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-generic/sections.h>
53#include <asm/cacheflush.h>
54#include <asm/errno.h>
55#include <asm/uaccess.h>
56
57#define KPROBE_HASH_BITS 6
58#define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
59
60
61/*
62 * Some oddball architectures like 64bit powerpc have function descriptors
63 * so this must be overridable.
64 */
65#ifndef kprobe_lookup_name
66#define kprobe_lookup_name(name, addr) \
67 addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
68#endif
69
70static int kprobes_initialized;
71static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
72static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
73
74/* NOTE: change this value only with kprobe_mutex held */
75static bool kprobes_all_disarmed;
76
77/* This protects kprobe_table and optimizing_list */
78static DEFINE_MUTEX(kprobe_mutex);
79static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
80static struct {
81 raw_spinlock_t lock ____cacheline_aligned_in_smp;
82} kretprobe_table_locks[KPROBE_TABLE_SIZE];
83
84static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
85{
86 return &(kretprobe_table_locks[hash].lock);
87}
88
89/*
90 * Normally, functions that we'd want to prohibit kprobes in, are marked
91 * __kprobes. But, there are cases where such functions already belong to
92 * a different section (__sched for preempt_schedule)
93 *
94 * For such cases, we now have a blacklist
95 */
96static struct kprobe_blackpoint kprobe_blacklist[] = {
97 {"preempt_schedule",},
98 {"native_get_debugreg",},
99 {"irq_entries_start",},
100 {"common_interrupt",},
101 {"mcount",}, /* mcount can be called from everywhere */
102 {NULL} /* Terminator */
103};
104
105#ifdef __ARCH_WANT_KPROBES_INSN_SLOT
106/*
107 * kprobe->ainsn.insn points to the copy of the instruction to be
108 * single-stepped. x86_64, POWER4 and above have no-exec support and
109 * stepping on the instruction on a vmalloced/kmalloced/data page
110 * is a recipe for disaster
111 */
112struct kprobe_insn_page {
113 struct list_head list;
114 kprobe_opcode_t *insns; /* Page of instruction slots */
115 int nused;
116 int ngarbage;
117 char slot_used[];
118};
119
120#define KPROBE_INSN_PAGE_SIZE(slots) \
121 (offsetof(struct kprobe_insn_page, slot_used) + \
122 (sizeof(char) * (slots)))
123
124struct kprobe_insn_cache {
125 struct list_head pages; /* list of kprobe_insn_page */
126 size_t insn_size; /* size of instruction slot */
127 int nr_garbage;
128};
129
130static int slots_per_page(struct kprobe_insn_cache *c)
131{
132 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
133}
134
135enum kprobe_slot_state {
136 SLOT_CLEAN = 0,
137 SLOT_DIRTY = 1,
138 SLOT_USED = 2,
139};
140
141static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_slots */
142static struct kprobe_insn_cache kprobe_insn_slots = {
143 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
144 .insn_size = MAX_INSN_SIZE,
145 .nr_garbage = 0,
146};
147static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c);
148
149/**
150 * __get_insn_slot() - Find a slot on an executable page for an instruction.
151 * We allocate an executable page if there's no room on existing ones.
152 */
153static kprobe_opcode_t __kprobes *__get_insn_slot(struct kprobe_insn_cache *c)
154{
155 struct kprobe_insn_page *kip;
156
157 retry:
158 list_for_each_entry(kip, &c->pages, list) {
159 if (kip->nused < slots_per_page(c)) {
160 int i;
161 for (i = 0; i < slots_per_page(c); i++) {
162 if (kip->slot_used[i] == SLOT_CLEAN) {
163 kip->slot_used[i] = SLOT_USED;
164 kip->nused++;
165 return kip->insns + (i * c->insn_size);
166 }
167 }
168 /* kip->nused is broken. Fix it. */
169 kip->nused = slots_per_page(c);
170 WARN_ON(1);
171 }
172 }
173
174 /* If there are any garbage slots, collect it and try again. */
175 if (c->nr_garbage && collect_garbage_slots(c) == 0)
176 goto retry;
177
178 /* All out of space. Need to allocate a new page. */
179 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
180 if (!kip)
181 return NULL;
182
183 /*
184 * Use module_alloc so this page is within +/- 2GB of where the
185 * kernel image and loaded module images reside. This is required
186 * so x86_64 can correctly handle the %rip-relative fixups.
187 */
188 kip->insns = module_alloc(PAGE_SIZE);
189 if (!kip->insns) {
190 kfree(kip);
191 return NULL;
192 }
193 INIT_LIST_HEAD(&kip->list);
194 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
195 kip->slot_used[0] = SLOT_USED;
196 kip->nused = 1;
197 kip->ngarbage = 0;
198 list_add(&kip->list, &c->pages);
199 return kip->insns;
200}
201
202
203kprobe_opcode_t __kprobes *get_insn_slot(void)
204{
205 kprobe_opcode_t *ret = NULL;
206
207 mutex_lock(&kprobe_insn_mutex);
208 ret = __get_insn_slot(&kprobe_insn_slots);
209 mutex_unlock(&kprobe_insn_mutex);
210
211 return ret;
212}
213
214/* Return 1 if all garbages are collected, otherwise 0. */
215static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
216{
217 kip->slot_used[idx] = SLOT_CLEAN;
218 kip->nused--;
219 if (kip->nused == 0) {
220 /*
221 * Page is no longer in use. Free it unless
222 * it's the last one. We keep the last one
223 * so as not to have to set it up again the
224 * next time somebody inserts a probe.
225 */
226 if (!list_is_singular(&kip->list)) {
227 list_del(&kip->list);
228 module_free(NULL, kip->insns);
229 kfree(kip);
230 }
231 return 1;
232 }
233 return 0;
234}
235
236static int __kprobes collect_garbage_slots(struct kprobe_insn_cache *c)
237{
238 struct kprobe_insn_page *kip, *next;
239
240 /* Ensure no-one is interrupted on the garbages */
241 synchronize_sched();
242
243 list_for_each_entry_safe(kip, next, &c->pages, list) {
244 int i;
245 if (kip->ngarbage == 0)
246 continue;
247 kip->ngarbage = 0; /* we will collect all garbages */
248 for (i = 0; i < slots_per_page(c); i++) {
249 if (kip->slot_used[i] == SLOT_DIRTY &&
250 collect_one_slot(kip, i))
251 break;
252 }
253 }
254 c->nr_garbage = 0;
255 return 0;
256}
257
258static void __kprobes __free_insn_slot(struct kprobe_insn_cache *c,
259 kprobe_opcode_t *slot, int dirty)
260{
261 struct kprobe_insn_page *kip;
262
263 list_for_each_entry(kip, &c->pages, list) {
264 long idx = ((long)slot - (long)kip->insns) /
265 (c->insn_size * sizeof(kprobe_opcode_t));
266 if (idx >= 0 && idx < slots_per_page(c)) {
267 WARN_ON(kip->slot_used[idx] != SLOT_USED);
268 if (dirty) {
269 kip->slot_used[idx] = SLOT_DIRTY;
270 kip->ngarbage++;
271 if (++c->nr_garbage > slots_per_page(c))
272 collect_garbage_slots(c);
273 } else
274 collect_one_slot(kip, idx);
275 return;
276 }
277 }
278 /* Could not free this slot. */
279 WARN_ON(1);
280}
281
282void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
283{
284 mutex_lock(&kprobe_insn_mutex);
285 __free_insn_slot(&kprobe_insn_slots, slot, dirty);
286 mutex_unlock(&kprobe_insn_mutex);
287}
288#ifdef CONFIG_OPTPROBES
289/* For optimized_kprobe buffer */
290static DEFINE_MUTEX(kprobe_optinsn_mutex); /* Protects kprobe_optinsn_slots */
291static struct kprobe_insn_cache kprobe_optinsn_slots = {
292 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
293 /* .insn_size is initialized later */
294 .nr_garbage = 0,
295};
296/* Get a slot for optimized_kprobe buffer */
297kprobe_opcode_t __kprobes *get_optinsn_slot(void)
298{
299 kprobe_opcode_t *ret = NULL;
300
301 mutex_lock(&kprobe_optinsn_mutex);
302 ret = __get_insn_slot(&kprobe_optinsn_slots);
303 mutex_unlock(&kprobe_optinsn_mutex);
304
305 return ret;
306}
307
308void __kprobes free_optinsn_slot(kprobe_opcode_t * slot, int dirty)
309{
310 mutex_lock(&kprobe_optinsn_mutex);
311 __free_insn_slot(&kprobe_optinsn_slots, slot, dirty);
312 mutex_unlock(&kprobe_optinsn_mutex);
313}
314#endif
315#endif
316
317/* We have preemption disabled.. so it is safe to use __ versions */
318static inline void set_kprobe_instance(struct kprobe *kp)
319{
320 __this_cpu_write(kprobe_instance, kp);
321}
322
323static inline void reset_kprobe_instance(void)
324{
325 __this_cpu_write(kprobe_instance, NULL);
326}
327
328/*
329 * This routine is called either:
330 * - under the kprobe_mutex - during kprobe_[un]register()
331 * OR
332 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
333 */
334struct kprobe __kprobes *get_kprobe(void *addr)
335{
336 struct hlist_head *head;
337 struct hlist_node *node;
338 struct kprobe *p;
339
340 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
341 hlist_for_each_entry_rcu(p, node, head, hlist) {
342 if (p->addr == addr)
343 return p;
344 }
345
346 return NULL;
347}
348
349static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
350
351/* Return true if the kprobe is an aggregator */
352static inline int kprobe_aggrprobe(struct kprobe *p)
353{
354 return p->pre_handler == aggr_pre_handler;
355}
356
357/* Return true(!0) if the kprobe is unused */
358static inline int kprobe_unused(struct kprobe *p)
359{
360 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
361 list_empty(&p->list);
362}
363
364/*
365 * Keep all fields in the kprobe consistent
366 */
367static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
368{
369 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
370 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
371}
372
373#ifdef CONFIG_OPTPROBES
374/* NOTE: change this value only with kprobe_mutex held */
375static bool kprobes_allow_optimization;
376
377/*
378 * Call all pre_handler on the list, but ignores its return value.
