1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Kernel Probes (KProbes)
4 * kernel/kprobes.c
5 *
6 * Copyright (C) IBM Corporation, 2002, 2004
7 *
8 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
9 * Probes initial implementation (includes suggestions from
10 * Rusty Russell).
11 * 2004-Aug Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
12 * hlists and exceptions notifier as suggested by Andi Kleen.
13 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
14 * interface to access function arguments.
15 * 2004-Sep Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
16 * exceptions notifier to be first on the priority list.
17 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
18 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
19 * <prasanna@in.ibm.com> added function-return probes.
20 */
21 #include <linux/kprobes.h>
22 #include <linux/hash.h>
23 #include <linux/init.h>
24 #include <linux/slab.h>
25 #include <linux/stddef.h>
26 #include <linux/export.h>
27 #include <linux/moduleloader.h>
28 #include <linux/kallsyms.h>
29 #include <linux/freezer.h>
30 #include <linux/seq_file.h>
31 #include <linux/debugfs.h>
32 #include <linux/sysctl.h>
33 #include <linux/kdebug.h>
34 #include <linux/memory.h>
35 #include <linux/ftrace.h>
36 #include <linux/cpu.h>
37 #include <linux/jump_label.h>
38 #include <linux/perf_event.h>
39 #include <linux/static_call.h>
40
41 #include <asm/sections.h>
42 #include <asm/cacheflush.h>
43 #include <asm/errno.h>
44 #include <linux/uaccess.h>
45
46 #define KPROBE_HASH_BITS 6
47 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
48
49
50 static int kprobes_initialized;
51 /* kprobe_table can be accessed by
52 * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
53 * Or
54 * - RCU hlist traversal under disabling preempt (breakpoint handlers)
55 */
56 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
57 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
58
59 /* NOTE: change this value only with kprobe_mutex held */
60 static bool kprobes_all_disarmed;
61
62 /* This protects kprobe_table and optimizing_list */
63 static DEFINE_MUTEX(kprobe_mutex);
64 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
65 static struct {
66 raw_spinlock_t lock ____cacheline_aligned_in_smp;
67 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
68
kprobe_lookup_name(const char * name,unsigned int __unused)69 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
70 unsigned int __unused)
71 {
72 return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
73 }
74
kretprobe_table_lock_ptr(unsigned long hash)75 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
76 {
77 return &(kretprobe_table_locks[hash].lock);
78 }
79
80 /* Blacklist -- list of struct kprobe_blacklist_entry */
81 static LIST_HEAD(kprobe_blacklist);
82
83 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
84 /*
85 * kprobe->ainsn.insn points to the copy of the instruction to be
86 * single-stepped. x86_64, POWER4 and above have no-exec support and
87 * stepping on the instruction on a vmalloced/kmalloced/data page
88 * is a recipe for disaster
89 */
90 struct kprobe_insn_page {
91 struct list_head list;
92 kprobe_opcode_t *insns; /* Page of instruction slots */
93 struct kprobe_insn_cache *cache;
94 int nused;
95 int ngarbage;
96 char slot_used[];
97 };
98
99 #define KPROBE_INSN_PAGE_SIZE(slots) \
100 (offsetof(struct kprobe_insn_page, slot_used) + \
101 (sizeof(char) * (slots)))
102
slots_per_page(struct kprobe_insn_cache * c)103 static int slots_per_page(struct kprobe_insn_cache *c)
104 {
105 return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
106 }
107
108 enum kprobe_slot_state {
109 SLOT_CLEAN = 0,
110 SLOT_DIRTY = 1,
111 SLOT_USED = 2,
112 };
113
alloc_insn_page(void)114 void __weak *alloc_insn_page(void)
115 {
116 return module_alloc(PAGE_SIZE);
117 }
118
free_insn_page(void * page)119 void __weak free_insn_page(void *page)
120 {
121 module_memfree(page);
122 }
123
124 struct kprobe_insn_cache kprobe_insn_slots = {
125 .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
126 .alloc = alloc_insn_page,
127 .free = free_insn_page,
128 .sym = KPROBE_INSN_PAGE_SYM,
129 .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
130 .insn_size = MAX_INSN_SIZE,
131 .nr_garbage = 0,
132 };
133 static int collect_garbage_slots(struct kprobe_insn_cache *c);
134
135 /**
136 * __get_insn_slot() - Find a slot on an executable page for an instruction.
137 * We allocate an executable page if there's no room on existing ones.
138 */
__get_insn_slot(struct kprobe_insn_cache * c)139 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
140 {
141 struct kprobe_insn_page *kip;
142 kprobe_opcode_t *slot = NULL;
143
144 /* Since the slot array is not protected by rcu, we need a mutex */
145 mutex_lock(&c->mutex);
146 retry:
147 rcu_read_lock();
148 list_for_each_entry_rcu(kip, &c->pages, list) {
149 if (kip->nused < slots_per_page(c)) {
150 int i;
151 for (i = 0; i < slots_per_page(c); i++) {
152 if (kip->slot_used[i] == SLOT_CLEAN) {
153 kip->slot_used[i] = SLOT_USED;
154 kip->nused++;
155 slot = kip->insns + (i * c->insn_size);
156 rcu_read_unlock();
157 goto out;
158 }
159 }
160 /* kip->nused is broken. Fix it. */
161 kip->nused = slots_per_page(c);
162 WARN_ON(1);
163 }
164 }
165 rcu_read_unlock();
166
167 /* If there are any garbage slots, collect it and try again. */
168 if (c->nr_garbage && collect_garbage_slots(c) == 0)
169 goto retry;
170
171 /* All out of space. Need to allocate a new page. */
172 kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
173 if (!kip)
174 goto out;
175
176 /*
177 * Use module_alloc so this page is within +/- 2GB of where the
178 * kernel image and loaded module images reside. This is required
179 * so x86_64 can correctly handle the %rip-relative fixups.
180 */
181 kip->insns = c->alloc();
182 if (!kip->insns) {
183 kfree(kip);
184 goto out;
185 }
186 INIT_LIST_HEAD(&kip->list);
187 memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
188 kip->slot_used[0] = SLOT_USED;
189 kip->nused = 1;
190 kip->ngarbage = 0;
191 kip->cache = c;
192 list_add_rcu(&kip->list, &c->pages);
193 slot = kip->insns;
194
195 /* Record the perf ksymbol register event after adding the page */
196 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
197 PAGE_SIZE, false, c->sym);
198 out:
199 mutex_unlock(&c->mutex);
200 return slot;
201 }
202
203 /* Return 1 if all garbages are collected, otherwise 0. */
collect_one_slot(struct kprobe_insn_page * kip,int idx)204 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
205 {
206 kip->slot_used[idx] = SLOT_CLEAN;
207 kip->nused--;
208 if (kip->nused == 0) {
209 /*
210 * Page is no longer in use. Free it unless
211 * it's the last one. We keep the last one
212 * so as not to have to set it up again the
213 * next time somebody inserts a probe.
214 */
215 if (!list_is_singular(&kip->list)) {
216 /*
217 * Record perf ksymbol unregister event before removing
218 * the page.
219 */
220 perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
221 (unsigned long)kip->insns, PAGE_SIZE, true,
222 kip->cache->sym);
223 list_del_rcu(&kip->list);
224 synchronize_rcu();
225 kip->cache->free(kip->insns);
226 kfree(kip);
227 }
228 return 1;
229 }
230 return 0;
231 }
232
collect_garbage_slots(struct kprobe_insn_cache * c)233 static int collect_garbage_slots(struct kprobe_insn_cache *c)
234 {
235 struct kprobe_insn_page *kip, *next;
236
237 /* Ensure no-one is interrupted on the garbages */
238 synchronize_rcu();
239
240 list_for_each_entry_safe(kip, next, &c->pages, list) {
241 int i;
242 if (kip->ngarbage == 0)
243 continue;
244 kip->ngarbage = 0; /* we will collect all garbages */
245 for (i = 0; i < slots_per_page(c); i++) {
246 if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
247 break;
248 }
249 }
250 c->nr_garbage = 0;
251 return 0;
252 }
253
__free_insn_slot(struct kprobe_insn_cache * c,kprobe_opcode_t * slot,int dirty)254 void __free_insn_slot(struct kprobe_insn_cache *c,
255 kprobe_opcode_t *slot, int dirty)
256 {
257 struct kprobe_insn_page *kip;
258 long idx;
259
260 mutex_lock(&c->mutex);
261 rcu_read_lock();
262 list_for_each_entry_rcu(kip, &c->pages, list) {
263 idx = ((long)slot - (long)kip->insns) /
264 (c->insn_size * sizeof(kprobe_opcode_t));
265 if (idx >= 0 && idx < slots_per_page(c))
266 goto out;
267 }
268 /* Could not find this slot. */
269 WARN_ON(1);
270 kip = NULL;
271 out:
272 rcu_read_unlock();
273 /* Mark and sweep: this may sleep */
274 if (kip) {
275 /* Check double free */
276 WARN_ON(kip->slot_used[idx] != SLOT_USED);
277 if (dirty) {
278 kip->slot_used[idx] = SLOT_DIRTY;
279 kip->ngarbage++;
280 if (++c->nr_garbage > slots_per_page(c))
281 collect_garbage_slots(c);
282 } else {
283 collect_one_slot(kip, idx);
284 }
285 }
286 mutex_unlock(&c->mutex);
287 }
288
289 /*
290 * Check given address is on the page of kprobe instruction slots.
291 * This will be used for checking whether the address on a stack
292 * is on a text area or not.
293 */
__is_insn_slot_addr(struct kprobe_insn_cache * c,unsigned long addr)294 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
295 {
296 struct kprobe_insn_page *kip;
297 bool ret = false;
298
299 rcu_read_lock();
300 list_for_each_entry_rcu(kip, &c->pages, list) {
301 if (addr >= (unsigned long)kip->insns &&
302 addr < (unsigned long)kip->insns + PAGE_SIZE) {
303 ret = true;
304 break;
305 }
306 }
307 rcu_read_unlock();
308
309 return ret;
310 }
311
kprobe_cache_get_kallsym(struct kprobe_insn_cache * c,unsigned int * symnum,unsigned long * value,char * type,char * sym)312 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
313 unsigned long *value, char *type, char *sym)
314 {
315 struct kprobe_insn_page *kip;
316 int ret = -ERANGE;
317
318 rcu_read_lock();
319 list_for_each_entry_rcu(kip, &c->pages, list) {
320 if ((*symnum)--)
321 continue;
322 strlcpy(sym, c->sym, KSYM_NAME_LEN);
323 *type = 't';
324 *value = (unsigned long)kip->insns;
325 ret = 0;
326 break;
327 }
328 rcu_read_unlock();
329
330 return ret;
331 }
332
333 #ifdef CONFIG_OPTPROBES
334 /* For optimized_kprobe buffer */
335 struct kprobe_insn_cache kprobe_optinsn_slots = {
336 .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
337 .alloc = alloc_insn_page,
338 .free = free_insn_page,
339 .sym = KPROBE_OPTINSN_PAGE_SYM,
340 .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
341 /* .insn_size is initialized later */
342 .nr_garbage = 0,
343 };
344 #endif
345 #endif
346
347 /* We have preemption disabled.. so it is safe to use __ versions */
set_kprobe_instance(struct kprobe * kp)348 static inline void set_kprobe_instance(struct kprobe *kp)
349 {
350 __this_cpu_write(kprobe_instance, kp);
351 }
352
reset_kprobe_instance(void)353 static inline void reset_kprobe_instance(void)
354 {
355 __this_cpu_write(kprobe_instance, NULL);
356 }
357
358 /*
359 * This routine is called either:
360 * - under the kprobe_mutex - during kprobe_[un]register()
361 * OR
362 * - with preemption disabled - from arch/xxx/kernel/kprobes.c
363 */
get_kprobe(void * addr)364 struct kprobe *get_kprobe(void *addr)
365 {
366 struct hlist_head *head;
367 struct kprobe *p;
368
369 head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
370 hlist_for_each_entry_rcu(p, head, hlist,
371 lockdep_is_held(&kprobe_mutex)) {
372 if (p->addr == addr)
373 return p;
374 }
375
376 return NULL;
377 }
378 NOKPROBE_SYMBOL(get_kprobe);
379
380 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
381
382 /* Return true if the kprobe is an aggregator */
kprobe_aggrprobe(struct kprobe * p)383 static inline int kprobe_aggrprobe(struct kprobe *p)
384 {
385 return p->pre_handler == aggr_pre_handler;
386 }
387
388 /* Return true(!0) if the kprobe is unused */
kprobe_unused(struct kprobe * p)389 static inline int kprobe_unused(struct kprobe *p)
390 {
391 return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
392 list_empty(&p->list);
393 }
394
395 /*
396 * Keep all fields in the kprobe consistent
397 */
copy_kprobe(struct kprobe * ap,struct kprobe * p)398 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
399 {
400 memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
401 memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
402 }
403
404 #ifdef CONFIG_OPTPROBES
405 /* NOTE: change this value only with kprobe_mutex held */
406 static bool kprobes_allow_optimization;
407
408 /*
409 * Call all pre_handler on the list, but ignores its return value.
