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_rcu();
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 list_del_init(&op->list);
550 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
551 /*
552 * This must not happen, but if there is a kprobe
553 * still in use, keep it on kprobes hash list.
554 */
555 continue;
556 }
557 free_aggr_kprobe(&op->kp);
558 }
559}
560
561/* Start optimizer after OPTIMIZE_DELAY passed */
562static void kick_kprobe_optimizer(void)
563{
564 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
565}
566
567/* Kprobe jump optimizer */
568static void kprobe_optimizer(struct work_struct *work)
569{
570 mutex_lock(&kprobe_mutex);
571 cpus_read_lock();
572 /* Lock modules while optimizing kprobes */
573 mutex_lock(&module_mutex);
574
575 /*
576 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
577 * kprobes before waiting for quiesence period.
578 */
579 do_unoptimize_kprobes();
580
581 /*
582 * Step 2: Wait for quiesence period to ensure all potentially
583 * preempted tasks to have normally scheduled. Because optprobe
584 * may modify multiple instructions, there is a chance that Nth
585 * instruction is preempted. In that case, such tasks can return
586 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
587 * Note that on non-preemptive kernel, this is transparently converted
588 * to synchronoze_sched() to wait for all interrupts to have completed.
589 */
590 synchronize_rcu_tasks();
591
592 /* Step 3: Optimize kprobes after quiesence period */
593 do_optimize_kprobes();
594
595 /* Step 4: Free cleaned kprobes after quiesence period */
596 do_free_cleaned_kprobes();
597
598 mutex_unlock(&module_mutex);
599 cpus_read_unlock();
600 mutex_unlock(&kprobe_mutex);
601
602 /* Step 5: Kick optimizer again if needed */
603 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
604 kick_kprobe_optimizer();
605}
606
607/* Wait for completing optimization and unoptimization */
608void wait_for_kprobe_optimizer(void)
609{
610 mutex_lock(&kprobe_mutex);
611
612 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
613 mutex_unlock(&kprobe_mutex);
614
615 /* this will also make optimizing_work execute immmediately */
616 flush_delayed_work(&optimizing_work);
617 /* @optimizing_work might not have been queued yet, relax */
618 cpu_relax();
619
620 mutex_lock(&kprobe_mutex);
621 }
622
623 mutex_unlock(&kprobe_mutex);
624}
625
626/* Optimize kprobe if p is ready to be optimized */
627static void optimize_kprobe(struct kprobe *p)
628{
629 struct optimized_kprobe *op;
630
631 /* Check if the kprobe is disabled or not ready for optimization. */
632 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
633 (kprobe_disabled(p) || kprobes_all_disarmed))
634 return;
635
636 /* kprobes with post_handler can not be optimized */
637 if (p->post_handler)
638 return;
639
640 op = container_of(p, struct optimized_kprobe, kp);
641
642 /* Check there is no other kprobes at the optimized instructions */
643 if (arch_check_optimized_kprobe(op) < 0)
644 return;
645
646 /* Check if it is already optimized. */
647 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
648 return;
649 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
650
651 if (!list_empty(&op->list))
652 /* This is under unoptimizing. Just dequeue the probe */
653 list_del_init(&op->list);
654 else {
655 list_add(&op->list, &optimizing_list);
656 kick_kprobe_optimizer();
657 }
658}
659
660/* Short cut to direct unoptimizing */
661static void force_unoptimize_kprobe(struct optimized_kprobe *op)
662{
663 lockdep_assert_cpus_held();
664 arch_unoptimize_kprobe(op);
665 if (kprobe_disabled(&op->kp))
666 arch_disarm_kprobe(&op->kp);
667}
668
669/* Unoptimize a kprobe if p is optimized */
670static void unoptimize_kprobe(struct kprobe *p, bool force)
671{
672 struct optimized_kprobe *op;
673
674 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
675 return; /* This is not an optprobe nor optimized */
676
677 op = container_of(p, struct optimized_kprobe, kp);
678 if (!kprobe_optimized(p)) {
679 /* Unoptimized or unoptimizing case */
680 if (force && !list_empty(&op->list)) {
681 /*
682 * Only if this is unoptimizing kprobe and forced,
683 * forcibly unoptimize it. (No need to unoptimize
684 * unoptimized kprobe again :)
685 */
686 list_del_init(&op->list);
687 force_unoptimize_kprobe(op);
688 }
689 return;
690 }
691
692 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
693 if (!list_empty(&op->list)) {
694 /* Dequeue from the optimization queue */
695 list_del_init(&op->list);
696 return;
697 }
698 /* Optimized kprobe case */
699 if (force)
700 /* Forcibly update the code: this is a special case */
701 force_unoptimize_kprobe(op);
702 else {
703 list_add(&op->list, &unoptimizing_list);
704 kick_kprobe_optimizer();
705 }
706}
707
708/* Cancel unoptimizing for reusing */
709static int reuse_unused_kprobe(struct kprobe *ap)
710{
711 struct optimized_kprobe *op;
712 int ret;
713
714 /*
715 * Unused kprobe MUST be on the way of delayed unoptimizing (means
716 * there is still a relative jump) and disabled.
717 */
718 op = container_of(ap, struct optimized_kprobe, kp);
719 WARN_ON_ONCE(list_empty(&op->list));
720 /* Enable the probe again */
721 ap->flags &= ~KPROBE_FLAG_DISABLED;
722 /* Optimize it again (remove from op->list) */
723 ret = kprobe_optready(ap);
724 if (ret)
725 return ret;
726
727 optimize_kprobe(ap);
728 return 0;
729}
730
731/* Remove optimized instructions */
732static void kill_optimized_kprobe(struct kprobe *p)
733{
734 struct optimized_kprobe *op;
735
736 op = container_of(p, struct optimized_kprobe, kp);
737 if (!list_empty(&op->list))
738 /* Dequeue from the (un)optimization queue */
739 list_del_init(&op->list);
740 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
741
742 if (kprobe_unused(p)) {
743 /* Enqueue if it is unused */
744 list_add(&op->list, &freeing_list);
745 /*
746 * Remove unused probes from the hash list. After waiting
747 * for synchronization, this probe is reclaimed.
748 * (reclaiming is done by do_free_cleaned_kprobes().)