379 * This must be called from arch-dep optimized caller.
380 */
381void __kprobes opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
382{
383 struct kprobe *kp;
384
385 list_for_each_entry_rcu(kp, &p->list, list) {
386 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
387 set_kprobe_instance(kp);
388 kp->pre_handler(kp, regs);
389 }
390 reset_kprobe_instance();
391 }
392}
393
394/* Free optimized instructions and optimized_kprobe */
395static __kprobes void free_aggr_kprobe(struct kprobe *p)
396{
397 struct optimized_kprobe *op;
398
399 op = container_of(p, struct optimized_kprobe, kp);
400 arch_remove_optimized_kprobe(op);
401 arch_remove_kprobe(p);
402 kfree(op);
403}
404
405/* Return true(!0) if the kprobe is ready for optimization. */
406static inline int kprobe_optready(struct kprobe *p)
407{
408 struct optimized_kprobe *op;
409
410 if (kprobe_aggrprobe(p)) {
411 op = container_of(p, struct optimized_kprobe, kp);
412 return arch_prepared_optinsn(&op->optinsn);
413 }
414
415 return 0;
416}
417
418/* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
419static inline int kprobe_disarmed(struct kprobe *p)
420{
421 struct optimized_kprobe *op;
422
423 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
424 if (!kprobe_aggrprobe(p))
425 return kprobe_disabled(p);
426
427 op = container_of(p, struct optimized_kprobe, kp);
428
429 return kprobe_disabled(p) && list_empty(&op->list);
430}
431
432/* Return true(!0) if the probe is queued on (un)optimizing lists */
433static int __kprobes kprobe_queued(struct kprobe *p)
434{
435 struct optimized_kprobe *op;
436
437 if (kprobe_aggrprobe(p)) {
438 op = container_of(p, struct optimized_kprobe, kp);
439 if (!list_empty(&op->list))
440 return 1;
441 }
442 return 0;
443}
444
445/*
446 * Return an optimized kprobe whose optimizing code replaces
447 * instructions including addr (exclude breakpoint).
448 */
449static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
450{
451 int i;
452 struct kprobe *p = NULL;
453 struct optimized_kprobe *op;
454
455 /* Don't check i == 0, since that is a breakpoint case. */
456 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
457 p = get_kprobe((void *)(addr - i));
458
459 if (p && kprobe_optready(p)) {
460 op = container_of(p, struct optimized_kprobe, kp);
461 if (arch_within_optimized_kprobe(op, addr))
462 return p;
463 }
464
465 return NULL;
466}
467
468/* Optimization staging list, protected by kprobe_mutex */
469static LIST_HEAD(optimizing_list);
470static LIST_HEAD(unoptimizing_list);
471
472static void kprobe_optimizer(struct work_struct *work);
473static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
474static DECLARE_COMPLETION(optimizer_comp);
475#define OPTIMIZE_DELAY 5
476
477/*
478 * Optimize (replace a breakpoint with a jump) kprobes listed on
479 * optimizing_list.
480 */
481static __kprobes void do_optimize_kprobes(void)
482{
483 /* Optimization never be done when disarmed */
484 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
485 list_empty(&optimizing_list))
486 return;
487
488 /*
489 * The optimization/unoptimization refers online_cpus via
490 * stop_machine() and cpu-hotplug modifies online_cpus.
491 * And same time, text_mutex will be held in cpu-hotplug and here.
492 * This combination can cause a deadlock (cpu-hotplug try to lock
493 * text_mutex but stop_machine can not be done because online_cpus
494 * has been changed)
495 * To avoid this deadlock, we need to call get_online_cpus()
496 * for preventing cpu-hotplug outside of text_mutex locking.
497 */
498 get_online_cpus();
499 mutex_lock(&text_mutex);
500 arch_optimize_kprobes(&optimizing_list);
501 mutex_unlock(&text_mutex);
502 put_online_cpus();
503}
504
505/*
506 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
507 * if need) kprobes listed on unoptimizing_list.
508 */
509static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
510{
511 struct optimized_kprobe *op, *tmp;
512
513 /* Unoptimization must be done anytime */
514 if (list_empty(&unoptimizing_list))
515 return;
516
517 /* Ditto to do_optimize_kprobes */
518 get_online_cpus();
519 mutex_lock(&text_mutex);
520 arch_unoptimize_kprobes(&unoptimizing_list, free_list);
521 /* Loop free_list for disarming */
522 list_for_each_entry_safe(op, tmp, free_list, list) {
523 /* Disarm probes if marked disabled */
524 if (kprobe_disabled(&op->kp))
525 arch_disarm_kprobe(&op->kp);
526 if (kprobe_unused(&op->kp)) {
527 /*
528 * Remove unused probes from hash list. After waiting
529 * for synchronization, these probes are reclaimed.
530 * (reclaiming is done by do_free_cleaned_kprobes.)
531 */
532 hlist_del_rcu(&op->kp.hlist);
533 } else
534 list_del_init(&op->list);
535 }
536 mutex_unlock(&text_mutex);
537 put_online_cpus();
538}
539
540/* Reclaim all kprobes on the free_list */
541static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
542{
543 struct optimized_kprobe *op, *tmp;
544
545 list_for_each_entry_safe(op, tmp, free_list, list) {
546 BUG_ON(!kprobe_unused(&op->kp));
547 list_del_init(&op->list);
548 free_aggr_kprobe(&op->kp);
549 }
550}
551
552/* Start optimizer after OPTIMIZE_DELAY passed */
553static __kprobes void kick_kprobe_optimizer(void)
554{
555 if (!delayed_work_pending(&optimizing_work))
556 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
557}
558
559/* Kprobe jump optimizer */
560static __kprobes void kprobe_optimizer(struct work_struct *work)
561{
562 LIST_HEAD(free_list);
563
564 /* Lock modules while optimizing kprobes */
565 mutex_lock(&module_mutex);
566 mutex_lock(&kprobe_mutex);
567
568 /*
569 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
570 * kprobes before waiting for quiesence period.
571 */
572 do_unoptimize_kprobes(&free_list);
573
574 /*
575 * Step 2: Wait for quiesence period to ensure all running interrupts
576 * are done. Because optprobe may modify multiple instructions
577 * there is a chance that Nth instruction is interrupted. In that
578 * case, running interrupt can return to 2nd-Nth byte of jump
579 * instruction. This wait is for avoiding it.