410 * This must be called from arch-dep optimized caller.
411 */
opt_pre_handler(struct kprobe * p,struct pt_regs * regs)412 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
413 {
414 struct kprobe *kp;
415
416 list_for_each_entry_rcu(kp, &p->list, list) {
417 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
418 set_kprobe_instance(kp);
419 kp->pre_handler(kp, regs);
420 }
421 reset_kprobe_instance();
422 }
423 }
424 NOKPROBE_SYMBOL(opt_pre_handler);
425
426 /* Free optimized instructions and optimized_kprobe */
free_aggr_kprobe(struct kprobe * p)427 static void free_aggr_kprobe(struct kprobe *p)
428 {
429 struct optimized_kprobe *op;
430
431 op = container_of(p, struct optimized_kprobe, kp);
432 arch_remove_optimized_kprobe(op);
433 arch_remove_kprobe(p);
434 kfree(op);
435 }
436
437 /* Return true(!0) if the kprobe is ready for optimization. */
kprobe_optready(struct kprobe * p)438 static inline int kprobe_optready(struct kprobe *p)
439 {
440 struct optimized_kprobe *op;
441
442 if (kprobe_aggrprobe(p)) {
443 op = container_of(p, struct optimized_kprobe, kp);
444 return arch_prepared_optinsn(&op->optinsn);
445 }
446
447 return 0;
448 }
449
450 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
kprobe_disarmed(struct kprobe * p)451 static inline int kprobe_disarmed(struct kprobe *p)
452 {
453 struct optimized_kprobe *op;
454
455 /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
456 if (!kprobe_aggrprobe(p))
457 return kprobe_disabled(p);
458
459 op = container_of(p, struct optimized_kprobe, kp);
460
461 return kprobe_disabled(p) && list_empty(&op->list);
462 }
463
464 /* Return true(!0) if the probe is queued on (un)optimizing lists */
kprobe_queued(struct kprobe * p)465 static int kprobe_queued(struct kprobe *p)
466 {
467 struct optimized_kprobe *op;
468
469 if (kprobe_aggrprobe(p)) {
470 op = container_of(p, struct optimized_kprobe, kp);
471 if (!list_empty(&op->list))
472 return 1;
473 }
474 return 0;
475 }
476
477 /*
478 * Return an optimized kprobe whose optimizing code replaces
479 * instructions including addr (exclude breakpoint).
480 */
get_optimized_kprobe(unsigned long addr)481 static struct kprobe *get_optimized_kprobe(unsigned long addr)
482 {
483 int i;
484 struct kprobe *p = NULL;
485 struct optimized_kprobe *op;
486
487 /* Don't check i == 0, since that is a breakpoint case. */
488 for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
489 p = get_kprobe((void *)(addr - i));
490
491 if (p && kprobe_optready(p)) {
492 op = container_of(p, struct optimized_kprobe, kp);
493 if (arch_within_optimized_kprobe(op, addr))
494 return p;
495 }
496
497 return NULL;
498 }
499
500 /* Optimization staging list, protected by kprobe_mutex */
501 static LIST_HEAD(optimizing_list);
502 static LIST_HEAD(unoptimizing_list);
503 static LIST_HEAD(freeing_list);
504
505 static void kprobe_optimizer(struct work_struct *work);
506 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
507 #define OPTIMIZE_DELAY 5
508
509 /*
510 * Optimize (replace a breakpoint with a jump) kprobes listed on
511 * optimizing_list.
512 */
do_optimize_kprobes(void)513 static void do_optimize_kprobes(void)
514 {
515 lockdep_assert_held(&text_mutex);
516 /*
517 * The optimization/unoptimization refers online_cpus via
518 * stop_machine() and cpu-hotplug modifies online_cpus.
519 * And same time, text_mutex will be held in cpu-hotplug and here.
520 * This combination can cause a deadlock (cpu-hotplug try to lock
521 * text_mutex but stop_machine can not be done because online_cpus
522 * has been changed)
523 * To avoid this deadlock, caller must have locked cpu hotplug
524 * for preventing cpu-hotplug outside of text_mutex locking.
525 */
526 lockdep_assert_cpus_held();
527
528 /* Optimization never be done when disarmed */
529 if (kprobes_all_disarmed || !kprobes_allow_optimization ||
530 list_empty(&optimizing_list))
531 return;
532
533 arch_optimize_kprobes(&optimizing_list);
534 }
535
536 /*
537 * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
538 * if need) kprobes listed on unoptimizing_list.
539 */
do_unoptimize_kprobes(void)540 static void do_unoptimize_kprobes(void)
541 {
542 struct optimized_kprobe *op, *tmp;
543
544 lockdep_assert_held(&text_mutex);
545 /* See comment in do_optimize_kprobes() */
546 lockdep_assert_cpus_held();
547
548 /* Unoptimization must be done anytime */
549 if (list_empty(&unoptimizing_list))
550 return;
551
552 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
553 /* Loop free_list for disarming */
554 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
555 /* Switching from detour code to origin */
556 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
557 /* Disarm probes if marked disabled */
558 if (kprobe_disabled(&op->kp))
559 arch_disarm_kprobe(&op->kp);
560 if (kprobe_unused(&op->kp)) {
561 /*
562 * Remove unused probes from hash list. After waiting
563 * for synchronization, these probes are reclaimed.
564 * (reclaiming is done by do_free_cleaned_kprobes.)
565 */
566 hlist_del_rcu(&op->kp.hlist);
567 } else
568 list_del_init(&op->list);
569 }
570 }
571
572 /* Reclaim all kprobes on the free_list */
do_free_cleaned_kprobes(void)573 static void do_free_cleaned_kprobes(void)
574 {
575 struct optimized_kprobe *op, *tmp;
576
577 list_for_each_entry_safe(op, tmp, &freeing_list, list) {
578 list_del_init(&op->list);
579 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
580 /*
581 * This must not happen, but if there is a kprobe
582 * still in use, keep it on kprobes hash list.
583 */
584 continue;
585 }
586 free_aggr_kprobe(&op->kp);
587 }
588 }
589
590 /* Start optimizer after OPTIMIZE_DELAY passed */
kick_kprobe_optimizer(void)591 static void kick_kprobe_optimizer(void)
592 {
593 schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
594 }
595
596 /* Kprobe jump optimizer */
kprobe_optimizer(struct work_struct * work)597 static void kprobe_optimizer(struct work_struct *work)
598 {
599 mutex_lock(&kprobe_mutex);
600 cpus_read_lock();
601 mutex_lock(&text_mutex);
602
603 /*
604 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
605 * kprobes before waiting for quiesence period.
606 */
607 do_unoptimize_kprobes();
608
609 /*
610 * Step 2: Wait for quiesence period to ensure all potentially
611 * preempted tasks to have normally scheduled. Because optprobe
612 * may modify multiple instructions, there is a chance that Nth
613 * instruction is preempted. In that case, such tasks can return
614 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
615 * Note that on non-preemptive kernel, this is transparently converted
616 * to synchronoze_sched() to wait for all interrupts to have completed.
617 */
618 synchronize_rcu_tasks();
619
620 /* Step 3: Optimize kprobes after quiesence period */
621 do_optimize_kprobes();
622
623 /* Step 4: Free cleaned kprobes after quiesence period */
624 do_free_cleaned_kprobes();
625
626 mutex_unlock(&text_mutex);
627 cpus_read_unlock();
628
629 /* Step 5: Kick optimizer again if needed */
630 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
631 kick_kprobe_optimizer();
632
633 mutex_unlock(&kprobe_mutex);
634 }
635
636 /* Wait for completing optimization and unoptimization */
wait_for_kprobe_optimizer(void)637 void wait_for_kprobe_optimizer(void)
638 {
639 mutex_lock(&kprobe_mutex);
640
641 while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
642 mutex_unlock(&kprobe_mutex);
643
644 /* this will also make optimizing_work execute immmediately */
645 flush_delayed_work(&optimizing_work);
646 /* @optimizing_work might not have been queued yet, relax */
647 cpu_relax();
648
649 mutex_lock(&kprobe_mutex);
650 }
651
652 mutex_unlock(&kprobe_mutex);
653 }
654
optprobe_queued_unopt(struct optimized_kprobe * op)655 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
656 {
657 struct optimized_kprobe *_op;
658
659 list_for_each_entry(_op, &unoptimizing_list, list) {
660 if (op == _op)
661 return true;
662 }
663
664 return false;
665 }
666
667 /* Optimize kprobe if p is ready to be optimized */
optimize_kprobe(struct kprobe * p)668 static void optimize_kprobe(struct kprobe *p)
669 {
670 struct optimized_kprobe *op;
671
672 /* Check if the kprobe is disabled or not ready for optimization. */
673 if (!kprobe_optready(p) || !kprobes_allow_optimization ||
674 (kprobe_disabled(p) || kprobes_all_disarmed))
675 return;
676
677 /* kprobes with post_handler can not be optimized */
678 if (p->post_handler)
679 return;
680
681 op = container_of(p, struct optimized_kprobe, kp);
682
683 /* Check there is no other kprobes at the optimized instructions */
684 if (arch_check_optimized_kprobe(op) < 0)
685 return;
686
687 /* Check if it is already optimized. */
688 if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
689 if (optprobe_queued_unopt(op)) {
690 /* This is under unoptimizing. Just dequeue the probe */
691 list_del_init(&op->list);
692 }
693 return;
694 }
695 op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
696
697 /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
698 if (WARN_ON_ONCE(!list_empty(&op->list)))
699 return;
700
701 list_add(&op->list, &optimizing_list);
702 kick_kprobe_optimizer();
703 }
704
705 /* Short cut to direct unoptimizing */
force_unoptimize_kprobe(struct optimized_kprobe * op)706 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
707 {
708 lockdep_assert_cpus_held();
709 arch_unoptimize_kprobe(op);
710 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
711 }
712
713 /* Unoptimize a kprobe if p is optimized */
unoptimize_kprobe(struct kprobe * p,bool force)714 static void unoptimize_kprobe(struct kprobe *p, bool force)
715 {
716 struct optimized_kprobe *op;
717
718 if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
719 return; /* This is not an optprobe nor optimized */
720
721 op = container_of(p, struct optimized_kprobe, kp);
722 if (!kprobe_optimized(p))
723 return;
724
725 if (!list_empty(&op->list)) {
726 if (optprobe_queued_unopt(op)) {
727 /* Queued in unoptimizing queue */
728 if (force) {
729 /*
730 * Forcibly unoptimize the kprobe here, and queue it
731 * in the freeing list for release afterwards.
732 */
733 force_unoptimize_kprobe(op);
734 list_move(&op->list, &freeing_list);
735 }
736 } else {
737 /* Dequeue from the optimizing queue */
738 list_del_init(&op->list);
739 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
740 }
741 return;
742 }
743
744 /* Optimized kprobe case */
745 if (force) {
746 /* Forcibly update the code: this is a special case */
747 force_unoptimize_kprobe(op);
748 } else {
749 list_add(&op->list, &unoptimizing_list);
750 kick_kprobe_optimizer();
751 }
752 }
753
754 /* Cancel unoptimizing for reusing */
reuse_unused_kprobe(struct kprobe * ap)755 static int reuse_unused_kprobe(struct kprobe *ap)
756 {
757 struct optimized_kprobe *op;
758
759 /*
760 * Unused kprobe MUST be on the way of delayed unoptimizing (means
761 * there is still a relative jump) and disabled.