749 */
750 hlist_del_rcu(&op->kp.hlist);
751 }
752
753 /* Don't touch the code, because it is already freed. */
754 arch_remove_optimized_kprobe(op);
755}
756
757static inline
758void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
759{
760 if (!kprobe_ftrace(p))
761 arch_prepare_optimized_kprobe(op, p);
762}
763
764/* Try to prepare optimized instructions */
765static void prepare_optimized_kprobe(struct kprobe *p)
766{
767 struct optimized_kprobe *op;
768
769 op = container_of(p, struct optimized_kprobe, kp);
770 __prepare_optimized_kprobe(op, p);
771}
772
773/* Allocate new optimized_kprobe and try to prepare optimized instructions */
774static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
775{
776 struct optimized_kprobe *op;
777
778 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
779 if (!op)
780 return NULL;
781
782 INIT_LIST_HEAD(&op->list);
783 op->kp.addr = p->addr;
784 __prepare_optimized_kprobe(op, p);
785
786 return &op->kp;
787}
788
789static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
790
791/*
792 * Prepare an optimized_kprobe and optimize it
793 * NOTE: p must be a normal registered kprobe
794 */
795static void try_to_optimize_kprobe(struct kprobe *p)
796{
797 struct kprobe *ap;
798 struct optimized_kprobe *op;
799
800 /* Impossible to optimize ftrace-based kprobe */
801 if (kprobe_ftrace(p))
802 return;
803
804 /* For preparing optimization, jump_label_text_reserved() is called */
805 cpus_read_lock();
806 jump_label_lock();
807 mutex_lock(&text_mutex);
808
809 ap = alloc_aggr_kprobe(p);
810 if (!ap)
811 goto out;
812
813 op = container_of(ap, struct optimized_kprobe, kp);
814 if (!arch_prepared_optinsn(&op->optinsn)) {
815 /* If failed to setup optimizing, fallback to kprobe */
816 arch_remove_optimized_kprobe(op);
817 kfree(op);
818 goto out;
819 }
820
821 init_aggr_kprobe(ap, p);
822 optimize_kprobe(ap); /* This just kicks optimizer thread */
823
824out:
825 mutex_unlock(&text_mutex);
826 jump_label_unlock();
827 cpus_read_unlock();
828}
829
830#ifdef CONFIG_SYSCTL
831static void optimize_all_kprobes(void)
832{
833 struct hlist_head *head;
834 struct kprobe *p;
835 unsigned int i;
836
837 mutex_lock(&kprobe_mutex);
838 /* If optimization is already allowed, just return */
839 if (kprobes_allow_optimization)
840 goto out;
841
842 cpus_read_lock();
843 kprobes_allow_optimization = true;
844 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
845 head = &kprobe_table[i];
846 hlist_for_each_entry_rcu(p, head, hlist)
847 if (!kprobe_disabled(p))
848 optimize_kprobe(p);
849 }
850 cpus_read_unlock();
851 printk(KERN_INFO "Kprobes globally optimized\n");
852out:
853 mutex_unlock(&kprobe_mutex);
854}
855
856static void unoptimize_all_kprobes(void)
857{
858 struct hlist_head *head;
859 struct kprobe *p;
860 unsigned int i;
861
862 mutex_lock(&kprobe_mutex);
863 /* If optimization is already prohibited, just return */
864 if (!kprobes_allow_optimization) {
865 mutex_unlock(&kprobe_mutex);
866 return;
867 }
868
869 cpus_read_lock();
870 kprobes_allow_optimization = false;
871 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
872 head = &kprobe_table[i];
873 hlist_for_each_entry_rcu(p, head, hlist) {
874 if (!kprobe_disabled(p))
875 unoptimize_kprobe(p, false);
876 }
877 }
878 cpus_read_unlock();
879 mutex_unlock(&kprobe_mutex);
880
881 /* Wait for unoptimizing completion */
882 wait_for_kprobe_optimizer();
883 printk(KERN_INFO "Kprobes globally unoptimized\n");
884}
885
886static DEFINE_MUTEX(kprobe_sysctl_mutex);
887int sysctl_kprobes_optimization;
888int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
889 void __user *buffer, size_t *length,
890 loff_t *ppos)
891{
892 int ret;
893
894 mutex_lock(&kprobe_sysctl_mutex);
895 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
896 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
897
898 if (sysctl_kprobes_optimization)
899 optimize_all_kprobes();
900 else
901 unoptimize_all_kprobes();
902 mutex_unlock(&kprobe_sysctl_mutex);
903
904 return ret;
905}
906#endif /* CONFIG_SYSCTL */
907
908/* Put a breakpoint for a probe. Must be called with text_mutex locked */
909static void __arm_kprobe(struct kprobe *p)
910{
911 struct kprobe *_p;
912
913 /* Check collision with other optimized kprobes */
914 _p = get_optimized_kprobe((unsigned long)p->addr);
915 if (unlikely(_p))
916 /* Fallback to unoptimized kprobe */
917 unoptimize_kprobe(_p, true);
918
919 arch_arm_kprobe(p);
920 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
921}
922
923/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
924static void __disarm_kprobe(struct kprobe *p, bool reopt)
925{
926 struct kprobe *_p;
927
928 /* Try to unoptimize */
929 unoptimize_kprobe(p, kprobes_all_disarmed);
930
931 if (!kprobe_queued(p)) {
932 arch_disarm_kprobe(p);
933 /* If another kprobe was blocked, optimize it. */
934 _p = get_optimized_kprobe((unsigned long)p->addr);
935 if (unlikely(_p) && reopt)
936 optimize_kprobe(_p);
937 }
938 /* TODO: reoptimize others after unoptimized this probe */
939}
940
941#else /* !CONFIG_OPTPROBES */
942
943#define optimize_kprobe(p) do {} while (0)
944#define unoptimize_kprobe(p, f) do {} while (0)
945#define kill_optimized_kprobe(p) do {} while (0)
946#define prepare_optimized_kprobe(p) do {} while (0)
947#define try_to_optimize_kprobe(p) do {} while (0)
948#define __arm_kprobe(p) arch_arm_kprobe(p)
949#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
950#define kprobe_disarmed(p) kprobe_disabled(p)
951#define wait_for_kprobe_optimizer() do {} while (0)
952
953static int reuse_unused_kprobe(struct kprobe *ap)
954{
955 /*
956 * If the optimized kprobe is NOT supported, the aggr kprobe is
957 * released at the same time that the last aggregated kprobe is
958 * unregistered.
959 * Thus there should be no chance to reuse unused kprobe.
960 */
961 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
962 return -EINVAL;
963}
964
965static void free_aggr_kprobe(struct kprobe *p)
966{
967 arch_remove_kprobe(p);
968 kfree(p);
969}
970
971static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
972{
973 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
974}
975#endif /* CONFIG_OPTPROBES */
976
977#ifdef CONFIG_KPROBES_ON_FTRACE
978static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
979 .func = kprobe_ftrace_handler,
980 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
981};
982static int kprobe_ftrace_enabled;
983
984/* Must ensure p->addr is really on ftrace */
985static int prepare_kprobe(struct kprobe *p)
986{
987 if (!kprobe_ftrace(p))
988 return arch_prepare_kprobe(p);
989
990 return arch_prepare_kprobe_ftrace(p);
991}
992
993/* Caller must lock kprobe_mutex */
994static int arm_kprobe_ftrace(struct kprobe *p)
995{
996 int ret = 0;
997
998 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
999 (unsigned long)p->addr, 0, 0);
1000 if (ret) {
1001 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1002 p->addr, ret);
1003 return ret;
1004 }
1005
1006 if (kprobe_ftrace_enabled == 0) {
1007 ret = register_ftrace_function(&kprobe_ftrace_ops);
1008 if (ret) {
1009 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1010 goto err_ftrace;
1011 }
1012 }
1013
1014 kprobe_ftrace_enabled++;
1015 return ret;
1016
1017err_ftrace:
1018 /*
1019 * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a
1020 * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental
1021 * empty filter_hash which would undesirably trace all functions.