580 */
581 synchronize_sched();
582
583 /* Step 3: Optimize kprobes after quiesence period */
584 do_optimize_kprobes();
585
586 /* Step 4: Free cleaned kprobes after quiesence period */
587 do_free_cleaned_kprobes(&free_list);
588
589 mutex_unlock(&kprobe_mutex);
590 mutex_unlock(&module_mutex);
591
592 /* Step 5: Kick optimizer again if needed */
593 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
594 kick_kprobe_optimizer();
595 else
596 /* Wake up all waiters */
597 complete_all(&optimizer_comp);
598}
599
600/* Wait for completing optimization and unoptimization */
601static __kprobes void wait_for_kprobe_optimizer(void)
602{
603 if (delayed_work_pending(&optimizing_work))
604 wait_for_completion(&optimizer_comp);
605}
606
607/* Optimize kprobe if p is ready to be optimized */
608static __kprobes void optimize_kprobe(struct kprobe *p)
609{
610 struct optimized_kprobe *op;
611
612 /* Check if the kprobe is disabled or not ready for optimization. */
613 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
614 (kprobe_disabled(p) || kprobes_all_disarmed))
615 return;
616
617 /* Both of break_handler and post_handler are not supported. */
618 if (p->break_handler || p->post_handler)
619 return;
620
621 op = container_of(p, struct optimized_kprobe, kp);
622
623 /* Check there is no other kprobes at the optimized instructions */
624 if (arch_check_optimized_kprobe(op) < 0)
625 return;
626
627 /* Check if it is already optimized. */
628 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
629 return;
630 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
631
632 if (!list_empty(&op->list))
633 /* This is under unoptimizing. Just dequeue the probe */
634 list_del_init(&op->list);
635 else {
636 list_add(&op->list, &optimizing_list);
637 kick_kprobe_optimizer();
638 }
639}
640
641/* Short cut to direct unoptimizing */
642static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
643{
644 get_online_cpus();
645 arch_unoptimize_kprobe(op);
646 put_online_cpus();
647 if (kprobe_disabled(&op->kp))
648 arch_disarm_kprobe(&op->kp);
649}
650
651/* Unoptimize a kprobe if p is optimized */
652static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
653{
654 struct optimized_kprobe *op;
655
656 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
657 return; /* This is not an optprobe nor optimized */
658
659 op = container_of(p, struct optimized_kprobe, kp);
660 if (!kprobe_optimized(p)) {
661 /* Unoptimized or unoptimizing case */
662 if (force && !list_empty(&op->list)) {
663 /*
664 * Only if this is unoptimizing kprobe and forced,
665 * forcibly unoptimize it. (No need to unoptimize
666 * unoptimized kprobe again :)
667 */
668 list_del_init(&op->list);
669 force_unoptimize_kprobe(op);
670 }
671 return;
672 }
673
674 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
675 if (!list_empty(&op->list)) {
676 /* Dequeue from the optimization queue */
677 list_del_init(&op->list);
678 return;
679 }
680 /* Optimized kprobe case */
681 if (force)
682 /* Forcibly update the code: this is a special case */
683 force_unoptimize_kprobe(op);
684 else {
685 list_add(&op->list, &unoptimizing_list);
686 kick_kprobe_optimizer();
687 }
688}
689
690/* Cancel unoptimizing for reusing */
691static void reuse_unused_kprobe(struct kprobe *ap)
692{
693 struct optimized_kprobe *op;
694
695 BUG_ON(!kprobe_unused(ap));
696 /*
697 * Unused kprobe MUST be on the way of delayed unoptimizing (means
698 * there is still a relative jump) and disabled.
699 */
700 op = container_of(ap, struct optimized_kprobe, kp);
701 if (unlikely(list_empty(&op->list)))
702 printk(KERN_WARNING "Warning: found a stray unused "
703 "aggrprobe@%p\n", ap->addr);
704 /* Enable the probe again */
705 ap->flags &= ~KPROBE_FLAG_DISABLED;
706 /* Optimize it again (remove from op->list) */
707 BUG_ON(!kprobe_optready(ap));
708 optimize_kprobe(ap);
709}
710
711/* Remove optimized instructions */
712static void __kprobes kill_optimized_kprobe(struct kprobe *p)
713{
714 struct optimized_kprobe *op;
715
716 op = container_of(p, struct optimized_kprobe, kp);
717 if (!list_empty(&op->list))
718 /* Dequeue from the (un)optimization queue */
719 list_del_init(&op->list);
720
721 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
722 /* Don't touch the code, because it is already freed. */
723 arch_remove_optimized_kprobe(op);
724}
725
726/* Try to prepare optimized instructions */
727static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
728{
729 struct optimized_kprobe *op;
730
731 op = container_of(p, struct optimized_kprobe, kp);
732 arch_prepare_optimized_kprobe(op);
733}
734
735/* Allocate new optimized_kprobe and try to prepare optimized instructions */
736static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
737{
738 struct optimized_kprobe *op;
739
740 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
741 if (!op)
742 return NULL;
743
744 INIT_LIST_HEAD(&op->list);
745 op->kp.addr = p->addr;
746 arch_prepare_optimized_kprobe(op);
747
748 return &op->kp;
749}
750
751static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
752
753/*
754 * Prepare an optimized_kprobe and optimize it
755 * NOTE: p must be a normal registered kprobe
756 */
757static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
758{
759 struct kprobe *ap;
760 struct optimized_kprobe *op;
761
762 ap = alloc_aggr_kprobe(p);
763 if (!ap)
764 return;
765
766 op = container_of(ap, struct optimized_kprobe, kp);
767 if (!arch_prepared_optinsn(&op->optinsn)) {
768 /* If failed to setup optimizing, fallback to kprobe */
769 arch_remove_optimized_kprobe(op);
770 kfree(op);
771 return;
772 }
773
774 init_aggr_kprobe(ap, p);
775 optimize_kprobe(ap);
776}
777
778#ifdef CONFIG_SYSCTL
779/* This should be called with kprobe_mutex locked */
780static void __kprobes optimize_all_kprobes(void)
781{
782 struct hlist_head *head;
783 struct hlist_node *node;
784 struct kprobe *p;
785 unsigned int i;
786
787 /* If optimization is already allowed, just return */
788 if (kprobes_allow_optimization)
789 return;
790
791 kprobes_allow_optimization = true;
792 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
793 head = &kprobe_table[i];
794 hlist_for_each_entry_rcu(p, node, head, hlist)
795 if (!kprobe_disabled(p))
796 optimize_kprobe(p);
797 }
798 printk(KERN_INFO "Kprobes globally optimized\n");
799}
800
801/* This should be called with kprobe_mutex locked */
802static void __kprobes unoptimize_all_kprobes(void)
803{
804 struct hlist_head *head;
805 struct hlist_node *node;
806 struct kprobe *p;
807 unsigned int i;
808
809 /* If optimization is already prohibited, just return */
810 if (!kprobes_allow_optimization)
811 return;
812
813 kprobes_allow_optimization = false;
814 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
815 head = &kprobe_table[i];
816 hlist_for_each_entry_rcu(p, node, head, hlist) {
817 if (!kprobe_disabled(p))
818 unoptimize_kprobe(p, false);
819 }
820 }
821 /* Wait for unoptimizing completion */
822 wait_for_kprobe_optimizer();
823 printk(KERN_INFO "Kprobes globally unoptimized\n");
824}
825
826int sysctl_kprobes_optimization;
827int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
828 void __user *buffer, size_t *length,
829 loff_t *ppos)
830{
831 int ret;
832
833 mutex_lock(&kprobe_mutex);
834 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
835 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
836
837 if (sysctl_kprobes_optimization)
838 optimize_all_kprobes();
839 else
840 unoptimize_all_kprobes();
841 mutex_unlock(&kprobe_mutex);
842
843 return ret;
844}
845#endif /* CONFIG_SYSCTL */
846
847/* Put a breakpoint for a probe. Must be called with text_mutex locked */
848static void __kprobes __arm_kprobe(struct kprobe *p)
849{
850 struct kprobe *_p;
851
852 /* Check collision with other optimized kprobes */
853 _p = get_optimized_kprobe((unsigned long)p->addr);
854 if (unlikely(_p))
855 /* Fallback to unoptimized kprobe */
856 unoptimize_kprobe(_p, true);
857
858 arch_arm_kprobe(p);
859 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
860}
861
862/* Remove the breakpoint of a probe. Must be called with text_mutex locked */
863static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
864{
865 struct kprobe *_p;
866
867 unoptimize_kprobe(p, false); /* Try to unoptimize */
868
869 if (!kprobe_queued(p)) {
870 arch_disarm_kprobe(p);
871 /* If another kprobe was blocked, optimize it. */
872 _p = get_optimized_kprobe((unsigned long)p->addr);
873 if (unlikely(_p) && reopt)
874 optimize_kprobe(_p);
875 }
876 /* TODO: reoptimize others after unoptimized this probe */
877}
878
879#else /* !CONFIG_OPTPROBES */
880
881#define optimize_kprobe(p) do {} while (0)
882#define unoptimize_kprobe(p, f) do {} while (0)
883#define kill_optimized_kprobe(p) do {} while (0)
884#define prepare_optimized_kprobe(p) do {} while (0)
885#define try_to_optimize_kprobe(p) do {} while (0)
886#define __arm_kprobe(p) arch_arm_kprobe(p)
887#define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
888#define kprobe_disarmed(p) kprobe_disabled(p)
889#define wait_for_kprobe_optimizer() do {} while (0)
890
891/* There should be no unused kprobes can be reused without optimization */
892static void reuse_unused_kprobe(struct kprobe *ap)
893{
894 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
895 BUG_ON(kprobe_unused(ap));
896}
897
898static __kprobes void free_aggr_kprobe(struct kprobe *p)
899{
900 arch_remove_kprobe(p);
901 kfree(p);
902}
903
904static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
905{
906 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
907}
908#endif /* CONFIG_OPTPROBES */
909
910/* Arm a kprobe with text_mutex */
911static void __kprobes arm_kprobe(struct kprobe *kp)
912{
913 /*
914 * Here, since __arm_kprobe() doesn't use stop_machine(),
915 * this doesn't cause deadlock on text_mutex. So, we don't
916 * need get_online_cpus().
