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