1022 */
1023 ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0);
1024 return ret;
1025}
1026
1027/* Caller must lock kprobe_mutex */
1028static int disarm_kprobe_ftrace(struct kprobe *p)
1029{
1030 int ret = 0;
1031
1032 if (kprobe_ftrace_enabled == 1) {
1033 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
1034 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1035 return ret;
1036 }
1037
1038 kprobe_ftrace_enabled--;
1039
1040 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
1041 (unsigned long)p->addr, 1, 0);
1042 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1043 p->addr, ret);
1044 return ret;
1045}
1046#else /* !CONFIG_KPROBES_ON_FTRACE */
1047#define prepare_kprobe(p) arch_prepare_kprobe(p)
1048#define arm_kprobe_ftrace(p) (-ENODEV)
1049#define disarm_kprobe_ftrace(p) (-ENODEV)
1050#endif
1051
1052/* Arm a kprobe with text_mutex */
1053static int arm_kprobe(struct kprobe *kp)
1054{
1055 if (unlikely(kprobe_ftrace(kp)))
1056 return arm_kprobe_ftrace(kp);
1057
1058 cpus_read_lock();
1059 mutex_lock(&text_mutex);
1060 __arm_kprobe(kp);
1061 mutex_unlock(&text_mutex);
1062 cpus_read_unlock();
1063
1064 return 0;
1065}
1066
1067/* Disarm a kprobe with text_mutex */
1068static int disarm_kprobe(struct kprobe *kp, bool reopt)
1069{
1070 if (unlikely(kprobe_ftrace(kp)))
1071 return disarm_kprobe_ftrace(kp);
1072
1073 cpus_read_lock();
1074 mutex_lock(&text_mutex);
1075 __disarm_kprobe(kp, reopt);
1076 mutex_unlock(&text_mutex);
1077 cpus_read_unlock();
1078
1079 return 0;
1080}
1081
1082/*
1083 * Aggregate handlers for multiple kprobes support - these handlers
1084 * take care of invoking the individual kprobe handlers on p->list
1085 */
1086static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1087{
1088 struct kprobe *kp;
1089
1090 list_for_each_entry_rcu(kp, &p->list, list) {
1091 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1092 set_kprobe_instance(kp);
1093 if (kp->pre_handler(kp, regs))
1094 return 1;
1095 }
1096 reset_kprobe_instance();
1097 }
1098 return 0;
1099}
1100NOKPROBE_SYMBOL(aggr_pre_handler);
1101
1102static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1103 unsigned long flags)
1104{
1105 struct kprobe *kp;
1106
1107 list_for_each_entry_rcu(kp, &p->list, list) {
1108 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1109 set_kprobe_instance(kp);
1110 kp->post_handler(kp, regs, flags);
1111 reset_kprobe_instance();
1112 }
1113 }
1114}
1115NOKPROBE_SYMBOL(aggr_post_handler);
1116
1117static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1118 int trapnr)
1119{
1120 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1121
1122 /*
1123 * if we faulted "during" the execution of a user specified
1124 * probe handler, invoke just that probe's fault handler
1125 */
1126 if (cur && cur->fault_handler) {
1127 if (cur->fault_handler(cur, regs, trapnr))
1128 return 1;
1129 }
1130 return 0;
1131}
1132NOKPROBE_SYMBOL(aggr_fault_handler);
1133
1134/* Walks the list and increments nmissed count for multiprobe case */
1135void kprobes_inc_nmissed_count(struct kprobe *p)
1136{
1137 struct kprobe *kp;
1138 if (!kprobe_aggrprobe(p)) {
1139 p->nmissed++;
1140 } else {
1141 list_for_each_entry_rcu(kp, &p->list, list)
1142 kp->nmissed++;
1143 }
1144 return;
1145}
1146NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1147
1148void recycle_rp_inst(struct kretprobe_instance *ri,
1149 struct hlist_head *head)
1150{
1151 struct kretprobe *rp = ri->rp;
1152
1153 /* remove rp inst off the rprobe_inst_table */
1154 hlist_del(&ri->hlist);
1155 INIT_HLIST_NODE(&ri->hlist);
1156 if (likely(rp)) {
1157 raw_spin_lock(&rp->lock);
1158 hlist_add_head(&ri->hlist, &rp->free_instances);
1159 raw_spin_unlock(&rp->lock);
1160 } else
1161 /* Unregistering */
1162 hlist_add_head(&ri->hlist, head);
1163}
1164NOKPROBE_SYMBOL(recycle_rp_inst);
1165
1166void kretprobe_hash_lock(struct task_struct *tsk,
1167 struct hlist_head **head, unsigned long *flags)
1168__acquires(hlist_lock)
1169{
1170 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1171 raw_spinlock_t *hlist_lock;
1172
1173 *head = &kretprobe_inst_table[hash];
1174 hlist_lock = kretprobe_table_lock_ptr(hash);
1175 raw_spin_lock_irqsave(hlist_lock, *flags);
1176}
1177NOKPROBE_SYMBOL(kretprobe_hash_lock);
1178
1179static void kretprobe_table_lock(unsigned long hash,
1180 unsigned long *flags)
1181__acquires(hlist_lock)
1182{
1183 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1184 raw_spin_lock_irqsave(hlist_lock, *flags);
1185}
1186NOKPROBE_SYMBOL(kretprobe_table_lock);
1187
1188void kretprobe_hash_unlock(struct task_struct *tsk,
1189 unsigned long *flags)
1190__releases(hlist_lock)
1191{
1192 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1193 raw_spinlock_t *hlist_lock;
1194
1195 hlist_lock = kretprobe_table_lock_ptr(hash);
1196 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1197}
1198NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1199
1200static void kretprobe_table_unlock(unsigned long hash,
1201 unsigned long *flags)
1202__releases(hlist_lock)
1203{
1204 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1205 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1206}
1207NOKPROBE_SYMBOL(kretprobe_table_unlock);
1208
1209/*
1210 * This function is called from finish_task_switch when task tk becomes dead,
1211 * so that we can recycle any function-return probe instances associated
1212 * with this task. These left over instances represent probed functions
1213 * that have been called but will never return.
1214 */
1215void kprobe_flush_task(struct task_struct *tk)
1216{
1217 struct kretprobe_instance *ri;
1218 struct hlist_head *head, empty_rp;
1219 struct hlist_node *tmp;
1220 unsigned long hash, flags = 0;
1221
1222 if (unlikely(!kprobes_initialized))
1223 /* Early boot. kretprobe_table_locks not yet initialized. */
1224 return;
1225
1226 INIT_HLIST_HEAD(&empty_rp);
1227 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1228 head = &kretprobe_inst_table[hash];
1229 kretprobe_table_lock(hash, &flags);
1230 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1231 if (ri->task == tk)
1232 recycle_rp_inst(ri, &empty_rp);
1233 }
1234 kretprobe_table_unlock(hash, &flags);
1235 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1236 hlist_del(&ri->hlist);
1237 kfree(ri);
1238 }
1239}
1240NOKPROBE_SYMBOL(kprobe_flush_task);
1241
1242static inline void free_rp_inst(struct kretprobe *rp)
1243{
1244 struct kretprobe_instance *ri;
1245 struct hlist_node *next;
1246
1247 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1248 hlist_del(&ri->hlist);
1249 kfree(ri);
1250 }
1251}
1252
1253static void cleanup_rp_inst(struct kretprobe *rp)
1254{
1255 unsigned long flags, hash;
1256 struct kretprobe_instance *ri;
1257 struct hlist_node *next;
1258 struct hlist_head *head;
1259
1260 /* No race here */
1261 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1262 kretprobe_table_lock(hash, &flags);
1263 head = &kretprobe_inst_table[hash];
1264 hlist_for_each_entry_safe(ri, next, head, hlist) {
1265 if (ri->rp == rp)
1266 ri->rp = NULL;
1267 }
1268 kretprobe_table_unlock(hash, &flags);
1269 }
1270 free_rp_inst(rp);
1271}
1272NOKPROBE_SYMBOL(cleanup_rp_inst);
1273
1274/* Add the new probe to ap->list */
1275static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1276{
1277 if (p->post_handler)
1278 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1279
1280 list_add_rcu(&p->list, &ap->list);
1281 if (p->post_handler && !ap->post_handler)
1282 ap->post_handler = aggr_post_handler;
1283
1284 return 0;
1285}
1286
1287/*
1288 * Fill in the required fields of the "manager kprobe". Replace the
1289 * earlier kprobe in the hlist with the manager kprobe
1290 */
1291static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1292{
1293 /* Copy p's insn slot to ap */
1294 copy_kprobe(p, ap);
1295 flush_insn_slot(ap);
1296 ap->addr = p->addr;
1297 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1298 ap->pre_handler = aggr_pre_handler;
1299 ap->fault_handler = aggr_fault_handler;
1300 /* We don't care the kprobe which has gone. */
1301 if (p->post_handler && !kprobe_gone(p))
1302 ap->post_handler = aggr_post_handler;
1303
1304 INIT_LIST_HEAD(&ap->list);
1305 INIT_HLIST_NODE(&ap->hlist);
1306
1307 list_add_rcu(&p->list, &ap->list);
1308 hlist_replace_rcu(&p->hlist, &ap->hlist);
1309}
1310
1311/*
1312 * This is the second or subsequent kprobe at the address - handle
1313 * the intricacies
1314 */
1315static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1316{
1317 int ret = 0;
1318 struct kprobe *ap = orig_p;
1319
1320 cpus_read_lock();
1321
1322 /* For preparing optimization, jump_label_text_reserved() is called */
1323 jump_label_lock();
1324 mutex_lock(&text_mutex);
1325
1326 if (!kprobe_aggrprobe(orig_p)) {
1327 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1328 ap = alloc_aggr_kprobe(orig_p);
1329 if (!ap) {
1330 ret = -ENOMEM;
1331 goto out;
1332 }
1333 init_aggr_kprobe(ap, orig_p);
1334 } else if (kprobe_unused(ap)) {
1335 /* This probe is going to die. Rescue it */
1336 ret = reuse_unused_kprobe(ap);
1337 if (ret)
1338 goto out;
1339 }
1340
1341 if (kprobe_gone(ap)) {
1342 /*
1343 * Attempting to insert new probe at the same location that
1344 * had a probe in the module vaddr area which already
1345 * freed. So, the instruction slot has already been
1346 * released. We need a new slot for the new probe.
