes_text_end;
1379}
1380
1381static bool __within_kprobe_blacklist(unsigned long addr)
1382{
1383 struct kprobe_blacklist_entry *ent;
1384
1385 if (arch_within_kprobe_blacklist(addr))
1386 return true;
1387 /*
1388 * If 'kprobe_blacklist' is defined, check the address and
1389 * reject any probe registration in the prohibited area.
1390 */
1391 list_for_each_entry(ent, &kprobe_blacklist, list) {
1392 if (addr >= ent->start_addr && addr < ent->end_addr)
1393 return true;
1394 }
1395 return false;
1396}
1397
1398bool within_kprobe_blacklist(unsigned long addr)
1399{
1400 char symname[KSYM_NAME_LEN], *p;
1401
1402 if (__within_kprobe_blacklist(addr))
1403 return true;
1404
1405 /* Check if the address is on a suffixed-symbol */
1406 if (!lookup_symbol_name(addr, symname)) {
1407 p = strchr(symname, '.');
1408 if (!p)
1409 return false;
1410 *p = '\0';
1411 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1412 if (addr)
1413 return __within_kprobe_blacklist(addr);
1414 }
1415 return false;
1416}
1417
1418/*
1419 * arch_adjust_kprobe_addr - adjust the address
1420 * @addr: symbol base address
1421 * @offset: offset within the symbol
1422 * @on_func_entry: was this @addr+@offset on the function entry
1423 *
1424 * Typically returns @addr + @offset, except for special cases where the
1425 * function might be prefixed by a CFI landing pad, in that case any offset
1426 * inside the landing pad is mapped to the first 'real' instruction of the
1427 * symbol.
1428 *
1429 * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1430 * instruction at +0.
1431 */
1432kprobe_opcode_t *__weak arch_adjust_kprobe_addr(unsigned long addr,
1433 unsigned long offset,
1434 bool *on_func_entry)
1435{
1436 *on_func_entry = !offset;
1437 return (kprobe_opcode_t *)(addr + offset);
1438}
1439
1440/*
1441 * If 'symbol_name' is specified, look it up and add the 'offset'
1442 * to it. This way, we can specify a relative address to a symbol.
1443 * This returns encoded errors if it fails to look up symbol or invalid
1444 * combination of parameters.
1445 */
1446static kprobe_opcode_t *
1447_kprobe_addr(kprobe_opcode_t *addr, const char *symbol_name,
1448 unsigned long offset, bool *on_func_entry)
1449{
1450 if ((symbol_name && addr) || (!symbol_name && !addr))
1451 goto invalid;
1452
1453 if (symbol_name) {
1454 /*
1455 * Input: @sym + @offset
1456 * Output: @addr + @offset
1457 *
1458 * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1459 * argument into it's output!
1460 */
1461 addr = kprobe_lookup_name(symbol_name, offset);
1462 if (!addr)
1463 return ERR_PTR(-ENOENT);
1464 }
1465
1466 /*
1467 * So here we have @addr + @offset, displace it into a new
1468 * @addr' + @offset' where @addr' is the symbol start address.
1469 */
1470 addr = (void *)addr + offset;
1471 if (!kallsyms_lookup_size_offset((unsigned long)addr, NULL, &offset))
1472 return ERR_PTR(-ENOENT);
1473 addr = (void *)addr - offset;
1474
1475 /*
1476 * Then ask the architecture to re-combine them, taking care of
1477 * magical function entry details while telling us if this was indeed
1478 * at the start of the function.
1479 */
1480 addr = arch_adjust_kprobe_addr((unsigned long)addr, offset, on_func_entry);
1481 if (addr)
1482 return addr;
1483
1484invalid:
1485 return ERR_PTR(-EINVAL);
1486}
1487
1488static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1489{
1490 bool on_func_entry;
1491 return _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1492}
1493
1494/*
1495 * Check the 'p' is valid and return the aggregator kprobe
1496 * at the same address.
1497 */
1498static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1499{
1500 struct kprobe *ap, *list_p;
1501
1502 lockdep_assert_held(&kprobe_mutex);
1503
1504 ap = get_kprobe(p->addr);
1505 if (unlikely(!ap))
1506 return NULL;
1507
1508 if (p != ap) {
1509 list_for_each_entry(list_p, &ap->list, list)
1510 if (list_p == p)
1511 /* kprobe p is a valid probe */
1512 goto valid;
1513 return NULL;
1514 }
1515valid:
1516 return ap;
1517}
1518
1519/*
1520 * Warn and return error if the kprobe is being re-registered since
1521 * there must be a software bug.
1522 */
1523static inline int warn_kprobe_rereg(struct kprobe *p)
1524{
1525 int ret = 0;
1526
1527 mutex_lock(&kprobe_mutex);
1528 if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1529 ret = -EINVAL;
1530 mutex_unlock(&kprobe_mutex);
1531
1532 return ret;
1533}
1534
1535static int check_ftrace_location(struct kprobe *p)
1536{
1537 unsigned long addr = (unsigned long)p->addr;
1538
1539 if (ftrace_location(addr) == addr) {
1540#ifdef CONFIG_KPROBES_ON_FTRACE
1541 p->flags |= KPROBE_FLAG_FTRACE;
1542#else
1543 return -EINVAL;
1544#endif
1545 }
1546 return 0;
1547}
1548
1549static bool is_cfi_preamble_symbol(unsigned long addr)
1550{
1551 char symbuf[KSYM_NAME_LEN];
1552
1553 if (lookup_symbol_name(addr, symbuf))
1554 return false;
1555
1556 return str_has_prefix(symbuf, "__cfi_") ||
1557 str_has_prefix(symbuf, "__pfx_");
1558}
1559
1560static int check_kprobe_address_safe(struct kprobe *p,
1561 struct module **probed_mod)
1562{
1563 int ret;
1564
1565 ret = check_ftrace_location(p);
1566 if (ret)
1567 return ret;
1568 jump_label_lock();
1569 preempt_disable();
1570
1571 /* Ensure the address is in a text area, and find a module if exists. */
1572 *probed_mod = NULL;
1573 if (!core_kernel_text((unsigned long) p->addr)) {
1574 *probed_mod = __module_text_address((unsigned long) p->addr);
1575 if (!(*probed_mod)) {
1576 ret = -EINVAL;
1577 goto out;
1578 }
1579 }
1580 /* Ensure it is not in reserved area. */
1581 if (in_gate_area_no_mm((unsigned long) p->addr) ||
1582 within_kprobe_blacklist((unsigned long) p->addr) ||
1583 jump_label_text_reserved(p->addr, p->addr) ||
1584 static_call_text_reserved(p->addr, p->addr) ||
1585 find_bug((unsigned long)p->addr) ||
1586 is_cfi_preamble_symbol((unsigned long)p->addr)) {
1587 ret = -EINVAL;
1588 goto out;
1589 }
1590
1591 /* Get module refcount and reject __init functions for loaded modules. */
1592 if (IS_ENABLED(CONFIG_MODULES) && *probed_mod) {
1593 /*
1594 * We must hold a refcount of the probed module while updating
1595 * its code to prohibit unexpected unloading.
1596 */
1597 if (unlikely(!try_module_get(*probed_mod))) {
1598 ret = -ENOENT;
1599 goto out;
1600 }
1601
1602 /*
1603 * If the module freed '.init.text', we couldn't insert
1604 * kprobes in there.
1605 */
1606 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1607 !module_is_coming(*probed_mod)) {
1608 module_put(*probed_mod);
1609 *probed_mod = NULL;
1610 ret = -ENOENT;
1611 }
1612 }
1613
1614out:
1615 preempt_enable();
1616 jump_label_unlock();
1617
1618 return ret;
1619}
1620
1621int register_kprobe(struct kprobe *p)
1622{
1623 int ret;
1624 struct kprobe *old_p;
1625 struct module *probed_mod;
1626 kprobe_opcode_t *addr;
1627 bool on_func_entry;
1628
1629 /* Adjust probe address from symbol */
1630 addr = _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1631 if (IS_ERR(addr))
1632 return PTR_ERR(addr);
1633 p->addr = addr;
1634
1635 ret = warn_kprobe_rereg(p);
1636 if (ret)
1637 return ret;
1638
1639 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1640 p->flags &= KPROBE_FLAG_DISABLED;
1641 p->nmissed = 0;
1642 INIT_LIST_HEAD(&p->list);
1643
1644 ret = check_kprobe_address_safe(p, &probed_mod);
1645 if (ret)
1646 return ret;
1647
1648 mutex_lock(&kprobe_mutex);
1649
1650 if (on_func_entry)
1651 p->flags |= KPROBE_FLAG_ON_FUNC_ENTRY;
1652
1653 old_p = get_kprobe(p->addr);
1654 if (old_p) {
1655 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1656 ret = register_aggr_kprobe(old_p, p);
1657 goto out;
1658 }
1659
1660 cpus_read_lock();
1661 /* Prevent text modification */
1662 mutex_lock(&text_mutex);
1663 ret = prepare_kprobe(p);
1664 mutex_unlock(&text_mutex);
1665 cpus_read_unlock();
1666 if (ret)
1667 goto out;
1668
1669 INIT_HLIST_NODE(&p->hlist);
1670 hlist_add_head_rcu(&p->hlist,
1671 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1672
1673 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1674 ret = arm_kprobe(p);
1675 if (ret) {
1676 hlist_del_rcu(&p->hlist);
1677 synchronize_rcu();
1678 goto out;
1679 }
1680 }
1681
1682 /* Try to optimize kprobe */
1683 try_to_optimize_kprobe(p);
1684out:
1685 mutex_unlock(&kprobe_mutex);
1686
1687 if (probed_mod)
1688 module_put(probed_mod);
1689
1690 return ret;
1691}
1692EXPORT_SYMBOL_GPL(register_kprobe);
1693
1694/* Check if all probes on the 'ap' are disabled. */
1695static bool aggr_kprobe_disabled(struct kprobe *ap)
1696{
1697 struct kprobe *kp;
1698
1699 lockdep_assert_held(&kprobe_mutex);
1700
1701 list_for_each_entry(kp, &ap->list, list)
1702 if (!kprobe_disabled(kp))
1703 /*
1704 * Since there is an active probe on the list,
1705 * we can't disable this 'ap'.
1706 */
1707 return false;
1708
1709 return true;
1710}
1711
1712static struct kprobe *__disable_kprobe(struct kprobe *p)
1713{
1714 struct kprobe *orig_p;
1715 int ret;
1716
1717 lockdep_assert_held(&kprobe_mutex);
1718
1719 /* Get an original kprobe for return */
1720 orig_p = __get_valid_kprobe(p);
1721 if (unlikely(orig_p == NULL))
1722 return ERR_PTR(-EINVAL);
1723
1724 if (kprobe_disabled(p))
1725 return orig_p;
1726
1727 /* Disable probe if it is a child probe */
1728 if (p != orig_p)
1729 p->flags |= KPROBE_FLAG_DISABLED;
1730
1731 /* Try to disarm and disable this/parent probe */
1732 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1733 /*
1734 * Don't be lazy here. Even if 'kprobes_all_disarmed'
1735 * is false, 'orig_p' might not have been armed yet.
1736 * Note arm_all_kprobes() __tries__ to arm all kprobes
1737 * on the best effort basis.
1738 */
1739 if (!kprobes_all_disarmed && !kprobe_disabled(orig_p)) {
1740 ret = disarm_kprobe(orig_p, true);
1741 if (ret) {
1742 p->flags &= ~KPROBE_FLAG_DISABLED;
1743 return ERR_PTR(ret);
1744 }
1745 }
1746 orig_p->flags |= KPROBE_FLAG_DISABLED;
1747 }
1748
1749 return orig_p;
1750}
1751
1752/*
1753 * Unregister a kprobe without a scheduler synchronization.
1754 */
1755static int __unregister_kprobe_top(struct kprobe *p)
1756{
1757 struct kprobe *ap, *list_p;
1758
1759 /* Disable kprobe. This will disarm it if needed. */
1760 ap = __disable_kprobe(p);
1761 if (IS_ERR(ap))
1762 return PTR_ERR(ap);
1763
1764 if (ap == p)
1765 /*
1766 * This probe is an independent(and non-optimized) kprobe
1767 * (not an aggrprobe). Remove from the hash list.
1768 */
1769 goto disarmed;
1770
1771 /* Following process expects this probe is an aggrprobe */
1772 WARN_ON(!kprobe_aggrprobe(ap));
1773
1774 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1775 /*
1776 * !disarmed could be happen if the probe is under delayed
1777 * unoptimizing.
1778 */
1779 goto disarmed;
1780 else {
1781 /* If disabling probe has special handlers, update aggrprobe */
1782 if (p->post_handler && !kprobe_gone(p)) {
1783 list_for_each_entry(list_p, &ap->list, list) {
1784 if ((list_p != p) && (list_p->post_handler))
1785 goto noclean;
1786 }
1787 /*
1788 * For the kprobe-on-ftrace case, we keep the
1789 * post_handler setting to identify this aggrprobe
1790 * armed with kprobe_ipmodify_ops.
1791 */
1792 if (!kprobe_ftrace(ap))
1793 ap->post_handler = NULL;
1794 }
1795noclean:
1796 /*
1797 * Remove from the aggrprobe: this path will do nothing in
1798 * __unregister_kprobe_bottom().
1799 */
1800 list_del_rcu(&p->list);
1801 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1802 /*
1803 * Try to optimize this probe again, because post
1804 * handler may have been changed.
1805 */
1806 optimize_kprobe(ap);
1807 }
1808 return 0;
1809
1810disarmed:
1811 hlist_del_rcu(&ap->hlist);
1812 return 0;
1813}
1814
1815static void __unregister_kprobe_bottom(struct kprobe *p)
1816{
1817 struct kprobe *ap;
1818
1819 if (list_empty(&p->list))
1820 /* This is an independent kprobe */
1821 arch_remove_kprobe(p);
1822 else if (list_is_singular(&p->list)) {
1823 /* This is the last child of an aggrprobe */
1824 ap = list_entry(p->list.next, struct kprobe, list);
1825 list_del(&p->list);
1826 free_aggr_kprobe(ap);
1827 }
1828 /* Otherwise, do nothing. */
1829}
1830
1831int register_kprobes(struct kprobe **kps, int num)
1832{
1833 int i, ret = 0;
1834
1835 if (num <= 0)
1836 return -EINVAL;
1837 for (i = 0; i < num; i++) {
1838 ret = register_kprobe(kps[i]);
1839 if (ret < 0) {
1840 if (i > 0)
1841 unregister_kprobes(kps, i);
1842 break;
1843 }
1844 }
1845 return ret;
1846}
1847EXPORT_SYMBOL_GPL(register_kprobes);
1848
1849void unregister_kprobe(struct kprobe *p)
1850{
1851 unregister_kprobes(&p, 1);
1852}
1853EXPORT_SYMBOL_GPL(unregister_kprobe);
1854
1855void unregister_kprobes(struct kprobe **kps, int num)
1856{
1857 int i;
1858
1859 if (num <= 0)
1860 return;
1861 mutex_lock(&kprobe_mutex);
1862 for (i = 0; i < num; i++)
1863 if (__unregister_kprobe_top(kps[i]) < 0)
1864 kps[i]->addr = NULL;
1865 mutex_unlock(&kprobe_mutex);
1866
1867 synchronize_rcu();
1868 for (i = 0; i < num; i++)
1869 if (kps[i]->addr)
1870 __unregister_kprobe_bottom(kps[i]);
1871}
1872EXPORT_SYMBOL_GPL(unregister_kprobes);
1873
1874int __weak kprobe_exceptions_notify(struct notifier_block *self,
1875 unsigned long val, void *data)
1876{
1877 return NOTIFY_DONE;
1878}
1879NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1880
1881static struct notifier_block kprobe_exceptions_nb = {
1882 .notifier_call = kprobe_exceptions_notify,
1883 .priority = 0x7fffffff /* we need to be notified first */
1884};
1885
1886#ifdef CONFIG_KRETPROBES
1887
1888#if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
1889
1890/* callbacks for objpool of kretprobe instances */
1891static int kretprobe_init_inst(void *nod, void *context)
1892{
1893 struct kretprobe_instance *ri = nod;
1894
1895 ri->rph = context;
1896 return 0;
1897}
1898static int kretprobe_fini_pool(struct objpool_head *head, void *context)
1899{
1900 kfree(context);
1901 return 0;
1902}
1903
1904static void free_rp_inst_rcu(struct rcu_head *head)
1905{
1906 struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1907 struct kretprobe_holder *rph = ri->rph;
1908
1909 objpool_drop(ri, &rph->pool);
1910}
1911NOKPROBE_SYMBOL(free_rp_inst_rcu);
1912
1913static void recycle_rp_inst(struct kretprobe_instance *ri)
1914{
1915 struct kretprobe *rp = get_kretprobe(ri);
1916
1917 if (likely(rp))
1918 objpool_push(ri, &rp->rph->pool);
1919 else
1920 call_rcu(&ri->rcu, free_rp_inst_rcu);
1921}
1922NOKPROBE_SYMBOL(recycle_rp_inst);
1923
1924/*
1925 * This function is called from delayed_put_task_struct() when a task is
1926 * dead and cleaned up to recycle any kretprobe instances associated with
1927 * this task. These left over instances represent probed functions that
1928 * have been called but will never return.
