xref: /openbmc/linux/kernel/kprobes.c (revision a4155dec)
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 
55 static 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  */
61 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
62 
63 /* NOTE: change this value only with 'kprobe_mutex' held */
64 static bool kprobes_all_disarmed;
65 
66 /* This protects 'kprobe_table' and 'optimizing_list' */
67 static DEFINE_MUTEX(kprobe_mutex);
68 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
69 
kprobe_lookup_name(const char * name,unsigned int __unused)70 kprobe_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  */
80 static 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  */
89 struct 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 
slots_per_page(struct kprobe_insn_cache * c)102 static int slots_per_page(struct kprobe_insn_cache *c)
103 {
104 	return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
105 }
106 
107 enum kprobe_slot_state {
108 	SLOT_CLEAN = 0,
109 	SLOT_DIRTY = 1,
110 	SLOT_USED = 2,
111 };
112 
alloc_insn_page(void)113 void __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 
free_insn_page(void * page)124 static void free_insn_page(void *page)
125 {
126 	module_memfree(page);
127 }
128 
129 struct 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 };
138 static 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  */
__get_insn_slot(struct kprobe_insn_cache * c)144 kprobe_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);
199 out:
200 	mutex_unlock(&c->mutex);
201 	return slot;
202 }
203 
204 /* Return true if all garbages are collected, otherwise false. */
collect_one_slot(struct kprobe_insn_page * kip,int idx)205 static 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 
collect_garbage_slots(struct kprobe_insn_cache * c)234 static 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 
__free_insn_slot(struct kprobe_insn_cache * c,kprobe_opcode_t * slot,int dirty)256 void __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;
273 out:
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  */
__is_insn_slot_addr(struct kprobe_insn_cache * c,unsigned long addr)296 bool __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 
kprobe_cache_get_kallsym(struct kprobe_insn_cache * c,unsigned int * symnum,unsigned long * value,char * type,char * sym)314 int 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
alloc_optinsn_page(void)336 void __weak *alloc_optinsn_page(void)
337 {
338 	return alloc_insn_page();
339 }
340 
free_optinsn_page(void * page)341 void __weak free_optinsn_page(void *page)
342 {
343 	free_insn_page(page);
344 }
345 
346 /* For optimized_kprobe buffer */
347 struct 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 */
set_kprobe_instance(struct kprobe * kp)360 static inline void set_kprobe_instance(struct kprobe *kp)
361 {
362 	__this_cpu_write(kprobe_instance, kp);
363 }
364 
reset_kprobe_instance(void)365 static 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  */
get_kprobe(void * addr)376 struct 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 }
390 NOKPROBE_SYMBOL(get_kprobe);
391 
392 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
393 
394 /* Return true if 'p' is an aggregator */
kprobe_aggrprobe(struct kprobe * p)395 static 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 */
kprobe_unused(struct kprobe * p)401 static 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. */
copy_kprobe(struct kprobe * ap,struct kprobe * p)408 static 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'. */
416 static 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  */
opt_pre_handler(struct kprobe * p,struct pt_regs * regs)422 void 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 }
434 NOKPROBE_SYMBOL(opt_pre_handler);
435 
436 /* Free optimized instructions and optimized_kprobe */
free_aggr_kprobe(struct kprobe * p)437 static 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. */
kprobe_optready(struct kprobe * p)448 static 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 */
kprobe_disarmed(struct kprobe * p)461 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 */
kprobe_queued(struct kprobe * p)475 static 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  */
get_optimized_kprobe(kprobe_opcode_t * addr)491 static 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' */
511 static LIST_HEAD(optimizing_list);
512 static LIST_HEAD(unoptimizing_list);
513 static LIST_HEAD(freeing_list);
514 
515 static void kprobe_optimizer(struct work_struct *work);
516 static 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  */
do_optimize_kprobes(void)523 static 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  */
do_unoptimize_kprobes(void)550 static void do_unoptimize_kprobes(void)
551 {
552 	struct optimized_kprobe *op, *tmp;
553 
554 	lockdep_assert_held(&text_mutex);
555 	/* See comment in do_optimize_kprobes() */
556 	lockdep_assert_cpus_held();
557 
558 	if (!list_empty(&unoptimizing_list))
559 		arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
560 
561 	/* Loop on 'freeing_list' for disarming and removing from kprobe hash list */
562 	list_for_each_entry_safe(op, tmp, &freeing_list, list) {
563 		/* Switching from detour code to origin */
564 		op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
565 		/* Disarm probes if marked disabled and not gone */
566 		if (kprobe_disabled(&op->kp) && !kprobe_gone(&op->kp))
567 			arch_disarm_kprobe(&op->kp);
568 		if (kprobe_unused(&op->kp)) {
569 			/*
570 			 * Remove unused probes from hash list. After waiting
571 			 * for synchronization, these probes are reclaimed.
572 			 * (reclaiming is done by do_free_cleaned_kprobes().)
573 			 */
574 			hlist_del_rcu(&op->kp.hlist);
575 		} else
576 			list_del_init(&op->list);
577 	}
578 }
579 
580 /* Reclaim all kprobes on the 'freeing_list' */
do_free_cleaned_kprobes(void)581 static void do_free_cleaned_kprobes(void)
582 {
583 	struct optimized_kprobe *op, *tmp;
584 
585 	list_for_each_entry_safe(op, tmp, &freeing_list, list) {
586 		list_del_init(&op->list);
587 		if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
588 			/*
589 			 * This must not happen, but if there is a kprobe
590 			 * still in use, keep it on kprobes hash list.