762 */
763 op = container_of(ap, struct optimized_kprobe, kp);
764 WARN_ON_ONCE(list_empty(&op->list));
765 /* Enable the probe again */
766 ap->flags &= ~KPROBE_FLAG_DISABLED;
767 /* Optimize it again (remove from op->list) */
768 if (!kprobe_optready(ap))
769 return -EINVAL;
770
771 optimize_kprobe(ap);
772 return 0;
773 }
774
775 /* Remove optimized instructions */
kill_optimized_kprobe(struct kprobe * p)776 static void kill_optimized_kprobe(struct kprobe *p)
777 {
778 struct optimized_kprobe *op;
779
780 op = container_of(p, struct optimized_kprobe, kp);
781 if (!list_empty(&op->list))
782 /* Dequeue from the (un)optimization queue */
783 list_del_init(&op->list);
784 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
785
786 if (kprobe_unused(p)) {
787 /* Enqueue if it is unused */
788 list_add(&op->list, &freeing_list);
789 /*
790 * Remove unused probes from the hash list. After waiting
791 * for synchronization, this probe is reclaimed.
792 * (reclaiming is done by do_free_cleaned_kprobes().)
793 */
794 hlist_del_rcu(&op->kp.hlist);
795 }
796
797 /* Don't touch the code, because it is already freed. */
798 arch_remove_optimized_kprobe(op);
799 }
800
801 static inline
__prepare_optimized_kprobe(struct optimized_kprobe * op,struct kprobe * p)802 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
803 {
804 if (!kprobe_ftrace(p))
805 arch_prepare_optimized_kprobe(op, p);
806 }
807
808 /* Try to prepare optimized instructions */
prepare_optimized_kprobe(struct kprobe * p)809 static void prepare_optimized_kprobe(struct kprobe *p)
810 {
811 struct optimized_kprobe *op;
812
813 op = container_of(p, struct optimized_kprobe, kp);
814 __prepare_optimized_kprobe(op, p);
815 }
816
817 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
alloc_aggr_kprobe(struct kprobe * p)818 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
819 {
820 struct optimized_kprobe *op;
821
822 op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
823 if (!op)
824 return NULL;
825
826 INIT_LIST_HEAD(&op->list);
827 op->kp.addr = p->addr;
828 __prepare_optimized_kprobe(op, p);
829
830 return &op->kp;
831 }
832
833 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
834
835 /*
836 * Prepare an optimized_kprobe and optimize it
837 * NOTE: p must be a normal registered kprobe
838 */
try_to_optimize_kprobe(struct kprobe * p)839 static void try_to_optimize_kprobe(struct kprobe *p)
840 {
841 struct kprobe *ap;
842 struct optimized_kprobe *op;
843
844 /* Impossible to optimize ftrace-based kprobe */
845 if (kprobe_ftrace(p))
846 return;
847
848 /* For preparing optimization, jump_label_text_reserved() is called */
849 cpus_read_lock();
850 jump_label_lock();
851 mutex_lock(&text_mutex);
852
853 ap = alloc_aggr_kprobe(p);
854 if (!ap)
855 goto out;
856
857 op = container_of(ap, struct optimized_kprobe, kp);
858 if (!arch_prepared_optinsn(&op->optinsn)) {
859 /* If failed to setup optimizing, fallback to kprobe */
860 arch_remove_optimized_kprobe(op);
861 kfree(op);
862 goto out;
863 }
864
865 init_aggr_kprobe(ap, p);
866 optimize_kprobe(ap); /* This just kicks optimizer thread */
867
868 out:
869 mutex_unlock(&text_mutex);
870 jump_label_unlock();
871 cpus_read_unlock();
872 }
873
optimize_all_kprobes(void)874 static void optimize_all_kprobes(void)
875 {
876 struct hlist_head *head;
877 struct kprobe *p;
878 unsigned int i;
879
880 mutex_lock(&kprobe_mutex);
881 /* If optimization is already allowed, just return */
882 if (kprobes_allow_optimization)
883 goto out;
884
885 cpus_read_lock();
886 kprobes_allow_optimization = true;
887 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
888 head = &kprobe_table[i];
889 hlist_for_each_entry(p, head, hlist)
890 if (!kprobe_disabled(p))
891 optimize_kprobe(p);
892 }
893 cpus_read_unlock();
894 printk(KERN_INFO "Kprobes globally optimized\n");
895 out:
896 mutex_unlock(&kprobe_mutex);
897 }
898
899 #ifdef CONFIG_SYSCTL
unoptimize_all_kprobes(void)900 static void unoptimize_all_kprobes(void)
901 {
902 struct hlist_head *head;
903 struct kprobe *p;
904 unsigned int i;
905
906 mutex_lock(&kprobe_mutex);
907 /* If optimization is already prohibited, just return */
908 if (!kprobes_allow_optimization) {
909 mutex_unlock(&kprobe_mutex);
910 return;
911 }
912
913 cpus_read_lock();
914 kprobes_allow_optimization = false;
915 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
916 head = &kprobe_table[i];
917 hlist_for_each_entry(p, head, hlist) {
918 if (!kprobe_disabled(p))
919 unoptimize_kprobe(p, false);
920 }
921 }
922 cpus_read_unlock();
923 mutex_unlock(&kprobe_mutex);
924
925 /* Wait for unoptimizing completion */
926 wait_for_kprobe_optimizer();
927 printk(KERN_INFO "Kprobes globally unoptimized\n");
928 }
929
930 static DEFINE_MUTEX(kprobe_sysctl_mutex);
931 int sysctl_kprobes_optimization;
proc_kprobes_optimization_handler(struct ctl_table * table,int write,void * buffer,size_t * length,loff_t * ppos)932 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
933 void *buffer, size_t *length,
934 loff_t *ppos)
935 {
936 int ret;
937
938 mutex_lock(&kprobe_sysctl_mutex);
939 sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
940 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
941
942 if (sysctl_kprobes_optimization)
943 optimize_all_kprobes();
944 else
945 unoptimize_all_kprobes();
946 mutex_unlock(&kprobe_sysctl_mutex);
947
948 return ret;
949 }
950 #endif /* CONFIG_SYSCTL */
951
952 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
__arm_kprobe(struct kprobe * p)953 static void __arm_kprobe(struct kprobe *p)
954 {
955 struct kprobe *_p;
956
957 /* Check collision with other optimized kprobes */
958 _p = get_optimized_kprobe((unsigned long)p->addr);
959 if (unlikely(_p))
960 /* Fallback to unoptimized kprobe */
961 unoptimize_kprobe(_p, true);
962
963 arch_arm_kprobe(p);
964 optimize_kprobe(p); /* Try to optimize (add kprobe to a list) */
965 }
966
967 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
__disarm_kprobe(struct kprobe * p,bool reopt)968 static void __disarm_kprobe(struct kprobe *p, bool reopt)
969 {
970 struct kprobe *_p;
971
972 /* Try to unoptimize */
973 unoptimize_kprobe(p, kprobes_all_disarmed);
974
975 if (!kprobe_queued(p)) {
976 arch_disarm_kprobe(p);
977 /* If another kprobe was blocked, optimize it. */
978 _p = get_optimized_kprobe((unsigned long)p->addr);
979 if (unlikely(_p) && reopt)
980 optimize_kprobe(_p);
981 }
982 /* TODO: reoptimize others after unoptimized this probe */
983 }
984
985 #else /* !CONFIG_OPTPROBES */
986
987 #define optimize_kprobe(p) do {} while (0)
988 #define unoptimize_kprobe(p, f) do {} while (0)
989 #define kill_optimized_kprobe(p) do {} while (0)
990 #define prepare_optimized_kprobe(p) do {} while (0)
991 #define try_to_optimize_kprobe(p) do {} while (0)
992 #define __arm_kprobe(p) arch_arm_kprobe(p)
993 #define __disarm_kprobe(p, o) arch_disarm_kprobe(p)
994 #define kprobe_disarmed(p) kprobe_disabled(p)
995 #define wait_for_kprobe_optimizer() do {} while (0)
996
reuse_unused_kprobe(struct kprobe * ap)997 static int reuse_unused_kprobe(struct kprobe *ap)
998 {
999 /*
1000 * If the optimized kprobe is NOT supported, the aggr kprobe is
1001 * released at the same time that the last aggregated kprobe is
1002 * unregistered.
1003 * Thus there should be no chance to reuse unused kprobe.
1004 */
1005 printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1006 return -EINVAL;
1007 }
1008
free_aggr_kprobe(struct kprobe * p)1009 static void free_aggr_kprobe(struct kprobe *p)
1010 {
1011 arch_remove_kprobe(p);
1012 kfree(p);
1013 }
1014
alloc_aggr_kprobe(struct kprobe * p)1015 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1016 {
1017 return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1018 }
1019 #endif /* CONFIG_OPTPROBES */
1020
1021 #ifdef CONFIG_KPROBES_ON_FTRACE
1022 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1023 .func = kprobe_ftrace_handler,
1024 .flags = FTRACE_OPS_FL_SAVE_REGS,
1025 };
1026
1027 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1028 .func = kprobe_ftrace_handler,
1029 .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1030 };
1031
1032 static int kprobe_ipmodify_enabled;
1033 static int kprobe_ftrace_enabled;
1034
1035 /* Must ensure p->addr is really on ftrace */
prepare_kprobe(struct kprobe * p)1036 static int prepare_kprobe(struct kprobe *p)
1037 {
1038 if (!kprobe_ftrace(p))
1039 return arch_prepare_kprobe(p);
1040
1041 return arch_prepare_kprobe_ftrace(p);
1042 }
1043
1044 /* Caller must lock kprobe_mutex */
__arm_kprobe_ftrace(struct kprobe * p,struct ftrace_ops * ops,int * cnt)1045 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1046 int *cnt)
1047 {
1048 int ret = 0;
1049
1050 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1051 if (ret) {
1052 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1053 p->addr, ret);
1054 return ret;
1055 }
1056
1057 if (*cnt == 0) {
1058 ret = register_ftrace_function(ops);
1059 if (ret) {
1060 pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1061 goto err_ftrace;
1062 }
1063 }
1064
1065 (*cnt)++;
1066 return ret;
1067
1068 err_ftrace:
1069 /*
1070 * At this point, sinec ops is not registered, we should be sefe from
1071 * registering empty filter.