1929 */
1930void kprobe_flush_task(struct task_struct *tk)
1931{
1932 struct kretprobe_instance *ri;
1933 struct llist_node *node;
1934
1935 /* Early boot, not yet initialized. */
1936 if (unlikely(!kprobes_initialized))
1937 return;
1938
1939 kprobe_busy_begin();
1940
1941 node = __llist_del_all(&tk->kretprobe_instances);
1942 while (node) {
1943 ri = container_of(node, struct kretprobe_instance, llist);
1944 node = node->next;
1945
1946 recycle_rp_inst(ri);
1947 }
1948
1949 kprobe_busy_end();
1950}
1951NOKPROBE_SYMBOL(kprobe_flush_task);
1952
1953static inline void free_rp_inst(struct kretprobe *rp)
1954{
1955 struct kretprobe_holder *rph = rp->rph;
1956
1957 if (!rph)
1958 return;
1959 rp->rph = NULL;
1960 objpool_fini(&rph->pool);
1961}
1962
1963/* This assumes the 'tsk' is the current task or the is not running. */
1964static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1965 struct llist_node **cur)
1966{
1967 struct kretprobe_instance *ri = NULL;
1968 struct llist_node *node = *cur;
1969
1970 if (!node)
1971 node = tsk->kretprobe_instances.first;
1972 else
1973 node = node->next;
1974
1975 while (node) {
1976 ri = container_of(node, struct kretprobe_instance, llist);
1977 if (ri->ret_addr != kretprobe_trampoline_addr()) {
1978 *cur = node;
1979 return ri->ret_addr;
1980 }
1981 node = node->next;
1982 }
1983 return NULL;
1984}
1985NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1986
1987/**
1988 * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1989 * @tsk: Target task
1990 * @fp: A frame pointer
1991 * @cur: a storage of the loop cursor llist_node pointer for next call
1992 *
1993 * Find the correct return address modified by a kretprobe on @tsk in unsigned
1994 * long type. If it finds the return address, this returns that address value,
1995 * or this returns 0.
1996 * The @tsk must be 'current' or a task which is not running. @fp is a hint
1997 * to get the currect return address - which is compared with the
1998 * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1999 * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
2000 * first call, but '@cur' itself must NOT NULL.
2001 */
2002unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
2003 struct llist_node **cur)
2004{
2005 struct kretprobe_instance *ri;
2006 kprobe_opcode_t *ret;
2007
2008 if (WARN_ON_ONCE(!cur))
2009 return 0;
2010
2011 do {
2012 ret = __kretprobe_find_ret_addr(tsk, cur);
2013 if (!ret)
2014 break;
2015 ri = container_of(*cur, struct kretprobe_instance, llist);
2016 } while (ri->fp != fp);
2017
2018 return (unsigned long)ret;
2019}
2020NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
2021
2022void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
2023 kprobe_opcode_t *correct_ret_addr)
2024{
2025 /*
2026 * Do nothing by default. Please fill this to update the fake return
2027 * address on the stack with the correct one on each arch if possible.
2028 */
2029}
2030
2031unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
2032 void *frame_pointer)
2033{
2034 struct kretprobe_instance *ri = NULL;
2035 struct llist_node *first, *node = NULL;
2036 kprobe_opcode_t *correct_ret_addr;
2037 struct kretprobe *rp;
2038
2039 /* Find correct address and all nodes for this frame. */
2040 correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
2041 if (!correct_ret_addr) {
2042 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2043 BUG_ON(1);
2044 }
2045
2046 /*
2047 * Set the return address as the instruction pointer, because if the
2048 * user handler calls stack_trace_save_regs() with this 'regs',
2049 * the stack trace will start from the instruction pointer.
2050 */
2051 instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
2052
2053 /* Run the user handler of the nodes. */
2054 first = current->kretprobe_instances.first;
2055 while (first) {
2056 ri = container_of(first, struct kretprobe_instance, llist);
2057
2058 if (WARN_ON_ONCE(ri->fp != frame_pointer))
2059 break;
2060
2061 rp = get_kretprobe(ri);
2062 if (rp && rp->handler) {
2063 struct kprobe *prev = kprobe_running();
2064
2065 __this_cpu_write(current_kprobe, &rp->kp);
2066 ri->ret_addr = correct_ret_addr;
2067 rp->handler(ri, regs);
2068 __this_cpu_write(current_kprobe, prev);
2069 }
2070 if (first == node)
2071 break;
2072
2073 first = first->next;
2074 }
2075
2076 arch_kretprobe_fixup_return(regs, correct_ret_addr);
2077
2078 /* Unlink all nodes for this frame. */
2079 first = current->kretprobe_instances.first;
2080 current->kretprobe_instances.first = node->next;
2081 node->next = NULL;
2082
2083 /* Recycle free instances. */
2084 while (first) {
2085 ri = container_of(first, struct kretprobe_instance, llist);
2086 first = first->next;
2087
2088 recycle_rp_inst(ri);
2089 }
2090
2091 return (unsigned long)correct_ret_addr;
2092}
2093NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2094
2095/*
2096 * This kprobe pre_handler is registered with every kretprobe. When probe
2097 * hits it will set up the return probe.
2098 */
2099static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2100{
2101 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2102 struct kretprobe_holder *rph = rp->rph;
2103 struct kretprobe_instance *ri;
2104
2105 ri = objpool_pop(&rph->pool);
2106 if (!ri) {
2107 rp->nmissed++;
2108 return 0;
2109 }
2110
2111 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2112 objpool_push(ri, &rph->pool);
2113 return 0;
2114 }
2115
2116 arch_prepare_kretprobe(ri, regs);
2117
2118 __llist_add(&ri->llist, ¤t->kretprobe_instances);
2119
2120 return 0;
2121}
2122NOKPROBE_SYMBOL(pre_handler_kretprobe);
2123#else /* CONFIG_KRETPROBE_ON_RETHOOK */
2124/*
2125 * This kprobe pre_handler is registered with every kretprobe. When probe
2126 * hits it will set up the return probe.
2127 */
2128static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2129{
2130 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2131 struct kretprobe_instance *ri;
2132 struct rethook_node *rhn;
2133
2134 rhn = rethook_try_get(rp->rh);
2135 if (!rhn) {
2136 rp->nmissed++;
2137 return 0;
2138 }
2139
2140 ri = container_of(rhn, struct kretprobe_instance, node);
2141
2142 if (rp->entry_handler && rp->entry_handler(ri, regs))
2143 rethook_recycle(rhn);
2144 else
2145 rethook_hook(rhn, regs, kprobe_ftrace(p));
2146
2147 return 0;
2148}
2149NOKPROBE_SYMBOL(pre_handler_kretprobe);
2150
2151static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
2152 unsigned long ret_addr,
2153 struct pt_regs *regs)
2154{
2155 struct kretprobe *rp = (struct kretprobe *)data;
2156 struct kretprobe_instance *ri;
2157 struct kprobe_ctlblk *kcb;
2158
2159 /* The data must NOT be null. This means rethook data structure is broken. */
2160 if (WARN_ON_ONCE(!data) || !rp->handler)
2161 return;
2162
2163 __this_cpu_write(current_kprobe, &rp->kp);
2164 kcb = get_kprobe_ctlblk();
2165 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
2166
2167 ri = container_of(rh, struct kretprobe_instance, node);
2168 rp->handler(ri, regs);
2169
2170 __this_cpu_write(current_kprobe, NULL);
2171}
2172NOKPROBE_SYMBOL(kretprobe_rethook_handler);
2173
2174#endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2175
2176/**
2177 * kprobe_on_func_entry() -- check whether given address is function entry
2178 * @addr: Target address
2179 * @sym: Target symbol name
2180 * @offset: The offset from the symbol or the address
2181 *
2182 * This checks whether the given @addr+@offset or @sym+@offset is on the
2183 * function entry address or not.
2184 * This returns 0 if it is the function entry, or -EINVAL if it is not.
2185 * And also it returns -ENOENT if it fails the symbol or address lookup.
2186 * Caller must pass @addr or @sym (either one must be NULL), or this
2187 * returns -EINVAL.
2188 */
2189int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2190{
2191 bool on_func_entry;
2192 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset, &on_func_entry);
2193
2194 if (IS_ERR(kp_addr))
2195 return PTR_ERR(kp_addr);
2196
2197 if (!on_func_entry)
2198 return -EINVAL;
2199
2200 return 0;
2201}
2202
2203int register_kretprobe(struct kretprobe *rp)
2204{
2205 int ret;
2206 int i;
2207 void *addr;
2208
2209 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2210 if (ret)
2211 return ret;
2212
2213 /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2214 if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2215 return -EINVAL;
2216
2217 if (kretprobe_blacklist_size) {
2218 addr = kprobe_addr(&rp->kp);
2219 if (IS_ERR(addr))
2220 return PTR_ERR(addr);
2221
2222 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2223 if (kretprobe_blacklist[i].addr == addr)
2224 return -EINVAL;
2225 }
2226 }
2227
2228 if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2229 return -E2BIG;
2230
2231 rp->kp.pre_handler = pre_handler_kretprobe;
2232 rp->kp.post_handler = NULL;
2233
2234 /* Pre-allocate memory for max kretprobe instances */
2235 if (rp->maxactive <= 0)
2236 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2237
2238#ifdef CONFIG_KRETPROBE_ON_RETHOOK
2239 rp->rh = rethook_alloc((void *)rp, kretprobe_rethook_handler,
2240 sizeof(struct kretprobe_instance) +
2241 rp->data_size, rp->maxactive);
2242 if (IS_ERR(rp->rh))
2243 return PTR_ERR(rp->rh);
2244
2245 rp->nmissed = 0;
2246 /* Establish function entry probe point */
2247 ret = register_kprobe(&rp->kp);
2248 if (ret != 0) {
2249 rethook_free(rp->rh);
2250 rp->rh = NULL;
2251 }
2252#else /* !CONFIG_KRETPROBE_ON_RETHOOK */
2253 rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2254 if (!rp->rph)
2255 return -ENOMEM;
2256
2257 if (objpool_init(&rp->rph->pool, rp->maxactive, rp->data_size +
2258 sizeof(struct kretprobe_instance), GFP_KERNEL,
2259 rp->rph, kretprobe_init_inst, kretprobe_fini_pool)) {
2260 kfree(rp->rph);
2261 rp->rph = NULL;
2262 return -ENOMEM;
2263 }
2264 rcu_assign_pointer(rp->rph->rp, rp);
2265 rp->nmissed = 0;
2266 /* Establish function entry probe point */
2267 ret = register_kprobe(&rp->kp);
2268 if (ret != 0)
2269 free_rp_inst(rp);
2270#endif
2271 return ret;
2272}
2273EXPORT_SYMBOL_GPL(register_kretprobe);
2274
2275int register_kretprobes(struct kretprobe **rps, int num)
2276{
2277 int ret = 0, i;
2278
2279 if (num <= 0)
2280 return -EINVAL;
2281 for (i = 0; i < num; i++) {
2282 ret = register_kretprobe(rps[i]);
2283 if (ret < 0) {
2284 if (i > 0)
2285 unregister_kretprobes(rps, i);
2286 break;
2287 }
2288 }
2289 return ret;
2290}
2291EXPORT_SYMBOL_GPL(register_kretprobes);
2292
2293void unregister_kretprobe(struct kretprobe *rp)
2294{
2295 unregister_kretprobes(&rp, 1);
2296}
2297EXPORT_SYMBOL_GPL(unregister_kretprobe);
2298
2299void unregister_kretprobes(struct kretprobe **rps, int num)
2300{
2301 int i;
2302
2303 if (num <= 0)
2304 return;
2305 mutex_lock(&kprobe_mutex);
2306 for (i = 0; i < num; i++) {
2307 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2308 rps[i]->kp.addr = NULL;
2309#ifdef CONFIG_KRETPROBE_ON_RETHOOK
2310 rethook_free(rps[i]->rh);
2311#else
2312 rcu_assign_pointer(rps[i]->rph->rp, NULL);
2313#endif
2314 }
2315 mutex_unlock(&kprobe_mutex);
2316
2317 synchronize_rcu();
2318 for (i = 0; i < num; i++) {
2319 if (rps[i]->kp.addr) {
2320 __unregister_kprobe_bottom(&rps[i]->kp);
2321#ifndef CONFIG_KRETPROBE_ON_RETHOOK
2322 free_rp_inst(rps[i]);
2323#endif
2324 }
2325 }
2326}
2327EXPORT_SYMBOL_GPL(unregister_kretprobes);
2328
2329#else /* CONFIG_KRETPROBES */
2330int register_kretprobe(struct kretprobe *rp)
2331{
2332 return -EOPNOTSUPP;
2333}
2334EXPORT_SYMBOL_GPL(register_kretprobe);
2335
2336int register_kretprobes(struct kretprobe **rps, int num)
2337{
2338 return -EOPNOTSUPP;
2339}
2340EXPORT_SYMBOL_GPL(register_kretprobes);
2341
2342void unregister_kretprobe(struct kretprobe *rp)
2343{
2344}
2345EXPORT_SYMBOL_GPL(unregister_kretprobe);
2346
2347void unregister_kretprobes(struct kretprobe **rps, int num)
2348{
2349}
2350EXPORT_SYMBOL_GPL(unregister_kretprobes);
2351
2352static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2353{
2354 return 0;
2355}
2356NOKPROBE_SYMBOL(pre_handler_kretprobe);
2357
2358#endif /* CONFIG_KRETPROBES */
2359
2360/* Set the kprobe gone and remove its instruction buffer. */
2361static void kill_kprobe(struct kprobe *p)
2362{
2363 struct kprobe *kp;
2364
2365 lockdep_assert_held(&kprobe_mutex);
2366
2367 /*
2368 * The module is going away. We should disarm the kprobe which
2369 * is using ftrace, because ftrace framework is still available at
2370 * 'MODULE_STATE_GOING' notification.
2371 */
2372 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2373 disarm_kprobe_ftrace(p);
2374
2375 p->flags |= KPROBE_FLAG_GONE;
2376 if (kprobe_aggrprobe(p)) {
2377 /*
2378 * If this is an aggr_kprobe, we have to list all the
2379 * chained probes and mark them GONE.
2380 */
2381 list_for_each_entry(kp, &p->list, list)
2382 kp->flags |= KPROBE_FLAG_GONE;
2383 p->post_handler = NULL;
2384 kill_optimized_kprobe(p);
2385 }
2386 /*
2387 * Here, we can remove insn_slot safely, because no thread calls
2388 * the original probed function (which will be freed soon) any more.