917 */
918 mutex_lock(&text_mutex);
919 __arm_kprobe(kp);
920 mutex_unlock(&text_mutex);
921}
922
923/* Disarm a kprobe with text_mutex */
924static void __kprobes disarm_kprobe(struct kprobe *kp)
925{
926 /* Ditto */
927 mutex_lock(&text_mutex);
928 __disarm_kprobe(kp, true);
929 mutex_unlock(&text_mutex);
930}
931
932/*
933 * Aggregate handlers for multiple kprobes support - these handlers
934 * take care of invoking the individual kprobe handlers on p->list
935 */
936static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
937{
938 struct kprobe *kp;
939
940 list_for_each_entry_rcu(kp, &p->list, list) {
941 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
942 set_kprobe_instance(kp);
943 if (kp->pre_handler(kp, regs))
944 return 1;
945 }
946 reset_kprobe_instance();
947 }
948 return 0;
949}
950
951static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
952 unsigned long flags)
953{
954 struct kprobe *kp;
955
956 list_for_each_entry_rcu(kp, &p->list, list) {
957 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
958 set_kprobe_instance(kp);
959 kp->post_handler(kp, regs, flags);
960 reset_kprobe_instance();
961 }
962 }
963}
964
965static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
966 int trapnr)
967{
968 struct kprobe *cur = __this_cpu_read(kprobe_instance);
969
970 /*
971 * if we faulted "during" the execution of a user specified
972 * probe handler, invoke just that probe's fault handler
973 */
974 if (cur && cur->fault_handler) {
975 if (cur->fault_handler(cur, regs, trapnr))
976 return 1;
977 }
978 return 0;
979}
980
981static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
982{
983 struct kprobe *cur = __this_cpu_read(kprobe_instance);
984 int ret = 0;
985
986 if (cur && cur->break_handler) {
987 if (cur->break_handler(cur, regs))
988 ret = 1;
989 }
990 reset_kprobe_instance();
991 return ret;
992}
993
994/* Walks the list and increments nmissed count for multiprobe case */
995void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
996{
997 struct kprobe *kp;
998 if (!kprobe_aggrprobe(p)) {
999 p->nmissed++;
1000 } else {
1001 list_for_each_entry_rcu(kp, &p->list, list)
1002 kp->nmissed++;
1003 }
1004 return;
1005}
1006
1007void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1008 struct hlist_head *head)
1009{
1010 struct kretprobe *rp = ri->rp;
1011
1012 /* remove rp inst off the rprobe_inst_table */
1013 hlist_del(&ri->hlist);
1014 INIT_HLIST_NODE(&ri->hlist);
1015 if (likely(rp)) {
1016 raw_spin_lock(&rp->lock);
1017 hlist_add_head(&ri->hlist, &rp->free_instances);
1018 raw_spin_unlock(&rp->lock);
1019 } else
1020 /* Unregistering */
1021 hlist_add_head(&ri->hlist, head);
1022}
1023
1024void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1025 struct hlist_head **head, unsigned long *flags)
1026__acquires(hlist_lock)
1027{
1028 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1029 raw_spinlock_t *hlist_lock;
1030
1031 *head = &kretprobe_inst_table[hash];
1032 hlist_lock = kretprobe_table_lock_ptr(hash);
1033 raw_spin_lock_irqsave(hlist_lock, *flags);
1034}
1035
1036static void __kprobes kretprobe_table_lock(unsigned long hash,
1037 unsigned long *flags)
1038__acquires(hlist_lock)
1039{
1040 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1041 raw_spin_lock_irqsave(hlist_lock, *flags);
1042}
1043
1044void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1045 unsigned long *flags)
1046__releases(hlist_lock)
1047{
1048 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1049 raw_spinlock_t *hlist_lock;
1050
1051 hlist_lock = kretprobe_table_lock_ptr(hash);
1052 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1053}
1054
1055static void __kprobes kretprobe_table_unlock(unsigned long hash,
1056 unsigned long *flags)
1057__releases(hlist_lock)
1058{
1059 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1060 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1061}
1062
1063/*
1064 * This function is called from finish_task_switch when task tk becomes dead,
1065 * so that we can recycle any function-return probe instances associated
1066 * with this task. These left over instances represent probed functions
1067 * that have been called but will never return.
1068 */
1069void __kprobes kprobe_flush_task(struct task_struct *tk)
1070{
1071 struct kretprobe_instance *ri;
1072 struct hlist_head *head, empty_rp;
1073 struct hlist_node *node, *tmp;
1074 unsigned long hash, flags = 0;
1075
1076 if (unlikely(!kprobes_initialized))
1077 /* Early boot. kretprobe_table_locks not yet initialized. */
1078 return;
1079
1080 INIT_HLIST_HEAD(&empty_rp);
1081 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1082 head = &kretprobe_inst_table[hash];
1083 kretprobe_table_lock(hash, &flags);
1084 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1085 if (ri->task == tk)
1086 recycle_rp_inst(ri, &empty_rp);
1087 }
1088 kretprobe_table_unlock(hash, &flags);
1089 hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1090 hlist_del(&ri->hlist);
1091 kfree(ri);
1092 }
1093}
1094
1095static inline void free_rp_inst(struct kretprobe *rp)
1096{
1097 struct kretprobe_instance *ri;
1098 struct hlist_node *pos, *next;
1099
1100 hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1101 hlist_del(&ri->hlist);
1102 kfree(ri);
1103 }
1104}
1105
1106static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1107{
1108 unsigned long flags, hash;
1109 struct kretprobe_instance *ri;
1110 struct hlist_node *pos, *next;
1111 struct hlist_head *head;
1112
1113 /* No race here */
1114 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1115 kretprobe_table_lock(hash, &flags);
1116 head = &kretprobe_inst_table[hash];
1117 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1118 if (ri->rp == rp)
1119 ri->rp = NULL;
1120 }
1121 kretprobe_table_unlock(hash, &flags);
1122 }
1123 free_rp_inst(rp);
1124}
1125
1126/*
1127* Add the new probe to ap->list. Fail if this is the
1128* second jprobe at the address - two jprobes can't coexist
1129*/
1130static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1131{
1132 BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1133
1134 if (p->break_handler || p->post_handler)
1135 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1136
1137 if (p->break_handler) {
1138 if (ap->break_handler)
1139 return -EEXIST;
1140 list_add_tail_rcu(&p->list, &ap->list);
1141 ap->break_handler = aggr_break_handler;
1142 } else
1143 list_add_rcu(&p->list, &ap->list);
1144 if (p->post_handler && !ap->post_handler)
1145 ap->post_handler = aggr_post_handler;
1146
1147 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1148 ap->flags &= ~KPROBE_FLAG_DISABLED;
1149 if (!kprobes_all_disarmed)
1150 /* Arm the breakpoint again. */
1151 __arm_kprobe(ap);
1152 }
1153 return 0;
1154}
1155
1156/*
1157 * Fill in the required fields of the "manager kprobe". Replace the
1158 * earlier kprobe in the hlist with the manager kprobe
1159 */
1160static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1161{
1162 /* Copy p's insn slot to ap */
1163 copy_kprobe(p, ap);
1164 flush_insn_slot(ap);
1165 ap->addr = p->addr;
1166 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1167 ap->pre_handler = aggr_pre_handler;
1168 ap->fault_handler = aggr_fault_handler;
1169 /* We don't care the kprobe which has gone. */
1170 if (p->post_handler && !kprobe_gone(p))
1171 ap->post_handler = aggr_post_handler;
1172 if (p->break_handler && !kprobe_gone(p))
1173 ap->break_handler = aggr_break_handler;
1174
1175 INIT_LIST_HEAD(&ap->list);
1176 INIT_HLIST_NODE(&ap->hlist);
1177
1178 list_add_rcu(&p->list, &ap->list);
1179 hlist_replace_rcu(&p->hlist, &ap->hlist);
1180}
1181
1182/*
1183 * This is the second or subsequent kprobe at the address - handle
1184 * the intricacies
1185 */
1186static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1187 struct kprobe *p)
1188{
1189 int ret = 0;
1190 struct kprobe *ap = orig_p;
1191
1192 if (!kprobe_aggrprobe(orig_p)) {
1193 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1194 ap = alloc_aggr_kprobe(orig_p);
1195 if (!ap)
1196 return -ENOMEM;
1197 init_aggr_kprobe(ap, orig_p);
1198 } else if (kprobe_unused(ap))
1199 /* This probe is going to die. Rescue it */
1200 reuse_unused_kprobe(ap);
1201
1202 if (kprobe_gone(ap)) {
1203 /*
1204 * Attempting to insert new probe at the same location that
1205 * had a probe in the module vaddr area which already
1206 * freed. So, the instruction slot has already been
1207 * released. We need a new slot for the new probe.