591 			 */
592 			continue;
593 		}
594 		free_aggr_kprobe(&op->kp);
595 	}
596 }
597 
598 /* Start optimizer after OPTIMIZE_DELAY passed */
kick_kprobe_optimizer(void)599 static void kick_kprobe_optimizer(void)
600 {
601 	schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
602 }
603 
604 /* Kprobe jump optimizer */
kprobe_optimizer(struct work_struct * work)605 static void kprobe_optimizer(struct work_struct *work)
606 {
607 	mutex_lock(&kprobe_mutex);
608 	cpus_read_lock();
609 	mutex_lock(&text_mutex);
610 
611 	/*
612 	 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
613 	 * kprobes before waiting for quiesence period.
614 	 */
615 	do_unoptimize_kprobes();
616 
617 	/*
618 	 * Step 2: Wait for quiesence period to ensure all potentially
619 	 * preempted tasks to have normally scheduled. Because optprobe
620 	 * may modify multiple instructions, there is a chance that Nth
621 	 * instruction is preempted. In that case, such tasks can return
622 	 * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
623 	 * Note that on non-preemptive kernel, this is transparently converted
624 	 * to synchronoze_sched() to wait for all interrupts to have completed.
625 	 */
626 	synchronize_rcu_tasks();
627 
628 	/* Step 3: Optimize kprobes after quiesence period */
629 	do_optimize_kprobes();
630 
631 	/* Step 4: Free cleaned kprobes after quiesence period */
632 	do_free_cleaned_kprobes();
633 
634 	mutex_unlock(&text_mutex);
635 	cpus_read_unlock();
636 
637 	/* Step 5: Kick optimizer again if needed */
638 	if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
639 		kick_kprobe_optimizer();
640 
641 	mutex_unlock(&kprobe_mutex);
642 }
643 
644 /* Wait for completing optimization and unoptimization */
wait_for_kprobe_optimizer(void)645 void wait_for_kprobe_optimizer(void)
646 {
647 	mutex_lock(&kprobe_mutex);
648 
649 	while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
650 		mutex_unlock(&kprobe_mutex);
651 
652 		/* This will also make 'optimizing_work' execute immmediately */
653 		flush_delayed_work(&optimizing_work);
654 		/* 'optimizing_work' might not have been queued yet, relax */
655 		cpu_relax();
656 
657 		mutex_lock(&kprobe_mutex);
658 	}
659 
660 	mutex_unlock(&kprobe_mutex);
661 }
662 
optprobe_queued_unopt(struct optimized_kprobe * op)663 bool optprobe_queued_unopt(struct optimized_kprobe *op)
664 {
665 	struct optimized_kprobe *_op;
666 
667 	list_for_each_entry(_op, &unoptimizing_list, list) {
668 		if (op == _op)
669 			return true;
670 	}
671 
672 	return false;
673 }
674 
675 /* Optimize kprobe if p is ready to be optimized */
optimize_kprobe(struct kprobe * p)676 static void optimize_kprobe(struct kprobe *p)
677 {
678 	struct optimized_kprobe *op;
679 
680 	/* Check if the kprobe is disabled or not ready for optimization. */
681 	if (!kprobe_optready(p) || !kprobes_allow_optimization ||
682 	    (kprobe_disabled(p) || kprobes_all_disarmed))
683 		return;
684 
685 	/* kprobes with 'post_handler' can not be optimized */
686 	if (p->post_handler)
687 		return;
688 
689 	op = container_of(p, struct optimized_kprobe, kp);
690 
691 	/* Check there is no other kprobes at the optimized instructions */
692 	if (arch_check_optimized_kprobe(op) < 0)
693 		return;
694 
695 	/* Check if it is already optimized. */
696 	if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
697 		if (optprobe_queued_unopt(op)) {
698 			/* This is under unoptimizing. Just dequeue the probe */
699 			list_del_init(&op->list);
700 		}
701 		return;
702 	}
703 	op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
704 
705 	/*
706 	 * On the 'unoptimizing_list' and 'optimizing_list',
707 	 * 'op' must have OPTIMIZED flag
708 	 */
709 	if (WARN_ON_ONCE(!list_empty(&op->list)))
710 		return;
711 
712 	list_add(&op->list, &optimizing_list);
713 	kick_kprobe_optimizer();
714 }
715 
716 /* Short cut to direct unoptimizing */
force_unoptimize_kprobe(struct optimized_kprobe * op)717 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
718 {
719 	lockdep_assert_cpus_held();
720 	arch_unoptimize_kprobe(op);
721 	op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
722 }
723 
724 /* Unoptimize a kprobe if p is optimized */
unoptimize_kprobe(struct kprobe * p,bool force)725 static void unoptimize_kprobe(struct kprobe *p, bool force)
726 {
727 	struct optimized_kprobe *op;
728 
729 	if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
730 		return; /* This is not an optprobe nor optimized */
731 
732 	op = container_of(p, struct optimized_kprobe, kp);
733 	if (!kprobe_optimized(p))
734 		return;
735 
736 	if (!list_empty(&op->list)) {
737 		if (optprobe_queued_unopt(op)) {
738 			/* Queued in unoptimizing queue */
739 			if (force) {
740 				/*
741 				 * Forcibly unoptimize the kprobe here, and queue it
742 				 * in the freeing list for release afterwards.
743 				 */
744 				force_unoptimize_kprobe(op);
745 				list_move(&op->list, &freeing_list);
746 			}
747 		} else {
748 			/* Dequeue from the optimizing queue */
749 			list_del_init(&op->list);
750 			op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
751 		}
752 		return;
753 	}
754 
755 	/* Optimized kprobe case */
756 	if (force) {
757 		/* Forcibly update the code: this is a special case */
758 		force_unoptimize_kprobe(op);
759 	} else {
760 		list_add(&op->list, &unoptimizing_list);
761 		kick_kprobe_optimizer();
762 	}
763 }
764 
765 /* Cancel unoptimizing for reusing */
reuse_unused_kprobe(struct kprobe * ap)766 static int reuse_unused_kprobe(struct kprobe *ap)
767 {
768 	struct optimized_kprobe *op;
769 
770 	/*
771 	 * Unused kprobe MUST be on the way of delayed unoptimizing (means
772 	 * there is still a relative jump) and disabled.