1347 */
1348 ret = arch_prepare_kprobe(ap);
1349 if (ret)
1350 /*
1351 * Even if fail to allocate new slot, don't need to
1352 * free aggr_probe. It will be used next time, or
1353 * freed by unregister_kprobe.
1354 */
1355 goto out;
1356
1357 /* Prepare optimized instructions if possible. */
1358 prepare_optimized_kprobe(ap);
1359
1360 /*
1361 * Clear gone flag to prevent allocating new slot again, and
1362 * set disabled flag because it is not armed yet.
1363 */
1364 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1365 | KPROBE_FLAG_DISABLED;
1366 }
1367
1368 /* Copy ap's insn slot to p */
1369 copy_kprobe(ap, p);
1370 ret = add_new_kprobe(ap, p);
1371
1372out:
1373 mutex_unlock(&text_mutex);
1374 jump_label_unlock();
1375 cpus_read_unlock();
1376
1377 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1378 ap->flags &= ~KPROBE_FLAG_DISABLED;
1379 if (!kprobes_all_disarmed) {
1380 /* Arm the breakpoint again. */
1381 ret = arm_kprobe(ap);
1382 if (ret) {
1383 ap->flags |= KPROBE_FLAG_DISABLED;
1384 list_del_rcu(&p->list);
1385 synchronize_rcu();
1386 }
1387 }
1388 }
1389 return ret;
1390}
1391
1392bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1393{
1394 /* The __kprobes marked functions and entry code must not be probed */
1395 return addr >= (unsigned long)__kprobes_text_start &&
1396 addr < (unsigned long)__kprobes_text_end;
1397}
1398
1399static bool __within_kprobe_blacklist(unsigned long addr)
1400{
1401 struct kprobe_blacklist_entry *ent;
1402
1403 if (arch_within_kprobe_blacklist(addr))
1404 return true;
1405 /*
1406 * If there exists a kprobe_blacklist, verify and
1407 * fail any probe registration in the prohibited area
1408 */
1409 list_for_each_entry(ent, &kprobe_blacklist, list) {
1410 if (addr >= ent->start_addr && addr < ent->end_addr)
1411 return true;
1412 }
1413 return false;
1414}
1415
1416bool within_kprobe_blacklist(unsigned long addr)
1417{
1418 char symname[KSYM_NAME_LEN], *p;
1419
1420 if (__within_kprobe_blacklist(addr))
1421 return true;
1422
1423 /* Check if the address is on a suffixed-symbol */
1424 if (!lookup_symbol_name(addr, symname)) {
1425 p = strchr(symname, '.');
1426 if (!p)
1427 return false;
1428 *p = '\0';
1429 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1430 if (addr)
1431 return __within_kprobe_blacklist(addr);
1432 }
1433 return false;
1434}
1435
1436/*
1437 * If we have a symbol_name argument, look it up and add the offset field
1438 * to it. This way, we can specify a relative address to a symbol.
1439 * This returns encoded errors if it fails to look up symbol or invalid
1440 * combination of parameters.
1441 */
1442static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1443 const char *symbol_name, unsigned int offset)
1444{
1445 if ((symbol_name && addr) || (!symbol_name && !addr))
1446 goto invalid;
1447
1448 if (symbol_name) {
1449 addr = kprobe_lookup_name(symbol_name, offset);
1450 if (!addr)
1451 return ERR_PTR(-ENOENT);
1452 }
1453
1454 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1455 if (addr)
1456 return addr;
1457
1458invalid:
1459 return ERR_PTR(-EINVAL);
1460}
1461
1462static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1463{
1464 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1465}
1466
1467/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1468static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1469{
1470 struct kprobe *ap, *list_p;
1471
1472 ap = get_kprobe(p->addr);
1473 if (unlikely(!ap))
1474 return NULL;
1475
1476 if (p != ap) {
1477 list_for_each_entry_rcu(list_p, &ap->list, list)
1478 if (list_p == p)
1479 /* kprobe p is a valid probe */
1480 goto valid;
1481 return NULL;
1482 }
1483valid:
1484 return ap;
1485}
1486
1487/* Return error if the kprobe is being re-registered */
1488static inline int check_kprobe_rereg(struct kprobe *p)
1489{
1490 int ret = 0;
1491
1492 mutex_lock(&kprobe_mutex);
1493 if (__get_valid_kprobe(p))
1494 ret = -EINVAL;
1495 mutex_unlock(&kprobe_mutex);
1496
1497 return ret;
1498}
1499
1500int __weak arch_check_ftrace_location(struct kprobe *p)
1501{
1502 unsigned long ftrace_addr;
1503
1504 ftrace_addr = ftrace_location((unsigned long)p->addr);
1505 if (ftrace_addr) {
1506#ifdef CONFIG_KPROBES_ON_FTRACE
1507 /* Given address is not on the instruction boundary */
1508 if ((unsigned long)p->addr != ftrace_addr)
1509 return -EILSEQ;
1510 p->flags |= KPROBE_FLAG_FTRACE;
1511#else /* !CONFIG_KPROBES_ON_FTRACE */
1512 return -EINVAL;
1513#endif
1514 }
1515 return 0;
1516}
1517
1518static int check_kprobe_address_safe(struct kprobe *p,
1519 struct module **probed_mod)
1520{
1521 int ret;
1522
1523 ret = arch_check_ftrace_location(p);
1524 if (ret)
1525 return ret;
1526 jump_label_lock();
1527 preempt_disable();
1528
1529 /* Ensure it is not in reserved area nor out of text */
1530 if (!kernel_text_address((unsigned long) p->addr) ||
1531 within_kprobe_blacklist((unsigned long) p->addr) ||
1532 jump_label_text_reserved(p->addr, p->addr)) {
1533 ret = -EINVAL;
1534 goto out;
1535 }
1536
1537 /* Check if are we probing a module */
1538 *probed_mod = __module_text_address((unsigned long) p->addr);
1539 if (*probed_mod) {
1540 /*
1541 * We must hold a refcount of the probed module while updating
1542 * its code to prohibit unexpected unloading.
1543 */
1544 if (unlikely(!try_module_get(*probed_mod))) {
1545 ret = -ENOENT;
1546 goto out;
1547 }
1548
1549 /*
1550 * If the module freed .init.text, we couldn't insert
1551 * kprobes in there.