2389 */
2390 arch_remove_kprobe(p);
2391}
2392
2393/* Disable one kprobe */
2394int disable_kprobe(struct kprobe *kp)
2395{
2396 int ret = 0;
2397 struct kprobe *p;
2398
2399 mutex_lock(&kprobe_mutex);
2400
2401 /* Disable this kprobe */
2402 p = __disable_kprobe(kp);
2403 if (IS_ERR(p))
2404 ret = PTR_ERR(p);
2405
2406 mutex_unlock(&kprobe_mutex);
2407 return ret;
2408}
2409EXPORT_SYMBOL_GPL(disable_kprobe);
2410
2411/* Enable one kprobe */
2412int enable_kprobe(struct kprobe *kp)
2413{
2414 int ret = 0;
2415 struct kprobe *p;
2416
2417 mutex_lock(&kprobe_mutex);
2418
2419 /* Check whether specified probe is valid. */
2420 p = __get_valid_kprobe(kp);
2421 if (unlikely(p == NULL)) {
2422 ret = -EINVAL;
2423 goto out;
2424 }
2425
2426 if (kprobe_gone(kp)) {
2427 /* This kprobe has gone, we couldn't enable it. */
2428 ret = -EINVAL;
2429 goto out;
2430 }
2431
2432 if (p != kp)
2433 kp->flags &= ~KPROBE_FLAG_DISABLED;
2434
2435 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2436 p->flags &= ~KPROBE_FLAG_DISABLED;
2437 ret = arm_kprobe(p);
2438 if (ret) {
2439 p->flags |= KPROBE_FLAG_DISABLED;
2440 if (p != kp)
2441 kp->flags |= KPROBE_FLAG_DISABLED;
2442 }
2443 }
2444out:
2445 mutex_unlock(&kprobe_mutex);
2446 return ret;
2447}
2448EXPORT_SYMBOL_GPL(enable_kprobe);
2449
2450/* Caller must NOT call this in usual path. This is only for critical case */
2451void dump_kprobe(struct kprobe *kp)
2452{
2453 pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2454 kp->symbol_name, kp->offset, kp->addr);
2455}
2456NOKPROBE_SYMBOL(dump_kprobe);
2457
2458int kprobe_add_ksym_blacklist(unsigned long entry)
2459{
2460 struct kprobe_blacklist_entry *ent;
2461 unsigned long offset = 0, size = 0;
2462
2463 if (!kernel_text_address(entry) ||
2464 !kallsyms_lookup_size_offset(entry, &size, &offset))
2465 return -EINVAL;
2466
2467 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2468 if (!ent)
2469 return -ENOMEM;
2470 ent->start_addr = entry;
2471 ent->end_addr = entry + size;
2472 INIT_LIST_HEAD(&ent->list);
2473 list_add_tail(&ent->list, &kprobe_blacklist);
2474
2475 return (int)size;
2476}
2477
2478/* Add all symbols in given area into kprobe blacklist */
2479int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2480{
2481 unsigned long entry;
2482 int ret = 0;
2483
2484 for (entry = start; entry < end; entry += ret) {
2485 ret = kprobe_add_ksym_blacklist(entry);
2486 if (ret < 0)
2487 return ret;
2488 if (ret == 0) /* In case of alias symbol */
2489 ret = 1;
2490 }
2491 return 0;
2492}
2493
2494int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2495 char *type, char *sym)
2496{
2497 return -ERANGE;
2498}
2499
2500int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2501 char *sym)
2502{
2503#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2504 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2505 return 0;
2506#ifdef CONFIG_OPTPROBES
2507 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2508 return 0;
2509#endif
2510#endif
2511 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2512 return 0;
2513 return -ERANGE;
2514}
2515
2516int __init __weak arch_populate_kprobe_blacklist(void)
2517{
2518 return 0;
2519}
2520
2521/*
2522 * Lookup and populate the kprobe_blacklist.
2523 *
2524 * Unlike the kretprobe blacklist, we'll need to determine
2525 * the range of addresses that belong to the said functions,
2526 * since a kprobe need not necessarily be at the beginning
2527 * of a function.
2528 */
2529static int __init populate_kprobe_blacklist(unsigned long *start,
2530 unsigned long *end)
2531{
2532 unsigned long entry;
2533 unsigned long *iter;
2534 int ret;
2535
2536 for (iter = start; iter < end; iter++) {
2537 entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2538 ret = kprobe_add_ksym_blacklist(entry);
2539 if (ret == -EINVAL)
2540 continue;
2541 if (ret < 0)
2542 return ret;
2543 }
2544
2545 /* Symbols in '__kprobes_text' are blacklisted */
2546 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2547 (unsigned long)__kprobes_text_end);
2548 if (ret)
2549 return ret;
2550
2551 /* Symbols in 'noinstr' section are blacklisted */
2552 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2553 (unsigned long)__noinstr_text_end);
2554
2555 return ret ? : arch_populate_kprobe_blacklist();
2556}
2557
2558#ifdef CONFIG_MODULES
2559/* Remove all symbols in given area from kprobe blacklist */
2560static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2561{
2562 struct kprobe_blacklist_entry *ent, *n;
2563
2564 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2565 if (ent->start_addr < start || ent->start_addr >= end)
2566 continue;
2567 list_del(&ent->list);
2568 kfree(ent);
2569 }
2570}
2571
2572static void kprobe_remove_ksym_blacklist(unsigned long entry)
2573{
2574 kprobe_remove_area_blacklist(entry, entry + 1);
2575}
2576
2577static void add_module_kprobe_blacklist(struct module *mod)
2578{
2579 unsigned long start, end;
2580 int i;
2581
2582 if (mod->kprobe_blacklist) {
2583 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2584 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2585 }
2586
2587 start = (unsigned long)mod->kprobes_text_start;
2588 if (start) {
2589 end = start + mod->kprobes_text_size;
2590 kprobe_add_area_blacklist(start, end);
2591 }
2592
2593 start = (unsigned long)mod->noinstr_text_start;
2594 if (start) {
2595 end = start + mod->noinstr_text_size;
2596 kprobe_add_area_blacklist(start, end);
2597 }
2598}
2599
2600static void remove_module_kprobe_blacklist(struct module *mod)
2601{
2602 unsigned long start, end;
2603 int i;
2604
2605 if (mod->kprobe_blacklist) {
2606 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2607 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2608 }
2609
2610 start = (unsigned long)mod->kprobes_text_start;
2611 if (start) {
2612 end = start + mod->kprobes_text_size;
2613 kprobe_remove_area_blacklist(start, end);
2614 }
2615
2616 start = (unsigned long)mod->noinstr_text_start;
2617 if (start) {
2618 end = start + mod->noinstr_text_size;
2619 kprobe_remove_area_blacklist(start, end);
2620 }
2621}
2622
2623/* Module notifier call back, checking kprobes on the module */
2624static int kprobes_module_callback(struct notifier_block *nb,
2625 unsigned long val, void *data)
2626{
2627 struct module *mod = data;
2628 struct hlist_head *head;
2629 struct kprobe *p;
2630 unsigned int i;
2631 int checkcore = (val == MODULE_STATE_GOING);
2632
2633 if (val == MODULE_STATE_COMING) {
2634 mutex_lock(&kprobe_mutex);
2635 add_module_kprobe_blacklist(mod);
2636 mutex_unlock(&kprobe_mutex);
2637 }
2638 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2639 return NOTIFY_DONE;
2640
2641 /*
2642 * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2643 * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2644 * notified, only '.init.text' section would be freed. We need to
2645 * disable kprobes which have been inserted in the sections.
2646 */
2647 mutex_lock(&kprobe_mutex);
2648 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2649 head = &kprobe_table[i];
2650 hlist_for_each_entry(p, head, hlist)
2651 if (within_module_init((unsigned long)p->addr, mod) ||
2652 (checkcore &&
2653 within_module_core((unsigned long)p->addr, mod))) {
2654 /*
2655 * The vaddr this probe is installed will soon
2656 * be vfreed buy not synced to disk. Hence,
2657 * disarming the breakpoint isn't needed.
2658 *
2659 * Note, this will also move any optimized probes
2660 * that are pending to be removed from their
2661 * corresponding lists to the 'freeing_list' and
2662 * will not be touched by the delayed
2663 * kprobe_optimizer() work handler.
2664 */
2665 kill_kprobe(p);
2666 }
2667 }
2668 if (val == MODULE_STATE_GOING)
2669 remove_module_kprobe_blacklist(mod);
2670 mutex_unlock(&kprobe_mutex);
2671 return NOTIFY_DONE;
2672}
2673
2674static struct notifier_block kprobe_module_nb = {
2675 .notifier_call = kprobes_module_callback,
2676 .priority = 0
2677};
2678
2679static int kprobe_register_module_notifier(void)
2680{
2681 return register_module_notifier(&kprobe_module_nb);
2682}
2683#else
2684static int kprobe_register_module_notifier(void)
2685{
2686 return 0;
2687}
2688#endif /* CONFIG_MODULES */
2689
2690void kprobe_free_init_mem(void)
2691{
2692 void *start = (void *)(&__init_begin);
2693 void *end = (void *)(&__init_end);
2694 struct hlist_head *head;
2695 struct kprobe *p;
2696 int i;
2697
2698 mutex_lock(&kprobe_mutex);
2699
2700 /* Kill all kprobes on initmem because the target code has been freed. */
2701 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2702 head = &kprobe_table[i];
2703 hlist_for_each_entry(p, head, hlist) {
2704 if (start <= (void *)p->addr && (void *)p->addr < end)
2705 kill_kprobe(p);
2706 }
2707 }
2708
2709 mutex_unlock(&kprobe_mutex);
2710}
2711
2712static int __init init_kprobes(void)
2713{
2714 int i, err;
2715
2716 /* FIXME allocate the probe table, currently defined statically */
2717 /* initialize all list heads */
2718 for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2719 INIT_HLIST_HEAD(&kprobe_table[i]);
2720
2721 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2722 __stop_kprobe_blacklist);
2723 if (err)
2724 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2725
2726 if (kretprobe_blacklist_size) {
2727 /* lookup the function address from its name */
2728 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2729 kretprobe_blacklist[i].addr =
2730 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2731 if (!kretprobe_blacklist[i].addr)
2732 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2733 kretprobe_blacklist[i].name);
2734 }
2735 }
2736
2737 /* By default, kprobes are armed */
2738 kprobes_all_disarmed = false;
2739
2740#if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2741 /* Init 'kprobe_optinsn_slots' for allocation */
2742 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2743#endif
2744
2745 err = arch_init_kprobes();
2746 if (!err)
2747 err = register_die_notifier(&kprobe_exceptions_nb);
2748 if (!err)
2749 err = kprobe_register_module_notifier();
2750
2751 kprobes_initialized = (err == 0);
2752 kprobe_sysctls_init();
2753 return err;
2754}
2755early_initcall(init_kprobes);
2756
2757#if defined(CONFIG_OPTPROBES)
2758static int __init init_optprobes(void)
2759{
2760 /*
2761 * Enable kprobe optimization - this kicks the optimizer which
2762 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2763 * not spawned in early initcall. So delay the optimization.
2764 */
2765 optimize_all_kprobes();
2766
2767 return 0;
2768}
2769subsys_initcall(init_optprobes);
2770#endif
2771
2772#ifdef CONFIG_DEBUG_FS
2773static void report_probe(struct seq_file *pi, struct kprobe *p,
2774 const char *sym, int offset, char *modname, struct kprobe *pp)
2775{
2776 char *kprobe_type;
2777 void *addr = p->addr;
2778
2779 if (p->pre_handler == pre_handler_kretprobe)
2780 kprobe_type = "r";
2781 else
2782 kprobe_type = "k";
2783
2784 if (!kallsyms_show_value(pi->file->f_cred))
2785 addr = NULL;
2786
2787 if (sym)
2788 seq_printf(pi, "%px %s %s+0x%x %s ",
2789 addr, kprobe_type, sym, offset,
2790 (modname ? modname : " "));
2791 else /* try to use %pS */
2792 seq_printf(pi, "%px %s %pS ",
2793 addr, kprobe_type, p->addr);
2794
2795 if (!pp)
2796 pp = p;
2797 seq_printf(pi, "%s%s%s%s\n",
2798 (kprobe_gone(p) ? "[GONE]" : ""),
2799 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2800 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2801 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2802}
2803
2804static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2805{
2806 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2807}
2808
2809static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2810{
2811 (*pos)++;
2812 if (*pos >= KPROBE_TABLE_SIZE)
2813 return NULL;
2814 return pos;
2815}
2816
2817static void kprobe_seq_stop(struct seq_file *f, void *v)
2818{
2819 /* Nothing to do */
2820}
2821
2822static int show_kprobe_addr(struct seq_file *pi, void *v)
2823{
2824 struct hlist_head *head;
2825 struct kprobe *p, *kp;
2826 const char *sym;
2827 unsigned int i = *(loff_t *) v;
2828 unsigned long offset = 0;
2829 char *modname, namebuf[KSYM_NAME_LEN];
2830
2831 head = &kprobe_table[i];
2832 preempt_disable();
2833 hlist_for_each_entry_rcu(p, head, hlist) {
2834 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2835 &offset, &modname, namebuf);
2836 if (kprobe_aggrprobe(p)) {
2837 list_for_each_entry_rcu(kp, &p->list, list)
2838 report_probe(pi, kp, sym, offset, modname, p);
2839 } else
2840 report_probe(pi, p, sym, offset, modname, NULL);
2841 }
2842 preempt_enable();
2843 return 0;
2844}
2845
2846static const struct seq_operations kprobes_sops = {
2847 .start = kprobe_seq_start,
2848 .next = kprobe_seq_next,
2849 .stop = kprobe_seq_stop,
2850 .show = show_kprobe_addr
2851};
2852
2853DEFINE_SEQ_ATTRIBUTE(kprobes);
2854
2855/* kprobes/blacklist -- shows which functions can not be probed */
2856static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2857{
2858 mutex_lock(&kprobe_mutex);
2859 return seq_list_start(&kprobe_blacklist, *pos);
2860}
2861
2862static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2863{
2864 return seq_list_next(v, &kprobe_blacklist, pos);
2865}
2866
2867static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2868{
2869 struct kprobe_blacklist_entry *ent =
2870 list_entry(v, struct kprobe_blacklist_entry, list);
2871
2872 /*
2873 * If '/proc/kallsyms' is not showing kernel address, we won't
2874 * show them here either.
2875 */
2876 if (!kallsyms_show_value(m->file->f_cred))
2877 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2878 (void *)ent->start_addr);
2879 else
2880 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2881 (void *)ent->end_addr, (void *)ent->start_addr);
2882 return 0;
2883}
2884
2885static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2886{
2887 mutex_unlock(&kprobe_mutex);
2888}
2889
2890static const struct seq_operations kprobe_blacklist_sops = {
2891 .start = kprobe_blacklist_seq_start,
2892 .next = kprobe_blacklist_seq_next,
2893 .stop = kprobe_blacklist_seq_stop,
2894 .show = kprobe_blacklist_seq_show,
2895};
2896DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2897
2898static int arm_all_kprobes(void)
2899{
2900 struct hlist_head *head;
2901 struct kprobe *p;
2902 unsigned int i, total = 0, errors = 0;
2903 int err, ret = 0;
2904
2905 mutex_lock(&kprobe_mutex);
2906
2907 /* If kprobes are armed, just return */
2908 if (!kprobes_all_disarmed)
2909 goto already_enabled;
2910
2911 /*
2912 * optimize_kprobe() called by arm_kprobe() checks
2913 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2914 * arm_kprobe.