1072 */
1073 ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1074 return ret;
1075 }
1076
arm_kprobe_ftrace(struct kprobe * p)1077 static int arm_kprobe_ftrace(struct kprobe *p)
1078 {
1079 bool ipmodify = (p->post_handler != NULL);
1080
1081 return __arm_kprobe_ftrace(p,
1082 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1083 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1084 }
1085
1086 /* Caller must lock kprobe_mutex */
__disarm_kprobe_ftrace(struct kprobe * p,struct ftrace_ops * ops,int * cnt)1087 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1088 int *cnt)
1089 {
1090 int ret = 0;
1091
1092 if (*cnt == 1) {
1093 ret = unregister_ftrace_function(ops);
1094 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1095 return ret;
1096 }
1097
1098 (*cnt)--;
1099
1100 ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1101 WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1102 p->addr, ret);
1103 return ret;
1104 }
1105
disarm_kprobe_ftrace(struct kprobe * p)1106 static int disarm_kprobe_ftrace(struct kprobe *p)
1107 {
1108 bool ipmodify = (p->post_handler != NULL);
1109
1110 return __disarm_kprobe_ftrace(p,
1111 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1112 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1113 }
1114 #else /* !CONFIG_KPROBES_ON_FTRACE */
prepare_kprobe(struct kprobe * p)1115 static inline int prepare_kprobe(struct kprobe *p)
1116 {
1117 return arch_prepare_kprobe(p);
1118 }
1119
arm_kprobe_ftrace(struct kprobe * p)1120 static inline int arm_kprobe_ftrace(struct kprobe *p)
1121 {
1122 return -ENODEV;
1123 }
1124
disarm_kprobe_ftrace(struct kprobe * p)1125 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1126 {
1127 return -ENODEV;
1128 }
1129 #endif
1130
1131 /* Arm a kprobe with text_mutex */
arm_kprobe(struct kprobe * kp)1132 static int arm_kprobe(struct kprobe *kp)
1133 {
1134 if (unlikely(kprobe_ftrace(kp)))
1135 return arm_kprobe_ftrace(kp);
1136
1137 cpus_read_lock();
1138 mutex_lock(&text_mutex);
1139 __arm_kprobe(kp);
1140 mutex_unlock(&text_mutex);
1141 cpus_read_unlock();
1142
1143 return 0;
1144 }
1145
1146 /* Disarm a kprobe with text_mutex */
disarm_kprobe(struct kprobe * kp,bool reopt)1147 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1148 {
1149 if (unlikely(kprobe_ftrace(kp)))
1150 return disarm_kprobe_ftrace(kp);
1151
1152 cpus_read_lock();
1153 mutex_lock(&text_mutex);
1154 __disarm_kprobe(kp, reopt);
1155 mutex_unlock(&text_mutex);
1156 cpus_read_unlock();
1157
1158 return 0;
1159 }
1160
1161 /*
1162 * Aggregate handlers for multiple kprobes support - these handlers
1163 * take care of invoking the individual kprobe handlers on p->list
1164 */
aggr_pre_handler(struct kprobe * p,struct pt_regs * regs)1165 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1166 {
1167 struct kprobe *kp;
1168
1169 list_for_each_entry_rcu(kp, &p->list, list) {
1170 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1171 set_kprobe_instance(kp);
1172 if (kp->pre_handler(kp, regs))
1173 return 1;
1174 }
1175 reset_kprobe_instance();
1176 }
1177 return 0;
1178 }
1179 NOKPROBE_SYMBOL(aggr_pre_handler);
1180
aggr_post_handler(struct kprobe * p,struct pt_regs * regs,unsigned long flags)1181 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1182 unsigned long flags)
1183 {
1184 struct kprobe *kp;
1185
1186 list_for_each_entry_rcu(kp, &p->list, list) {
1187 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1188 set_kprobe_instance(kp);
1189 kp->post_handler(kp, regs, flags);
1190 reset_kprobe_instance();
1191 }
1192 }
1193 }
1194 NOKPROBE_SYMBOL(aggr_post_handler);
1195
aggr_fault_handler(struct kprobe * p,struct pt_regs * regs,int trapnr)1196 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1197 int trapnr)
1198 {
1199 struct kprobe *cur = __this_cpu_read(kprobe_instance);
1200
1201 /*
1202 * if we faulted "during" the execution of a user specified
1203 * probe handler, invoke just that probe's fault handler
1204 */
1205 if (cur && cur->fault_handler) {
1206 if (cur->fault_handler(cur, regs, trapnr))
1207 return 1;
1208 }
1209 return 0;
1210 }
1211 NOKPROBE_SYMBOL(aggr_fault_handler);
1212
1213 /* Walks the list and increments nmissed count for multiprobe case */
kprobes_inc_nmissed_count(struct kprobe * p)1214 void kprobes_inc_nmissed_count(struct kprobe *p)
1215 {
1216 struct kprobe *kp;
1217 if (!kprobe_aggrprobe(p)) {
1218 p->nmissed++;
1219 } else {
1220 list_for_each_entry_rcu(kp, &p->list, list)
1221 kp->nmissed++;
1222 }
1223 return;
1224 }
1225 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1226
recycle_rp_inst(struct kretprobe_instance * ri)1227 static void recycle_rp_inst(struct kretprobe_instance *ri)
1228 {
1229 struct kretprobe *rp = ri->rp;
1230
1231 /* remove rp inst off the rprobe_inst_table */
1232 hlist_del(&ri->hlist);
1233 INIT_HLIST_NODE(&ri->hlist);
1234 if (likely(rp)) {
1235 raw_spin_lock(&rp->lock);
1236 hlist_add_head(&ri->hlist, &rp->free_instances);
1237 raw_spin_unlock(&rp->lock);
1238 } else
1239 kfree_rcu(ri, rcu);
1240 }
1241 NOKPROBE_SYMBOL(recycle_rp_inst);
1242
kretprobe_hash_lock(struct task_struct * tsk,struct hlist_head ** head,unsigned long * flags)1243 static void kretprobe_hash_lock(struct task_struct *tsk,
1244 struct hlist_head **head, unsigned long *flags)
1245 __acquires(hlist_lock)
1246 {
1247 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1248 raw_spinlock_t *hlist_lock;
1249
1250 *head = &kretprobe_inst_table[hash];
1251 hlist_lock = kretprobe_table_lock_ptr(hash);
1252 /*
1253 * Nested is a workaround that will soon not be needed.
1254 * There's other protections that make sure the same lock
1255 * is not taken on the same CPU that lockdep is unaware of.
1256 * Differentiate when it is taken in NMI context.
1257 */
1258 raw_spin_lock_irqsave_nested(hlist_lock, *flags, !!in_nmi());
1259 }
1260 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1261
kretprobe_table_lock(unsigned long hash,unsigned long * flags)1262 static void kretprobe_table_lock(unsigned long hash,
1263 unsigned long *flags)
1264 __acquires(hlist_lock)
1265 {
1266 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1267 /*
1268 * Nested is a workaround that will soon not be needed.
1269 * There's other protections that make sure the same lock
1270 * is not taken on the same CPU that lockdep is unaware of.
1271 * Differentiate when it is taken in NMI context.
1272 */
1273 raw_spin_lock_irqsave_nested(hlist_lock, *flags, !!in_nmi());
1274 }
1275 NOKPROBE_SYMBOL(kretprobe_table_lock);
1276
kretprobe_hash_unlock(struct task_struct * tsk,unsigned long * flags)1277 static void kretprobe_hash_unlock(struct task_struct *tsk,
1278 unsigned long *flags)
1279 __releases(hlist_lock)
1280 {
1281 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1282 raw_spinlock_t *hlist_lock;
1283
1284 hlist_lock = kretprobe_table_lock_ptr(hash);
1285 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1286 }
1287 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1288
kretprobe_table_unlock(unsigned long hash,unsigned long * flags)1289 static void kretprobe_table_unlock(unsigned long hash,
1290 unsigned long *flags)
1291 __releases(hlist_lock)
1292 {
1293 raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1294 raw_spin_unlock_irqrestore(hlist_lock, *flags);
1295 }
1296 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1297
1298 static struct kprobe kprobe_busy = {
1299 .addr = (void *) get_kprobe,
1300 };
1301
kprobe_busy_begin(void)1302 void kprobe_busy_begin(void)
1303 {
1304 struct kprobe_ctlblk *kcb;
1305
1306 preempt_disable();
1307 __this_cpu_write(current_kprobe, &kprobe_busy);
1308 kcb = get_kprobe_ctlblk();
1309 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1310 }
1311
kprobe_busy_end(void)1312 void kprobe_busy_end(void)
1313 {
1314 __this_cpu_write(current_kprobe, NULL);
1315 preempt_enable();
1316 }
1317
1318 /*
1319 * This function is called from finish_task_switch when task tk becomes dead,
1320 * so that we can recycle any function-return probe instances associated
1321 * with this task. These left over instances represent probed functions
1322 * that have been called but will never return.
1323 */
kprobe_flush_task(struct task_struct * tk)1324 void kprobe_flush_task(struct task_struct *tk)
1325 {
1326 struct kretprobe_instance *ri;
1327 struct hlist_head *head;
1328 struct hlist_node *tmp;
1329 unsigned long hash, flags = 0;
1330
1331 if (unlikely(!kprobes_initialized))
1332 /* Early boot. kretprobe_table_locks not yet initialized. */
1333 return;
1334
1335 kprobe_busy_begin();
1336
1337 hash = hash_ptr(tk, KPROBE_HASH_BITS);
1338 head = &kretprobe_inst_table[hash];
1339 kretprobe_table_lock(hash, &flags);
1340 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1341 if (ri->task == tk)
1342 recycle_rp_inst(ri);
1343 }
1344 kretprobe_table_unlock(hash, &flags);
1345
1346 kprobe_busy_end();
1347 }
1348 NOKPROBE_SYMBOL(kprobe_flush_task);
1349
free_rp_inst(struct kretprobe * rp)1350 static inline void free_rp_inst(struct kretprobe *rp)
1351 {
1352 struct kretprobe_instance *ri;
1353 struct hlist_node *next;
1354
1355 hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1356 hlist_del(&ri->hlist);
1357 kfree(ri);
1358 }
1359 }
1360
cleanup_rp_inst(struct kretprobe * rp)1361 static void cleanup_rp_inst(struct kretprobe *rp)
1362 {
1363 unsigned long flags, hash;
1364 struct kretprobe_instance *ri;
1365 struct hlist_node *next;
1366 struct hlist_head *head;
1367
1368 /* To avoid recursive kretprobe by NMI, set kprobe busy here */
1369 kprobe_busy_begin();
1370 for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1371 kretprobe_table_lock(hash, &flags);
1372 head = &kretprobe_inst_table[hash];
1373 hlist_for_each_entry_safe(ri, next, head, hlist) {
1374 if (ri->rp == rp)
1375 ri->rp = NULL;
1376 }
1377 kretprobe_table_unlock(hash, &flags);
1378 }
1379 kprobe_busy_end();
1380
1381 free_rp_inst(rp);
1382 }
1383 NOKPROBE_SYMBOL(cleanup_rp_inst);
1384
1385 /* Add the new probe to ap->list */
add_new_kprobe(struct kprobe * ap,struct kprobe * p)1386 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1387 {
1388 if (p->post_handler)
1389 unoptimize_kprobe(ap, true); /* Fall back to normal kprobe */
1390
1391 list_add_rcu(&p->list, &ap->list);
1392 if (p->post_handler && !ap->post_handler)
1393 ap->post_handler = aggr_post_handler;
1394
1395 return 0;
1396 }
1397
1398 /*
1399 * Fill in the required fields of the "manager kprobe". Replace the
1400 * earlier kprobe in the hlist with the manager kprobe
1401 */
init_aggr_kprobe(struct kprobe * ap,struct kprobe * p)1402 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1403 {
1404 /* Copy p's insn slot to ap */
1405 copy_kprobe(p, ap);
1406 flush_insn_slot(ap);
1407 ap->addr = p->addr;
1408 ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1409 ap->pre_handler = aggr_pre_handler;
1410 ap->fault_handler = aggr_fault_handler;
1411 /* We don't care the kprobe which has gone. */
1412 if (p->post_handler && !kprobe_gone(p))
1413 ap->post_handler = aggr_post_handler;
1414
1415 INIT_LIST_HEAD(&ap->list);
1416 INIT_HLIST_NODE(&ap->hlist);
1417
1418 list_add_rcu(&p->list, &ap->list);
1419 hlist_replace_rcu(&p->hlist, &ap->hlist);
1420 }
1421
1422 /*
1423 * This is the second or subsequent kprobe at the address - handle
1424 * the intricacies
1425 */
register_aggr_kprobe(struct kprobe * orig_p,struct kprobe * p)1426 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1427 {
1428 int ret = 0;
1429 struct kprobe *ap = orig_p;
1430
1431 cpus_read_lock();
1432
1433 /* For preparing optimization, jump_label_text_reserved() is called */
1434 jump_label_lock();
1435 mutex_lock(&text_mutex);
1436
1437 if (!kprobe_aggrprobe(orig_p)) {
1438 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1439 ap = alloc_aggr_kprobe(orig_p);
1440 if (!ap) {
1441 ret = -ENOMEM;
1442 goto out;
1443 }
1444 init_aggr_kprobe(ap, orig_p);
1445 } else if (kprobe_unused(ap)) {
1446 /* This probe is going to die. Rescue it */
1447 ret = reuse_unused_kprobe(ap);
1448 if (ret)
1449 goto out;
1450 }
1451
1452 if (kprobe_gone(ap)) {
1453 /*
1454 * Attempting to insert new probe at the same location that
1455 * had a probe in the module vaddr area which already
1456 * freed. So, the instruction slot has already been
1457 * released. We need a new slot for the new probe.
1458 */
1459 ret = arch_prepare_kprobe(ap);
1460 if (ret)
1461 /*
1462 * Even if fail to allocate new slot, don't need to
1463 * free aggr_probe. It will be used next time, or
1464 * freed by unregister_kprobe.
1465 */
1466 goto out;
1467
1468 /* Prepare optimized instructions if possible. */
1469 prepare_optimized_kprobe(ap);
1470
1471 /*
1472 * Clear gone flag to prevent allocating new slot again, and
1473 * set disabled flag because it is not armed yet.