1552 */
1553 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1554 (*probed_mod)->state != MODULE_STATE_COMING) {
1555 module_put(*probed_mod);
1556 *probed_mod = NULL;
1557 ret = -ENOENT;
1558 }
1559 }
1560out:
1561 preempt_enable();
1562 jump_label_unlock();
1563
1564 return ret;
1565}
1566
1567int register_kprobe(struct kprobe *p)
1568{
1569 int ret;
1570 struct kprobe *old_p;
1571 struct module *probed_mod;
1572 kprobe_opcode_t *addr;
1573
1574 /* Adjust probe address from symbol */
1575 addr = kprobe_addr(p);
1576 if (IS_ERR(addr))
1577 return PTR_ERR(addr);
1578 p->addr = addr;
1579
1580 ret = check_kprobe_rereg(p);
1581 if (ret)
1582 return ret;
1583
1584 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1585 p->flags &= KPROBE_FLAG_DISABLED;
1586 p->nmissed = 0;
1587 INIT_LIST_HEAD(&p->list);
1588
1589 ret = check_kprobe_address_safe(p, &probed_mod);
1590 if (ret)
1591 return ret;
1592
1593 mutex_lock(&kprobe_mutex);
1594
1595 old_p = get_kprobe(p->addr);
1596 if (old_p) {
1597 /* Since this may unoptimize old_p, locking text_mutex. */
1598 ret = register_aggr_kprobe(old_p, p);
1599 goto out;
1600 }
1601
1602 cpus_read_lock();
1603 /* Prevent text modification */
1604 mutex_lock(&text_mutex);
1605 ret = prepare_kprobe(p);
1606 mutex_unlock(&text_mutex);
1607 cpus_read_unlock();
1608 if (ret)
1609 goto out;
1610
1611 INIT_HLIST_NODE(&p->hlist);
1612 hlist_add_head_rcu(&p->hlist,
1613 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1614
1615 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1616 ret = arm_kprobe(p);
1617 if (ret) {
1618 hlist_del_rcu(&p->hlist);
1619 synchronize_rcu();
1620 goto out;
1621 }
1622 }
1623
1624 /* Try to optimize kprobe */
1625 try_to_optimize_kprobe(p);
1626out:
1627 mutex_unlock(&kprobe_mutex);
1628
1629 if (probed_mod)
1630 module_put(probed_mod);
1631
1632 return ret;
1633}
1634EXPORT_SYMBOL_GPL(register_kprobe);
1635
1636/* Check if all probes on the aggrprobe are disabled */
1637static int aggr_kprobe_disabled(struct kprobe *ap)
1638{
1639 struct kprobe *kp;
1640
1641 list_for_each_entry_rcu(kp, &ap->list, list)
1642 if (!kprobe_disabled(kp))
1643 /*
1644 * There is an active probe on the list.
1645 * We can't disable this ap.
1646 */
1647 return 0;
1648
1649 return 1;
1650}
1651
1652/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1653static struct kprobe *__disable_kprobe(struct kprobe *p)
1654{
1655 struct kprobe *orig_p;
1656 int ret;
1657
1658 /* Get an original kprobe for return */
1659 orig_p = __get_valid_kprobe(p);
1660 if (unlikely(orig_p == NULL))
1661 return ERR_PTR(-EINVAL);
1662
1663 if (!kprobe_disabled(p)) {
1664 /* Disable probe if it is a child probe */
1665 if (p != orig_p)
1666 p->flags |= KPROBE_FLAG_DISABLED;
1667
1668 /* Try to disarm and disable this/parent probe */
1669 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1670 /*
1671 * If kprobes_all_disarmed is set, orig_p
1672 * should have already been disarmed, so
1673 * skip unneed disarming process.
1674 */
1675 if (!kprobes_all_disarmed) {
1676 ret = disarm_kprobe(orig_p, true);
1677 if (ret) {
1678 p->flags &= ~KPROBE_FLAG_DISABLED;
1679 return ERR_PTR(ret);
1680 }
1681 }
1682 orig_p->flags |= KPROBE_FLAG_DISABLED;
1683 }
1684 }
1685
1686 return orig_p;
1687}
1688
1689/*
1690 * Unregister a kprobe without a scheduler synchronization.
1691 */
1692static int __unregister_kprobe_top(struct kprobe *p)
1693{
1694 struct kprobe *ap, *list_p;
1695
1696 /* Disable kprobe. This will disarm it if needed. */
1697 ap = __disable_kprobe(p);
1698 if (IS_ERR(ap))
1699 return PTR_ERR(ap);
1700
1701 if (ap == p)
1702 /*
1703 * This probe is an independent(and non-optimized) kprobe
1704 * (not an aggrprobe). Remove from the hash list.
1705 */
1706 goto disarmed;
1707
1708 /* Following process expects this probe is an aggrprobe */
1709 WARN_ON(!kprobe_aggrprobe(ap));
1710
1711 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1712 /*
1713 * !disarmed could be happen if the probe is under delayed
1714 * unoptimizing.
1715 */
1716 goto disarmed;
1717 else {
1718 /* If disabling probe has special handlers, update aggrprobe */
1719 if (p->post_handler && !kprobe_gone(p)) {
1720 list_for_each_entry_rcu(list_p, &ap->list, list) {
1721 if ((list_p != p) && (list_p->post_handler))
1722 goto noclean;
1723 }
1724 ap->post_handler = NULL;
1725 }
1726noclean:
1727 /*
1728 * Remove from the aggrprobe: this path will do nothing in
1729 * __unregister_kprobe_bottom().
1730 */
1731 list_del_rcu(&p->list);
1732 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1733 /*
1734 * Try to optimize this probe again, because post
1735 * handler may have been changed.
1736 */
1737 optimize_kprobe(ap);
1738 }
1739 return 0;
1740
1741disarmed:
1742 hlist_del_rcu(&ap->hlist);
1743 return 0;
1744}
1745
1746static void __unregister_kprobe_bottom(struct kprobe *p)
1747{
1748 struct kprobe *ap;
1749
1750 if (list_empty(&p->list))
1751 /* This is an independent kprobe */
1752 arch_remove_kprobe(p);
1753 else if (list_is_singular(&p->list)) {
1754 /* This is the last child of an aggrprobe */
1755 ap = list_entry(p->list.next, struct kprobe, list);
1756 list_del(&p->list);
1757 free_aggr_kprobe(ap);
1758 }
1759 /* Otherwise, do nothing. */
1760}
1761
1762int register_kprobes(struct kprobe **kps, int num)
1763{
1764 int i, ret = 0;
1765
1766 if (num <= 0)
1767 return -EINVAL;
1768 for (i = 0; i < num; i++) {
1769 ret = register_kprobe(kps[i]);
1770 if (ret < 0) {
1771 if (i > 0)
1772 unregister_kprobes(kps, i);
1773 break;
1774 }
1775 }
1776 return ret;
1777}
1778EXPORT_SYMBOL_GPL(register_kprobes);
1779
1780void unregister_kprobe(struct kprobe *p)
1781{
1782 unregister_kprobes(&p, 1);
1783}
1784EXPORT_SYMBOL_GPL(unregister_kprobe);
1785
1786void unregister_kprobes(struct kprobe **kps, int num)
1787{
1788 int i;
1789
1790 if (num <= 0)
1791 return;
1792 mutex_lock(&kprobe_mutex);
1793 for (i = 0; i < num; i++)
1794 if (__unregister_kprobe_top(kps[i]) < 0)
1795 kps[i]->addr = NULL;
1796 mutex_unlock(&kprobe_mutex);
1797
1798 synchronize_rcu();
1799 for (i = 0; i < num; i++)
1800 if (kps[i]->addr)
1801 __unregister_kprobe_bottom(kps[i]);
1802}
1803EXPORT_SYMBOL_GPL(unregister_kprobes);
1804
1805int __weak kprobe_exceptions_notify(struct notifier_block *self,
1806 unsigned long val, void *data)
1807{
1808 return NOTIFY_DONE;
1809}
1810NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1811
1812static struct notifier_block kprobe_exceptions_nb = {
1813 .notifier_call = kprobe_exceptions_notify,
1814 .priority = 0x7fffffff /* we need to be notified first */
1815};
1816
1817unsigned long __weak arch_deref_entry_point(void *entry)
1818{
1819 return (unsigned long)entry;
1820}
1821
1822#ifdef CONFIG_KRETPROBES
1823/*
1824 * This kprobe pre_handler is registered with every kretprobe. When probe
1825 * hits it will set up the return probe.