1208 */
1209 ret = arch_prepare_kprobe(ap);
1210 if (ret)
1211 /*
1212 * Even if fail to allocate new slot, don't need to
1213 * free aggr_probe. It will be used next time, or
1214 * freed by unregister_kprobe.
1215 */
1216 return ret;
1217
1218 /* Prepare optimized instructions if possible. */
1219 prepare_optimized_kprobe(ap);
1220
1221 /*
1222 * Clear gone flag to prevent allocating new slot again, and
1223 * set disabled flag because it is not armed yet.
1224 */
1225 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1226 | KPROBE_FLAG_DISABLED;
1227 }
1228
1229 /* Copy ap's insn slot to p */
1230 copy_kprobe(ap, p);
1231 return add_new_kprobe(ap, p);
1232}
1233
1234static int __kprobes in_kprobes_functions(unsigned long addr)
1235{
1236 struct kprobe_blackpoint *kb;
1237
1238 if (addr >= (unsigned long)__kprobes_text_start &&
1239 addr < (unsigned long)__kprobes_text_end)
1240 return -EINVAL;
1241 /*
1242 * If there exists a kprobe_blacklist, verify and
1243 * fail any probe registration in the prohibited area
1244 */
1245 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1246 if (kb->start_addr) {
1247 if (addr >= kb->start_addr &&
1248 addr < (kb->start_addr + kb->range))
1249 return -EINVAL;
1250 }
1251 }
1252 return 0;
1253}
1254
1255/*
1256 * If we have a symbol_name argument, look it up and add the offset field
1257 * to it. This way, we can specify a relative address to a symbol.
1258 * This returns encoded errors if it fails to look up symbol or invalid
1259 * combination of parameters.
1260 */
1261static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1262{
1263 kprobe_opcode_t *addr = p->addr;
1264
1265 if ((p->symbol_name && p->addr) ||
1266 (!p->symbol_name && !p->addr))
1267 goto invalid;
1268
1269 if (p->symbol_name) {
1270 kprobe_lookup_name(p->symbol_name, addr);
1271 if (!addr)
1272 return ERR_PTR(-ENOENT);
1273 }
1274
1275 addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1276 if (addr)
1277 return addr;
1278
1279invalid:
1280 return ERR_PTR(-EINVAL);
1281}
1282
1283/* Check passed kprobe is valid and return kprobe in kprobe_table. */
1284static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1285{
1286 struct kprobe *ap, *list_p;
1287
1288 ap = get_kprobe(p->addr);
1289 if (unlikely(!ap))
1290 return NULL;
1291
1292 if (p != ap) {
1293 list_for_each_entry_rcu(list_p, &ap->list, list)
1294 if (list_p == p)
1295 /* kprobe p is a valid probe */
1296 goto valid;
1297 return NULL;
1298 }
1299valid:
1300 return ap;
1301}
1302
1303/* Return error if the kprobe is being re-registered */
1304static inline int check_kprobe_rereg(struct kprobe *p)
1305{
1306 int ret = 0;
1307
1308 mutex_lock(&kprobe_mutex);
1309 if (__get_valid_kprobe(p))
1310 ret = -EINVAL;
1311 mutex_unlock(&kprobe_mutex);
1312
1313 return ret;
1314}
1315
1316int __kprobes register_kprobe(struct kprobe *p)
1317{
1318 int ret = 0;
1319 struct kprobe *old_p;
1320 struct module *probed_mod;
1321 kprobe_opcode_t *addr;
1322
1323 addr = kprobe_addr(p);
1324 if (IS_ERR(addr))
1325 return PTR_ERR(addr);
1326 p->addr = addr;
1327
1328 ret = check_kprobe_rereg(p);
1329 if (ret)
1330 return ret;
1331
1332 jump_label_lock();
1333 preempt_disable();
1334 if (!kernel_text_address((unsigned long) p->addr) ||
1335 in_kprobes_functions((unsigned long) p->addr) ||
1336 ftrace_text_reserved(p->addr, p->addr) ||
1337 jump_label_text_reserved(p->addr, p->addr)) {
1338 ret = -EINVAL;
1339 goto cannot_probe;
1340 }
1341
1342 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1343 p->flags &= KPROBE_FLAG_DISABLED;
1344
1345 /*
1346 * Check if are we probing a module.
1347 */
1348 probed_mod = __module_text_address((unsigned long) p->addr);
1349 if (probed_mod) {
1350 /* Return -ENOENT if fail. */
1351 ret = -ENOENT;
1352 /*
1353 * We must hold a refcount of the probed module while updating
1354 * its code to prohibit unexpected unloading.
1355 */
1356 if (unlikely(!try_module_get(probed_mod)))
1357 goto cannot_probe;
1358
1359 /*
1360 * If the module freed .init.text, we couldn't insert
1361 * kprobes in there.
1362 */
1363 if (within_module_init((unsigned long)p->addr, probed_mod) &&
1364 probed_mod->state != MODULE_STATE_COMING) {
1365 module_put(probed_mod);
1366 goto cannot_probe;
1367 }
1368 /* ret will be updated by following code */
1369 }
1370 preempt_enable();
1371 jump_label_unlock();
1372
1373 p->nmissed = 0;
1374 INIT_LIST_HEAD(&p->list);
1375 mutex_lock(&kprobe_mutex);
1376
1377 jump_label_lock(); /* needed to call jump_label_text_reserved() */
1378
1379 get_online_cpus(); /* For avoiding text_mutex deadlock. */
1380 mutex_lock(&text_mutex);
1381
1382 old_p = get_kprobe(p->addr);
1383 if (old_p) {
1384 /* Since this may unoptimize old_p, locking text_mutex. */
1385 ret = register_aggr_kprobe(old_p, p);
1386 goto out;
1387 }
1388
1389 ret = arch_prepare_kprobe(p);
1390 if (ret)
1391 goto out;
1392
1393 INIT_HLIST_NODE(&p->hlist);
1394 hlist_add_head_rcu(&p->hlist,
1395 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1396
1397 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1398 __arm_kprobe(p);
1399
1400 /* Try to optimize kprobe */
1401 try_to_optimize_kprobe(p);
1402
1403out:
1404 mutex_unlock(&text_mutex);
1405 put_online_cpus();
1406 jump_label_unlock();
1407 mutex_unlock(&kprobe_mutex);
1408
1409 if (probed_mod)
1410 module_put(probed_mod);
1411
1412 return ret;
1413
1414cannot_probe:
1415 preempt_enable();
1416 jump_label_unlock();
1417 return ret;
1418}
1419EXPORT_SYMBOL_GPL(register_kprobe);
1420
1421/* Check if all probes on the aggrprobe are disabled */
1422static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1423{
1424 struct kprobe *kp;
1425
1426 list_for_each_entry_rcu(kp, &ap->list, list)
1427 if (!kprobe_disabled(kp))
1428 /*
1429 * There is an active probe on the list.
1430 * We can't disable this ap.
1431 */
1432 return 0;
1433
1434 return 1;
1435}
1436
1437/* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1438static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1439{
1440 struct kprobe *orig_p;
1441
1442 /* Get an original kprobe for return */
1443 orig_p = __get_valid_kprobe(p);
1444 if (unlikely(orig_p == NULL))
1445 return NULL;
1446
1447 if (!kprobe_disabled(p)) {
1448 /* Disable probe if it is a child probe */
1449 if (p != orig_p)
1450 p->flags |= KPROBE_FLAG_DISABLED;
1451
1452 /* Try to disarm and disable this/parent probe */
1453 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1454 disarm_kprobe(orig_p);
1455 orig_p->flags |= KPROBE_FLAG_DISABLED;
1456 }
1457 }
1458
1459 return orig_p;
1460}
1461
1462/*
1463 * Unregister a kprobe without a scheduler synchronization.