2915 */
2916 kprobes_all_disarmed = false;
2917 /* Arming kprobes doesn't optimize kprobe itself */
2918 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2919 head = &kprobe_table[i];
2920 /* Arm all kprobes on a best-effort basis */
2921 hlist_for_each_entry(p, head, hlist) {
2922 if (!kprobe_disabled(p)) {
2923 err = arm_kprobe(p);
2924 if (err) {
2925 errors++;
2926 ret = err;
2927 }
2928 total++;
2929 }
2930 }
2931 }
2932
2933 if (errors)
2934 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2935 errors, total);
2936 else
2937 pr_info("Kprobes globally enabled\n");
2938
2939already_enabled:
2940 mutex_unlock(&kprobe_mutex);
2941 return ret;
2942}
2943
2944static int disarm_all_kprobes(void)
2945{
2946 struct hlist_head *head;
2947 struct kprobe *p;
2948 unsigned int i, total = 0, errors = 0;
2949 int err, ret = 0;
2950
2951 mutex_lock(&kprobe_mutex);
2952
2953 /* If kprobes are already disarmed, just return */
2954 if (kprobes_all_disarmed) {
2955 mutex_unlock(&kprobe_mutex);
2956 return 0;
2957 }
2958
2959 kprobes_all_disarmed = true;
2960
2961 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2962 head = &kprobe_table[i];
2963 /* Disarm all kprobes on a best-effort basis */
2964 hlist_for_each_entry(p, head, hlist) {
2965 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2966 err = disarm_kprobe(p, false);
2967 if (err) {
2968 errors++;
2969 ret = err;
2970 }
2971 total++;
2972 }
2973 }
2974 }
2975
2976 if (errors)
2977 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2978 errors, total);
2979 else
2980 pr_info("Kprobes globally disabled\n");
2981
2982 mutex_unlock(&kprobe_mutex);
2983
2984 /* Wait for disarming all kprobes by optimizer */
2985 wait_for_kprobe_optimizer();
2986
2987 return ret;
2988}
2989
2990/*
2991 * XXX: The debugfs bool file interface doesn't allow for callbacks
2992 * when the bool state is switched. We can reuse that facility when
2993 * available
2994 */
2995static ssize_t read_enabled_file_bool(struct file *file,
2996 char __user *user_buf, size_t count, loff_t *ppos)
2997{
2998 char buf[3];
2999
3000 if (!kprobes_all_disarmed)
3001 buf[0] = '1';
3002 else
3003 buf[0] = '0';
3004 buf[1] = '\n';
3005 buf[2] = 0x00;
3006 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
3007}
3008
3009static ssize_t write_enabled_file_bool(struct file *file,
3010 const char __user *user_buf, size_t count, loff_t *ppos)
3011{
3012 bool enable;
3013 int ret;
3014
3015 ret = kstrtobool_from_user(user_buf, count, &enable);
3016 if (ret)
3017 return ret;
3018
3019 ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
3020 if (ret)
3021 return ret;
3022
3023 return count;
3024}
3025
3026static const struct file_operations fops_kp = {
3027 .read = read_enabled_file_bool,
3028 .write = write_enabled_file_bool,
3029 .llseek = default_llseek,
3030};
3031
3032static int __init debugfs_kprobe_init(void)
3033{
3034 struct dentry *dir;
3035
3036 dir = debugfs_create_dir("kprobes", NULL);
3037
3038 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
3039
3040 debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
3041
3042 debugfs_create_file("blacklist", 0400, dir, NULL,
3043 &kprobe_blacklist_fops);
3044
3045 return 0;
3046}
3047
3048late_initcall(debugfs_kprobe_init);
3049#endif /* CONFIG_DEBUG_FS */
1/*
2 * Kernel Probes (KProbes)
3 * kernel/kprobes.c
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 *
19 * Copyright (C) IBM Corporation, 2002, 2004
20 *
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation (includes suggestions from
23 * Rusty Russell).
24 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25 * hlists and exceptions notifier as suggested by Andi Kleen.
26 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27 * interface to access function arguments.
28 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29 * exceptions notifier to be first on the priority list.
30 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32 * <prasanna@in.ibm.com> added function-return probes.
33 */
34#include <linux/kprobes.h>
35#include <linux/hash.h>
36#include <linux/init.h>
37#include <linux/slab.h>
38#include <linux/stddef.h>
39#include <linux/export.h>
40#include <linux/moduleloader.h>
41#include <linux/kallsyms.h>
42#include <linux/freezer.h>
43#include <linux/seq_file.h>
44#include <linux/debugfs.h>
45#include <linux/sysctl.h>
46#include <linux/kdebug.h>
47#include <linux/memory.h>
48#include <linux/ftrace.h>
49#include <linux/cpu.h>
50#include <linux/jump_label.h>
51
52#include <asm-generic/sections.h>
53#include <asm/cacheflush.h>
54#include <asm/errno.h>
55#include <asm/uaccess.h>
56
57#define KPROBE_HASH_BITS 6
58#define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
59
60
61/*
62 * Some oddball architectures like 64bit powerpc have function descriptors
63 * so this must be overridable.
64 */
65#ifndef kprobe_lookup_name
66#define kprobe_lookup_name(name, addr) \
67 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
68#endif
69
70static int kprobes_initialized;
71static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
72static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
73
74/* NOTE: change this value only with kprobe_mutex held */
75static bool kprobes_all_disarmed;
76
77/* This protects kprobe_table and optimizing_list */
78static DEFINE_MUTEX(kprobe_mutex);
79static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
80static struct {
81 raw_spinlock_t lock ____cacheline_aligned_in_smp;
82} kretprobe_table_locks[KPROBE_TABLE_SIZE];
83
84static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
85{
86 return &(kretprobe_table_locks[hash].lock);
87}
88
89/* Blacklist -- list of struct kprobe_blacklist_entry */
90static LIST_HEAD(kprobe_blacklist);
91
92#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
93/*
94 * kprobe->ainsn.insn points to the copy of the instruction to be
95 * single-stepped. x86_64, POWER4 and above have no-exec support and
96 * stepping on the instruction on a vmalloced/kmalloced/data page
97 * is a recipe for disaster
98 */
99struct kprobe_insn_page {
100 struct list_head list;
101 kprobe_opcode_t *insns; /* Page of instruction slots */
102 struct kprobe_insn_cache *cache;
103 int nused;
104 int ngarbage;
105 char slot_used[];
106};
107
108#define KPROBE_INSN_PAGE_SIZE(slots) \
109 (offsetof(struct kprobe_insn_page, slot_used) + \
110 (sizeof(char) * (slots)))
111
112static int slots_per_page(struct kprobe_insn_cache *c)
113{
114 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
115}
116
117enum kprobe_slot_state {
118 SLOT_CLEAN = 0,
119 SLOT_DIRTY = 1,
120 SLOT_USED = 2,
121};
122
123static void *alloc_insn_page(void)
124{
125 return module_alloc(PAGE_SIZE);
126}
127
128static void free_insn_page(void *page)
129{
130 module_memfree(page);
131}
132
133struct kprobe_insn_cache kprobe_insn_slots = {
134 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
135 .alloc = alloc_insn_page,
136 .free = free_insn_page,
137 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
138 .insn_size = MAX_INSN_SIZE,
139 .nr_garbage = 0,
140};
141static int collect_garbage_slots(struct kprobe_insn_cache *c);
142
143/**
144 * __get_insn_slot() - Find a slot on an executable page for an instruction.
145 * We allocate an executable page if there's no room on existing ones.
146 */
147kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
148{
149 struct kprobe_insn_page *kip;
150 kprobe_opcode_t *slot = NULL;
151
152 mutex_lock(&c->mutex);
153 retry:
154 list_for_each_entry(kip, &c->pages, list) {
155 if (kip->nused < slots_per_page(c)) {
156 int i;
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;
160 kip->nused++;
161 slot = kip->insns + (i * c->insn_size);
162 goto out;
163 }
164 }
165 /* kip->nused is broken. Fix it. */
166 kip->nused = slots_per_page(c);
167 WARN_ON(1);
168 }
169 }
170
171 /* If there are any garbage slots, collect it and try again. */
172 if (c->nr_garbage && collect_garbage_slots(c) == 0)
173 goto retry;
174
175 /* All out of space. Need to allocate a new page. */
176 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
177 if (!kip)
178 goto out;
179
180 /*
181 * Use module_alloc so this page is within +/- 2GB of where the
182 * kernel image and loaded module images reside. This is required
183 * so x86_64 can correctly handle the %rip-relative fixups.
184 */
185 kip->insns = c->alloc();
186 if (!kip->insns) {
187 kfree(kip);
188 goto out;
189 }
190 INIT_LIST_HEAD(&kip->list);
191 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
192 kip->slot_used[0] = SLOT_USED;
193 kip->nused = 1;
194 kip->ngarbage = 0;
195 kip->cache = c;
196 list_add(&kip->list, &c->pages);
197 slot = kip->insns;
198out:
199 mutex_unlock(&c->mutex);
200 return slot;
201}
202
203/* Return 1 if all garbages are collected, otherwise 0. */
204static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
205{
206 kip->slot_used[idx] = SLOT_CLEAN;
207 kip->nused--;
208 if (kip->nused == 0) {
209 /*
210 * Page is no longer in use. Free it unless
211 * it's the last one. We keep the last one
212 * so as not to have to set it up again the
213 * next time somebody inserts a probe.
214 */
215 if (!list_is_singular(&kip->list)) {
216 list_del(&kip->list);
217 kip->cache->free(kip->insns);
218 kfree(kip);
219 }
220 return 1;
221 }
222 return 0;
223}
224
225static int collect_garbage_slots(struct kprobe_insn_cache *c)
226{
227 struct kprobe_insn_page *kip, *next;
228
229 /* Ensure no-one is interrupted on the garbages */
230 synchronize_sched();
231
232 list_for_each_entry_safe(kip, next, &c->pages, list) {
233 int i;
234 if (kip->ngarbage == 0)
235 continue;
236 kip->ngarbage = 0; /* we will collect all garbages */
237 for (i = 0; i < slots_per_page(c); i++) {
238 if (kip->slot_used[i] == SLOT_DIRTY &&
239 collect_one_slot(kip, i))
240 break;
241 }
242 }
243 c->nr_garbage = 0;
244 return 0;
245}
246
247void __free_insn_slot(struct kprobe_insn_cache *c,
248 kprobe_opcode_t *slot, int dirty)
249{
250 struct kprobe_insn_page *kip;
251
252 mutex_lock(&c->mutex);
253 list_for_each_entry(kip, &c->pages, list) {
254 long idx = ((long)slot - (long)kip->insns) /
255 (c->insn_size * sizeof(kprobe_opcode_t));
256 if (idx >= 0 && idx < slots_per_page(c)) {
257 WARN_ON(kip->slot_used[idx] != SLOT_USED);
258 if (dirty) {
259 kip->slot_used[idx] = SLOT_DIRTY;
260 kip->ngarbage++;
261 if (++c->nr_garbage > slots_per_page(c))
262 collect_garbage_slots(c);
263 } else
264 collect_one_slot(kip, idx);
265 goto out;
266 }
267 }
268 /* Could not free this slot. */
269 WARN_ON(1);
270out:
271 mutex_unlock(&c->mutex);
272}
273
274#ifdef CONFIG_OPTPROBES
275/* For optimized_kprobe buffer */
276struct kprobe_insn_cache kprobe_optinsn_slots = {
277 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
278 .alloc = alloc_insn_page,
279 .free = free_insn_page,
280 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
281 /* .insn_size is initialized later */
282 .nr_garbage = 0,
283};
284#endif
285#endif
286
287/* We have preemption disabled.. so it is safe to use __ versions */
288static inline void set_kprobe_instance(struct kprobe *kp)
289{
290 __this_cpu_write(kprobe_instance, kp);
291}
292
293static inline void reset_kprobe_instance(void)
294{
295 __this_cpu_write(kprobe_instance, NULL);
296}
297
298/*
299 * This routine is called either:
300 * - under the kprobe_mutex - during kprobe_[un]register()
301 * OR
302 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
303 */
304struct kprobe *get_kprobe(void *addr)
305{
306 struct hlist_head *head;
307 struct kprobe *p;
308
309 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
310 hlist_for_each_entry_rcu(p, head, hlist) {
311 if (p->addr == addr)
312 return p;
313 }
314
315 return NULL;
316}
317NOKPROBE_SYMBOL(get_kprobe);
318
319static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
320
321/* Return true if the kprobe is an aggregator */
322static inline int kprobe_aggrprobe(struct kprobe *p)
323{
324 return p->pre_handler == aggr_pre_handler;
325}
326
327/* Return true(!0) if the kprobe is unused */
328static inline int kprobe_unused(struct kprobe *p)
329{
330 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
331 list_empty(&p->list);
332}
333
334/*
335 * Keep all fields in the kprobe consistent
336 */
337static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
338{
339 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
340 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
341}
342
343#ifdef CONFIG_OPTPROBES
344/* NOTE: change this value only with kprobe_mutex held */
345static bool kprobes_allow_optimization;
346
347/*
348 * Call all pre_handler on the list, but ignores its return value.
349 * This must be called from arch-dep optimized caller.
350 */
351void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
352{
353 struct kprobe *kp;
354
355 list_for_each_entry_rcu(kp, &p->list, list) {
356 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
357 set_kprobe_instance(kp);
358 kp->pre_handler(kp, regs);
359 }
360 reset_kprobe_instance();
361 }
362}
363NOKPROBE_SYMBOL(opt_pre_handler);
364
365/* Free optimized instructions and optimized_kprobe */
366static void free_aggr_kprobe(struct kprobe *p)
367{
368 struct optimized_kprobe *op;
369
370 op = container_of(p, struct optimized_kprobe, kp);
371 arch_remove_optimized_kprobe(op);
372 arch_remove_kprobe(p);
373 kfree(op);
374}
375
376/* Return true(!0) if the kprobe is ready for optimization. */
377static inline int kprobe_optready(struct kprobe *p)
378{
379 struct optimized_kprobe *op;
380
381 if (kprobe_aggrprobe(p)) {
382 op = container_of(p, struct optimized_kprobe, kp);
383 return arch_prepared_optinsn(&op->optinsn);
384 }
385
386 return 0;
387}
388
389/* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
390static inline int kprobe_disarmed(struct kprobe *p)
391{
392 struct optimized_kprobe *op;
393
394 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
395 if (!kprobe_aggrprobe(p))
396 return kprobe_disabled(p);
397
398 op = container_of(p, struct optimized_kprobe, kp);
399
400 return kprobe_disabled(p) && list_empty(&op->list);
401}
402
403/* Return true(!0) if the probe is queued on (un)optimizing lists */
404static int kprobe_queued(struct kprobe *p)
405{
406 struct optimized_kprobe *op;
407
408 if (kprobe_aggrprobe(p)) {
409 op = container_of(p, struct optimized_kprobe, kp);
410 if (!list_empty(&op->list))
411 return 1;
412 }
413 return 0;
414}
415
416/*
417 * Return an optimized kprobe whose optimizing code replaces
418 * instructions including addr (exclude breakpoint).
419 */
420static struct kprobe *get_optimized_kprobe(unsigned long addr)
421{
422 int i;
423 struct kprobe *p = NULL;
424 struct optimized_kprobe *op;
425
426 /* Don't check i == 0, since that is a breakpoint case. */
427 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
428 p = get_kprobe((void *)(addr - i));
429
430 if (p && kprobe_optready(p)) {
431 op = container_of(p, struct optimized_kprobe, kp);
432 if (arch_within_optimized_kprobe(op, addr))
433 return p;
434 }
435
436 return NULL;
437}
438
439/* Optimization staging list, protected by kprobe_mutex */
440static LIST_HEAD(optimizing_list);
441static LIST_HEAD(unoptimizing_list);
442static LIST_HEAD(freeing_list);
443
444static void kprobe_optimizer(struct work_struct *work);
445static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
446#define OPTIMIZE_DELAY 5
447
448/*
449 * Optimize (replace a breakpoint with a jump) kprobes listed on
450 * optimizing_list.