1826 */
1827static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1828{
1829 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1830 unsigned long hash, flags = 0;
1831 struct kretprobe_instance *ri;
1832
1833 /*
1834 * To avoid deadlocks, prohibit return probing in NMI contexts,
1835 * just skip the probe and increase the (inexact) 'nmissed'
1836 * statistical counter, so that the user is informed that
1837 * something happened:
1838 */
1839 if (unlikely(in_nmi())) {
1840 rp->nmissed++;
1841 return 0;
1842 }
1843
1844 /* TODO: consider to only swap the RA after the last pre_handler fired */
1845 hash = hash_ptr(current, KPROBE_HASH_BITS);
1846 raw_spin_lock_irqsave(&rp->lock, flags);
1847 if (!hlist_empty(&rp->free_instances)) {
1848 ri = hlist_entry(rp->free_instances.first,
1849 struct kretprobe_instance, hlist);
1850 hlist_del(&ri->hlist);
1851 raw_spin_unlock_irqrestore(&rp->lock, flags);
1852
1853 ri->rp = rp;
1854 ri->task = current;
1855
1856 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1857 raw_spin_lock_irqsave(&rp->lock, flags);
1858 hlist_add_head(&ri->hlist, &rp->free_instances);
1859 raw_spin_unlock_irqrestore(&rp->lock, flags);
1860 return 0;
1861 }
1862
1863 arch_prepare_kretprobe(ri, regs);
1864
1865 /* XXX(hch): why is there no hlist_move_head? */
1866 INIT_HLIST_NODE(&ri->hlist);
1867 kretprobe_table_lock(hash, &flags);
1868 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1869 kretprobe_table_unlock(hash, &flags);
1870 } else {
1871 rp->nmissed++;
1872 raw_spin_unlock_irqrestore(&rp->lock, flags);
1873 }
1874 return 0;
1875}
1876NOKPROBE_SYMBOL(pre_handler_kretprobe);
1877
1878bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1879{
1880 return !offset;
1881}
1882
1883bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1884{
1885 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1886
1887 if (IS_ERR(kp_addr))
1888 return false;
1889
1890 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) ||
1891 !arch_kprobe_on_func_entry(offset))
1892 return false;
1893
1894 return true;
1895}
1896
1897int register_kretprobe(struct kretprobe *rp)
1898{
1899 int ret = 0;
1900 struct kretprobe_instance *inst;
1901 int i;
1902 void *addr;
1903
1904 if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset))
1905 return -EINVAL;
1906
1907 if (kretprobe_blacklist_size) {
1908 addr = kprobe_addr(&rp->kp);
1909 if (IS_ERR(addr))
1910 return PTR_ERR(addr);
1911
1912 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1913 if (kretprobe_blacklist[i].addr == addr)
1914 return -EINVAL;
1915 }
1916 }
1917
1918 rp->kp.pre_handler = pre_handler_kretprobe;
1919 rp->kp.post_handler = NULL;
1920 rp->kp.fault_handler = NULL;
1921
1922 /* Pre-allocate memory for max kretprobe instances */
1923 if (rp->maxactive <= 0) {
1924#ifdef CONFIG_PREEMPT
1925 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1926#else
1927 rp->maxactive = num_possible_cpus();
1928#endif
1929 }
1930 raw_spin_lock_init(&rp->lock);
1931 INIT_HLIST_HEAD(&rp->free_instances);
1932 for (i = 0; i < rp->maxactive; i++) {
1933 inst = kmalloc(sizeof(struct kretprobe_instance) +
1934 rp->data_size, GFP_KERNEL);
1935 if (inst == NULL) {
1936 free_rp_inst(rp);
1937 return -ENOMEM;
1938 }
1939 INIT_HLIST_NODE(&inst->hlist);
1940 hlist_add_head(&inst->hlist, &rp->free_instances);
1941 }
1942
1943 rp->nmissed = 0;
1944 /* Establish function entry probe point */
1945 ret = register_kprobe(&rp->kp);
1946 if (ret != 0)
1947 free_rp_inst(rp);
1948 return ret;
1949}
1950EXPORT_SYMBOL_GPL(register_kretprobe);
1951
1952int register_kretprobes(struct kretprobe **rps, int num)
1953{
1954 int ret = 0, i;
1955
1956 if (num <= 0)
1957 return -EINVAL;
1958 for (i = 0; i < num; i++) {
1959 ret = register_kretprobe(rps[i]);
1960 if (ret < 0) {
1961 if (i > 0)
1962 unregister_kretprobes(rps, i);
1963 break;
1964 }
1965 }
1966 return ret;
1967}
1968EXPORT_SYMBOL_GPL(register_kretprobes);
1969
1970void unregister_kretprobe(struct kretprobe *rp)
1971{
1972 unregister_kretprobes(&rp, 1);
1973}
1974EXPORT_SYMBOL_GPL(unregister_kretprobe);
1975
1976void unregister_kretprobes(struct kretprobe **rps, int num)
1977{
1978 int i;
1979
1980 if (num <= 0)
1981 return;
1982 mutex_lock(&kprobe_mutex);
1983 for (i = 0; i < num; i++)
1984 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1985 rps[i]->kp.addr = NULL;
1986 mutex_unlock(&kprobe_mutex);
1987
1988 synchronize_rcu();
1989 for (i = 0; i < num; i++) {
1990 if (rps[i]->kp.addr) {
1991 __unregister_kprobe_bottom(&rps[i]->kp);
1992 cleanup_rp_inst(rps[i]);
1993 }
1994 }
1995}
1996EXPORT_SYMBOL_GPL(unregister_kretprobes);
1997
1998#else /* CONFIG_KRETPROBES */
1999int register_kretprobe(struct kretprobe *rp)
2000{
2001 return -ENOSYS;
2002}
2003EXPORT_SYMBOL_GPL(register_kretprobe);
2004
2005int register_kretprobes(struct kretprobe **rps, int num)
2006{
2007 return -ENOSYS;
2008}
2009EXPORT_SYMBOL_GPL(register_kretprobes);
2010
2011void unregister_kretprobe(struct kretprobe *rp)
2012{
2013}
2014EXPORT_SYMBOL_GPL(unregister_kretprobe);
2015
2016void unregister_kretprobes(struct kretprobe **rps, int num)
2017{
2018}
2019EXPORT_SYMBOL_GPL(unregister_kretprobes);
2020
2021static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2022{
2023 return 0;
2024}
2025NOKPROBE_SYMBOL(pre_handler_kretprobe);
2026
2027#endif /* CONFIG_KRETPROBES */
2028
2029/* Set the kprobe gone and remove its instruction buffer. */
2030static void kill_kprobe(struct kprobe *p)
2031{
2032 struct kprobe *kp;
2033
2034 p->flags |= KPROBE_FLAG_GONE;
2035 if (kprobe_aggrprobe(p)) {
2036 /*
2037 * If this is an aggr_kprobe, we have to list all the
2038 * chained probes and mark them GONE.
2039 */
2040 list_for_each_entry_rcu(kp, &p->list, list)
2041 kp->flags |= KPROBE_FLAG_GONE;
2042 p->post_handler = NULL;
2043 kill_optimized_kprobe(p);
2044 }
2045 /*
2046 * Here, we can remove insn_slot safely, because no thread calls
2047 * the original probed function (which will be freed soon) any more.