1474 */
1475 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1476 | KPROBE_FLAG_DISABLED;
1477 }
1478
1479 /* Copy ap's insn slot to p */
1480 copy_kprobe(ap, p);
1481 ret = add_new_kprobe(ap, p);
1482
1483 out:
1484 mutex_unlock(&text_mutex);
1485 jump_label_unlock();
1486 cpus_read_unlock();
1487
1488 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1489 ap->flags &= ~KPROBE_FLAG_DISABLED;
1490 if (!kprobes_all_disarmed) {
1491 /* Arm the breakpoint again. */
1492 ret = arm_kprobe(ap);
1493 if (ret) {
1494 ap->flags |= KPROBE_FLAG_DISABLED;
1495 list_del_rcu(&p->list);
1496 synchronize_rcu();
1497 }
1498 }
1499 }
1500 return ret;
1501 }
1502
arch_within_kprobe_blacklist(unsigned long addr)1503 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1504 {
1505 /* The __kprobes marked functions and entry code must not be probed */
1506 return addr >= (unsigned long)__kprobes_text_start &&
1507 addr < (unsigned long)__kprobes_text_end;
1508 }
1509
__within_kprobe_blacklist(unsigned long addr)1510 static bool __within_kprobe_blacklist(unsigned long addr)
1511 {
1512 struct kprobe_blacklist_entry *ent;
1513
1514 if (arch_within_kprobe_blacklist(addr))
1515 return true;
1516 /*
1517 * If there exists a kprobe_blacklist, verify and
1518 * fail any probe registration in the prohibited area
1519 */
1520 list_for_each_entry(ent, &kprobe_blacklist, list) {
1521 if (addr >= ent->start_addr && addr < ent->end_addr)
1522 return true;
1523 }
1524 return false;
1525 }
1526
within_kprobe_blacklist(unsigned long addr)1527 bool within_kprobe_blacklist(unsigned long addr)
1528 {
1529 char symname[KSYM_NAME_LEN], *p;
1530
1531 if (__within_kprobe_blacklist(addr))
1532 return true;
1533
1534 /* Check if the address is on a suffixed-symbol */
1535 if (!lookup_symbol_name(addr, symname)) {
1536 p = strchr(symname, '.');
1537 if (!p)
1538 return false;
1539 *p = '\0';
1540 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1541 if (addr)
1542 return __within_kprobe_blacklist(addr);
1543 }
1544 return false;
1545 }
1546
1547 /*
1548 * If we have a symbol_name argument, look it up and add the offset field
1549 * to it. This way, we can specify a relative address to a symbol.
1550 * This returns encoded errors if it fails to look up symbol or invalid
1551 * combination of parameters.
1552 */
_kprobe_addr(kprobe_opcode_t * addr,const char * symbol_name,unsigned int offset)1553 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1554 const char *symbol_name, unsigned int offset)
1555 {
1556 if ((symbol_name && addr) || (!symbol_name && !addr))
1557 goto invalid;
1558
1559 if (symbol_name) {
1560 addr = kprobe_lookup_name(symbol_name, offset);
1561 if (!addr)
1562 return ERR_PTR(-ENOENT);
1563 }
1564
1565 addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1566 if (addr)
1567 return addr;
1568
1569 invalid:
1570 return ERR_PTR(-EINVAL);
1571 }
1572
kprobe_addr(struct kprobe * p)1573 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1574 {
1575 return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1576 }
1577
1578 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
__get_valid_kprobe(struct kprobe * p)1579 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1580 {
1581 struct kprobe *ap, *list_p;
1582
1583 lockdep_assert_held(&kprobe_mutex);
1584
1585 ap = get_kprobe(p->addr);
1586 if (unlikely(!ap))
1587 return NULL;
1588
1589 if (p != ap) {
1590 list_for_each_entry(list_p, &ap->list, list)
1591 if (list_p == p)
1592 /* kprobe p is a valid probe */
1593 goto valid;
1594 return NULL;
1595 }
1596 valid:
1597 return ap;
1598 }
1599
1600 /* Return error if the kprobe is being re-registered */
check_kprobe_rereg(struct kprobe * p)1601 static inline int check_kprobe_rereg(struct kprobe *p)
1602 {
1603 int ret = 0;
1604
1605 mutex_lock(&kprobe_mutex);
1606 if (__get_valid_kprobe(p))
1607 ret = -EINVAL;
1608 mutex_unlock(&kprobe_mutex);
1609
1610 return ret;
1611 }
1612
arch_check_ftrace_location(struct kprobe * p)1613 int __weak arch_check_ftrace_location(struct kprobe *p)
1614 {
1615 unsigned long ftrace_addr;
1616
1617 ftrace_addr = ftrace_location((unsigned long)p->addr);
1618 if (ftrace_addr) {
1619 #ifdef CONFIG_KPROBES_ON_FTRACE
1620 /* Given address is not on the instruction boundary */
1621 if ((unsigned long)p->addr != ftrace_addr)
1622 return -EILSEQ;
1623 p->flags |= KPROBE_FLAG_FTRACE;
1624 #else /* !CONFIG_KPROBES_ON_FTRACE */
1625 return -EINVAL;
1626 #endif
1627 }
1628 return 0;
1629 }
1630
check_kprobe_address_safe(struct kprobe * p,struct module ** probed_mod)1631 static int check_kprobe_address_safe(struct kprobe *p,
1632 struct module **probed_mod)
1633 {
1634 int ret;
1635
1636 ret = arch_check_ftrace_location(p);
1637 if (ret)
1638 return ret;
1639 jump_label_lock();
1640 preempt_disable();
1641
1642 /* Ensure it is not in reserved area nor out of text */
1643 if (!kernel_text_address((unsigned long) p->addr) ||
1644 within_kprobe_blacklist((unsigned long) p->addr) ||
1645 jump_label_text_reserved(p->addr, p->addr) ||
1646 static_call_text_reserved(p->addr, p->addr) ||
1647 find_bug((unsigned long)p->addr)) {
1648 ret = -EINVAL;
1649 goto out;
1650 }
1651
1652 /* Check if are we probing a module */
1653 *probed_mod = __module_text_address((unsigned long) p->addr);
1654 if (*probed_mod) {
1655 /*
1656 * We must hold a refcount of the probed module while updating
1657 * its code to prohibit unexpected unloading.
1658 */
1659 if (unlikely(!try_module_get(*probed_mod))) {
1660 ret = -ENOENT;
1661 goto out;
1662 }
1663
1664 /*
1665 * If the module freed .init.text, we couldn't insert
1666 * kprobes in there.
1667 */
1668 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1669 (*probed_mod)->state != MODULE_STATE_COMING) {
1670 module_put(*probed_mod);
1671 *probed_mod = NULL;
1672 ret = -ENOENT;
1673 }
1674 }
1675 out:
1676 preempt_enable();
1677 jump_label_unlock();
1678
1679 return ret;
1680 }
1681
register_kprobe(struct kprobe * p)1682 int register_kprobe(struct kprobe *p)
1683 {
1684 int ret;
1685 struct kprobe *old_p;
1686 struct module *probed_mod;
1687 kprobe_opcode_t *addr;
1688
1689 /* Adjust probe address from symbol */
1690 addr = kprobe_addr(p);
1691 if (IS_ERR(addr))
1692 return PTR_ERR(addr);
1693 p->addr = addr;
1694
1695 ret = check_kprobe_rereg(p);
1696 if (ret)
1697 return ret;
1698
1699 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1700 p->flags &= KPROBE_FLAG_DISABLED;
1701 p->nmissed = 0;
1702 INIT_LIST_HEAD(&p->list);
1703
1704 ret = check_kprobe_address_safe(p, &probed_mod);
1705 if (ret)
1706 return ret;
1707
1708 mutex_lock(&kprobe_mutex);
1709
1710 old_p = get_kprobe(p->addr);
1711 if (old_p) {
1712 /* Since this may unoptimize old_p, locking text_mutex. */
1713 ret = register_aggr_kprobe(old_p, p);
1714 goto out;
1715 }
1716
1717 cpus_read_lock();
1718 /* Prevent text modification */
1719 mutex_lock(&text_mutex);
1720 ret = prepare_kprobe(p);
1721 mutex_unlock(&text_mutex);
1722 cpus_read_unlock();
1723 if (ret)
1724 goto out;
1725
1726 INIT_HLIST_NODE(&p->hlist);
1727 hlist_add_head_rcu(&p->hlist,
1728 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1729
1730 if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1731 ret = arm_kprobe(p);
1732 if (ret) {
1733 hlist_del_rcu(&p->hlist);
1734 synchronize_rcu();
1735 goto out;
1736 }
1737 }
1738
1739 /* Try to optimize kprobe */
1740 try_to_optimize_kprobe(p);
1741 out:
1742 mutex_unlock(&kprobe_mutex);
1743
1744 if (probed_mod)
1745 module_put(probed_mod);
1746
1747 return ret;
1748 }
1749 EXPORT_SYMBOL_GPL(register_kprobe);
1750
1751 /* Check if all probes on the aggrprobe are disabled */
aggr_kprobe_disabled(struct kprobe * ap)1752 static int aggr_kprobe_disabled(struct kprobe *ap)
1753 {
1754 struct kprobe *kp;
1755
1756 lockdep_assert_held(&kprobe_mutex);
1757
1758 list_for_each_entry(kp, &ap->list, list)
1759 if (!kprobe_disabled(kp))
1760 /*
1761 * There is an active probe on the list.
1762 * We can't disable this ap.
1763 */
1764 return 0;
1765
1766 return 1;
1767 }
1768
1769 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
__disable_kprobe(struct kprobe * p)1770 static struct kprobe *__disable_kprobe(struct kprobe *p)
1771 {
1772 struct kprobe *orig_p;
1773 int ret;
1774
1775 /* Get an original kprobe for return */
1776 orig_p = __get_valid_kprobe(p);
1777 if (unlikely(orig_p == NULL))
1778 return ERR_PTR(-EINVAL);
1779
1780 if (!kprobe_disabled(p)) {
1781 /* Disable probe if it is a child probe */
1782 if (p != orig_p)
1783 p->flags |= KPROBE_FLAG_DISABLED;
1784
1785 /* Try to disarm and disable this/parent probe */
1786 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1787 /*
1788 * If kprobes_all_disarmed is set, orig_p
1789 * should have already been disarmed, so
1790 * skip unneed disarming process.
1791 */
1792 if (!kprobes_all_disarmed) {
1793 ret = disarm_kprobe(orig_p, true);
1794 if (ret) {
1795 p->flags &= ~KPROBE_FLAG_DISABLED;
1796 return ERR_PTR(ret);
1797 }
1798 }
1799 orig_p->flags |= KPROBE_FLAG_DISABLED;
1800 }
1801 }
1802
1803 return orig_p;
1804 }
1805
1806 /*
1807 * Unregister a kprobe without a scheduler synchronization.
1808 */
__unregister_kprobe_top(struct kprobe * p)1809 static int __unregister_kprobe_top(struct kprobe *p)
1810 {
1811 struct kprobe *ap, *list_p;
1812
1813 /* Disable kprobe. This will disarm it if needed. */
1814 ap = __disable_kprobe(p);
1815 if (IS_ERR(ap))
1816 return PTR_ERR(ap);
1817
1818 if (ap == p)
1819 /*
1820 * This probe is an independent(and non-optimized) kprobe
1821 * (not an aggrprobe). Remove from the hash list.
1822 */
1823 goto disarmed;
1824
1825 /* Following process expects this probe is an aggrprobe */
1826 WARN_ON(!kprobe_aggrprobe(ap));
1827
1828 if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1829 /*
1830 * !disarmed could be happen if the probe is under delayed
1831 * unoptimizing.
1832 */
1833 goto disarmed;
1834 else {
1835 /* If disabling probe has special handlers, update aggrprobe */
1836 if (p->post_handler && !kprobe_gone(p)) {
1837 list_for_each_entry(list_p, &ap->list, list) {
1838 if ((list_p != p) && (list_p->post_handler))
1839 goto noclean;
1840 }
1841 ap->post_handler = NULL;
1842 }
1843 noclean:
1844 /*
1845 * Remove from the aggrprobe: this path will do nothing in
1846 * __unregister_kprobe_bottom().
1847 */
1848 list_del_rcu(&p->list);
1849 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1850 /*
1851 * Try to optimize this probe again, because post
1852 * handler may have been changed.