773 	 */
774 	op = container_of(ap, struct optimized_kprobe, kp);
775 	WARN_ON_ONCE(list_empty(&op->list));
776 	/* Enable the probe again */
777 	ap->flags &= ~KPROBE_FLAG_DISABLED;
778 	/* Optimize it again. (remove from 'op->list') */
779 	if (!kprobe_optready(ap))
780 		return -EINVAL;
781 
782 	optimize_kprobe(ap);
783 	return 0;
784 }
785 
786 /* Remove optimized instructions */
kill_optimized_kprobe(struct kprobe * p)787 static void kill_optimized_kprobe(struct kprobe *p)
788 {
789 	struct optimized_kprobe *op;
790 
791 	op = container_of(p, struct optimized_kprobe, kp);
792 	if (!list_empty(&op->list))
793 		/* Dequeue from the (un)optimization queue */
794 		list_del_init(&op->list);
795 	op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
796 
797 	if (kprobe_unused(p)) {
798 		/*
799 		 * Unused kprobe is on unoptimizing or freeing list. We move it
800 		 * to freeing_list and let the kprobe_optimizer() remove it from
801 		 * the kprobe hash list and free it.
802 		 */
803 		if (optprobe_queued_unopt(op))
804 			list_move(&op->list, &freeing_list);
805 	}
806 
807 	/* Don't touch the code, because it is already freed. */
808 	arch_remove_optimized_kprobe(op);
809 }
810 
811 static inline
__prepare_optimized_kprobe(struct optimized_kprobe * op,struct kprobe * p)812 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
813 {
814 	if (!kprobe_ftrace(p))
815 		arch_prepare_optimized_kprobe(op, p);
816 }
817 
818 /* Try to prepare optimized instructions */
prepare_optimized_kprobe(struct kprobe * p)819 static void prepare_optimized_kprobe(struct kprobe *p)
820 {
821 	struct optimized_kprobe *op;
822 
823 	op = container_of(p, struct optimized_kprobe, kp);
824 	__prepare_optimized_kprobe(op, p);
825 }
826 
827 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
alloc_aggr_kprobe(struct kprobe * p)828 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
829 {
830 	struct optimized_kprobe *op;
831 
832 	op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
833 	if (!op)
834 		return NULL;
835 
836 	INIT_LIST_HEAD(&op->list);
837 	op->kp.addr = p->addr;
838 	__prepare_optimized_kprobe(op, p);
839 
840 	return &op->kp;
841 }
842 
843 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
844 
845 /*
846  * Prepare an optimized_kprobe and optimize it.
847  * NOTE: 'p' must be a normal registered kprobe.
848  */
try_to_optimize_kprobe(struct kprobe * p)849 static void try_to_optimize_kprobe(struct kprobe *p)
850 {
851 	struct kprobe *ap;
852 	struct optimized_kprobe *op;
853 
854 	/* Impossible to optimize ftrace-based kprobe. */
855 	if (kprobe_ftrace(p))
856 		return;
857 
858 	/* For preparing optimization, jump_label_text_reserved() is called. */
859 	cpus_read_lock();
860 	jump_label_lock();
861 	mutex_lock(&text_mutex);
862 
863 	ap = alloc_aggr_kprobe(p);
864 	if (!ap)
865 		goto out;
866 
867 	op = container_of(ap, struct optimized_kprobe, kp);
868 	if (!arch_prepared_optinsn(&op->optinsn)) {
869 		/* If failed to setup optimizing, fallback to kprobe. */
870 		arch_remove_optimized_kprobe(op);
871 		kfree(op);
872 		goto out;
873 	}
874 
875 	init_aggr_kprobe(ap, p);
876 	optimize_kprobe(ap);	/* This just kicks optimizer thread. */
877 
878 out:
879 	mutex_unlock(&text_mutex);
880 	jump_label_unlock();
881 	cpus_read_unlock();
882 }
883 
optimize_all_kprobes(void)884 static void optimize_all_kprobes(void)
885 {
886 	struct hlist_head *head;
887 	struct kprobe *p;
888 	unsigned int i;
889 
890 	mutex_lock(&kprobe_mutex);
891 	/* If optimization is already allowed, just return. */
892 	if (kprobes_allow_optimization)
893 		goto out;
894 
895 	cpus_read_lock();
896 	kprobes_allow_optimization = true;
897 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
898 		head = &kprobe_table[i];
899 		hlist_for_each_entry(p, head, hlist)
900 			if (!kprobe_disabled(p))
901 				optimize_kprobe(p);
902 	}
903 	cpus_read_unlock();
904 	pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
905 out:
906 	mutex_unlock(&kprobe_mutex);
907 }
908 
909 #ifdef CONFIG_SYSCTL
unoptimize_all_kprobes(void)910 static void unoptimize_all_kprobes(void)
911 {
912 	struct hlist_head *head;
913 	struct kprobe *p;
914 	unsigned int i;
915 
916 	mutex_lock(&kprobe_mutex);
917 	/* If optimization is already prohibited, just return. */
918 	if (!kprobes_allow_optimization) {
919 		mutex_unlock(&kprobe_mutex);
920 		return;
921 	}
922 
923 	cpus_read_lock();
924 	kprobes_allow_optimization = false;
925 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
926 		head = &kprobe_table[i];
927 		hlist_for_each_entry(p, head, hlist) {
928 			if (!kprobe_disabled(p))
929 				unoptimize_kprobe(p, false);
930 		}
931 	}
932 	cpus_read_unlock();
933 	mutex_unlock(&kprobe_mutex);
934 
935 	/* Wait for unoptimizing completion. */
936 	wait_for_kprobe_optimizer();
937 	pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
938 }
939 
940 static DEFINE_MUTEX(kprobe_sysctl_mutex);
941 static int sysctl_kprobes_optimization;