451 */
452static void do_optimize_kprobes(void)
453{
454 /* Optimization never be done when disarmed */
455 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
456 list_empty(&optimizing_list))
457 return;
458
459 /*
460 * The optimization/unoptimization refers online_cpus via
461 * stop_machine() and cpu-hotplug modifies online_cpus.
462 * And same time, text_mutex will be held in cpu-hotplug and here.
463 * This combination can cause a deadlock (cpu-hotplug try to lock
464 * text_mutex but stop_machine can not be done because online_cpus
465 * has been changed)
466 * To avoid this deadlock, we need to call get_online_cpus()
467 * for preventing cpu-hotplug outside of text_mutex locking.
468 */
469 get_online_cpus();
470 mutex_lock(&text_mutex);
471 arch_optimize_kprobes(&optimizing_list);
472 mutex_unlock(&text_mutex);
473 put_online_cpus();
474}
475
476/*
477 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
478 * if need) kprobes listed on unoptimizing_list.
479 */
480static void do_unoptimize_kprobes(void)
481{
482 struct optimized_kprobe *op, *tmp;
483
484 /* Unoptimization must be done anytime */
485 if (list_empty(&unoptimizing_list))
486 return;
487
488 /* Ditto to do_optimize_kprobes */
489 get_online_cpus();
490 mutex_lock(&text_mutex);
491 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
492 /* Loop free_list for disarming */
493 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
494 /* Disarm probes if marked disabled */
495 if (kprobe_disabled(&op->kp))
496 arch_disarm_kprobe(&op->kp);
497 if (kprobe_unused(&op->kp)) {
498 /*
499 * Remove unused probes from hash list. After waiting
500 * for synchronization, these probes are reclaimed.
501 * (reclaiming is done by do_free_cleaned_kprobes.)
502 */
503 hlist_del_rcu(&op->kp.hlist);
504 } else
505 list_del_init(&op->list);
506 }
507 mutex_unlock(&text_mutex);
508 put_online_cpus();
509}
510
511/* Reclaim all kprobes on the free_list */
512static void do_free_cleaned_kprobes(void)
513{
514 struct optimized_kprobe *op, *tmp;
515
516 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
517 BUG_ON(!kprobe_unused(&op->kp));
518 list_del_init(&op->list);
519 free_aggr_kprobe(&op->kp);
520 }
521}
522
523/* Start optimizer after OPTIMIZE_DELAY passed */
524static void kick_kprobe_optimizer(void)
525{
526 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
527}
528
529/* Kprobe jump optimizer */
530static void kprobe_optimizer(struct work_struct *work)
531{
532 mutex_lock(&kprobe_mutex);
533 /* Lock modules while optimizing kprobes */
534 mutex_lock(&module_mutex);
535
536 /*
537 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
538 * kprobes before waiting for quiesence period.
539 */
540 do_unoptimize_kprobes();
541
542 /*
543 * Step 2: Wait for quiesence period to ensure all running interrupts
544 * are done. Because optprobe may modify multiple instructions
545 * there is a chance that Nth instruction is interrupted. In that
546 * case, running interrupt can return to 2nd-Nth byte of jump
547 * instruction. This wait is for avoiding it.
548 */
549 synchronize_sched();
550
551 /* Step 3: Optimize kprobes after quiesence period */
552 do_optimize_kprobes();
553
554 /* Step 4: Free cleaned kprobes after quiesence period */
555 do_free_cleaned_kprobes();
556
557 mutex_unlock(&module_mutex);
558 mutex_unlock(&kprobe_mutex);
559
560 /* Step 5: Kick optimizer again if needed */
561 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
562 kick_kprobe_optimizer();
563}
564
565/* Wait for completing optimization and unoptimization */
566static void wait_for_kprobe_optimizer(void)
567{
568 mutex_lock(&kprobe_mutex);
569
570 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
571 mutex_unlock(&kprobe_mutex);
572
573 /* this will also make optimizing_work execute immmediately */
574 flush_delayed_work(&optimizing_work);
575 /* @optimizing_work might not have been queued yet, relax */
576 cpu_relax();
577
578 mutex_lock(&kprobe_mutex);
579 }
580
581 mutex_unlock(&kprobe_mutex);
582}
583
584/* Optimize kprobe if p is ready to be optimized */
585static void optimize_kprobe(struct kprobe *p)
586{
587 struct optimized_kprobe *op;
588
589 /* Check if the kprobe is disabled or not ready for optimization. */
590 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
591 (kprobe_disabled(p) || kprobes_all_disarmed))
592 return;
593
594 /* Both of break_handler and post_handler are not supported. */
595 if (p->break_handler || p->post_handler)
596 return;
597
598 op = container_of(p, struct optimized_kprobe, kp);
599
600 /* Check there is no other kprobes at the optimized instructions */
601 if (arch_check_optimized_kprobe(op) < 0)
602 return;
603
604 /* Check if it is already optimized. */
605 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
606 return;
607 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
608
609 if (!list_empty(&op->list))
610 /* This is under unoptimizing. Just dequeue the probe */
611 list_del_init(&op->list);
612 else {
613 list_add(&op->list, &optimizing_list);
614 kick_kprobe_optimizer();
615 }
616}
617
618/* Short cut to direct unoptimizing */
619static void force_unoptimize_kprobe(struct optimized_kprobe *op)
620{
621 get_online_cpus();
622 arch_unoptimize_kprobe(op);
623 put_online_cpus();
624 if (kprobe_disabled(&op->kp))
625 arch_disarm_kprobe(&op->kp);
626}
627
628/* Unoptimize a kprobe if p is optimized */
629static void unoptimize_kprobe(struct kprobe *p, bool force)
630{
631 struct optimized_kprobe *op;
632
633 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
634 return; /* This is not an optprobe nor optimized */
635
636 op = container_of(p, struct optimized_kprobe, kp);
637 if (!kprobe_optimized(p)) {
638 /* Unoptimized or unoptimizing case */
639 if (force && !list_empty(&op->list)) {
640 /*
641 * Only if this is unoptimizing kprobe and forced,
642 * forcibly unoptimize it. (No need to unoptimize
643 * unoptimized kprobe again :)
644 */
645 list_del_init(&op->list);
646 force_unoptimize_kprobe(op);
647 }
648 return;
649 }
650
651 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
652 if (!list_empty(&op->list)) {
653 /* Dequeue from the optimization queue */
654 list_del_init(&op->list);
655 return;
656 }
657 /* Optimized kprobe case */
658 if (force)
659 /* Forcibly update the code: this is a special case */
660 force_unoptimize_kprobe(op);
661 else {
662 list_add(&op->list, &unoptimizing_list);
663 kick_kprobe_optimizer();
664 }
665}
666
667/* Cancel unoptimizing for reusing */
668static void reuse_unused_kprobe(struct kprobe *ap)
669{
670 struct optimized_kprobe *op;
671
672 BUG_ON(!kprobe_unused(ap));
673 /*
674 * Unused kprobe MUST be on the way of delayed unoptimizing (means
675 * there is still a relative jump) and disabled.
676 */
677 op = container_of(ap, struct optimized_kprobe, kp);
678 if (unlikely(list_empty(&op->list)))
679 printk(KERN_WARNING "Warning: found a stray unused "
680 "aggrprobe@%p\n", ap->addr);
681 /* Enable the probe again */
682 ap->flags &= ~KPROBE_FLAG_DISABLED;
683 /* Optimize it again (remove from op->list) */
684 BUG_ON(!kprobe_optready(ap));
685 optimize_kprobe(ap);
686}
687
688/* Remove optimized instructions */
689static void kill_optimized_kprobe(struct kprobe *p)
690{
691 struct optimized_kprobe *op;
692
693 op = container_of(p, struct optimized_kprobe, kp);
694 if (!list_empty(&op->list))
695 /* Dequeue from the (un)optimization queue */
696 list_del_init(&op->list);
697 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
698
699 if (kprobe_unused(p)) {
700 /* Enqueue if it is unused */
701 list_add(&op->list, &freeing_list);
702 /*
703 * Remove unused probes from the hash list. After waiting
704 * for synchronization, this probe is reclaimed.
705 * (reclaiming is done by do_free_cleaned_kprobes().)
706 */
707 hlist_del_rcu(&op->kp.hlist);
708 }
709
710 /* Don't touch the code, because it is already freed. */
711 arch_remove_optimized_kprobe(op);
712}
713
714/* Try to prepare optimized instructions */
715static void prepare_optimized_kprobe(struct kprobe *p)
716{
717 struct optimized_kprobe *op;
718
719 op = container_of(p, struct optimized_kprobe, kp);
720 arch_prepare_optimized_kprobe(op, p);
721}
722
723/* Allocate new optimized_kprobe and try to prepare optimized instructions */
724static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
725{
726 struct optimized_kprobe *op;
727
728 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
729 if (!op)
730 return NULL;
731
732 INIT_LIST_HEAD(&op->list);
733 op->kp.addr = p->addr;
734 arch_prepare_optimized_kprobe(op, p);
735
736 return &op->kp;
737}
738
739static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
740
741/*
742 * Prepare an optimized_kprobe and optimize it
743 * NOTE: p must be a normal registered kprobe
744 */
745static void try_to_optimize_kprobe(struct kprobe *p)
746{
747 struct kprobe *ap;
748 struct optimized_kprobe *op;
749
750 /* Impossible to optimize ftrace-based kprobe */
751 if (kprobe_ftrace(p))
752 return;
753
754 /* For preparing optimization, jump_label_text_reserved() is called */
755 jump_label_lock();
756 mutex_lock(&text_mutex);
757
758 ap = alloc_aggr_kprobe(p);
759 if (!ap)
760 goto out;
761
762 op = container_of(ap, struct optimized_kprobe, kp);
763 if (!arch_prepared_optinsn(&op->optinsn)) {
764 /* If failed to setup optimizing, fallback to kprobe */
765 arch_remove_optimized_kprobe(op);
766 kfree(op);
767 goto out;
768 }
769
770 init_aggr_kprobe(ap, p);
771 optimize_kprobe(ap); /* This just kicks optimizer thread */
772
773out:
774 mutex_unlock(&text_mutex);
775 jump_label_unlock();
776}
777
778#ifdef CONFIG_SYSCTL
779static void optimize_all_kprobes(void)
780{
781 struct hlist_head *head;
782 struct kprobe *p;
783 unsigned int i;
784
785 mutex_lock(&kprobe_mutex);
786 /* If optimization is already allowed, just return */
787 if (kprobes_allow_optimization)
788 goto out;
789
790 kprobes_allow_optimization = true;
791 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
792 head = &kprobe_table[i];
793 hlist_for_each_entry_rcu(p, head, hlist)
794 if (!kprobe_disabled(p))
795 optimize_kprobe(p);
796 }
797 printk(KERN_INFO "Kprobes globally optimized\n");
798out:
799 mutex_unlock(&kprobe_mutex);
800}
801
802static void unoptimize_all_kprobes(void)
803{
804 struct hlist_head *head;
805 struct kprobe *p;
806 unsigned int i;
807
808 mutex_lock(&kprobe_mutex);
809 /* If optimization is already prohibited, just return */
810 if (!kprobes_allow_optimization) {
811 mutex_unlock(&kprobe_mutex);
812 return;
813 }
814
815 kprobes_allow_optimization = false;
816 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
817 head = &kprobe_table[i];
818 hlist_for_each_entry_rcu(p, head, hlist) {
819 if (!kprobe_disabled(p))
820 unoptimize_kprobe(p, false);
821 }
822 }
823 mutex_unlock(&kprobe_mutex);
824
825 /* Wait for unoptimizing completion */
826 wait_for_kprobe_optimizer();
827 printk(KERN_INFO "Kprobes globally unoptimized\n");
828}
829
830static DEFINE_MUTEX(kprobe_sysctl_mutex);
831int sysctl_kprobes_optimization;
832int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
833 void __user *buffer, size_t *length,
834 loff_t *ppos)
835{
836 int ret;
837
838 mutex_lock(&kprobe_sysctl_mutex);
839 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
840 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
841
842 if (sysctl_kprobes_optimization)
843 optimize_all_kprobes();
844 else
845 unoptimize_all_kprobes();
846 mutex_unlock(&kprobe_sysctl_mutex);
847
848 return ret;
849}
850#endif /* CONFIG_SYSCTL */
851
852/* Put a breakpoint for a probe. Must be called with text_mutex locked */
853static void __arm_kprobe(struct kprobe *p)
854{
855 struct kprobe *_p;
856
857 /* Check collision with other optimized kprobes */
858 _p = get_optimized_kprobe((unsigned long)p->addr);
859 if (unlikely(_p))
860 /* Fallback to unoptimized kprobe */
861 unoptimize_kprobe(_p, true);
862
863 arch_arm_kprobe(p);
864 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
865}
866
867/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
868static void __disarm_kprobe(struct kprobe *p, bool reopt)
869{
870 struct kprobe *_p;
871
872 /* Try to unoptimize */
873 unoptimize_kprobe(p, kprobes_all_disarmed);
874
875 if (!kprobe_queued(p)) {
876 arch_disarm_kprobe(p);
877 /* If another kprobe was blocked, optimize it. */
878 _p = get_optimized_kprobe((unsigned long)p->addr);
879 if (unlikely(_p) && reopt)
880 optimize_kprobe(_p);
881 }
882 /* TODO: reoptimize others after unoptimized this probe */
883}
884
885#else /* !CONFIG_OPTPROBES */
886
887#define optimize_kprobe(p) do {} while (0)
888#define unoptimize_kprobe(p, f) do {} while (0)
889#define kill_optimized_kprobe(p) do {} while (0)
890#define prepare_optimized_kprobe(p) do {} while (0)
891#define try_to_optimize_kprobe(p) do {} while (0)
892#define __arm_kprobe(p) arch_arm_kprobe(p)
893#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
894#define kprobe_disarmed(p) kprobe_disabled(p)
895#define wait_for_kprobe_optimizer() do {} while (0)
896
897/* There should be no unused kprobes can be reused without optimization */
898static void reuse_unused_kprobe(struct kprobe *ap)
899{
900 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
901 BUG_ON(kprobe_unused(ap));
902}
903
904static void free_aggr_kprobe(struct kprobe *p)
905{
906 arch_remove_kprobe(p);
907 kfree(p);
908}
909
910static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
911{
912 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
913}
914#endif /* CONFIG_OPTPROBES */
915
916#ifdef CONFIG_KPROBES_ON_FTRACE
917static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
918 .func = kprobe_ftrace_handler,
919 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
920};
921static int kprobe_ftrace_enabled;
922
923/* Must ensure p->addr is really on ftrace */
924static int prepare_kprobe(struct kprobe *p)
925{
926 if (!kprobe_ftrace(p))
927 return arch_prepare_kprobe(p);
928
929 return arch_prepare_kprobe_ftrace(p);
930}
931
932/* Caller must lock kprobe_mutex */
933static void arm_kprobe_ftrace(struct kprobe *p)
934{
935 int ret;
936
937 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
938 (unsigned long)p->addr, 0, 0);
939 WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
940 kprobe_ftrace_enabled++;
941 if (kprobe_ftrace_enabled == 1) {
942 ret = register_ftrace_function(&kprobe_ftrace_ops);
943 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
944 }
945}
946
947/* Caller must lock kprobe_mutex */
948static void disarm_kprobe_ftrace(struct kprobe *p)
949{
950 int ret;
951
952 kprobe_ftrace_enabled--;
953 if (kprobe_ftrace_enabled == 0) {
954 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
955 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
956 }
957 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
958 (unsigned long)p->addr, 1, 0);
959 WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
960}
961#else /* !CONFIG_KPROBES_ON_FTRACE */
962#define prepare_kprobe(p) arch_prepare_kprobe(p)
963#define arm_kprobe_ftrace(p) do {} while (0)
964#define disarm_kprobe_ftrace(p) do {} while (0)
965#endif
966
967/* Arm a kprobe with text_mutex */
968static void arm_kprobe(struct kprobe *kp)
969{
970 if (unlikely(kprobe_ftrace(kp))) {
971 arm_kprobe_ftrace(kp);
972 return;
973 }
974 /*
975 * Here, since __arm_kprobe() doesn't use stop_machine(),
976 * this doesn't cause deadlock on text_mutex. So, we don't
977 * need get_online_cpus().