1853 */
1854 optimize_kprobe(ap);
1855 }
1856 return 0;
1857
1858 disarmed:
1859 hlist_del_rcu(&ap->hlist);
1860 return 0;
1861 }
1862
__unregister_kprobe_bottom(struct kprobe * p)1863 static void __unregister_kprobe_bottom(struct kprobe *p)
1864 {
1865 struct kprobe *ap;
1866
1867 if (list_empty(&p->list))
1868 /* This is an independent kprobe */
1869 arch_remove_kprobe(p);
1870 else if (list_is_singular(&p->list)) {
1871 /* This is the last child of an aggrprobe */
1872 ap = list_entry(p->list.next, struct kprobe, list);
1873 list_del(&p->list);
1874 free_aggr_kprobe(ap);
1875 }
1876 /* Otherwise, do nothing. */
1877 }
1878
register_kprobes(struct kprobe ** kps,int num)1879 int register_kprobes(struct kprobe **kps, int num)
1880 {
1881 int i, ret = 0;
1882
1883 if (num <= 0)
1884 return -EINVAL;
1885 for (i = 0; i < num; i++) {
1886 ret = register_kprobe(kps[i]);
1887 if (ret < 0) {
1888 if (i > 0)
1889 unregister_kprobes(kps, i);
1890 break;
1891 }
1892 }
1893 return ret;
1894 }
1895 EXPORT_SYMBOL_GPL(register_kprobes);
1896
unregister_kprobe(struct kprobe * p)1897 void unregister_kprobe(struct kprobe *p)
1898 {
1899 unregister_kprobes(&p, 1);
1900 }
1901 EXPORT_SYMBOL_GPL(unregister_kprobe);
1902
unregister_kprobes(struct kprobe ** kps,int num)1903 void unregister_kprobes(struct kprobe **kps, int num)
1904 {
1905 int i;
1906
1907 if (num <= 0)
1908 return;
1909 mutex_lock(&kprobe_mutex);
1910 for (i = 0; i < num; i++)
1911 if (__unregister_kprobe_top(kps[i]) < 0)
1912 kps[i]->addr = NULL;
1913 mutex_unlock(&kprobe_mutex);
1914
1915 synchronize_rcu();
1916 for (i = 0; i < num; i++)
1917 if (kps[i]->addr)
1918 __unregister_kprobe_bottom(kps[i]);
1919 }
1920 EXPORT_SYMBOL_GPL(unregister_kprobes);
1921
kprobe_exceptions_notify(struct notifier_block * self,unsigned long val,void * data)1922 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1923 unsigned long val, void *data)
1924 {
1925 return NOTIFY_DONE;
1926 }
1927 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1928
1929 static struct notifier_block kprobe_exceptions_nb = {
1930 .notifier_call = kprobe_exceptions_notify,
1931 .priority = 0x7fffffff /* we need to be notified first */
1932 };
1933
arch_deref_entry_point(void * entry)1934 unsigned long __weak arch_deref_entry_point(void *entry)
1935 {
1936 return (unsigned long)entry;
1937 }
1938
1939 #ifdef CONFIG_KRETPROBES
1940
__kretprobe_trampoline_handler(struct pt_regs * regs,void * trampoline_address,void * frame_pointer)1941 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1942 void *trampoline_address,
1943 void *frame_pointer)
1944 {
1945 struct kretprobe_instance *ri = NULL, *last = NULL;
1946 struct hlist_head *head;
1947 struct hlist_node *tmp;
1948 unsigned long flags;
1949 kprobe_opcode_t *correct_ret_addr = NULL;
1950 bool skipped = false;
1951
1952 kretprobe_hash_lock(current, &head, &flags);
1953
1954 /*
1955 * It is possible to have multiple instances associated with a given
1956 * task either because multiple functions in the call path have
1957 * return probes installed on them, and/or more than one
1958 * return probe was registered for a target function.
1959 *
1960 * We can handle this because:
1961 * - instances are always pushed into the head of the list
1962 * - when multiple return probes are registered for the same
1963 * function, the (chronologically) first instance's ret_addr
1964 * will be the real return address, and all the rest will
1965 * point to kretprobe_trampoline.
1966 */
1967 hlist_for_each_entry(ri, head, hlist) {
1968 if (ri->task != current)
1969 /* another task is sharing our hash bucket */
1970 continue;
1971 /*
1972 * Return probes must be pushed on this hash list correct
1973 * order (same as return order) so that it can be popped
1974 * correctly. However, if we find it is pushed it incorrect
1975 * order, this means we find a function which should not be
1976 * probed, because the wrong order entry is pushed on the
1977 * path of processing other kretprobe itself.
1978 */
1979 if (ri->fp != frame_pointer) {
1980 if (!skipped)
1981 pr_warn("kretprobe is stacked incorrectly. Trying to fixup.\n");
1982 skipped = true;
1983 continue;
1984 }
1985
1986 correct_ret_addr = ri->ret_addr;
1987 if (skipped)
1988 pr_warn("%ps must be blacklisted because of incorrect kretprobe order\n",
1989 ri->rp->kp.addr);
1990
1991 if (correct_ret_addr != trampoline_address)
1992 /*
1993 * This is the real return address. Any other
1994 * instances associated with this task are for
1995 * other calls deeper on the call stack
1996 */
1997 break;
1998 }
1999
2000 BUG_ON(!correct_ret_addr || (correct_ret_addr == trampoline_address));
2001 last = ri;
2002
2003 hlist_for_each_entry_safe(ri, tmp, head, hlist) {
2004 if (ri->task != current)
2005 /* another task is sharing our hash bucket */
2006 continue;
2007 if (ri->fp != frame_pointer)
2008 continue;
2009
2010 if (ri->rp && ri->rp->handler) {
2011 struct kprobe *prev = kprobe_running();
2012
2013 __this_cpu_write(current_kprobe, &ri->rp->kp);
2014 ri->ret_addr = correct_ret_addr;
2015 ri->rp->handler(ri, regs);
2016 __this_cpu_write(current_kprobe, prev);
2017 }
2018
2019 recycle_rp_inst(ri);
2020
2021 if (ri == last)
2022 break;
2023 }
2024
2025 kretprobe_hash_unlock(current, &flags);
2026
2027 return (unsigned long)correct_ret_addr;
2028 }
NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)2029 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2030
2031 /*
2032 * This kprobe pre_handler is registered with every kretprobe. When probe
2033 * hits it will set up the return probe.
2034 */
2035 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2036 {
2037 struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2038 unsigned long hash, flags = 0;
2039 struct kretprobe_instance *ri;
2040
2041 /* TODO: consider to only swap the RA after the last pre_handler fired */
2042 hash = hash_ptr(current, KPROBE_HASH_BITS);
2043 /*
2044 * Nested is a workaround that will soon not be needed.
2045 * There's other protections that make sure the same lock
2046 * is not taken on the same CPU that lockdep is unaware of.
2047 */
2048 raw_spin_lock_irqsave_nested(&rp->lock, flags, 1);
2049 if (!hlist_empty(&rp->free_instances)) {
2050 ri = hlist_entry(rp->free_instances.first,
2051 struct kretprobe_instance, hlist);
2052 hlist_del(&ri->hlist);
2053 raw_spin_unlock_irqrestore(&rp->lock, flags);
2054
2055 ri->rp = rp;
2056 ri->task = current;
2057
2058 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2059 raw_spin_lock_irqsave_nested(&rp->lock, flags, 1);
2060 hlist_add_head(&ri->hlist, &rp->free_instances);
2061 raw_spin_unlock_irqrestore(&rp->lock, flags);
2062 return 0;
2063 }
2064
2065 arch_prepare_kretprobe(ri, regs);
2066
2067 /* XXX(hch): why is there no hlist_move_head? */
2068 INIT_HLIST_NODE(&ri->hlist);
2069 kretprobe_table_lock(hash, &flags);
2070 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
2071 kretprobe_table_unlock(hash, &flags);
2072 } else {
2073 rp->nmissed++;
2074 raw_spin_unlock_irqrestore(&rp->lock, flags);
2075 }
2076 return 0;
2077 }
2078 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2079
arch_kprobe_on_func_entry(unsigned long offset)2080 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
2081 {
2082 return !offset;
2083 }
2084
2085 /**
2086 * kprobe_on_func_entry() -- check whether given address is function entry
2087 * @addr: Target address
2088 * @sym: Target symbol name
2089 * @offset: The offset from the symbol or the address
2090 *
2091 * This checks whether the given @addr+@offset or @sym+@offset is on the
2092 * function entry address or not.
2093 * This returns 0 if it is the function entry, or -EINVAL if it is not.
2094 * And also it returns -ENOENT if it fails the symbol or address lookup.
2095 * Caller must pass @addr or @sym (either one must be NULL), or this
2096 * returns -EINVAL.
2097 */
kprobe_on_func_entry(kprobe_opcode_t * addr,const char * sym,unsigned long offset)2098 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2099 {
2100 kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
2101
2102 if (IS_ERR(kp_addr))
2103 return PTR_ERR(kp_addr);
2104
2105 if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
2106 return -ENOENT;
2107
2108 if (!arch_kprobe_on_func_entry(offset))
2109 return -EINVAL;
2110
2111 return 0;
2112 }
2113
register_kretprobe(struct kretprobe * rp)2114 int register_kretprobe(struct kretprobe *rp)
2115 {
2116 int ret;
2117 struct kretprobe_instance *inst;
2118 int i;
2119 void *addr;
2120
2121 ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2122 if (ret)
2123 return ret;
2124
2125 /* If only rp->kp.addr is specified, check reregistering kprobes */
2126 if (rp->kp.addr && check_kprobe_rereg(&rp->kp))
2127 return -EINVAL;
2128
2129 if (kretprobe_blacklist_size) {
2130 addr = kprobe_addr(&rp->kp);
2131 if (IS_ERR(addr))
2132 return PTR_ERR(addr);
2133
2134 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2135 if (kretprobe_blacklist[i].addr == addr)
2136 return -EINVAL;
2137 }
2138 }
2139
2140 if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2141 return -E2BIG;
2142
2143 rp->kp.pre_handler = pre_handler_kretprobe;
2144 rp->kp.post_handler = NULL;
2145 rp->kp.fault_handler = NULL;
2146
2147 /* Pre-allocate memory for max kretprobe instances */
2148 if (rp->maxactive <= 0) {
2149 #ifdef CONFIG_PREEMPTION
2150 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2151 #else
2152 rp->maxactive = num_possible_cpus();
2153 #endif
2154 }
2155 raw_spin_lock_init(&rp->lock);
2156 INIT_HLIST_HEAD(&rp->free_instances);
2157 for (i = 0; i < rp->maxactive; i++) {
2158 inst = kmalloc(sizeof(struct kretprobe_instance) +
2159 rp->data_size, GFP_KERNEL);
2160 if (inst == NULL) {
2161 free_rp_inst(rp);
2162 return -ENOMEM;
2163 }
2164 INIT_HLIST_NODE(&inst->hlist);
2165 hlist_add_head(&inst->hlist, &rp->free_instances);
2166 }
2167
2168 rp->nmissed = 0;
2169 /* Establish function entry probe point */
2170 ret = register_kprobe(&rp->kp);
2171 if (ret != 0)
2172 free_rp_inst(rp);
2173 return ret;
2174 }
2175 EXPORT_SYMBOL_GPL(register_kretprobe);
2176
register_kretprobes(struct kretprobe ** rps,int num)2177 int register_kretprobes(struct kretprobe **rps, int num)
2178 {
2179 int ret = 0, i;
2180
2181 if (num <= 0)
2182 return -EINVAL;
2183 for (i = 0; i < num; i++) {
2184 ret = register_kretprobe(rps[i]);
2185 if (ret < 0) {
2186 if (i > 0)
2187 unregister_kretprobes(rps, i);
2188 break;
2189 }
2190 }
2191 return ret;
2192 }
2193 EXPORT_SYMBOL_GPL(register_kretprobes);
2194
unregister_kretprobe(struct kretprobe * rp)2195 void unregister_kretprobe(struct kretprobe *rp)
2196 {
2197 unregister_kretprobes(&rp, 1);
2198 }
2199 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2200
unregister_kretprobes(struct kretprobe ** rps,int num)2201 void unregister_kretprobes(struct kretprobe **rps, int num)
2202 {
2203 int i;
2204
2205 if (num <= 0)
2206 return;
2207 mutex_lock(&kprobe_mutex);
2208 for (i = 0; i < num; i++)
2209 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2210 rps[i]->kp.addr = NULL;
2211 mutex_unlock(&kprobe_mutex);
2212
2213 synchronize_rcu();
2214 for (i = 0; i < num; i++) {
2215 if (rps[i]->kp.addr) {
2216 __unregister_kprobe_bottom(&rps[i]->kp);
2217 cleanup_rp_inst(rps[i]);
2218 }
2219 }
2220 }
2221 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2222
2223 #else /* CONFIG_KRETPROBES */
register_kretprobe(struct kretprobe * rp)2224 int register_kretprobe(struct kretprobe *rp)
2225 {
2226 return -ENOSYS;
2227 }
2228 EXPORT_SYMBOL_GPL(register_kretprobe);
2229
register_kretprobes(struct kretprobe ** rps,int num)2230 int register_kretprobes(struct kretprobe **rps, int num)
2231 {
2232 return -ENOSYS;
2233 }
2234 EXPORT_SYMBOL_GPL(register_kretprobes);
2235
unregister_kretprobe(struct kretprobe * rp)2236 void unregister_kretprobe(struct kretprobe *rp)
2237 {
2238 }
2239 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2240
unregister_kretprobes(struct kretprobe ** rps,int num)2241 void unregister_kretprobes(struct kretprobe **rps, int num)
2242 {
2243 }
2244 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2245
pre_handler_kretprobe(struct kprobe * p,struct pt_regs * regs)2246 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2247 {
2248 return 0;
2249 }
2250 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2251
2252 #endif /* CONFIG_KRETPROBES */
2253
2254 /* Set the kprobe gone and remove its instruction buffer. */
kill_kprobe(struct kprobe * p)2255 static void kill_kprobe(struct kprobe *p)
2256 {
2257 struct kprobe *kp;
2258
2259 lockdep_assert_held(&kprobe_mutex);
2260
2261 if (WARN_ON_ONCE(kprobe_gone(p)))
2262 return;
2263
2264 p->flags |= KPROBE_FLAG_GONE;
2265 if (kprobe_aggrprobe(p)) {
2266 /*
2267 * If this is an aggr_kprobe, we have to list all the
2268 * chained probes and mark them GONE.