2048 */
2049 arch_remove_kprobe(p);
2050}
2051
2052/* Disable one kprobe */
2053int disable_kprobe(struct kprobe *kp)
2054{
2055 int ret = 0;
2056 struct kprobe *p;
2057
2058 mutex_lock(&kprobe_mutex);
2059
2060 /* Disable this kprobe */
2061 p = __disable_kprobe(kp);
2062 if (IS_ERR(p))
2063 ret = PTR_ERR(p);
2064
2065 mutex_unlock(&kprobe_mutex);
2066 return ret;
2067}
2068EXPORT_SYMBOL_GPL(disable_kprobe);
2069
2070/* Enable one kprobe */
2071int enable_kprobe(struct kprobe *kp)
2072{
2073 int ret = 0;
2074 struct kprobe *p;
2075
2076 mutex_lock(&kprobe_mutex);
2077
2078 /* Check whether specified probe is valid. */
2079 p = __get_valid_kprobe(kp);
2080 if (unlikely(p == NULL)) {
2081 ret = -EINVAL;
2082 goto out;
2083 }
2084
2085 if (kprobe_gone(kp)) {
2086 /* This kprobe has gone, we couldn't enable it. */
2087 ret = -EINVAL;
2088 goto out;
2089 }
2090
2091 if (p != kp)
2092 kp->flags &= ~KPROBE_FLAG_DISABLED;
2093
2094 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2095 p->flags &= ~KPROBE_FLAG_DISABLED;
2096 ret = arm_kprobe(p);
2097 if (ret)
2098 p->flags |= KPROBE_FLAG_DISABLED;
2099 }
2100out:
2101 mutex_unlock(&kprobe_mutex);
2102 return ret;
2103}
2104EXPORT_SYMBOL_GPL(enable_kprobe);
2105
2106/* Caller must NOT call this in usual path. This is only for critical case */
2107void dump_kprobe(struct kprobe *kp)
2108{
2109 pr_err("Dumping kprobe:\n");
2110 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2111 kp->symbol_name, kp->offset, kp->addr);
2112}
2113NOKPROBE_SYMBOL(dump_kprobe);
2114
2115int kprobe_add_ksym_blacklist(unsigned long entry)
2116{
2117 struct kprobe_blacklist_entry *ent;
2118 unsigned long offset = 0, size = 0;
2119
2120 if (!kernel_text_address(entry) ||
2121 !kallsyms_lookup_size_offset(entry, &size, &offset))
2122 return -EINVAL;
2123
2124 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2125 if (!ent)
2126 return -ENOMEM;
2127 ent->start_addr = entry;
2128 ent->end_addr = entry + size;
2129 INIT_LIST_HEAD(&ent->list);
2130 list_add_tail(&ent->list, &kprobe_blacklist);
2131
2132 return (int)size;
2133}
2134
2135/* Add all symbols in given area into kprobe blacklist */
2136int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2137{
2138 unsigned long entry;
2139 int ret = 0;
2140
2141 for (entry = start; entry < end; entry += ret) {
2142 ret = kprobe_add_ksym_blacklist(entry);
2143 if (ret < 0)
2144 return ret;
2145 if (ret == 0) /* In case of alias symbol */
2146 ret = 1;
2147 }
2148 return 0;
2149}
2150
2151int __init __weak arch_populate_kprobe_blacklist(void)
2152{
2153 return 0;
2154}
2155
2156/*
2157 * Lookup and populate the kprobe_blacklist.
2158 *
2159 * Unlike the kretprobe blacklist, we'll need to determine
2160 * the range of addresses that belong to the said functions,
2161 * since a kprobe need not necessarily be at the beginning
2162 * of a function.
2163 */
2164static int __init populate_kprobe_blacklist(unsigned long *start,
2165 unsigned long *end)
2166{
2167 unsigned long entry;
2168 unsigned long *iter;
2169 int ret;
2170
2171 for (iter = start; iter < end; iter++) {
2172 entry = arch_deref_entry_point((void *)*iter);
2173 ret = kprobe_add_ksym_blacklist(entry);
2174 if (ret == -EINVAL)
2175 continue;
2176 if (ret < 0)
2177 return ret;
2178 }
2179
2180 /* Symbols in __kprobes_text are blacklisted */
2181 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2182 (unsigned long)__kprobes_text_end);
2183
2184 return ret ? : arch_populate_kprobe_blacklist();
2185}
2186
2187/* Module notifier call back, checking kprobes on the module */
2188static int kprobes_module_callback(struct notifier_block *nb,
2189 unsigned long val, void *data)
2190{
2191 struct module *mod = data;
2192 struct hlist_head *head;
2193 struct kprobe *p;
2194 unsigned int i;
2195 int checkcore = (val == MODULE_STATE_GOING);
2196
2197 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2198 return NOTIFY_DONE;
2199
2200 /*
2201 * When MODULE_STATE_GOING was notified, both of module .text and
2202 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2203 * notified, only .init.text section would be freed. We need to
2204 * disable kprobes which have been inserted in the sections.
2205 */
2206 mutex_lock(&kprobe_mutex);
2207 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2208 head = &kprobe_table[i];
2209 hlist_for_each_entry_rcu(p, head, hlist)
2210 if (within_module_init((unsigned long)p->addr, mod) ||
2211 (checkcore &&
2212 within_module_core((unsigned long)p->addr, mod))) {
2213 /*
2214 * The vaddr this probe is installed will soon
2215 * be vfreed buy not synced to disk. Hence,
2216 * disarming the breakpoint isn't needed.
2217 *
2218 * Note, this will also move any optimized probes
2219 * that are pending to be removed from their
2220 * corresponding lists to the freeing_list and
2221 * will not be touched by the delayed
2222 * kprobe_optimizer work handler.
2223 */
2224 kill_kprobe(p);
2225 }
2226 }
2227 mutex_unlock(&kprobe_mutex);
2228 return NOTIFY_DONE;
2229}
2230
2231static struct notifier_block kprobe_module_nb = {
2232 .notifier_call = kprobes_module_callback,
2233 .priority = 0
2234};
2235
2236/* Markers of _kprobe_blacklist section */
2237extern unsigned long __start_kprobe_blacklist[];
2238extern unsigned long __stop_kprobe_blacklist[];
2239
2240static int __init init_kprobes(void)
2241{
2242 int i, err = 0;
2243
2244 /* FIXME allocate the probe table, currently defined statically */
2245 /* initialize all list heads */
2246 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2247 INIT_HLIST_HEAD(&kprobe_table[i]);
2248 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2249 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2250 }
2251
2252 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2253 __stop_kprobe_blacklist);
2254 if (err) {
2255 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2256 pr_err("Please take care of using kprobes.\n");
2257 }
2258
2259 if (kretprobe_blacklist_size) {
2260 /* lookup the function address from its name */
2261 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2262 kretprobe_blacklist[i].addr =
2263 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2264 if (!kretprobe_blacklist[i].addr)
2265 printk("kretprobe: lookup failed: %s\n",
2266 kretprobe_blacklist[i].name);
2267 }
2268 }
2269
2270#if defined(CONFIG_OPTPROBES)
2271#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2272 /* Init kprobe_optinsn_slots */
2273 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2274#endif
2275 /* By default, kprobes can be optimized */
2276 kprobes_allow_optimization = true;
2277#endif
2278
2279 /* By default, kprobes are armed */
2280 kprobes_all_disarmed = false;
2281
2282 err = arch_init_kprobes();
2283 if (!err)
2284 err = register_die_notifier(&kprobe_exceptions_nb);
2285 if (!err)
2286 err = register_module_notifier(&kprobe_module_nb);
2287
2288 kprobes_initialized = (err == 0);
2289
2290 if (!err)
2291 init_test_probes();
2292 return err;
2293}
2294
2295#ifdef CONFIG_DEBUG_FS
2296static void report_probe(struct seq_file *pi, struct kprobe *p,
2297 const char *sym, int offset, char *modname, struct kprobe *pp)
2298{
2299 char *kprobe_type;
2300 void *addr = p->addr;
2301
2302 if (p->pre_handler == pre_handler_kretprobe)
2303 kprobe_type = "r";
2304 else
2305 kprobe_type = "k";
2306
2307 if (!kallsyms_show_value())
2308 addr = NULL;
2309
2310 if (sym)
2311 seq_printf(pi, "%px %s %s+0x%x %s ",
2312 addr, kprobe_type, sym, offset,
2313 (modname ? modname : " "));
2314 else /* try to use %pS */
2315 seq_printf(pi, "%px %s %pS ",
2316 addr, kprobe_type, p->addr);
2317
2318 if (!pp)
2319 pp = p;
2320 seq_printf(pi, "%s%s%s%s\n",
2321 (kprobe_gone(p) ? "[GONE]" : ""),
2322 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2323 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2324 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2325}
2326
2327static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2328{
2329 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2330}
2331
2332static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2333{
2334 (*pos)++;
2335 if (*pos >= KPROBE_TABLE_SIZE)
2336 return NULL;
2337 return pos;
2338}
2339
2340static void kprobe_seq_stop(struct seq_file *f, void *v)
2341{
2342 /* Nothing to do */
2343}
2344
2345static int show_kprobe_addr(struct seq_file *pi, void *v)
2346{
2347 struct hlist_head *head;
2348 struct kprobe *p, *kp;
2349 const char *sym = NULL;
2350 unsigned int i = *(loff_t *) v;
2351 unsigned long offset = 0;
2352 char *modname, namebuf[KSYM_NAME_LEN];
2353
2354 head = &kprobe_table[i];
2355 preempt_disable();
2356 hlist_for_each_entry_rcu(p, head, hlist) {
2357 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2358 &offset, &modname, namebuf);
2359 if (kprobe_aggrprobe(p)) {
2360 list_for_each_entry_rcu(kp, &p->list, list)
2361 report_probe(pi, kp, sym, offset, modname, p);
2362 } else
2363 report_probe(pi, p, sym, offset, modname, NULL);
2364 }
2365 preempt_enable();
2366 return 0;
2367}
2368
2369static const struct seq_operations kprobes_seq_ops = {
2370 .start = kprobe_seq_start,
2371 .next = kprobe_seq_next,
2372 .stop = kprobe_seq_stop,
2373 .show = show_kprobe_addr
2374};
2375
2376static int kprobes_open(struct inode *inode, struct file *filp)
2377{
2378 return seq_open(filp, &kprobes_seq_ops);
2379}
2380
2381static const struct file_operations debugfs_kprobes_operations = {
2382 .open = kprobes_open,
2383 .read = seq_read,
2384 .llseek = seq_lseek,
2385 .release = seq_release,
2386};
2387
2388/* kprobes/blacklist -- shows which functions can not be probed */
2389static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2390{
2391 return seq_list_start(&kprobe_blacklist, *pos);
2392}
2393
2394static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2395{
2396 return seq_list_next(v, &kprobe_blacklist, pos);
2397}
2398
2399static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2400{
2401 struct kprobe_blacklist_entry *ent =
2402 list_entry(v, struct kprobe_blacklist_entry, list);
2403
2404 /*
2405 * If /proc/kallsyms is not showing kernel address, we won't
2406 * show them here either.