978 */
979 mutex_lock(&text_mutex);
980 __arm_kprobe(kp);
981 mutex_unlock(&text_mutex);
982}
983
984/* Disarm a kprobe with text_mutex */
985static void disarm_kprobe(struct kprobe *kp, bool reopt)
986{
987 if (unlikely(kprobe_ftrace(kp))) {
988 disarm_kprobe_ftrace(kp);
989 return;
990 }
991 /* Ditto */
992 mutex_lock(&text_mutex);
993 __disarm_kprobe(kp, reopt);
994 mutex_unlock(&text_mutex);
995}
996
997/*
998 * Aggregate handlers for multiple kprobes support - these handlers
999 * take care of invoking the individual kprobe handlers on p->list
1000 */
1001static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1002{
1003 struct kprobe *kp;
1004
1005 list_for_each_entry_rcu(kp, &p->list, list) {
1006 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1007 set_kprobe_instance(kp);
1008 if (kp->pre_handler(kp, regs))
1009 return 1;
1010 }
1011 reset_kprobe_instance();
1012 }
1013 return 0;
1014}
1015NOKPROBE_SYMBOL(aggr_pre_handler);
1016
1017static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1018 unsigned long flags)
1019{
1020 struct kprobe *kp;
1021
1022 list_for_each_entry_rcu(kp, &p->list, list) {
1023 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1024 set_kprobe_instance(kp);
1025 kp->post_handler(kp, regs, flags);
1026 reset_kprobe_instance();
1027 }
1028 }
1029}
1030NOKPROBE_SYMBOL(aggr_post_handler);
1031
1032static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1033 int trapnr)
1034{
1035 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1036
1037 /*
1038 * if we faulted "during" the execution of a user specified
1039 * probe handler, invoke just that probe's fault handler
1040 */
1041 if (cur && cur->fault_handler) {
1042 if (cur->fault_handler(cur, regs, trapnr))
1043 return 1;
1044 }
1045 return 0;
1046}
1047NOKPROBE_SYMBOL(aggr_fault_handler);
1048
1049static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1050{
1051 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1052 int ret = 0;
1053
1054 if (cur && cur->break_handler) {
1055 if (cur->break_handler(cur, regs))
1056 ret = 1;
1057 }
1058 reset_kprobe_instance();
1059 return ret;
1060}
1061NOKPROBE_SYMBOL(aggr_break_handler);
1062
1063/* Walks the list and increments nmissed count for multiprobe case */
1064void kprobes_inc_nmissed_count(struct kprobe *p)
1065{
1066 struct kprobe *kp;
1067 if (!kprobe_aggrprobe(p)) {
1068 p->nmissed++;
1069 } else {
1070 list_for_each_entry_rcu(kp, &p->list, list)
1071 kp->nmissed++;
1072 }
1073 return;
1074}
1075NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1076
1077void recycle_rp_inst(struct kretprobe_instance *ri,
1078 struct hlist_head *head)
1079{
1080 struct kretprobe *rp = ri->rp;
1081
1082 /* remove rp inst off the rprobe_inst_table */
1083 hlist_del(&ri->hlist);
1084 INIT_HLIST_NODE(&ri->hlist);
1085 if (likely(rp)) {
1086 raw_spin_lock(&rp->lock);
1087 hlist_add_head(&ri->hlist, &rp->free_instances);
1088 raw_spin_unlock(&rp->lock);
1089 } else
1090 /* Unregistering */
1091 hlist_add_head(&ri->hlist, head);
1092}
1093NOKPROBE_SYMBOL(recycle_rp_inst);
1094
1095void kretprobe_hash_lock(struct task_struct *tsk,
1096 struct hlist_head **head, unsigned long *flags)
1097__acquires(hlist_lock)
1098{
1099 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1100 raw_spinlock_t *hlist_lock;
1101
1102 *head = &kretprobe_inst_table[hash];
1103 hlist_lock = kretprobe_table_lock_ptr(hash);
1104 raw_spin_lock_irqsave(hlist_lock, *flags);
1105}
1106NOKPROBE_SYMBOL(kretprobe_hash_lock);
1107
1108static void kretprobe_table_lock(unsigned long hash,
1109 unsigned long *flags)
1110__acquires(hlist_lock)
1111{
1112 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1113 raw_spin_lock_irqsave(hlist_lock, *flags);
1114}
1115NOKPROBE_SYMBOL(kretprobe_table_lock);
1116
1117void kretprobe_hash_unlock(struct task_struct *tsk,
1118 unsigned long *flags)
1119__releases(hlist_lock)
1120{
1121 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1122 raw_spinlock_t *hlist_lock;
1123
1124 hlist_lock = kretprobe_table_lock_ptr(hash);
1125 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1126}
1127NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1128
1129static void kretprobe_table_unlock(unsigned long hash,
1130 unsigned long *flags)
1131__releases(hlist_lock)
1132{
1133 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1134 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1135}
1136NOKPROBE_SYMBOL(kretprobe_table_unlock);
1137
1138/*
1139 * This function is called from finish_task_switch when task tk becomes dead,
1140 * so that we can recycle any function-return probe instances associated
1141 * with this task. These left over instances represent probed functions
1142 * that have been called but will never return.
1143 */
1144void kprobe_flush_task(struct task_struct *tk)
1145{
1146 struct kretprobe_instance *ri;
1147 struct hlist_head *head, empty_rp;
1148 struct hlist_node *tmp;
1149 unsigned long hash, flags = 0;
1150
1151 if (unlikely(!kprobes_initialized))
1152 /* Early boot. kretprobe_table_locks not yet initialized. */
1153 return;
1154
1155 INIT_HLIST_HEAD(&empty_rp);
1156 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1157 head = &kretprobe_inst_table[hash];
1158 kretprobe_table_lock(hash, &flags);
1159 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1160 if (ri->task == tk)
1161 recycle_rp_inst(ri, &empty_rp);
1162 }
1163 kretprobe_table_unlock(hash, &flags);
1164 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1165 hlist_del(&ri->hlist);
1166 kfree(ri);
1167 }
1168}
1169NOKPROBE_SYMBOL(kprobe_flush_task);
1170
1171static inline void free_rp_inst(struct kretprobe *rp)
1172{
1173 struct kretprobe_instance *ri;
1174 struct hlist_node *next;
1175
1176 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1177 hlist_del(&ri->hlist);
1178 kfree(ri);
1179 }
1180}
1181
1182static void cleanup_rp_inst(struct kretprobe *rp)
1183{
1184 unsigned long flags, hash;
1185 struct kretprobe_instance *ri;
1186 struct hlist_node *next;
1187 struct hlist_head *head;
1188
1189 /* No race here */
1190 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1191 kretprobe_table_lock(hash, &flags);
1192 head = &kretprobe_inst_table[hash];
1193 hlist_for_each_entry_safe(ri, next, head, hlist) {
1194 if (ri->rp == rp)
1195 ri->rp = NULL;
1196 }
1197 kretprobe_table_unlock(hash, &flags);
1198 }
1199 free_rp_inst(rp);
1200}
1201NOKPROBE_SYMBOL(cleanup_rp_inst);
1202
1203/*
1204* Add the new probe to ap->list. Fail if this is the
1205* second jprobe at the address - two jprobes can't coexist
1206*/
1207static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1208{
1209 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1210
1211 if (p->break_handler || p->post_handler)
1212 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1213
1214 if (p->break_handler) {
1215 if (ap->break_handler)
1216 return -EEXIST;
1217 list_add_tail_rcu(&p->list, &ap->list);
1218 ap->break_handler = aggr_break_handler;
1219 } else
1220 list_add_rcu(&p->list, &ap->list);
1221 if (p->post_handler && !ap->post_handler)
1222 ap->post_handler = aggr_post_handler;
1223
1224 return 0;
1225}
1226
1227/*
1228 * Fill in the required fields of the "manager kprobe". Replace the
1229 * earlier kprobe in the hlist with the manager kprobe
1230 */
1231static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1232{
1233 /* Copy p's insn slot to ap */
1234 copy_kprobe(p, ap);
1235 flush_insn_slot(ap);
1236 ap->addr = p->addr;
1237 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1238 ap->pre_handler = aggr_pre_handler;
1239 ap->fault_handler = aggr_fault_handler;
1240 /* We don't care the kprobe which has gone. */
1241 if (p->post_handler && !kprobe_gone(p))
1242 ap->post_handler = aggr_post_handler;
1243 if (p->break_handler && !kprobe_gone(p))
1244 ap->break_handler = aggr_break_handler;
1245
1246 INIT_LIST_HEAD(&ap->list);
1247 INIT_HLIST_NODE(&ap->hlist);
1248
1249 list_add_rcu(&p->list, &ap->list);
1250 hlist_replace_rcu(&p->hlist, &ap->hlist);
1251}
1252
1253/*
1254 * This is the second or subsequent kprobe at the address - handle
1255 * the intricacies
1256 */
1257static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1258{
1259 int ret = 0;
1260 struct kprobe *ap = orig_p;
1261
1262 /* For preparing optimization, jump_label_text_reserved() is called */
1263 jump_label_lock();
1264 /*
1265 * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1266 * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1267 */
1268 get_online_cpus();
1269 mutex_lock(&text_mutex);
1270
1271 if (!kprobe_aggrprobe(orig_p)) {
1272 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1273 ap = alloc_aggr_kprobe(orig_p);
1274 if (!ap) {
1275 ret = -ENOMEM;
1276 goto out;
1277 }
1278 init_aggr_kprobe(ap, orig_p);
1279 } else if (kprobe_unused(ap))
1280 /* This probe is going to die. Rescue it */
1281 reuse_unused_kprobe(ap);
1282
1283 if (kprobe_gone(ap)) {
1284 /*
1285 * Attempting to insert new probe at the same location that
1286 * had a probe in the module vaddr area which already
1287 * freed. So, the instruction slot has already been
1288 * released. We need a new slot for the new probe.
1289 */
1290 ret = arch_prepare_kprobe(ap);
1291 if (ret)
1292 /*
1293 * Even if fail to allocate new slot, don't need to
1294 * free aggr_probe. It will be used next time, or
1295 * freed by unregister_kprobe.
1296 */
1297 goto out;
1298
1299 /* Prepare optimized instructions if possible. */
1300 prepare_optimized_kprobe(ap);
1301
1302 /*
1303 * Clear gone flag to prevent allocating new slot again, and
1304 * set disabled flag because it is not armed yet.
1305 */
1306 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1307 | KPROBE_FLAG_DISABLED;
1308 }
1309
1310 /* Copy ap's insn slot to p */
1311 copy_kprobe(ap, p);
1312 ret = add_new_kprobe(ap, p);
1313
1314out:
1315 mutex_unlock(&text_mutex);
1316 put_online_cpus();
1317 jump_label_unlock();
1318
1319 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1320 ap->flags &= ~KPROBE_FLAG_DISABLED;
1321 if (!kprobes_all_disarmed)
1322 /* Arm the breakpoint again. */
1323 arm_kprobe(ap);
1324 }
1325 return ret;
1326}
1327
1328bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1329{
1330 /* The __kprobes marked functions and entry code must not be probed */
1331 return addr >= (unsigned long)__kprobes_text_start &&
1332 addr < (unsigned long)__kprobes_text_end;
1333}
1334
1335bool within_kprobe_blacklist(unsigned long addr)
1336{
1337 struct kprobe_blacklist_entry *ent;
1338
1339 if (arch_within_kprobe_blacklist(addr))
1340 return true;
1341 /*
1342 * If there exists a kprobe_blacklist, verify and
1343 * fail any probe registration in the prohibited area
1344 */
1345 list_for_each_entry(ent, &kprobe_blacklist, list) {
1346 if (addr >= ent->start_addr && addr < ent->end_addr)
1347 return true;
1348 }
1349
1350 return false;
1351}
1352
1353/*
1354 * If we have a symbol_name argument, look it up and add the offset field
1355 * to it. This way, we can specify a relative address to a symbol.
1356 * This returns encoded errors if it fails to look up symbol or invalid
1357 * combination of parameters.
1358 */
1359static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1360{
1361 kprobe_opcode_t *addr = p->addr;
1362
1363 if ((p->symbol_name && p->addr) ||
1364 (!p->symbol_name && !p->addr))
1365 goto invalid;
1366
1367 if (p->symbol_name) {
1368 kprobe_lookup_name(p->symbol_name, addr);
1369 if (!addr)
1370 return ERR_PTR(-ENOENT);
1371 }
1372
1373 addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1374 if (addr)
1375 return addr;
1376
1377invalid:
1378 return ERR_PTR(-EINVAL);
1379}
1380
1381/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1382static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1383{
1384 struct kprobe *ap, *list_p;
1385
1386 ap = get_kprobe(p->addr);
1387 if (unlikely(!ap))
1388 return NULL;
1389
1390 if (p != ap) {
1391 list_for_each_entry_rcu(list_p, &ap->list, list)
1392 if (list_p == p)
1393 /* kprobe p is a valid probe */
1394 goto valid;
1395 return NULL;
1396 }
1397valid:
1398 return ap;
1399}
1400
1401/* Return error if the kprobe is being re-registered */
1402static inline int check_kprobe_rereg(struct kprobe *p)
1403{
1404 int ret = 0;
1405
1406 mutex_lock(&kprobe_mutex);
1407 if (__get_valid_kprobe(p))
1408 ret = -EINVAL;
1409 mutex_unlock(&kprobe_mutex);
1410
1411 return ret;
1412}
1413
1414int __weak arch_check_ftrace_location(struct kprobe *p)
1415{
1416 unsigned long ftrace_addr;
1417
1418 ftrace_addr = ftrace_location((unsigned long)p->addr);
1419 if (ftrace_addr) {
1420#ifdef CONFIG_KPROBES_ON_FTRACE
1421 /* Given address is not on the instruction boundary */
1422 if ((unsigned long)p->addr != ftrace_addr)
1423 return -EILSEQ;
1424 p->flags |= KPROBE_FLAG_FTRACE;
1425#else /* !CONFIG_KPROBES_ON_FTRACE */
1426 return -EINVAL;
1427#endif
1428 }
1429 return 0;
1430}
1431
1432static int check_kprobe_address_safe(struct kprobe *p,
1433 struct module **probed_mod)
1434{
1435 int ret;
1436
1437 ret = arch_check_ftrace_location(p);
1438 if (ret)
1439 return ret;
1440 jump_label_lock();
1441 preempt_disable();
1442
1443 /* Ensure it is not in reserved area nor out of text */
1444 if (!kernel_text_address((unsigned long) p->addr) ||
1445 within_kprobe_blacklist((unsigned long) p->addr) ||
1446 jump_label_text_reserved(p->addr, p->addr)) {
1447 ret = -EINVAL;
1448 goto out;
1449 }
1450
1451 /* Check if are we probing a module */
1452 *probed_mod = __module_text_address((unsigned long) p->addr);
1453 if (*probed_mod) {
1454 /*
1455 * We must hold a refcount of the probed module while updating
1456 * its code to prohibit unexpected unloading.
1457 */
1458 if (unlikely(!try_module_get(*probed_mod))) {
1459 ret = -ENOENT;
1460 goto out;
1461 }
1462
1463 /*
1464 * If the module freed .init.text, we couldn't insert
1465 * kprobes in there.