proc_kprobes_optimization_handler(struct ctl_table * table,int write,void * buffer,size_t * length,loff_t * ppos)942 static int proc_kprobes_optimization_handler(struct ctl_table *table,
943 					     int write, void *buffer,
944 					     size_t *length, loff_t *ppos)
945 {
946 	int ret;
947 
948 	mutex_lock(&kprobe_sysctl_mutex);
949 	sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
950 	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
951 
952 	if (sysctl_kprobes_optimization)
953 		optimize_all_kprobes();
954 	else
955 		unoptimize_all_kprobes();
956 	mutex_unlock(&kprobe_sysctl_mutex);
957 
958 	return ret;
959 }
960 
961 static struct ctl_table kprobe_sysctls[] = {
962 	{
963 		.procname	= "kprobes-optimization",
964 		.data		= &sysctl_kprobes_optimization,
965 		.maxlen		= sizeof(int),
966 		.mode		= 0644,
967 		.proc_handler	= proc_kprobes_optimization_handler,
968 		.extra1		= SYSCTL_ZERO,
969 		.extra2		= SYSCTL_ONE,
970 	},
971 	{}
972 };
973 
kprobe_sysctls_init(void)974 static void __init kprobe_sysctls_init(void)
975 {
976 	register_sysctl_init("debug", kprobe_sysctls);
977 }
978 #endif /* CONFIG_SYSCTL */
979 
980 /* Put a breakpoint for a probe. */
__arm_kprobe(struct kprobe * p)981 static void __arm_kprobe(struct kprobe *p)
982 {
983 	struct kprobe *_p;
984 
985 	lockdep_assert_held(&text_mutex);
986 
987 	/* Find the overlapping optimized kprobes. */
988 	_p = get_optimized_kprobe(p->addr);
989 	if (unlikely(_p))
990 		/* Fallback to unoptimized kprobe */
991 		unoptimize_kprobe(_p, true);
992 
993 	arch_arm_kprobe(p);
994 	optimize_kprobe(p);	/* Try to optimize (add kprobe to a list) */
995 }
996 
997 /* Remove the breakpoint of a probe. */
__disarm_kprobe(struct kprobe * p,bool reopt)998 static void __disarm_kprobe(struct kprobe *p, bool reopt)
999 {
1000 	struct kprobe *_p;
1001 
1002 	lockdep_assert_held(&text_mutex);
1003 
1004 	/* Try to unoptimize */
1005 	unoptimize_kprobe(p, kprobes_all_disarmed);
1006 
1007 	if (!kprobe_queued(p)) {
1008 		arch_disarm_kprobe(p);
1009 		/* If another kprobe was blocked, re-optimize it. */
1010 		_p = get_optimized_kprobe(p->addr);
1011 		if (unlikely(_p) && reopt)
1012 			optimize_kprobe(_p);
1013 	}
1014 	/*
1015 	 * TODO: Since unoptimization and real disarming will be done by
1016 	 * the worker thread, we can not check whether another probe are
1017 	 * unoptimized because of this probe here. It should be re-optimized
1018 	 * by the worker thread.
1019 	 */
1020 }
1021 
1022 #else /* !CONFIG_OPTPROBES */
1023 
1024 #define optimize_kprobe(p)			do {} while (0)
1025 #define unoptimize_kprobe(p, f)			do {} while (0)
1026 #define kill_optimized_kprobe(p)		do {} while (0)
1027 #define prepare_optimized_kprobe(p)		do {} while (0)
1028 #define try_to_optimize_kprobe(p)		do {} while (0)
1029 #define __arm_kprobe(p)				arch_arm_kprobe(p)
1030 #define __disarm_kprobe(p, o)			arch_disarm_kprobe(p)
1031 #define kprobe_disarmed(p)			kprobe_disabled(p)
1032 #define wait_for_kprobe_optimizer()		do {} while (0)
1033 
reuse_unused_kprobe(struct kprobe * ap)1034 static int reuse_unused_kprobe(struct kprobe *ap)
1035 {
1036 	/*
1037 	 * If the optimized kprobe is NOT supported, the aggr kprobe is
1038 	 * released at the same time that the last aggregated kprobe is
1039 	 * unregistered.
1040 	 * Thus there should be no chance to reuse unused kprobe.
1041 	 */
1042 	WARN_ON_ONCE(1);
1043 	return -EINVAL;
1044 }
1045 
free_aggr_kprobe(struct kprobe * p)1046 static void free_aggr_kprobe(struct kprobe *p)
1047 {
1048 	arch_remove_kprobe(p);
1049 	kfree(p);
1050 }
1051 
alloc_aggr_kprobe(struct kprobe * p)1052 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1053 {
1054 	return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1055 }
1056 #endif /* CONFIG_OPTPROBES */
1057 
1058 #ifdef CONFIG_KPROBES_ON_FTRACE
1059 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1060 	.func = kprobe_ftrace_handler,
1061 	.flags = FTRACE_OPS_FL_SAVE_REGS,
1062 };
1063 
1064 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1065 	.func = kprobe_ftrace_handler,
1066 	.flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1067 };
1068 
1069 static int kprobe_ipmodify_enabled;
1070 static int kprobe_ftrace_enabled;
1071 bool kprobe_ftrace_disabled;
1072 
__arm_kprobe_ftrace(struct kprobe * p,struct ftrace_ops * ops,int * cnt)1073 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1074 			       int *cnt)
1075 {
1076 	int ret;
1077 
1078 	lockdep_assert_held(&kprobe_mutex);
1079 
1080 	ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1081 	if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1082 		return ret;
1083 
1084 	if (*cnt == 0) {
1085 		ret = register_ftrace_function(ops);
1086 		if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1087 			goto err_ftrace;
1088 	}
1089 
1090 	(*cnt)++;
1091 	return ret;
1092 
1093 err_ftrace:
1094 	/*
1095 	 * At this point, sinec ops is not registered, we should be sefe from
1096 	 * registering empty filter.