1464 */
1465static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1466{
1467 struct kprobe *ap, *list_p;
1468
1469 /* Disable kprobe. This will disarm it if needed. */
1470 ap = __disable_kprobe(p);
1471 if (ap == NULL)
1472 return -EINVAL;
1473
1474 if (ap == p)
1475 /*
1476 * This probe is an independent(and non-optimized) kprobe
1477 * (not an aggrprobe). Remove from the hash list.
1478 */
1479 goto disarmed;
1480
1481 /* Following process expects this probe is an aggrprobe */
1482 WARN_ON(!kprobe_aggrprobe(ap));
1483
1484 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1485 /*
1486 * !disarmed could be happen if the probe is under delayed
1487 * unoptimizing.
1488 */
1489 goto disarmed;
1490 else {
1491 /* If disabling probe has special handlers, update aggrprobe */
1492 if (p->break_handler && !kprobe_gone(p))
1493 ap->break_handler = NULL;
1494 if (p->post_handler && !kprobe_gone(p)) {
1495 list_for_each_entry_rcu(list_p, &ap->list, list) {
1496 if ((list_p != p) && (list_p->post_handler))
1497 goto noclean;
1498 }
1499 ap->post_handler = NULL;
1500 }
1501noclean:
1502 /*
1503 * Remove from the aggrprobe: this path will do nothing in
1504 * __unregister_kprobe_bottom().
1505 */
1506 list_del_rcu(&p->list);
1507 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1508 /*
1509 * Try to optimize this probe again, because post
1510 * handler may have been changed.
1511 */
1512 optimize_kprobe(ap);
1513 }
1514 return 0;
1515
1516disarmed:
1517 BUG_ON(!kprobe_disarmed(ap));
1518 hlist_del_rcu(&ap->hlist);
1519 return 0;
1520}
1521
1522static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1523{
1524 struct kprobe *ap;
1525
1526 if (list_empty(&p->list))
1527 /* This is an independent kprobe */
1528 arch_remove_kprobe(p);
1529 else if (list_is_singular(&p->list)) {
1530 /* This is the last child of an aggrprobe */
1531 ap = list_entry(p->list.next, struct kprobe, list);
1532 list_del(&p->list);
1533 free_aggr_kprobe(ap);
1534 }
1535 /* Otherwise, do nothing. */
1536}
1537
1538int __kprobes register_kprobes(struct kprobe **kps, int num)
1539{
1540 int i, ret = 0;
1541
1542 if (num <= 0)
1543 return -EINVAL;
1544 for (i = 0; i < num; i++) {
1545 ret = register_kprobe(kps[i]);
1546 if (ret < 0) {
1547 if (i > 0)
1548 unregister_kprobes(kps, i);
1549 break;
1550 }
1551 }
1552 return ret;
1553}
1554EXPORT_SYMBOL_GPL(register_kprobes);
1555
1556void __kprobes unregister_kprobe(struct kprobe *p)
1557{
1558 unregister_kprobes(&p, 1);
1559}
1560EXPORT_SYMBOL_GPL(unregister_kprobe);
1561
1562void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1563{
1564 int i;
1565
1566 if (num <= 0)
1567 return;
1568 mutex_lock(&kprobe_mutex);
1569 for (i = 0; i < num; i++)
1570 if (__unregister_kprobe_top(kps[i]) < 0)
1571 kps[i]->addr = NULL;
1572 mutex_unlock(&kprobe_mutex);
1573
1574 synchronize_sched();
1575 for (i = 0; i < num; i++)
1576 if (kps[i]->addr)
1577 __unregister_kprobe_bottom(kps[i]);
1578}
1579EXPORT_SYMBOL_GPL(unregister_kprobes);
1580
1581static struct notifier_block kprobe_exceptions_nb = {
1582 .notifier_call = kprobe_exceptions_notify,
1583 .priority = 0x7fffffff /* we need to be notified first */
1584};
1585
1586unsigned long __weak arch_deref_entry_point(void *entry)
1587{
1588 return (unsigned long)entry;
1589}
1590
1591int __kprobes register_jprobes(struct jprobe **jps, int num)
1592{
1593 struct jprobe *jp;
1594 int ret = 0, i;
1595
1596 if (num <= 0)
1597 return -EINVAL;
1598 for (i = 0; i < num; i++) {
1599 unsigned long addr, offset;
1600 jp = jps[i];
1601 addr = arch_deref_entry_point(jp->entry);
1602
1603 /* Verify probepoint is a function entry point */
1604 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1605 offset == 0) {
1606 jp->kp.pre_handler = setjmp_pre_handler;
1607 jp->kp.break_handler = longjmp_break_handler;
1608 ret = register_kprobe(&jp->kp);
1609 } else
1610 ret = -EINVAL;
1611
1612 if (ret < 0) {
1613 if (i > 0)
1614 unregister_jprobes(jps, i);
1615 break;
1616 }
1617 }
1618 return ret;
1619}
1620EXPORT_SYMBOL_GPL(register_jprobes);
1621
1622int __kprobes register_jprobe(struct jprobe *jp)
1623{
1624 return register_jprobes(&jp, 1);
1625}
1626EXPORT_SYMBOL_GPL(register_jprobe);
1627
1628void __kprobes unregister_jprobe(struct jprobe *jp)
1629{
1630 unregister_jprobes(&jp, 1);
1631}
1632EXPORT_SYMBOL_GPL(unregister_jprobe);
1633
1634void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1635{
1636 int i;
1637
1638 if (num <= 0)
1639 return;
1640 mutex_lock(&kprobe_mutex);
1641 for (i = 0; i < num; i++)
1642 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1643 jps[i]->kp.addr = NULL;
1644 mutex_unlock(&kprobe_mutex);
1645
1646 synchronize_sched();
1647 for (i = 0; i < num; i++) {
1648 if (jps[i]->kp.addr)
1649 __unregister_kprobe_bottom(&jps[i]->kp);
1650 }
1651}
1652EXPORT_SYMBOL_GPL(unregister_jprobes);
1653
1654#ifdef CONFIG_KRETPROBES
1655/*
1656 * This kprobe pre_handler is registered with every kretprobe. When probe
1657 * hits it will set up the return probe.