2269 */
2270 list_for_each_entry(kp, &p->list, list)
2271 kp->flags |= KPROBE_FLAG_GONE;
2272 p->post_handler = NULL;
2273 kill_optimized_kprobe(p);
2274 }
2275 /*
2276 * Here, we can remove insn_slot safely, because no thread calls
2277 * the original probed function (which will be freed soon) any more.
2278 */
2279 arch_remove_kprobe(p);
2280
2281 /*
2282 * The module is going away. We should disarm the kprobe which
2283 * is using ftrace, because ftrace framework is still available at
2284 * MODULE_STATE_GOING notification.
2285 */
2286 if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2287 disarm_kprobe_ftrace(p);
2288 }
2289
2290 /* Disable one kprobe */
disable_kprobe(struct kprobe * kp)2291 int disable_kprobe(struct kprobe *kp)
2292 {
2293 int ret = 0;
2294 struct kprobe *p;
2295
2296 mutex_lock(&kprobe_mutex);
2297
2298 /* Disable this kprobe */
2299 p = __disable_kprobe(kp);
2300 if (IS_ERR(p))
2301 ret = PTR_ERR(p);
2302
2303 mutex_unlock(&kprobe_mutex);
2304 return ret;
2305 }
2306 EXPORT_SYMBOL_GPL(disable_kprobe);
2307
2308 /* Enable one kprobe */
enable_kprobe(struct kprobe * kp)2309 int enable_kprobe(struct kprobe *kp)
2310 {
2311 int ret = 0;
2312 struct kprobe *p;
2313
2314 mutex_lock(&kprobe_mutex);
2315
2316 /* Check whether specified probe is valid. */
2317 p = __get_valid_kprobe(kp);
2318 if (unlikely(p == NULL)) {
2319 ret = -EINVAL;
2320 goto out;
2321 }
2322
2323 if (kprobe_gone(kp)) {
2324 /* This kprobe has gone, we couldn't enable it. */
2325 ret = -EINVAL;
2326 goto out;
2327 }
2328
2329 if (p != kp)
2330 kp->flags &= ~KPROBE_FLAG_DISABLED;
2331
2332 if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2333 p->flags &= ~KPROBE_FLAG_DISABLED;
2334 ret = arm_kprobe(p);
2335 if (ret)
2336 p->flags |= KPROBE_FLAG_DISABLED;
2337 }
2338 out:
2339 mutex_unlock(&kprobe_mutex);
2340 return ret;
2341 }
2342 EXPORT_SYMBOL_GPL(enable_kprobe);
2343
2344 /* Caller must NOT call this in usual path. This is only for critical case */
dump_kprobe(struct kprobe * kp)2345 void dump_kprobe(struct kprobe *kp)
2346 {
2347 pr_err("Dumping kprobe:\n");
2348 pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2349 kp->symbol_name, kp->offset, kp->addr);
2350 }
2351 NOKPROBE_SYMBOL(dump_kprobe);
2352
kprobe_add_ksym_blacklist(unsigned long entry)2353 int kprobe_add_ksym_blacklist(unsigned long entry)
2354 {
2355 struct kprobe_blacklist_entry *ent;
2356 unsigned long offset = 0, size = 0;
2357
2358 if (!kernel_text_address(entry) ||
2359 !kallsyms_lookup_size_offset(entry, &size, &offset))
2360 return -EINVAL;
2361
2362 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2363 if (!ent)
2364 return -ENOMEM;
2365 ent->start_addr = entry;
2366 ent->end_addr = entry + size;
2367 INIT_LIST_HEAD(&ent->list);
2368 list_add_tail(&ent->list, &kprobe_blacklist);
2369
2370 return (int)size;
2371 }
2372
2373 /* Add all symbols in given area into kprobe blacklist */
kprobe_add_area_blacklist(unsigned long start,unsigned long end)2374 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2375 {
2376 unsigned long entry;
2377 int ret = 0;
2378
2379 for (entry = start; entry < end; entry += ret) {
2380 ret = kprobe_add_ksym_blacklist(entry);
2381 if (ret < 0)
2382 return ret;
2383 if (ret == 0) /* In case of alias symbol */
2384 ret = 1;
2385 }
2386 return 0;
2387 }
2388
2389 /* Remove all symbols in given area from kprobe blacklist */
kprobe_remove_area_blacklist(unsigned long start,unsigned long end)2390 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2391 {
2392 struct kprobe_blacklist_entry *ent, *n;
2393
2394 list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2395 if (ent->start_addr < start || ent->start_addr >= end)
2396 continue;
2397 list_del(&ent->list);
2398 kfree(ent);
2399 }
2400 }
2401
kprobe_remove_ksym_blacklist(unsigned long entry)2402 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2403 {
2404 kprobe_remove_area_blacklist(entry, entry + 1);
2405 }
2406
arch_kprobe_get_kallsym(unsigned int * symnum,unsigned long * value,char * type,char * sym)2407 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2408 char *type, char *sym)
2409 {
2410 return -ERANGE;
2411 }
2412
kprobe_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)2413 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2414 char *sym)
2415 {
2416 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2417 if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2418 return 0;
2419 #ifdef CONFIG_OPTPROBES
2420 if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2421 return 0;
2422 #endif
2423 #endif
2424 if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2425 return 0;
2426 return -ERANGE;
2427 }
2428
arch_populate_kprobe_blacklist(void)2429 int __init __weak arch_populate_kprobe_blacklist(void)
2430 {
2431 return 0;
2432 }
2433
2434 /*
2435 * Lookup and populate the kprobe_blacklist.
2436 *
2437 * Unlike the kretprobe blacklist, we'll need to determine
2438 * the range of addresses that belong to the said functions,
2439 * since a kprobe need not necessarily be at the beginning
2440 * of a function.
2441 */
populate_kprobe_blacklist(unsigned long * start,unsigned long * end)2442 static int __init populate_kprobe_blacklist(unsigned long *start,
2443 unsigned long *end)
2444 {
2445 unsigned long entry;
2446 unsigned long *iter;
2447 int ret;
2448
2449 for (iter = start; iter < end; iter++) {
2450 entry = arch_deref_entry_point((void *)*iter);
2451 ret = kprobe_add_ksym_blacklist(entry);
2452 if (ret == -EINVAL)
2453 continue;
2454 if (ret < 0)
2455 return ret;
2456 }
2457
2458 /* Symbols in __kprobes_text are blacklisted */
2459 ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2460 (unsigned long)__kprobes_text_end);
2461 if (ret)
2462 return ret;
2463
2464 /* Symbols in noinstr section are blacklisted */
2465 ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2466 (unsigned long)__noinstr_text_end);
2467
2468 return ret ? : arch_populate_kprobe_blacklist();
2469 }
2470
add_module_kprobe_blacklist(struct module * mod)2471 static void add_module_kprobe_blacklist(struct module *mod)
2472 {
2473 unsigned long start, end;
2474 int i;
2475
2476 if (mod->kprobe_blacklist) {
2477 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2478 kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2479 }
2480
2481 start = (unsigned long)mod->kprobes_text_start;
2482 if (start) {
2483 end = start + mod->kprobes_text_size;
2484 kprobe_add_area_blacklist(start, end);
2485 }
2486
2487 start = (unsigned long)mod->noinstr_text_start;
2488 if (start) {
2489 end = start + mod->noinstr_text_size;
2490 kprobe_add_area_blacklist(start, end);
2491 }
2492 }
2493
remove_module_kprobe_blacklist(struct module * mod)2494 static void remove_module_kprobe_blacklist(struct module *mod)
2495 {
2496 unsigned long start, end;
2497 int i;
2498
2499 if (mod->kprobe_blacklist) {
2500 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2501 kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2502 }
2503
2504 start = (unsigned long)mod->kprobes_text_start;
2505 if (start) {
2506 end = start + mod->kprobes_text_size;
2507 kprobe_remove_area_blacklist(start, end);
2508 }
2509
2510 start = (unsigned long)mod->noinstr_text_start;
2511 if (start) {
2512 end = start + mod->noinstr_text_size;
2513 kprobe_remove_area_blacklist(start, end);
2514 }
2515 }
2516
2517 /* Module notifier call back, checking kprobes on the module */
kprobes_module_callback(struct notifier_block * nb,unsigned long val,void * data)2518 static int kprobes_module_callback(struct notifier_block *nb,
2519 unsigned long val, void *data)
2520 {
2521 struct module *mod = data;
2522 struct hlist_head *head;
2523 struct kprobe *p;
2524 unsigned int i;
2525 int checkcore = (val == MODULE_STATE_GOING);
2526
2527 if (val == MODULE_STATE_COMING) {
2528 mutex_lock(&kprobe_mutex);
2529 add_module_kprobe_blacklist(mod);
2530 mutex_unlock(&kprobe_mutex);
2531 }
2532 if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2533 return NOTIFY_DONE;
2534
2535 /*
2536 * When MODULE_STATE_GOING was notified, both of module .text and
2537 * .init.text sections would be freed. When MODULE_STATE_LIVE was
2538 * notified, only .init.text section would be freed. We need to
2539 * disable kprobes which have been inserted in the sections.
2540 */
2541 mutex_lock(&kprobe_mutex);
2542 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2543 head = &kprobe_table[i];
2544 hlist_for_each_entry(p, head, hlist) {
2545 if (kprobe_gone(p))
2546 continue;
2547
2548 if (within_module_init((unsigned long)p->addr, mod) ||
2549 (checkcore &&
2550 within_module_core((unsigned long)p->addr, mod))) {
2551 /*
2552 * The vaddr this probe is installed will soon
2553 * be vfreed buy not synced to disk. Hence,
2554 * disarming the breakpoint isn't needed.
2555 *
2556 * Note, this will also move any optimized probes
2557 * that are pending to be removed from their
2558 * corresponding lists to the freeing_list and
2559 * will not be touched by the delayed
2560 * kprobe_optimizer work handler.