1466 */
1467 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1468 (*probed_mod)->state != MODULE_STATE_COMING) {
1469 module_put(*probed_mod);
1470 *probed_mod = NULL;
1471 ret = -ENOENT;
1472 }
1473 }
1474out:
1475 preempt_enable();
1476 jump_label_unlock();
1477
1478 return ret;
1479}
1480
1481int register_kprobe(struct kprobe *p)
1482{
1483 int ret;
1484 struct kprobe *old_p;
1485 struct module *probed_mod;
1486 kprobe_opcode_t *addr;
1487
1488 /* Adjust probe address from symbol */
1489 addr = kprobe_addr(p);
1490 if (IS_ERR(addr))
1491 return PTR_ERR(addr);
1492 p->addr = addr;
1493
1494 ret = check_kprobe_rereg(p);
1495 if (ret)
1496 return ret;
1497
1498 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1499 p->flags &= KPROBE_FLAG_DISABLED;
1500 p->nmissed = 0;
1501 INIT_LIST_HEAD(&p->list);
1502
1503 ret = check_kprobe_address_safe(p, &probed_mod);
1504 if (ret)
1505 return ret;
1506
1507 mutex_lock(&kprobe_mutex);
1508
1509 old_p = get_kprobe(p->addr);
1510 if (old_p) {
1511 /* Since this may unoptimize old_p, locking text_mutex. */
1512 ret = register_aggr_kprobe(old_p, p);
1513 goto out;
1514 }
1515
1516 mutex_lock(&text_mutex); /* Avoiding text modification */
1517 ret = prepare_kprobe(p);
1518 mutex_unlock(&text_mutex);
1519 if (ret)
1520 goto out;
1521
1522 INIT_HLIST_NODE(&p->hlist);
1523 hlist_add_head_rcu(&p->hlist,
1524 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1525
1526 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1527 arm_kprobe(p);
1528
1529 /* Try to optimize kprobe */
1530 try_to_optimize_kprobe(p);
1531
1532out:
1533 mutex_unlock(&kprobe_mutex);
1534
1535 if (probed_mod)
1536 module_put(probed_mod);
1537
1538 return ret;
1539}
1540EXPORT_SYMBOL_GPL(register_kprobe);
1541
1542/* Check if all probes on the aggrprobe are disabled */
1543static int aggr_kprobe_disabled(struct kprobe *ap)
1544{
1545 struct kprobe *kp;
1546
1547 list_for_each_entry_rcu(kp, &ap->list, list)
1548 if (!kprobe_disabled(kp))
1549 /*
1550 * There is an active probe on the list.
1551 * We can't disable this ap.
1552 */
1553 return 0;
1554
1555 return 1;
1556}
1557
1558/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1559static struct kprobe *__disable_kprobe(struct kprobe *p)
1560{
1561 struct kprobe *orig_p;
1562
1563 /* Get an original kprobe for return */
1564 orig_p = __get_valid_kprobe(p);
1565 if (unlikely(orig_p == NULL))
1566 return NULL;
1567
1568 if (!kprobe_disabled(p)) {
1569 /* Disable probe if it is a child probe */
1570 if (p != orig_p)
1571 p->flags |= KPROBE_FLAG_DISABLED;
1572
1573 /* Try to disarm and disable this/parent probe */
1574 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1575 /*
1576 * If kprobes_all_disarmed is set, orig_p
1577 * should have already been disarmed, so
1578 * skip unneed disarming process.
1579 */
1580 if (!kprobes_all_disarmed)
1581 disarm_kprobe(orig_p, true);
1582 orig_p->flags |= KPROBE_FLAG_DISABLED;
1583 }
1584 }
1585
1586 return orig_p;
1587}
1588
1589/*
1590 * Unregister a kprobe without a scheduler synchronization.
1591 */
1592static int __unregister_kprobe_top(struct kprobe *p)
1593{
1594 struct kprobe *ap, *list_p;
1595
1596 /* Disable kprobe. This will disarm it if needed. */
1597 ap = __disable_kprobe(p);
1598 if (ap == NULL)
1599 return -EINVAL;
1600
1601 if (ap == p)
1602 /*
1603 * This probe is an independent(and non-optimized) kprobe
1604 * (not an aggrprobe). Remove from the hash list.
1605 */
1606 goto disarmed;
1607
1608 /* Following process expects this probe is an aggrprobe */
1609 WARN_ON(!kprobe_aggrprobe(ap));
1610
1611 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1612 /*
1613 * !disarmed could be happen if the probe is under delayed
1614 * unoptimizing.
1615 */
1616 goto disarmed;
1617 else {
1618 /* If disabling probe has special handlers, update aggrprobe */
1619 if (p->break_handler && !kprobe_gone(p))
1620 ap->break_handler = NULL;
1621 if (p->post_handler && !kprobe_gone(p)) {
1622 list_for_each_entry_rcu(list_p, &ap->list, list) {
1623 if ((list_p != p) && (list_p->post_handler))
1624 goto noclean;
1625 }
1626 ap->post_handler = NULL;
1627 }
1628noclean:
1629 /*
1630 * Remove from the aggrprobe: this path will do nothing in
1631 * __unregister_kprobe_bottom().
1632 */
1633 list_del_rcu(&p->list);
1634 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1635 /*
1636 * Try to optimize this probe again, because post
1637 * handler may have been changed.
1638 */
1639 optimize_kprobe(ap);
1640 }
1641 return 0;
1642
1643disarmed:
1644 BUG_ON(!kprobe_disarmed(ap));
1645 hlist_del_rcu(&ap->hlist);
1646 return 0;
1647}
1648
1649static void __unregister_kprobe_bottom(struct kprobe *p)
1650{
1651 struct kprobe *ap;
1652
1653 if (list_empty(&p->list))
1654 /* This is an independent kprobe */
1655 arch_remove_kprobe(p);
1656 else if (list_is_singular(&p->list)) {
1657 /* This is the last child of an aggrprobe */
1658 ap = list_entry(p->list.next, struct kprobe, list);
1659 list_del(&p->list);
1660 free_aggr_kprobe(ap);
1661 }
1662 /* Otherwise, do nothing. */
1663}
1664
1665int register_kprobes(struct kprobe **kps, int num)
1666{
1667 int i, ret = 0;
1668
1669 if (num <= 0)
1670 return -EINVAL;
1671 for (i = 0; i < num; i++) {
1672 ret = register_kprobe(kps[i]);
1673 if (ret < 0) {
1674 if (i > 0)
1675 unregister_kprobes(kps, i);
1676 break;
1677 }
1678 }
1679 return ret;
1680}
1681EXPORT_SYMBOL_GPL(register_kprobes);
1682
1683void unregister_kprobe(struct kprobe *p)
1684{
1685 unregister_kprobes(&p, 1);
1686}
1687EXPORT_SYMBOL_GPL(unregister_kprobe);
1688
1689void unregister_kprobes(struct kprobe **kps, int num)
1690{
1691 int i;
1692
1693 if (num <= 0)
1694 return;
1695 mutex_lock(&kprobe_mutex);
1696 for (i = 0; i < num; i++)
1697 if (__unregister_kprobe_top(kps[i]) < 0)
1698 kps[i]->addr = NULL;
1699 mutex_unlock(&kprobe_mutex);
1700
1701 synchronize_sched();
1702 for (i = 0; i < num; i++)
1703 if (kps[i]->addr)
1704 __unregister_kprobe_bottom(kps[i]);
1705}
1706EXPORT_SYMBOL_GPL(unregister_kprobes);
1707
1708static struct notifier_block kprobe_exceptions_nb = {
1709 .notifier_call = kprobe_exceptions_notify,
1710 .priority = 0x7fffffff /* we need to be notified first */
1711};
1712
1713unsigned long __weak arch_deref_entry_point(void *entry)
1714{
1715 return (unsigned long)entry;
1716}
1717
1718int register_jprobes(struct jprobe **jps, int num)
1719{
1720 struct jprobe *jp;
1721 int ret = 0, i;
1722
1723 if (num <= 0)
1724 return -EINVAL;
1725 for (i = 0; i < num; i++) {
1726 unsigned long addr, offset;
1727 jp = jps[i];
1728 addr = arch_deref_entry_point(jp->entry);
1729
1730 /* Verify probepoint is a function entry point */
1731 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1732 offset == 0) {
1733 jp->kp.pre_handler = setjmp_pre_handler;
1734 jp->kp.break_handler = longjmp_break_handler;
1735 ret = register_kprobe(&jp->kp);
1736 } else
1737 ret = -EINVAL;
1738
1739 if (ret < 0) {
1740 if (i > 0)
1741 unregister_jprobes(jps, i);
1742 break;
1743 }
1744 }
1745 return ret;
1746}
1747EXPORT_SYMBOL_GPL(register_jprobes);
1748
1749int register_jprobe(struct jprobe *jp)
1750{
1751 return register_jprobes(&jp, 1);
1752}
1753EXPORT_SYMBOL_GPL(register_jprobe);
1754
1755void unregister_jprobe(struct jprobe *jp)
1756{
1757 unregister_jprobes(&jp, 1);
1758}
1759EXPORT_SYMBOL_GPL(unregister_jprobe);
1760
1761void unregister_jprobes(struct jprobe **jps, int num)
1762{
1763 int i;
1764
1765 if (num <= 0)
1766 return;
1767 mutex_lock(&kprobe_mutex);
1768 for (i = 0; i < num; i++)
1769 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1770 jps[i]->kp.addr = NULL;
1771 mutex_unlock(&kprobe_mutex);
1772
1773 synchronize_sched();
1774 for (i = 0; i < num; i++) {
1775 if (jps[i]->kp.addr)
1776 __unregister_kprobe_bottom(&jps[i]->kp);
1777 }
1778}
1779EXPORT_SYMBOL_GPL(unregister_jprobes);
1780
1781#ifdef CONFIG_KRETPROBES
1782/*
1783 * This kprobe pre_handler is registered with every kretprobe. When probe
1784 * hits it will set up the return probe.
1785 */
1786static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1787{
1788 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1789 unsigned long hash, flags = 0;
1790 struct kretprobe_instance *ri;
1791
1792 /*
1793 * To avoid deadlocks, prohibit return probing in NMI contexts,
1794 * just skip the probe and increase the (inexact) 'nmissed'
1795 * statistical counter, so that the user is informed that
1796 * something happened:
1797 */
1798 if (unlikely(in_nmi())) {
1799 rp->nmissed++;
1800 return 0;
1801 }
1802
1803 /* TODO: consider to only swap the RA after the last pre_handler fired */
1804 hash = hash_ptr(current, KPROBE_HASH_BITS);
1805 raw_spin_lock_irqsave(&rp->lock, flags);
1806 if (!hlist_empty(&rp->free_instances)) {
1807 ri = hlist_entry(rp->free_instances.first,
1808 struct kretprobe_instance, hlist);
1809 hlist_del(&ri->hlist);
1810 raw_spin_unlock_irqrestore(&rp->lock, flags);
1811
1812 ri->rp = rp;
1813 ri->task = current;
1814
1815 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1816 raw_spin_lock_irqsave(&rp->lock, flags);
1817 hlist_add_head(&ri->hlist, &rp->free_instances);
1818 raw_spin_unlock_irqrestore(&rp->lock, flags);
1819 return 0;
1820 }
1821
1822 arch_prepare_kretprobe(ri, regs);
1823
1824 /* XXX(hch): why is there no hlist_move_head? */
1825 INIT_HLIST_NODE(&ri->hlist);
1826 kretprobe_table_lock(hash, &flags);
1827 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1828 kretprobe_table_unlock(hash, &flags);
1829 } else {
1830 rp->nmissed++;
1831 raw_spin_unlock_irqrestore(&rp->lock, flags);
1832 }
1833 return 0;
1834}
1835NOKPROBE_SYMBOL(pre_handler_kretprobe);
1836
1837int register_kretprobe(struct kretprobe *rp)
1838{
1839 int ret = 0;
1840 struct kretprobe_instance *inst;
1841 int i;
1842 void *addr;
1843
1844 if (kretprobe_blacklist_size) {
1845 addr = kprobe_addr(&rp->kp);
1846 if (IS_ERR(addr))
1847 return PTR_ERR(addr);
1848
1849 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1850 if (kretprobe_blacklist[i].addr == addr)
1851 return -EINVAL;
1852 }
1853 }
1854
1855 rp->kp.pre_handler = pre_handler_kretprobe;
1856 rp->kp.post_handler = NULL;
1857 rp->kp.fault_handler = NULL;
1858 rp->kp.break_handler = NULL;
1859
1860 /* Pre-allocate memory for max kretprobe instances */
1861 if (rp->maxactive <= 0) {
1862#ifdef CONFIG_PREEMPT
1863 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1864#else
1865 rp->maxactive = num_possible_cpus();
1866#endif
1867 }
1868 raw_spin_lock_init(&rp->lock);
1869 INIT_HLIST_HEAD(&rp->free_instances);
1870 for (i = 0; i < rp->maxactive; i++) {
1871 inst = kmalloc(sizeof(struct kretprobe_instance) +
1872 rp->data_size, GFP_KERNEL);
1873 if (inst == NULL) {
1874 free_rp_inst(rp);
1875 return -ENOMEM;
1876 }
1877 INIT_HLIST_NODE(&inst->hlist);
1878 hlist_add_head(&inst->hlist, &rp->free_instances);
1879 }
1880
1881 rp->nmissed = 0;
1882 /* Establish function entry probe point */
1883 ret = register_kprobe(&rp->kp);
1884 if (ret != 0)
1885 free_rp_inst(rp);
1886 return ret;
1887}
1888EXPORT_SYMBOL_GPL(register_kretprobe);
1889
1890int register_kretprobes(struct kretprobe **rps, int num)
1891{
1892 int ret = 0, i;
1893
1894 if (num <= 0)
1895 return -EINVAL;
1896 for (i = 0; i < num; i++) {
1897 ret = register_kretprobe(rps[i]);
1898 if (ret < 0) {
1899 if (i > 0)
1900 unregister_kretprobes(rps, i);
1901 break;
1902 }
1903 }
1904 return ret;
1905}
1906EXPORT_SYMBOL_GPL(register_kretprobes);
1907
1908void unregister_kretprobe(struct kretprobe *rp)
1909{
1910 unregister_kretprobes(&rp, 1);
1911}
1912EXPORT_SYMBOL_GPL(unregister_kretprobe);
1913
1914void unregister_kretprobes(struct kretprobe **rps, int num)
1915{
1916 int i;
1917
1918 if (num <= 0)
1919 return;
1920 mutex_lock(&kprobe_mutex);
1921 for (i = 0; i < num; i++)
1922 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1923 rps[i]->kp.addr = NULL;
1924 mutex_unlock(&kprobe_mutex);
1925
1926 synchronize_sched();
1927 for (i = 0; i < num; i++) {
1928 if (rps[i]->kp.addr) {
1929 __unregister_kprobe_bottom(&rps[i]->kp);
1930 cleanup_rp_inst(rps[i]);
1931 }
1932 }
1933}
1934EXPORT_SYMBOL_GPL(unregister_kretprobes);
1935
1936#else /* CONFIG_KRETPROBES */
1937int register_kretprobe(struct kretprobe *rp)
1938{
1939 return -ENOSYS;
1940}
1941EXPORT_SYMBOL_GPL(register_kretprobe);
1942
1943int register_kretprobes(struct kretprobe **rps, int num)
1944{
1945 return -ENOSYS;
1946}
1947EXPORT_SYMBOL_GPL(register_kretprobes);
1948
1949void unregister_kretprobe(struct kretprobe *rp)
1950{
1951}
1952EXPORT_SYMBOL_GPL(unregister_kretprobe);
1953
1954void unregister_kretprobes(struct kretprobe **rps, int num)
1955{
1956}
1957EXPORT_SYMBOL_GPL(unregister_kretprobes);
1958
1959static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1960{
1961 return 0;
1962}
1963NOKPROBE_SYMBOL(pre_handler_kretprobe);
1964
1965#endif /* CONFIG_KRETPROBES */
1966
1967/* Set the kprobe gone and remove its instruction buffer. */
1968static void kill_kprobe(struct kprobe *p)
1969{
1970 struct kprobe *kp;
1971
1972 p->flags |= KPROBE_FLAG_GONE;
1973 if (kprobe_aggrprobe(p)) {
1974 /*
1975 * If this is an aggr_kprobe, we have to list all the
1976 * chained probes and mark them GONE.