1658 */
1659static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1660 struct pt_regs *regs)
1661{
1662 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1663 unsigned long hash, flags = 0;
1664 struct kretprobe_instance *ri;
1665
1666 /*TODO: consider to only swap the RA after the last pre_handler fired */
1667 hash = hash_ptr(current, KPROBE_HASH_BITS);
1668 raw_spin_lock_irqsave(&rp->lock, flags);
1669 if (!hlist_empty(&rp->free_instances)) {
1670 ri = hlist_entry(rp->free_instances.first,
1671 struct kretprobe_instance, hlist);
1672 hlist_del(&ri->hlist);
1673 raw_spin_unlock_irqrestore(&rp->lock, flags);
1674
1675 ri->rp = rp;
1676 ri->task = current;
1677
1678 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1679 raw_spin_lock_irqsave(&rp->lock, flags);
1680 hlist_add_head(&ri->hlist, &rp->free_instances);
1681 raw_spin_unlock_irqrestore(&rp->lock, flags);
1682 return 0;
1683 }
1684
1685 arch_prepare_kretprobe(ri, regs);
1686
1687 /* XXX(hch): why is there no hlist_move_head? */
1688 INIT_HLIST_NODE(&ri->hlist);
1689 kretprobe_table_lock(hash, &flags);
1690 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1691 kretprobe_table_unlock(hash, &flags);
1692 } else {
1693 rp->nmissed++;
1694 raw_spin_unlock_irqrestore(&rp->lock, flags);
1695 }
1696 return 0;
1697}
1698
1699int __kprobes register_kretprobe(struct kretprobe *rp)
1700{
1701 int ret = 0;
1702 struct kretprobe_instance *inst;
1703 int i;
1704 void *addr;
1705
1706 if (kretprobe_blacklist_size) {
1707 addr = kprobe_addr(&rp->kp);
1708 if (IS_ERR(addr))
1709 return PTR_ERR(addr);
1710
1711 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1712 if (kretprobe_blacklist[i].addr == addr)
1713 return -EINVAL;
1714 }
1715 }
1716
1717 rp->kp.pre_handler = pre_handler_kretprobe;
1718 rp->kp.post_handler = NULL;
1719 rp->kp.fault_handler = NULL;
1720 rp->kp.break_handler = NULL;
1721
1722 /* Pre-allocate memory for max kretprobe instances */
1723 if (rp->maxactive <= 0) {
1724#ifdef CONFIG_PREEMPT
1725 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1726#else
1727 rp->maxactive = num_possible_cpus();
1728#endif
1729 }
1730 raw_spin_lock_init(&rp->lock);
1731 INIT_HLIST_HEAD(&rp->free_instances);
1732 for (i = 0; i < rp->maxactive; i++) {
1733 inst = kmalloc(sizeof(struct kretprobe_instance) +
1734 rp->data_size, GFP_KERNEL);
1735 if (inst == NULL) {
1736 free_rp_inst(rp);
1737 return -ENOMEM;
1738 }
1739 INIT_HLIST_NODE(&inst->hlist);
1740 hlist_add_head(&inst->hlist, &rp->free_instances);
1741 }
1742
1743 rp->nmissed = 0;
1744 /* Establish function entry probe point */
1745 ret = register_kprobe(&rp->kp);
1746 if (ret != 0)
1747 free_rp_inst(rp);
1748 return ret;
1749}
1750EXPORT_SYMBOL_GPL(register_kretprobe);
1751
1752int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1753{
1754 int ret = 0, i;
1755
1756 if (num <= 0)
1757 return -EINVAL;
1758 for (i = 0; i < num; i++) {
1759 ret = register_kretprobe(rps[i]);
1760 if (ret < 0) {
1761 if (i > 0)
1762 unregister_kretprobes(rps, i);
1763 break;
1764 }
1765 }
1766 return ret;
1767}
1768EXPORT_SYMBOL_GPL(register_kretprobes);
1769
1770void __kprobes unregister_kretprobe(struct kretprobe *rp)
1771{
1772 unregister_kretprobes(&rp, 1);
1773}
1774EXPORT_SYMBOL_GPL(unregister_kretprobe);
1775
1776void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1777{
1778 int i;
1779
1780 if (num <= 0)
1781 return;
1782 mutex_lock(&kprobe_mutex);
1783 for (i = 0; i < num; i++)
1784 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1785 rps[i]->kp.addr = NULL;
1786 mutex_unlock(&kprobe_mutex);
1787
1788 synchronize_sched();
1789 for (i = 0; i < num; i++) {
1790 if (rps[i]->kp.addr) {
1791 __unregister_kprobe_bottom(&rps[i]->kp);
1792 cleanup_rp_inst(rps[i]);
1793 }
1794 }
1795}
1796EXPORT_SYMBOL_GPL(unregister_kretprobes);
1797
1798#else /* CONFIG_KRETPROBES */
1799int __kprobes register_kretprobe(struct kretprobe *rp)
1800{
1801 return -ENOSYS;
1802}
1803EXPORT_SYMBOL_GPL(register_kretprobe);
1804
1805int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1806{
1807 return -ENOSYS;
1808}
1809EXPORT_SYMBOL_GPL(register_kretprobes);
1810
1811void __kprobes unregister_kretprobe(struct kretprobe *rp)
1812{
1813}
1814EXPORT_SYMBOL_GPL(unregister_kretprobe);
1815
1816void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1817{
1818}
1819EXPORT_SYMBOL_GPL(unregister_kretprobes);
1820
1821static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1822 struct pt_regs *regs)
1823{
1824 return 0;
1825}
1826
1827#endif /* CONFIG_KRETPROBES */
1828
1829/* Set the kprobe gone and remove its instruction buffer. */
1830static void __kprobes kill_kprobe(struct kprobe *p)
1831{
1832 struct kprobe *kp;
1833
1834 p->flags |= KPROBE_FLAG_GONE;
1835 if (kprobe_aggrprobe(p)) {
1836 /*
1837 * If this is an aggr_kprobe, we have to list all the
1838 * chained probes and mark them GONE.
1839 */
1840 list_for_each_entry_rcu(kp, &p->list, list)
1841 kp->flags |= KPROBE_FLAG_GONE;
1842 p->post_handler = NULL;
1843 p->break_handler = NULL;
1844 kill_optimized_kprobe(p);
1845 }
1846 /*
1847 * Here, we can remove insn_slot safely, because no thread calls
1848 * the original probed function (which will be freed soon) any more.
1849 */
1850 arch_remove_kprobe(p);
1851}
1852
1853/* Disable one kprobe */
1854int __kprobes disable_kprobe(struct kprobe *kp)
1855{
1856 int ret = 0;
1857
1858 mutex_lock(&kprobe_mutex);
1859
1860 /* Disable this kprobe */
1861 if (__disable_kprobe(kp) == NULL)
1862 ret = -EINVAL;
1863
1864 mutex_unlock(&kprobe_mutex);
1865 return ret;
1866}
1867EXPORT_SYMBOL_GPL(disable_kprobe);
1868
1869/* Enable one kprobe */
1870int __kprobes enable_kprobe(struct kprobe *kp)
1871{
1872 int ret = 0;
1873 struct kprobe *p;
1874
1875 mutex_lock(&kprobe_mutex);
1876
1877 /* Check whether specified probe is valid. */
1878 p = __get_valid_kprobe(kp);
1879 if (unlikely(p == NULL)) {
1880 ret = -EINVAL;
1881 goto out;
1882 }
1883
1884 if (kprobe_gone(kp)) {
1885 /* This kprobe has gone, we couldn't enable it. */
1886 ret = -EINVAL;
1887 goto out;
1888 }
1889
1890 if (p != kp)
1891 kp->flags &= ~KPROBE_FLAG_DISABLED;
1892
1893 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
1894 p->flags &= ~KPROBE_FLAG_DISABLED;
1895 arm_kprobe(p);
1896 }
1897out:
1898 mutex_unlock(&kprobe_mutex);
1899 return ret;
1900}
1901EXPORT_SYMBOL_GPL(enable_kprobe);
1902
1903void __kprobes dump_kprobe(struct kprobe *kp)
1904{
1905 printk(KERN_WARNING "Dumping kprobe:\n");
1906 printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
1907 kp->symbol_name, kp->addr, kp->offset);
1908}
1909
1910/* Module notifier call back, checking kprobes on the module */
1911static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1912 unsigned long val, void *data)
1913{
1914 struct module *mod = data;
1915 struct hlist_head *head;
1916 struct hlist_node *node;
1917 struct kprobe *p;
1918 unsigned int i;
1919 int checkcore = (val == MODULE_STATE_GOING);
1920
1921 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1922 return NOTIFY_DONE;
1923
1924 /*
1925 * When MODULE_STATE_GOING was notified, both of module .text and
1926 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1927 * notified, only .init.text section would be freed. We need to
1928 * disable kprobes which have been inserted in the sections.
1929 */
1930 mutex_lock(&kprobe_mutex);
1931 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1932 head = &kprobe_table[i];
1933 hlist_for_each_entry_rcu(p, node, head, hlist)
1934 if (within_module_init((unsigned long)p->addr, mod) ||
1935 (checkcore &&
1936 within_module_core((unsigned long)p->addr, mod))) {
1937 /*
1938 * The vaddr this probe is installed will soon
1939 * be vfreed buy not synced to disk. Hence,
1940 * disarming the breakpoint isn't needed.
1941 */
1942 kill_kprobe(p);
1943 }
1944 }
1945 mutex_unlock(&kprobe_mutex);
1946 return NOTIFY_DONE;
1947}
1948
1949static struct notifier_block kprobe_module_nb = {
1950 .notifier_call = kprobes_module_callback,
1951 .priority = 0
1952};
1953
1954static int __init init_kprobes(void)
1955{
1956 int i, err = 0;
1957 unsigned long offset = 0, size = 0;
1958 char *modname, namebuf[128];
1959 const char *symbol_name;
1960 void *addr;
1961 struct kprobe_blackpoint *kb;
1962
1963 /* FIXME allocate the probe table, currently defined statically */
1964 /* initialize all list heads */
1965 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1966 INIT_HLIST_HEAD(&kprobe_table[i]);
1967 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1968 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
1969 }
1970
1971 /*
1972 * Lookup and populate the kprobe_blacklist.
1973 *
1974 * Unlike the kretprobe blacklist, we'll need to determine
1975 * the range of addresses that belong to the said functions,
1976 * since a kprobe need not necessarily be at the beginning
1977 * of a function.