1097 	 */
1098 	ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1099 	return ret;
1100 }
1101 
arm_kprobe_ftrace(struct kprobe * p)1102 static int arm_kprobe_ftrace(struct kprobe *p)
1103 {
1104 	bool ipmodify = (p->post_handler != NULL);
1105 
1106 	return __arm_kprobe_ftrace(p,
1107 		ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1108 		ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1109 }
1110 
__disarm_kprobe_ftrace(struct kprobe * p,struct ftrace_ops * ops,int * cnt)1111 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1112 				  int *cnt)
1113 {
1114 	int ret;
1115 
1116 	lockdep_assert_held(&kprobe_mutex);
1117 
1118 	if (*cnt == 1) {
1119 		ret = unregister_ftrace_function(ops);
1120 		if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1121 			return ret;
1122 	}
1123 
1124 	(*cnt)--;
1125 
1126 	ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1127 	WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1128 		  p->addr, ret);
1129 	return ret;
1130 }
1131 
disarm_kprobe_ftrace(struct kprobe * p)1132 static int disarm_kprobe_ftrace(struct kprobe *p)
1133 {
1134 	bool ipmodify = (p->post_handler != NULL);
1135 
1136 	return __disarm_kprobe_ftrace(p,
1137 		ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1138 		ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1139 }
1140 
kprobe_ftrace_kill(void)1141 void kprobe_ftrace_kill(void)
1142 {
1143 	kprobe_ftrace_disabled = true;
1144 }
1145 #else	/* !CONFIG_KPROBES_ON_FTRACE */
arm_kprobe_ftrace(struct kprobe * p)1146 static inline int arm_kprobe_ftrace(struct kprobe *p)
1147 {
1148 	return -ENODEV;
1149 }
1150 
disarm_kprobe_ftrace(struct kprobe * p)1151 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1152 {
1153 	return -ENODEV;
1154 }
1155 #endif
1156 
prepare_kprobe(struct kprobe * p)1157 static int prepare_kprobe(struct kprobe *p)
1158 {
1159 	/* Must ensure p->addr is really on ftrace */
1160 	if (kprobe_ftrace(p))
1161 		return arch_prepare_kprobe_ftrace(p);
1162 
1163 	return arch_prepare_kprobe(p);
1164 }
1165 
arm_kprobe(struct kprobe * kp)1166 static int arm_kprobe(struct kprobe *kp)
1167 {
1168 	if (unlikely(kprobe_ftrace(kp)))
1169 		return arm_kprobe_ftrace(kp);
1170 
1171 	cpus_read_lock();
1172 	mutex_lock(&text_mutex);
1173 	__arm_kprobe(kp);
1174 	mutex_unlock(&text_mutex);
1175 	cpus_read_unlock();
1176 
1177 	return 0;
1178 }
1179 
disarm_kprobe(struct kprobe * kp,bool reopt)1180 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1181 {
1182 	if (unlikely(kprobe_ftrace(kp)))
1183 		return disarm_kprobe_ftrace(kp);
1184 
1185 	cpus_read_lock();
1186 	mutex_lock(&text_mutex);
1187 	__disarm_kprobe(kp, reopt);
1188 	mutex_unlock(&text_mutex);
1189 	cpus_read_unlock();
1190 
1191 	return 0;
1192 }
1193 
1194 /*
1195  * Aggregate handlers for multiple kprobes support - these handlers
1196  * take care of invoking the individual kprobe handlers on p->list
1197  */
aggr_pre_handler(struct kprobe * p,struct pt_regs * regs)1198 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1199 {
1200 	struct kprobe *kp;
1201 
1202 	list_for_each_entry_rcu(kp, &p->list, list) {
1203 		if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1204 			set_kprobe_instance(kp);
1205 			if (kp->pre_handler(kp, regs))
1206 				return 1;
1207 		}
1208 		reset_kprobe_instance();
1209 	}
1210 	return 0;
1211 }
1212 NOKPROBE_SYMBOL(aggr_pre_handler);
1213 
aggr_post_handler(struct kprobe * p,struct pt_regs * regs,unsigned long flags)1214 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1215 			      unsigned long flags)
1216 {
1217 	struct kprobe *kp;
1218 
1219 	list_for_each_entry_rcu(kp, &p->list, list) {
1220 		if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1221 			set_kprobe_instance(kp);
1222 			kp->post_handler(kp, regs, flags);
1223 			reset_kprobe_instance();
1224 		}
1225 	}
1226 }
1227 NOKPROBE_SYMBOL(aggr_post_handler);
1228 
1229 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
kprobes_inc_nmissed_count(struct kprobe * p)1230 void kprobes_inc_nmissed_count(struct kprobe *p)
1231 {
1232 	struct kprobe *kp;
1233 
1234 	if (!kprobe_aggrprobe(p)) {
1235 		p->nmissed++;
1236 	} else {
1237 		list_for_each_entry_rcu(kp, &p->list, list)
1238 			kp->nmissed++;
1239 	}
1240 }
1241 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1242 
1243 static struct kprobe kprobe_busy = {
1244 	.addr = (void *) get_kprobe,
1245 };
1246 
kprobe_busy_begin(void)1247 void kprobe_busy_begin(void)
1248 {
1249 	struct kprobe_ctlblk *kcb;
1250 
1251 	preempt_disable();
1252 	__this_cpu_write(current_kprobe, &kprobe_busy);
1253 	kcb = get_kprobe_ctlblk();
1254 	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1255 }
1256 
kprobe_busy_end(void)1257 void kprobe_busy_end(void)
1258 {
1259 	__this_cpu_write(current_kprobe, NULL);
1260 	preempt_enable();
1261 }
1262 
1263 /* Add the new probe to 'ap->list'. */
add_new_kprobe(struct kprobe * ap,struct kprobe * p)1264 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1265 {
1266 	if (p->post_handler)
1267 		unoptimize_kprobe(ap, true);	/* Fall back to normal kprobe */
1268 
1269 	list_add_rcu(&p->list, &ap->list);
1270 	if (p->post_handler && !ap->post_handler)
1271 		ap->post_handler = aggr_post_handler;
1272 
1273 	return 0;
1274 }
1275 
1276 /*
1277  * Fill in the required fields of the aggregator kprobe. Replace the
1278  * earlier kprobe in the hlist with the aggregator kprobe.