2561 */
2562 kill_kprobe(p);
2563 }
2564 }
2565 }
2566 if (val == MODULE_STATE_GOING)
2567 remove_module_kprobe_blacklist(mod);
2568 mutex_unlock(&kprobe_mutex);
2569 return NOTIFY_DONE;
2570 }
2571
2572 static struct notifier_block kprobe_module_nb = {
2573 .notifier_call = kprobes_module_callback,
2574 .priority = 0
2575 };
2576
2577 /* Markers of _kprobe_blacklist section */
2578 extern unsigned long __start_kprobe_blacklist[];
2579 extern unsigned long __stop_kprobe_blacklist[];
2580
kprobe_free_init_mem(void)2581 void kprobe_free_init_mem(void)
2582 {
2583 void *start = (void *)(&__init_begin);
2584 void *end = (void *)(&__init_end);
2585 struct hlist_head *head;
2586 struct kprobe *p;
2587 int i;
2588
2589 mutex_lock(&kprobe_mutex);
2590
2591 /* Kill all kprobes on initmem */
2592 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2593 head = &kprobe_table[i];
2594 hlist_for_each_entry(p, head, hlist) {
2595 if (start <= (void *)p->addr && (void *)p->addr < end)
2596 kill_kprobe(p);
2597 }
2598 }
2599
2600 mutex_unlock(&kprobe_mutex);
2601 }
2602
init_kprobes(void)2603 static int __init init_kprobes(void)
2604 {
2605 int i, err = 0;
2606
2607 /* FIXME allocate the probe table, currently defined statically */
2608 /* initialize all list heads */
2609 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2610 INIT_HLIST_HEAD(&kprobe_table[i]);
2611 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2612 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2613 }
2614
2615 err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2616 __stop_kprobe_blacklist);
2617 if (err) {
2618 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2619 pr_err("Please take care of using kprobes.\n");
2620 }
2621
2622 if (kretprobe_blacklist_size) {
2623 /* lookup the function address from its name */
2624 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2625 kretprobe_blacklist[i].addr =
2626 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2627 if (!kretprobe_blacklist[i].addr)
2628 printk("kretprobe: lookup failed: %s\n",
2629 kretprobe_blacklist[i].name);
2630 }
2631 }
2632
2633 /* By default, kprobes are armed */
2634 kprobes_all_disarmed = false;
2635
2636 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2637 /* Init kprobe_optinsn_slots for allocation */
2638 kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2639 #endif
2640
2641 err = arch_init_kprobes();
2642 if (!err)
2643 err = register_die_notifier(&kprobe_exceptions_nb);
2644 if (!err)
2645 err = register_module_notifier(&kprobe_module_nb);
2646
2647 kprobes_initialized = (err == 0);
2648
2649 if (!err)
2650 init_test_probes();
2651 return err;
2652 }
2653 early_initcall(init_kprobes);
2654
2655 #if defined(CONFIG_OPTPROBES)
init_optprobes(void)2656 static int __init init_optprobes(void)
2657 {
2658 /*
2659 * Enable kprobe optimization - this kicks the optimizer which
2660 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2661 * not spawned in early initcall. So delay the optimization.
2662 */
2663 optimize_all_kprobes();
2664
2665 return 0;
2666 }
2667 subsys_initcall(init_optprobes);
2668 #endif
2669
2670 #ifdef CONFIG_DEBUG_FS
report_probe(struct seq_file * pi,struct kprobe * p,const char * sym,int offset,char * modname,struct kprobe * pp)2671 static void report_probe(struct seq_file *pi, struct kprobe *p,
2672 const char *sym, int offset, char *modname, struct kprobe *pp)
2673 {
2674 char *kprobe_type;
2675 void *addr = p->addr;
2676
2677 if (p->pre_handler == pre_handler_kretprobe)
2678 kprobe_type = "r";
2679 else
2680 kprobe_type = "k";
2681
2682 if (!kallsyms_show_value(pi->file->f_cred))
2683 addr = NULL;
2684
2685 if (sym)
2686 seq_printf(pi, "%px %s %s+0x%x %s ",
2687 addr, kprobe_type, sym, offset,
2688 (modname ? modname : " "));
2689 else /* try to use %pS */
2690 seq_printf(pi, "%px %s %pS ",
2691 addr, kprobe_type, p->addr);
2692
2693 if (!pp)
2694 pp = p;
2695 seq_printf(pi, "%s%s%s%s\n",
2696 (kprobe_gone(p) ? "[GONE]" : ""),
2697 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2698 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2699 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2700 }
2701
kprobe_seq_start(struct seq_file * f,loff_t * pos)2702 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2703 {
2704 return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2705 }
2706
kprobe_seq_next(struct seq_file * f,void * v,loff_t * pos)2707 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2708 {
2709 (*pos)++;
2710 if (*pos >= KPROBE_TABLE_SIZE)
2711 return NULL;
2712 return pos;
2713 }
2714
kprobe_seq_stop(struct seq_file * f,void * v)2715 static void kprobe_seq_stop(struct seq_file *f, void *v)
2716 {
2717 /* Nothing to do */
2718 }
2719
show_kprobe_addr(struct seq_file * pi,void * v)2720 static int show_kprobe_addr(struct seq_file *pi, void *v)
2721 {
2722 struct hlist_head *head;
2723 struct kprobe *p, *kp;
2724 const char *sym = NULL;
2725 unsigned int i = *(loff_t *) v;
2726 unsigned long offset = 0;
2727 char *modname, namebuf[KSYM_NAME_LEN];
2728
2729 head = &kprobe_table[i];
2730 preempt_disable();
2731 hlist_for_each_entry_rcu(p, head, hlist) {
2732 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2733 &offset, &modname, namebuf);
2734 if (kprobe_aggrprobe(p)) {
2735 list_for_each_entry_rcu(kp, &p->list, list)
2736 report_probe(pi, kp, sym, offset, modname, p);
2737 } else
2738 report_probe(pi, p, sym, offset, modname, NULL);
2739 }
2740 preempt_enable();
2741 return 0;
2742 }
2743
2744 static const struct seq_operations kprobes_sops = {
2745 .start = kprobe_seq_start,
2746 .next = kprobe_seq_next,
2747 .stop = kprobe_seq_stop,
2748 .show = show_kprobe_addr
2749 };
2750
2751 DEFINE_SEQ_ATTRIBUTE(kprobes);
2752
2753 /* kprobes/blacklist -- shows which functions can not be probed */
kprobe_blacklist_seq_start(struct seq_file * m,loff_t * pos)2754 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2755 {
2756 mutex_lock(&kprobe_mutex);
2757 return seq_list_start(&kprobe_blacklist, *pos);
2758 }
2759
kprobe_blacklist_seq_next(struct seq_file * m,void * v,loff_t * pos)2760 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2761 {
2762 return seq_list_next(v, &kprobe_blacklist, pos);
2763 }
2764
kprobe_blacklist_seq_show(struct seq_file * m,void * v)2765 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2766 {
2767 struct kprobe_blacklist_entry *ent =
2768 list_entry(v, struct kprobe_blacklist_entry, list);
2769
2770 /*
2771 * If /proc/kallsyms is not showing kernel address, we won't
2772 * show them here either.
2773 */
2774 if (!kallsyms_show_value(m->file->f_cred))
2775 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2776 (void *)ent->start_addr);
2777 else
2778 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2779 (void *)ent->end_addr, (void *)ent->start_addr);
2780 return 0;
2781 }
2782
kprobe_blacklist_seq_stop(struct seq_file * f,void * v)2783 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2784 {
2785 mutex_unlock(&kprobe_mutex);
2786 }
2787
2788 static const struct seq_operations kprobe_blacklist_sops = {
2789 .start = kprobe_blacklist_seq_start,
2790 .next = kprobe_blacklist_seq_next,
2791 .stop = kprobe_blacklist_seq_stop,
2792 .show = kprobe_blacklist_seq_show,
2793 };
2794 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2795
arm_all_kprobes(void)2796 static int arm_all_kprobes(void)
2797 {
2798 struct hlist_head *head;
2799 struct kprobe *p;
2800 unsigned int i, total = 0, errors = 0;
2801 int err, ret = 0;
2802
2803 mutex_lock(&kprobe_mutex);
2804
2805 /* If kprobes are armed, just return */
2806 if (!kprobes_all_disarmed)
2807 goto already_enabled;
2808
2809 /*
2810 * optimize_kprobe() called by arm_kprobe() checks
2811 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2812 * arm_kprobe.
2813 */
2814 kprobes_all_disarmed = false;
2815 /* Arming kprobes doesn't optimize kprobe itself */
2816 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2817 head = &kprobe_table[i];
2818 /* Arm all kprobes on a best-effort basis */
2819 hlist_for_each_entry(p, head, hlist) {
2820 if (!kprobe_disabled(p)) {
2821 err = arm_kprobe(p);
2822 if (err) {
2823 errors++;
2824 ret = err;
2825 }
2826 total++;
2827 }
2828 }
2829 }
2830
2831 if (errors)
2832 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2833 errors, total);
2834 else
2835 pr_info("Kprobes globally enabled\n");
2836
2837 already_enabled:
2838 mutex_unlock(&kprobe_mutex);
2839 return ret;
2840 }
2841
disarm_all_kprobes(void)2842 static int disarm_all_kprobes(void)
2843 {
2844 struct hlist_head *head;
2845 struct kprobe *p;
2846 unsigned int i, total = 0, errors = 0;
2847 int err, ret = 0;
2848
2849 mutex_lock(&kprobe_mutex);
2850
2851 /* If kprobes are already disarmed, just return */
2852 if (kprobes_all_disarmed) {
2853 mutex_unlock(&kprobe_mutex);
2854 return 0;
2855 }
2856
2857 kprobes_all_disarmed = true;
2858
2859 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2860 head = &kprobe_table[i];
2861 /* Disarm all kprobes on a best-effort basis */
2862 hlist_for_each_entry(p, head, hlist) {
2863 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2864 err = disarm_kprobe(p, false);
2865 if (err) {
2866 errors++;
2867 ret = err;
2868 }
2869 total++;
2870 }
2871 }
2872 }
2873
2874 if (errors)
2875 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2876 errors, total);
2877 else
2878 pr_info("Kprobes globally disabled\n");
2879
2880 mutex_unlock(&kprobe_mutex);
2881
2882 /* Wait for disarming all kprobes by optimizer */
2883 wait_for_kprobe_optimizer();
2884
2885 return ret;
2886 }
2887
2888 /*
2889 * XXX: The debugfs bool file interface doesn't allow for callbacks
2890 * when the bool state is switched. We can reuse that facility when
2891 * available
2892 */
read_enabled_file_bool(struct file * file,char __user * user_buf,size_t count,loff_t * ppos)2893 static ssize_t read_enabled_file_bool(struct file *file,
2894 char __user *user_buf, size_t count, loff_t *ppos)
2895 {
2896 char buf[3];
2897
2898 if (!kprobes_all_disarmed)
2899 buf[0] = '1';
2900 else
2901 buf[0] = '0';
2902 buf[1] = '\n';
2903 buf[2] = 0x00;
2904 return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2905 }
2906
write_enabled_file_bool(struct file * file,const char __user * user_buf,size_t count,loff_t * ppos)2907 static ssize_t write_enabled_file_bool(struct file *file,
2908 const char __user *user_buf, size_t count, loff_t *ppos)
2909 {
2910 char buf[32];
2911 size_t buf_size;
2912 int ret = 0;
2913
2914 buf_size = min(count, (sizeof(buf)-1));
2915 if (copy_from_user(buf, user_buf, buf_size))
2916 return -EFAULT;
2917
2918 buf[buf_size] = '\0';
2919 switch (buf[0]) {
2920 case 'y':
2921 case 'Y':
2922 case '1':
2923 ret = arm_all_kprobes();
2924 break;
2925 case 'n':
2926 case 'N':
2927 case '0':
2928 ret = disarm_all_kprobes();
2929 break;
2930 default:
2931 return -EINVAL;
2932 }
2933
2934 if (ret)
2935 return ret;
2936
2937 return count;
2938 }
2939
2940 static const struct file_operations fops_kp = {
2941 .read = read_enabled_file_bool,
2942 .write = write_enabled_file_bool,
2943 .llseek = default_llseek,
2944 };
2945
debugfs_kprobe_init(void)2946 static int __init debugfs_kprobe_init(void)
2947 {
2948 struct dentry *dir;
2949
2950 dir = debugfs_create_dir("kprobes", NULL);
2951
2952 debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2953
2954 debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
2955
2956 debugfs_create_file("blacklist", 0400, dir, NULL,
2957 &kprobe_blacklist_fops);
2958
2959 return 0;
2960 }
2961
2962 late_initcall(debugfs_kprobe_init);
2963 #endif /* CONFIG_DEBUG_FS */
2964