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