1978 */
1979 for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1980 kprobe_lookup_name(kb->name, addr);
1981 if (!addr)
1982 continue;
1983
1984 kb->start_addr = (unsigned long)addr;
1985 symbol_name = kallsyms_lookup(kb->start_addr,
1986 &size, &offset, &modname, namebuf);
1987 if (!symbol_name)
1988 kb->range = 0;
1989 else
1990 kb->range = size;
1991 }
1992
1993 if (kretprobe_blacklist_size) {
1994 /* lookup the function address from its name */
1995 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1996 kprobe_lookup_name(kretprobe_blacklist[i].name,
1997 kretprobe_blacklist[i].addr);
1998 if (!kretprobe_blacklist[i].addr)
1999 printk("kretprobe: lookup failed: %s\n",
2000 kretprobe_blacklist[i].name);
2001 }
2002 }
2003
2004#if defined(CONFIG_OPTPROBES)
2005#if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2006 /* Init kprobe_optinsn_slots */
2007 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2008#endif
2009 /* By default, kprobes can be optimized */
2010 kprobes_allow_optimization = true;
2011#endif
2012
2013 /* By default, kprobes are armed */
2014 kprobes_all_disarmed = false;
2015
2016 err = arch_init_kprobes();
2017 if (!err)
2018 err = register_die_notifier(&kprobe_exceptions_nb);
2019 if (!err)
2020 err = register_module_notifier(&kprobe_module_nb);
2021
2022 kprobes_initialized = (err == 0);
2023
2024 if (!err)
2025 init_test_probes();
2026 return err;
2027}
2028
2029#ifdef CONFIG_DEBUG_FS
2030static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2031 const char *sym, int offset, char *modname, struct kprobe *pp)
2032{
2033 char *kprobe_type;
2034
2035 if (p->pre_handler == pre_handler_kretprobe)
2036 kprobe_type = "r";
2037 else if (p->pre_handler == setjmp_pre_handler)
2038 kprobe_type = "j";
2039 else
2040 kprobe_type = "k";
2041
2042 if (sym)
2043 seq_printf(pi, "%p %s %s+0x%x %s ",
2044 p->addr, kprobe_type, sym, offset,
2045 (modname ? modname : " "));
2046 else
2047 seq_printf(pi, "%p %s %p ",
2048 p->addr, kprobe_type, p->addr);
2049
2050 if (!pp)
2051 pp = p;
2052 seq_printf(pi, "%s%s%s\n",
2053 (kprobe_gone(p) ? "[GONE]" : ""),
2054 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2055 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""));
2056}
2057
2058static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2059{
2060 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2061}
2062
2063static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2064{
2065 (*pos)++;
2066 if (*pos >= KPROBE_TABLE_SIZE)
2067 return NULL;
2068 return pos;
2069}
2070
2071static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2072{
2073 /* Nothing to do */
2074}
2075
2076static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2077{
2078 struct hlist_head *head;
2079 struct hlist_node *node;
2080 struct kprobe *p, *kp;
2081 const char *sym = NULL;
2082 unsigned int i = *(loff_t *) v;
2083 unsigned long offset = 0;
2084 char *modname, namebuf[128];
2085
2086 head = &kprobe_table[i];
2087 preempt_disable();
2088 hlist_for_each_entry_rcu(p, node, head, hlist) {
2089 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2090 &offset, &modname, namebuf);
2091 if (kprobe_aggrprobe(p)) {
2092 list_for_each_entry_rcu(kp, &p->list, list)
2093 report_probe(pi, kp, sym, offset, modname, p);
2094 } else
2095 report_probe(pi, p, sym, offset, modname, NULL);
2096 }
2097 preempt_enable();
2098 return 0;
2099}
2100
2101static const struct seq_operations kprobes_seq_ops = {
2102 .start = kprobe_seq_start,
2103 .next = kprobe_seq_next,
2104 .stop = kprobe_seq_stop,
2105 .show = show_kprobe_addr
2106};
2107
2108static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2109{
2110 return seq_open(filp, &kprobes_seq_ops);
2111}
2112
2113static const struct file_operations debugfs_kprobes_operations = {
2114 .open = kprobes_open,
2115 .read = seq_read,
2116 .llseek = seq_lseek,
2117 .release = seq_release,
2118};
2119
2120static void __kprobes arm_all_kprobes(void)
2121{
2122 struct hlist_head *head;
2123 struct hlist_node *node;
2124 struct kprobe *p;
2125 unsigned int i;
2126
2127 mutex_lock(&kprobe_mutex);
2128
2129 /* If kprobes are armed, just return */
2130 if (!kprobes_all_disarmed)
2131 goto already_enabled;
2132
2133 /* Arming kprobes doesn't optimize kprobe itself */
2134 mutex_lock(&text_mutex);
2135 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2136 head = &kprobe_table[i];
2137 hlist_for_each_entry_rcu(p, node, head, hlist)
2138 if (!kprobe_disabled(p))
2139 __arm_kprobe(p);
2140 }
2141 mutex_unlock(&text_mutex);
2142
2143 kprobes_all_disarmed = false;
2144 printk(KERN_INFO "Kprobes globally enabled\n");
2145
2146already_enabled:
2147 mutex_unlock(&kprobe_mutex);
2148 return;
2149}
2150
2151static void __kprobes disarm_all_kprobes(void)
2152{
2153 struct hlist_head *head;
2154 struct hlist_node *node;
2155 struct kprobe *p;
2156 unsigned int i;
2157
2158 mutex_lock(&kprobe_mutex);
2159
2160 /* If kprobes are already disarmed, just return */
2161 if (kprobes_all_disarmed) {
2162 mutex_unlock(&kprobe_mutex);
2163 return;
2164 }
2165
2166 kprobes_all_disarmed = true;
2167 printk(KERN_INFO "Kprobes globally disabled\n");
2168
2169 mutex_lock(&text_mutex);
2170 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2171 head = &kprobe_table[i];
2172 hlist_for_each_entry_rcu(p, node, head, hlist) {
2173 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2174 __disarm_kprobe(p, false);
2175 }
2176 }
2177 mutex_unlock(&text_mutex);
2178 mutex_unlock(&kprobe_mutex);
2179
2180 /* Wait for disarming all kprobes by optimizer */
2181 wait_for_kprobe_optimizer();
2182}
2183
2184/*
2185 * XXX: The debugfs bool file interface doesn't allow for callbacks
2186 * when the bool state is switched. We can reuse that facility when
2187 * available
2188 */
2189static ssize_t read_enabled_file_bool(struct file *file,
2190 char __user *user_buf, size_t count, loff_t *ppos)
2191{
2192 char buf[3];
2193
2194 if (!kprobes_all_disarmed)
2195 buf[0] = '1';
2196 else
2197 buf[0] = '0';
2198 buf[1] = '\n';
2199 buf[2] = 0x00;
2200 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2201}
2202
2203static ssize_t write_enabled_file_bool(struct file *file,
2204 const char __user *user_buf, size_t count, loff_t *ppos)
2205{
2206 char buf[32];
2207 size_t buf_size;
2208
2209 buf_size = min(count, (sizeof(buf)-1));
2210 if (copy_from_user(buf, user_buf, buf_size))
2211 return -EFAULT;
2212
2213 switch (buf[0]) {
2214 case 'y':
2215 case 'Y':
2216 case '1':
2217 arm_all_kprobes();
2218 break;
2219 case 'n':
2220 case 'N':
2221 case '0':
2222 disarm_all_kprobes();
2223 break;
2224 }
2225
2226 return count;
2227}
2228
2229static const struct file_operations fops_kp = {
2230 .read = read_enabled_file_bool,
2231 .write = write_enabled_file_bool,
2232 .llseek = default_llseek,
2233};
2234
2235static int __kprobes debugfs_kprobe_init(void)
2236{
2237 struct dentry *dir, *file;
2238 unsigned int value = 1;
2239
2240 dir = debugfs_create_dir("kprobes", NULL);
2241 if (!dir)
2242 return -ENOMEM;
2243
2244 file = debugfs_create_file("list", 0444, dir, NULL,
2245 &debugfs_kprobes_operations);
2246 if (!file) {
2247 debugfs_remove(dir);
2248 return -ENOMEM;
2249 }
2250
2251 file = debugfs_create_file("enabled", 0600, dir,
2252 &value, &fops_kp);
2253 if (!file) {
2254 debugfs_remove(dir);
2255 return -ENOMEM;
2256 }
2257
2258 return 0;
2259}
2260
2261late_initcall(debugfs_kprobe_init);
2262#endif /* CONFIG_DEBUG_FS */
2263
2264module_init(init_kprobes);
2265
2266/* defined in arch/.../kernel/kprobes.c */
2267EXPORT_SYMBOL_GPL(jprobe_return);