2407 */
2408 if (!kallsyms_show_value())
2409 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2410 (void *)ent->start_addr);
2411 else
2412 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2413 (void *)ent->end_addr, (void *)ent->start_addr);
2414 return 0;
2415}
2416
2417static const struct seq_operations kprobe_blacklist_seq_ops = {
2418 .start = kprobe_blacklist_seq_start,
2419 .next = kprobe_blacklist_seq_next,
2420 .stop = kprobe_seq_stop, /* Reuse void function */
2421 .show = kprobe_blacklist_seq_show,
2422};
2423
2424static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2425{
2426 return seq_open(filp, &kprobe_blacklist_seq_ops);
2427}
2428
2429static const struct file_operations debugfs_kprobe_blacklist_ops = {
2430 .open = kprobe_blacklist_open,
2431 .read = seq_read,
2432 .llseek = seq_lseek,
2433 .release = seq_release,
2434};
2435
2436static int arm_all_kprobes(void)
2437{
2438 struct hlist_head *head;
2439 struct kprobe *p;
2440 unsigned int i, total = 0, errors = 0;
2441 int err, ret = 0;
2442
2443 mutex_lock(&kprobe_mutex);
2444
2445 /* If kprobes are armed, just return */
2446 if (!kprobes_all_disarmed)
2447 goto already_enabled;
2448
2449 /*
2450 * optimize_kprobe() called by arm_kprobe() checks
2451 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2452 * arm_kprobe.
2453 */
2454 kprobes_all_disarmed = false;
2455 /* Arming kprobes doesn't optimize kprobe itself */
2456 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2457 head = &kprobe_table[i];
2458 /* Arm all kprobes on a best-effort basis */
2459 hlist_for_each_entry_rcu(p, head, hlist) {
2460 if (!kprobe_disabled(p)) {
2461 err = arm_kprobe(p);
2462 if (err) {
2463 errors++;
2464 ret = err;
2465 }
2466 total++;
2467 }
2468 }
2469 }
2470
2471 if (errors)
2472 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2473 errors, total);
2474 else
2475 pr_info("Kprobes globally enabled\n");
2476
2477already_enabled:
2478 mutex_unlock(&kprobe_mutex);
2479 return ret;
2480}
2481
2482static int disarm_all_kprobes(void)
2483{
2484 struct hlist_head *head;
2485 struct kprobe *p;
2486 unsigned int i, total = 0, errors = 0;
2487 int err, ret = 0;
2488
2489 mutex_lock(&kprobe_mutex);
2490
2491 /* If kprobes are already disarmed, just return */
2492 if (kprobes_all_disarmed) {
2493 mutex_unlock(&kprobe_mutex);
2494 return 0;
2495 }
2496
2497 kprobes_all_disarmed = true;
2498
2499 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2500 head = &kprobe_table[i];
2501 /* Disarm all kprobes on a best-effort basis */
2502 hlist_for_each_entry_rcu(p, head, hlist) {
2503 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2504 err = disarm_kprobe(p, false);
2505 if (err) {
2506 errors++;
2507 ret = err;
2508 }
2509 total++;
2510 }
2511 }
2512 }
2513
2514 if (errors)
2515 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2516 errors, total);
2517 else
2518 pr_info("Kprobes globally disabled\n");
2519
2520 mutex_unlock(&kprobe_mutex);
2521
2522 /* Wait for disarming all kprobes by optimizer */
2523 wait_for_kprobe_optimizer();
2524
2525 return ret;
2526}
2527
2528/*
2529 * XXX: The debugfs bool file interface doesn't allow for callbacks
2530 * when the bool state is switched. We can reuse that facility when
2531 * available
2532 */
2533static ssize_t read_enabled_file_bool(struct file *file,
2534 char __user *user_buf, size_t count, loff_t *ppos)
2535{
2536 char buf[3];
2537
2538 if (!kprobes_all_disarmed)
2539 buf[0] = '1';
2540 else
2541 buf[0] = '0';
2542 buf[1] = '\n';
2543 buf[2] = 0x00;
2544 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2545}
2546
2547static ssize_t write_enabled_file_bool(struct file *file,
2548 const char __user *user_buf, size_t count, loff_t *ppos)
2549{
2550 char buf[32];
2551 size_t buf_size;
2552 int ret = 0;
2553
2554 buf_size = min(count, (sizeof(buf)-1));
2555 if (copy_from_user(buf, user_buf, buf_size))
2556 return -EFAULT;
2557
2558 buf[buf_size] = '\0';
2559 switch (buf[0]) {
2560 case 'y':
2561 case 'Y':
2562 case '1':
2563 ret = arm_all_kprobes();
2564 break;
2565 case 'n':
2566 case 'N':
2567 case '0':
2568 ret = disarm_all_kprobes();
2569 break;
2570 default:
2571 return -EINVAL;
2572 }
2573
2574 if (ret)
2575 return ret;
2576
2577 return count;
2578}
2579
2580static const struct file_operations fops_kp = {
2581 .read = read_enabled_file_bool,
2582 .write = write_enabled_file_bool,
2583 .llseek = default_llseek,
2584};
2585
2586static int __init debugfs_kprobe_init(void)
2587{
2588 struct dentry *dir, *file;
2589 unsigned int value = 1;
2590
2591 dir = debugfs_create_dir("kprobes", NULL);
2592 if (!dir)
2593 return -ENOMEM;
2594
2595 file = debugfs_create_file("list", 0400, dir, NULL,
2596 &debugfs_kprobes_operations);
2597 if (!file)
2598 goto error;
2599
2600 file = debugfs_create_file("enabled", 0600, dir,
2601 &value, &fops_kp);
2602 if (!file)
2603 goto error;
2604
2605 file = debugfs_create_file("blacklist", 0400, dir, NULL,
2606 &debugfs_kprobe_blacklist_ops);
2607 if (!file)
2608 goto error;
2609
2610 return 0;
2611
2612error:
2613 debugfs_remove(dir);
2614 return -ENOMEM;
2615}
2616
2617late_initcall(debugfs_kprobe_init);
2618#endif /* CONFIG_DEBUG_FS */
2619
2620module_init(init_kprobes);
2621