1279  */
init_aggr_kprobe(struct kprobe * ap,struct kprobe * p)1280 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1281 {
1282 	/* Copy the insn slot of 'p' to 'ap'. */
1283 	copy_kprobe(p, ap);
1284 	flush_insn_slot(ap);
1285 	ap->addr = p->addr;
1286 	ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1287 	ap->pre_handler = aggr_pre_handler;
1288 	/* We don't care the kprobe which has gone. */
1289 	if (p->post_handler && !kprobe_gone(p))
1290 		ap->post_handler = aggr_post_handler;
1291 
1292 	INIT_LIST_HEAD(&ap->list);
1293 	INIT_HLIST_NODE(&ap->hlist);
1294 
1295 	list_add_rcu(&p->list, &ap->list);
1296 	hlist_replace_rcu(&p->hlist, &ap->hlist);
1297 }
1298 
1299 /*
1300  * This registers the second or subsequent kprobe at the same address.
1301  */
register_aggr_kprobe(struct kprobe * orig_p,struct kprobe * p)1302 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1303 {
1304 	int ret = 0;
1305 	struct kprobe *ap = orig_p;
1306 
1307 	cpus_read_lock();
1308 
1309 	/* For preparing optimization, jump_label_text_reserved() is called */
1310 	jump_label_lock();
1311 	mutex_lock(&text_mutex);
1312 
1313 	if (!kprobe_aggrprobe(orig_p)) {
1314 		/* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1315 		ap = alloc_aggr_kprobe(orig_p);
1316 		if (!ap) {
1317 			ret = -ENOMEM;
1318 			goto out;
1319 		}
1320 		init_aggr_kprobe(ap, orig_p);
1321 	} else if (kprobe_unused(ap)) {
1322 		/* This probe is going to die. Rescue it */
1323 		ret = reuse_unused_kprobe(ap);
1324 		if (ret)
1325 			goto out;
1326 	}
1327 
1328 	if (kprobe_gone(ap)) {
1329 		/*
1330 		 * Attempting to insert new probe at the same location that
1331 		 * had a probe in the module vaddr area which already
1332 		 * freed. So, the instruction slot has already been
1333 		 * released. We need a new slot for the new probe.
1334 		 */
1335 		ret = arch_prepare_kprobe(ap);
1336 		if (ret)
1337 			/*
1338 			 * Even if fail to allocate new slot, don't need to
1339 			 * free the 'ap'. It will be used next time, or
1340 			 * freed by unregister_kprobe().
1341 			 */
1342 			goto out;
1343 
1344 		/* Prepare optimized instructions if possible. */
1345 		prepare_optimized_kprobe(ap);
1346 
1347 		/*
1348 		 * Clear gone flag to prevent allocating new slot again, and
1349 		 * set disabled flag because it is not armed yet.
1350 		 */
1351 		ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1352 			    | KPROBE_FLAG_DISABLED;
1353 	}
1354 
1355 	/* Copy the insn slot of 'p' to 'ap'. */
1356 	copy_kprobe(ap, p);
1357 	ret = add_new_kprobe(ap, p);
1358 
1359 out:
1360 	mutex_unlock(&text_mutex);
1361 	jump_label_unlock();
1362 	cpus_read_unlock();
1363 
1364 	if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1365 		ap->flags &= ~KPROBE_FLAG_DISABLED;
1366 		if (!kprobes_all_disarmed) {
1367 			/* Arm the breakpoint again. */
1368 			ret = arm_kprobe(ap);
1369 			if (ret) {
1370 				ap->flags |= KPROBE_FLAG_DISABLED;
1371 				list_del_rcu(&p->list);
1372 				synchronize_rcu();
1373 			}
1374 		}
1375 	}
1376 	return ret;
1377 }
1378 
arch_within_kprobe_blacklist(unsigned long addr)1379 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1380 {
1381 	/* The '__kprobes' functions and entry code must not be probed. */
1382 	return addr >= (unsigned long)__kprobes_text_start &&
1383 	       addr < (unsigned long)__kprobes_text_end;
1384 }
1385 
__within_kprobe_blacklist(unsigned long addr)1386 static bool __within_kprobe_blacklist(unsigned long addr)
1387 {
1388 	struct kprobe_blacklist_entry *ent;
1389 
1390 	if (arch_within_kprobe_blacklist(addr))
1391 		return true;
1392 	/*
1393 	 * If 'kprobe_blacklist' is defined, check the address and
1394 	 * reject any probe registration in the prohibited area.
1395 	 */
1396 	list_for_each_entry(ent, &kprobe_blacklist, list) {
1397 		if (addr >= ent->start_addr && addr < ent->end_addr)
1398 			return true;
1399 	}
1400 	return false;
1401 }
1402 
within_kprobe_blacklist(unsigned long addr)1403 bool within_kprobe_blacklist(unsigned long addr)
1404 {
1405 	char symname[KSYM_NAME_LEN], *p;
1406 
1407 	if (__within_kprobe_blacklist(addr))
1408 		return true;
1409 
1410 	/* Check if the address is on a suffixed-symbol */
1411 	if (!lookup_symbol_name(addr, symname)) {
1412 		p = strchr(symname, '.');
1413 		if (!p)
1414 			return false;
1415 		*p = '\0';
1416 		addr = (unsigned long)kprobe_lookup_name(symname, 0);
1417 		if (addr)
1418 			return __within_kprobe_blacklist(addr);
1419 	}
1420 	return false;
1421 }
1422 
1423 /*
1424  * arch_adjust_kprobe_addr - adjust the address
1425  * @addr: symbol base address
1426  * @offset: offset within the symbol
1427  * @on_func_entry: was this @addr+@offset on the function entry
1428  *
1429  * Typically returns @addr + @offset, except for special cases where the
1430  * function might be prefixed by a CFI landing pad, in that case any offset
1431  * inside the landing pad is mapped to the first 'real' instruction of the
1432  * symbol.
