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