1977 */
1978 list_for_each_entry_rcu(kp, &p->list, list)
1979 kp->flags |= KPROBE_FLAG_GONE;
1980 p->post_handler = NULL;
1981 p->break_handler = NULL;
1982 kill_optimized_kprobe(p);
1983 }
1984 /*
1985 * Here, we can remove insn_slot safely, because no thread calls
1986 * the original probed function (which will be freed soon) any more.
1987 */
1988 arch_remove_kprobe(p);
1989}
1990
1991/* Disable one kprobe */
1992int disable_kprobe(struct kprobe *kp)
1993{
1994 int ret = 0;
1995
1996 mutex_lock(&kprobe_mutex);
1997
1998 /* Disable this kprobe */
1999 if (__disable_kprobe(kp) == NULL)
2000 ret = -EINVAL;
2001
2002 mutex_unlock(&kprobe_mutex);
2003 return ret;
2004}
2005EXPORT_SYMBOL_GPL(disable_kprobe);
2006
2007/* Enable one kprobe */
2008int enable_kprobe(struct kprobe *kp)
2009{
2010 int ret = 0;
2011 struct kprobe *p;
2012
2013 mutex_lock(&kprobe_mutex);
2014
2015 /* Check whether specified probe is valid. */
2016 p = __get_valid_kprobe(kp);
2017 if (unlikely(p == NULL)) {
2018 ret = -EINVAL;
2019 goto out;
2020 }
2021
2022 if (kprobe_gone(kp)) {
2023 /* This kprobe has gone, we couldn't enable it. */
2024 ret = -EINVAL;
2025 goto out;
2026 }
2027
2028 if (p != kp)
2029 kp->flags &= ~KPROBE_FLAG_DISABLED;
2030
2031 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2032 p->flags &= ~KPROBE_FLAG_DISABLED;
2033 arm_kprobe(p);
2034 }
2035out:
2036 mutex_unlock(&kprobe_mutex);
2037 return ret;
2038}
2039EXPORT_SYMBOL_GPL(enable_kprobe);
2040
2041void dump_kprobe(struct kprobe *kp)
2042{
2043 printk(KERN_WARNING "Dumping kprobe:\n");
2044 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2045 kp->symbol_name, kp->addr, kp->offset);
2046}
2047NOKPROBE_SYMBOL(dump_kprobe);
2048
2049/*
2050 * Lookup and populate the kprobe_blacklist.
2051 *
2052 * Unlike the kretprobe blacklist, we'll need to determine
2053 * the range of addresses that belong to the said functions,
2054 * since a kprobe need not necessarily be at the beginning
2055 * of a function.
2056 */
2057static int __init populate_kprobe_blacklist(unsigned long *start,
2058 unsigned long *end)
2059{
2060 unsigned long *iter;
2061 struct kprobe_blacklist_entry *ent;
2062 unsigned long entry, offset = 0, size = 0;
2063
2064 for (iter = start; iter < end; iter++) {
2065 entry = arch_deref_entry_point((void *)*iter);
2066
2067 if (!kernel_text_address(entry) ||
2068 !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2069 pr_err("Failed to find blacklist at %p\n",
2070 (void *)entry);
2071 continue;
2072 }
2073
2074 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2075 if (!ent)
2076 return -ENOMEM;
2077 ent->start_addr = entry;
2078 ent->end_addr = entry + size;
2079 INIT_LIST_HEAD(&ent->list);
2080 list_add_tail(&ent->list, &kprobe_blacklist);
2081 }
2082 return 0;
2083}
2084
2085/* Module notifier call back, checking kprobes on the module */
2086static int kprobes_module_callback(struct notifier_block *nb,
2087 unsigned long val, void *data)
2088{
2089 struct module *mod = data;
2090 struct hlist_head *head;
2091 struct kprobe *p;
2092 unsigned int i;
2093 int checkcore = (val == MODULE_STATE_GOING);
2094
2095 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2096 return NOTIFY_DONE;
2097
2098 /*
2099 * When MODULE_STATE_GOING was notified, both of module .text and
2100 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2101 * notified, only .init.text section would be freed. We need to
2102 * disable kprobes which have been inserted in the sections.
2103 */
2104 mutex_lock(&kprobe_mutex);
2105 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2106 head = &kprobe_table[i];
2107 hlist_for_each_entry_rcu(p, head, hlist)
2108 if (within_module_init((unsigned long)p->addr, mod) ||
2109 (checkcore &&
2110 within_module_core((unsigned long)p->addr, mod))) {
2111 /*
2112 * The vaddr this probe is installed will soon
2113 * be vfreed buy not synced to disk. Hence,
2114 * disarming the breakpoint isn't needed.
2115 */
2116 kill_kprobe(p);
2117 }
2118 }
2119 mutex_unlock(&kprobe_mutex);
2120 return NOTIFY_DONE;
2121}
2122
2123static struct notifier_block kprobe_module_nb = {
2124 .notifier_call = kprobes_module_callback,
2125 .priority = 0
2126};
2127
2128/* Markers of _kprobe_blacklist section */
2129extern unsigned long __start_kprobe_blacklist[];
2130extern unsigned long __stop_kprobe_blacklist[];
2131
2132static int __init init_kprobes(void)
2133{
2134 int i, err = 0;
2135
2136 /* FIXME allocate the probe table, currently defined statically */
2137 /* initialize all list heads */
2138 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2139 INIT_HLIST_HEAD(&kprobe_table[i]);
2140 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2141 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2142 }
2143
2144 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2145 __stop_kprobe_blacklist);
2146 if (err) {
2147 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2148 pr_err("Please take care of using kprobes.\n");
2149 }
2150
2151 if (kretprobe_blacklist_size) {
2152 /* lookup the function address from its name */
2153 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2154 kprobe_lookup_name(kretprobe_blacklist[i].name,
2155 kretprobe_blacklist[i].addr);
2156 if (!kretprobe_blacklist[i].addr)
2157 printk("kretprobe: lookup failed: %s\n",
2158 kretprobe_blacklist[i].name);
2159 }
2160 }
2161
2162#if defined(CONFIG_OPTPROBES)
2163#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2164 /* Init kprobe_optinsn_slots */
2165 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2166#endif
2167 /* By default, kprobes can be optimized */
2168 kprobes_allow_optimization = true;
2169#endif
2170
2171 /* By default, kprobes are armed */
2172 kprobes_all_disarmed = false;
2173
2174 err = arch_init_kprobes();
2175 if (!err)
2176 err = register_die_notifier(&kprobe_exceptions_nb);
2177 if (!err)
2178 err = register_module_notifier(&kprobe_module_nb);
2179
2180 kprobes_initialized = (err == 0);
2181
2182 if (!err)
2183 init_test_probes();
2184 return err;
2185}
2186
2187#ifdef CONFIG_DEBUG_FS
2188static void report_probe(struct seq_file *pi, struct kprobe *p,
2189 const char *sym, int offset, char *modname, struct kprobe *pp)
2190{
2191 char *kprobe_type;
2192
2193 if (p->pre_handler == pre_handler_kretprobe)
2194 kprobe_type = "r";
2195 else if (p->pre_handler == setjmp_pre_handler)
2196 kprobe_type = "j";
2197 else
2198 kprobe_type = "k";
2199
2200 if (sym)
2201 seq_printf(pi, "%p %s %s+0x%x %s ",
2202 p->addr, kprobe_type, sym, offset,
2203 (modname ? modname : " "));
2204 else
2205 seq_printf(pi, "%p %s %p ",
2206 p->addr, kprobe_type, p->addr);
2207
2208 if (!pp)
2209 pp = p;
2210 seq_printf(pi, "%s%s%s%s\n",
2211 (kprobe_gone(p) ? "[GONE]" : ""),
2212 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2213 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2214 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2215}
2216
2217static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2218{
2219 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2220}
2221
2222static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2223{
2224 (*pos)++;
2225 if (*pos >= KPROBE_TABLE_SIZE)
2226 return NULL;
2227 return pos;
2228}
2229
2230static void kprobe_seq_stop(struct seq_file *f, void *v)
2231{
2232 /* Nothing to do */
2233}
2234
2235static int show_kprobe_addr(struct seq_file *pi, void *v)
2236{
2237 struct hlist_head *head;
2238 struct kprobe *p, *kp;
2239 const char *sym = NULL;
2240 unsigned int i = *(loff_t *) v;
2241 unsigned long offset = 0;
2242 char *modname, namebuf[KSYM_NAME_LEN];
2243
2244 head = &kprobe_table[i];
2245 preempt_disable();
2246 hlist_for_each_entry_rcu(p, head, hlist) {
2247 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2248 &offset, &modname, namebuf);
2249 if (kprobe_aggrprobe(p)) {
2250 list_for_each_entry_rcu(kp, &p->list, list)
2251 report_probe(pi, kp, sym, offset, modname, p);
2252 } else
2253 report_probe(pi, p, sym, offset, modname, NULL);
2254 }
2255 preempt_enable();
2256 return 0;
2257}
2258
2259static const struct seq_operations kprobes_seq_ops = {
2260 .start = kprobe_seq_start,
2261 .next = kprobe_seq_next,
2262 .stop = kprobe_seq_stop,
2263 .show = show_kprobe_addr
2264};
2265
2266static int kprobes_open(struct inode *inode, struct file *filp)
2267{
2268 return seq_open(filp, &kprobes_seq_ops);
2269}
2270
2271static const struct file_operations debugfs_kprobes_operations = {
2272 .open = kprobes_open,
2273 .read = seq_read,
2274 .llseek = seq_lseek,
2275 .release = seq_release,
2276};
2277
2278/* kprobes/blacklist -- shows which functions can not be probed */
2279static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2280{
2281 return seq_list_start(&kprobe_blacklist, *pos);
2282}
2283
2284static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2285{
2286 return seq_list_next(v, &kprobe_blacklist, pos);
2287}
2288
2289static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2290{
2291 struct kprobe_blacklist_entry *ent =
2292 list_entry(v, struct kprobe_blacklist_entry, list);
2293
2294 seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2295 (void *)ent->end_addr, (void *)ent->start_addr);
2296 return 0;
2297}
2298
2299static const struct seq_operations kprobe_blacklist_seq_ops = {
2300 .start = kprobe_blacklist_seq_start,
2301 .next = kprobe_blacklist_seq_next,
2302 .stop = kprobe_seq_stop, /* Reuse void function */
2303 .show = kprobe_blacklist_seq_show,
2304};
2305
2306static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2307{
2308 return seq_open(filp, &kprobe_blacklist_seq_ops);
2309}
2310
2311static const struct file_operations debugfs_kprobe_blacklist_ops = {
2312 .open = kprobe_blacklist_open,
2313 .read = seq_read,
2314 .llseek = seq_lseek,
2315 .release = seq_release,
2316};
2317
2318static void arm_all_kprobes(void)
2319{
2320 struct hlist_head *head;
2321 struct kprobe *p;
2322 unsigned int i;
2323
2324 mutex_lock(&kprobe_mutex);
2325
2326 /* If kprobes are armed, just return */
2327 if (!kprobes_all_disarmed)
2328 goto already_enabled;
2329
2330 /*
2331 * optimize_kprobe() called by arm_kprobe() checks
2332 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2333 * arm_kprobe.
2334 */
2335 kprobes_all_disarmed = false;
2336 /* Arming kprobes doesn't optimize kprobe itself */
2337 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2338 head = &kprobe_table[i];
2339 hlist_for_each_entry_rcu(p, head, hlist)
2340 if (!kprobe_disabled(p))
2341 arm_kprobe(p);
2342 }
2343
2344 printk(KERN_INFO "Kprobes globally enabled\n");
2345
2346already_enabled:
2347 mutex_unlock(&kprobe_mutex);
2348 return;
2349}
2350
2351static void disarm_all_kprobes(void)
2352{
2353 struct hlist_head *head;
2354 struct kprobe *p;
2355 unsigned int i;
2356
2357 mutex_lock(&kprobe_mutex);
2358
2359 /* If kprobes are already disarmed, just return */
2360 if (kprobes_all_disarmed) {
2361 mutex_unlock(&kprobe_mutex);
2362 return;
2363 }
2364
2365 kprobes_all_disarmed = true;
2366 printk(KERN_INFO "Kprobes globally disabled\n");
2367
2368 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2369 head = &kprobe_table[i];
2370 hlist_for_each_entry_rcu(p, head, hlist) {
2371 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2372 disarm_kprobe(p, false);
2373 }
2374 }
2375 mutex_unlock(&kprobe_mutex);
2376
2377 /* Wait for disarming all kprobes by optimizer */
2378 wait_for_kprobe_optimizer();
2379}
2380
2381/*
2382 * XXX: The debugfs bool file interface doesn't allow for callbacks
2383 * when the bool state is switched. We can reuse that facility when
2384 * available
2385 */
2386static ssize_t read_enabled_file_bool(struct file *file,
2387 char __user *user_buf, size_t count, loff_t *ppos)
2388{
2389 char buf[3];
2390
2391 if (!kprobes_all_disarmed)
2392 buf[0] = '1';
2393 else
2394 buf[0] = '0';
2395 buf[1] = '\n';
2396 buf[2] = 0x00;
2397 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2398}
2399
2400static ssize_t write_enabled_file_bool(struct file *file,
2401 const char __user *user_buf, size_t count, loff_t *ppos)
2402{
2403 char buf[32];
2404 size_t buf_size;
2405
2406 buf_size = min(count, (sizeof(buf)-1));
2407 if (copy_from_user(buf, user_buf, buf_size))
2408 return -EFAULT;
2409
2410 buf[buf_size] = '\0';
2411 switch (buf[0]) {
2412 case 'y':
2413 case 'Y':
2414 case '1':
2415 arm_all_kprobes();
2416 break;
2417 case 'n':
2418 case 'N':
2419 case '0':
2420 disarm_all_kprobes();
2421 break;
2422 default:
2423 return -EINVAL;
2424 }
2425
2426 return count;
2427}
2428
2429static const struct file_operations fops_kp = {
2430 .read = read_enabled_file_bool,
2431 .write = write_enabled_file_bool,
2432 .llseek = default_llseek,
2433};
2434
2435static int __init debugfs_kprobe_init(void)
2436{
2437 struct dentry *dir, *file;
2438 unsigned int value = 1;
2439
2440 dir = debugfs_create_dir("kprobes", NULL);
2441 if (!dir)
2442 return -ENOMEM;
2443
2444 file = debugfs_create_file("list", 0444, dir, NULL,
2445 &debugfs_kprobes_operations);
2446 if (!file)
2447 goto error;
2448
2449 file = debugfs_create_file("enabled", 0600, dir,
2450 &value, &fops_kp);
2451 if (!file)
2452 goto error;
2453
2454 file = debugfs_create_file("blacklist", 0444, dir, NULL,
2455 &debugfs_kprobe_blacklist_ops);
2456 if (!file)
2457 goto error;
2458
2459 return 0;
2460
2461error:
2462 debugfs_remove(dir);
2463 return -ENOMEM;
2464}
2465
2466late_initcall(debugfs_kprobe_init);
2467#endif /* CONFIG_DEBUG_FS */
2468
2469module_init(init_kprobes);
2470
2471/* defined in arch/.../kernel/kprobes.c */
2472EXPORT_SYMBOL_GPL(jprobe_return);