1433  *
1434  * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1435  * instruction at +0.
1436  */
arch_adjust_kprobe_addr(unsigned long addr,unsigned long offset,bool * on_func_entry)1437 kprobe_opcode_t *__weak arch_adjust_kprobe_addr(unsigned long addr,
1438 						unsigned long offset,
1439 						bool *on_func_entry)
1440 {
1441 	*on_func_entry = !offset;
1442 	return (kprobe_opcode_t *)(addr + offset);
1443 }
1444 
1445 /*
1446  * If 'symbol_name' is specified, look it up and add the 'offset'
1447  * to it. This way, we can specify a relative address to a symbol.
1448  * This returns encoded errors if it fails to look up symbol or invalid
1449  * combination of parameters.
1450  */
1451 static kprobe_opcode_t *
_kprobe_addr(kprobe_opcode_t * addr,const char * symbol_name,unsigned long offset,bool * on_func_entry)1452 _kprobe_addr(kprobe_opcode_t *addr, const char *symbol_name,
1453 	     unsigned long offset, bool *on_func_entry)
1454 {
1455 	if ((symbol_name && addr) || (!symbol_name && !addr))
1456 		goto invalid;
1457 
1458 	if (symbol_name) {
1459 		/*
1460 		 * Input: @sym + @offset
1461 		 * Output: @addr + @offset
1462 		 *
1463 		 * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1464 		 *       argument into it's output!
1465 		 */
1466 		addr = kprobe_lookup_name(symbol_name, offset);
1467 		if (!addr)
1468 			return ERR_PTR(-ENOENT);
1469 	}
1470 
1471 	/*
1472 	 * So here we have @addr + @offset, displace it into a new
1473 	 * @addr' + @offset' where @addr' is the symbol start address.
1474 	 */
1475 	addr = (void *)addr + offset;
1476 	if (!kallsyms_lookup_size_offset((unsigned long)addr, NULL, &offset))
1477 		return ERR_PTR(-ENOENT);
1478 	addr = (void *)addr - offset;
1479 
1480 	/*
1481 	 * Then ask the architecture to re-combine them, taking care of
1482 	 * magical function entry details while telling us if this was indeed
1483 	 * at the start of the function.
1484 	 */
1485 	addr = arch_adjust_kprobe_addr((unsigned long)addr, offset, on_func_entry);
1486 	if (addr)
1487 		return addr;
1488 
1489 invalid:
1490 	return ERR_PTR(-EINVAL);
1491 }
1492 
kprobe_addr(struct kprobe * p)1493 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1494 {
1495 	bool on_func_entry;
1496 	return _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1497 }
1498 
1499 /*
1500  * Check the 'p' is valid and return the aggregator kprobe
1501  * at the same address.
1502  */
__get_valid_kprobe(struct kprobe * p)1503 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1504 {
1505 	struct kprobe *ap, *list_p;
1506 
1507 	lockdep_assert_held(&kprobe_mutex);
1508 
1509 	ap = get_kprobe(p->addr);
1510 	if (unlikely(!ap))
1511 		return NULL;
1512 
1513 	if (p != ap) {
1514 		list_for_each_entry(list_p, &ap->list, list)
1515 			if (list_p == p)
1516 			/* kprobe p is a valid probe */
1517 				goto valid;
1518 		return NULL;
1519 	}
1520 valid:
1521 	return ap;
1522 }
1523 
1524 /*
1525  * Warn and return error if the kprobe is being re-registered since
1526  * there must be a software bug.
1527  */
warn_kprobe_rereg(struct kprobe * p)1528 static inline int warn_kprobe_rereg(struct kprobe *p)
1529 {
1530 	int ret = 0;
1531 
1532 	mutex_lock(&kprobe_mutex);
1533 	if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1534 		ret = -EINVAL;
1535 	mutex_unlock(&kprobe_mutex);
1536 
1537 	return ret;
1538 }
1539 
check_ftrace_location(struct kprobe * p)1540 static int check_ftrace_location(struct kprobe *p)
1541 {
1542 	unsigned long addr = (unsigned long)p->addr;
1543 
1544 	if (ftrace_location(addr) == addr) {
1545 #ifdef CONFIG_KPROBES_ON_FTRACE
1546 		p->flags |= KPROBE_FLAG_FTRACE;
1547 #else	/* !CONFIG_KPROBES_ON_FTRACE */
1548 		return -EINVAL;
1549 #endif
1550 	}
1551 	return 0;
1552 }
1553 
is_cfi_preamble_symbol(unsigned long addr)1554 static bool is_cfi_preamble_symbol(unsigned long addr)
1555 {
1556 	char symbuf[KSYM_NAME_LEN];
1557 
1558 	if (lookup_symbol_name(addr, symbuf))
1559 		return false;
1560 
1561 	return str_has_prefix(symbuf, "__cfi_") ||
1562 		str_has_prefix(symbuf, "__pfx_");
1563 }
1564 
check_kprobe_address_safe(struct kprobe * p,struct module ** probed_mod)1565 static int check_kprobe_address_safe(struct kprobe *p,
1566 				     struct module **probed_mod)
1567 {
1568 	int ret;
1569 
1570 	ret = check_ftrace_location(p);
1571 	if (ret)
1572 		return ret;
1573 	jump_label_lock();
1574 	preempt_disable();
1575 
1576 	/* Ensure the address is in a text area, and find a module if exists. */
1577 	*probed_mod = NULL;
1578 	if (!core_kernel_text((unsigned long) p->addr)) {
1579 		*probed_mod = __module_text_address((unsigned long) p->addr);
1580 		if (!(*probed_mod)) {
1581 			ret = -EINVAL;
1582 			goto out;
1583 		}
1584 	}
1585 	/* Ensure it is not in reserved area. */
1586 	if (in_gate_area_no_mm((unsigned long) p->addr) ||
1587 	    within_kprobe_blacklist((unsigned long) p->addr) ||
1588 	    jump_label_text_reserved(p->addr, p->addr) ||
1589 	    static_call_text_reserved(p->addr, p->addr) ||
1590 	    find_bug((unsigned long)p->addr) ||
1591 	    is_cfi_preamble_symbol((unsigned long)p->addr)) {
1592 		ret = -EINVAL;
1593 		goto out;
1594 	}
1595 
1596 	/* Get module refcount and reject __init functions for loaded modules. */
1597 	if (*probed_mod) {
1598 		/*
1599 		 * We must hold a refcount of the probed module while updating
1600 		 * its code to prohibit unexpected unloading.
1601 		 */
1602 		if (unlikely(!try_module_get(*probed_mod))) {
1603 			ret = -ENOENT;
1604 			goto out;
1605 		}
1606 
1607 		/*
1608 		 * If the module freed '.init.text', we couldn't insert
1609 		 * kprobes in there.
1610 		 */
1611 		if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1612 		    (*probed_mod)->state != MODULE_STATE_COMING) {
1613 			module_put(*probed_mod);
1614 			*probed_mod = NULL;
1615 			ret = -ENOENT;
1616 		}
1617 	}
1618 out:
1619 	preempt_enable();
1620 	jump_label_unlock();
1621 
1622 	return ret;
1623 }
1624 
register_kprobe(struct kprobe * p)1625 int register_kprobe(struct kprobe *p)
1626 {
1627 	int ret;
1628 	struct kprobe *old_p;
1629 	struct module *probed_mod;
1630 	kprobe_opcode_t *addr;
1631 	bool on_func_entry;
1632 
1633 	/* Adjust probe address from symbol */
1634 	addr = _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1635 	if (IS_ERR(addr))
1636 		return PTR_ERR(addr);
1637 	p->addr = addr;
1638 
1639 	ret = warn_kprobe_rereg(p);
1640 	if (ret)
1641 		return ret;
1642 
1643 	/* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1644 	p->flags &= KPROBE_FLAG_DISABLED;
1645 	p->nmissed = 0;
1646 	INIT_LIST_HEAD(&p->list);
1647 
1648 	ret = check_kprobe_address_safe(p, &probed_mod);
1649 	if (ret)
1650 		return ret;
1651 
1652 	mutex_lock(&kprobe_mutex);
1653 
1654 	if (on_func_entry)
1655 		p->flags |= KPROBE_FLAG_ON_FUNC_ENTRY;
1656 
1657 	old_p = get_kprobe(p->addr);
1658 	if (old_p) {
1659 		/* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1660 		ret = register_aggr_kprobe(old_p, p);
1661 		goto out;
1662 	}
1663 
1664 	cpus_read_lock();
1665 	/* Prevent text modification */
1666 	mutex_lock(&text_mutex);
1667 	ret = prepare_kprobe(p);
1668 	mutex_unlock(&text_mutex);
1669 	cpus_read_unlock();
1670 	if (ret)
1671 		goto out;
1672 
1673 	INIT_HLIST_NODE(&p->hlist);
1674 	hlist_add_head_rcu(&p->hlist,
1675 		       &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1676 
1677 	if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1678 		ret = arm_kprobe(p);
1679 		if (ret) {
1680 			hlist_del_rcu(&p->hlist);
1681 			synchronize_rcu();
1682 			goto out;
1683 		}
1684 	}
1685 
1686 	/* Try to optimize kprobe */
1687 	try_to_optimize_kprobe(p);
1688 out:
1689 	mutex_unlock(&kprobe_mutex);
1690 
1691 	if (probed_mod)
1692 		module_put(probed_mod);
1693 
1694 	return ret;
1695 }
1696 EXPORT_SYMBOL_GPL(register_kprobe);
1697 
1698 /* Check if all probes on the 'ap' are disabled. */
aggr_kprobe_disabled(struct kprobe * ap)1699 static bool aggr_kprobe_disabled(struct kprobe *ap)
1700 {
1701 	struct kprobe *kp;
1702 
1703 	lockdep_assert_held(&kprobe_mutex);
1704 
1705 	list_for_each_entry(kp, &ap->list, list)
1706 		if (!kprobe_disabled(kp))
1707 			/*
1708 			 * Since there is an active probe on the list,
1709 			 * we can't disable this 'ap'.
1710 			 */
1711 			return false;
1712 
1713 	return true;
1714 }
1715 
__disable_kprobe(struct kprobe * p)1716 static struct kprobe *__disable_kprobe(struct kprobe *p)
1717 {
1718 	struct kprobe *orig_p;
1719 	int ret;
1720 
1721 	lockdep_assert_held(&kprobe_mutex);
1722 
1723 	/* Get an original kprobe for return */
1724 	orig_p = __get_valid_kprobe(p);
1725 	if (unlikely(orig_p == NULL))
1726 		return ERR_PTR(-EINVAL);
1727 
1728 	if (!kprobe_disabled(p)) {
1729 		/* Disable probe if it is a child probe */
1730 		if (p != orig_p)
1731 			p->flags |= KPROBE_FLAG_DISABLED;
1732 
1733 		/* Try to disarm and disable this/parent probe */
1734 		if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1735 			/*
1736 			 * Don't be lazy here.  Even if 'kprobes_all_disarmed'
1737 			 * is false, 'orig_p' might not have been armed yet.
1738 			 * Note arm_all_kprobes() __tries__ to arm all kprobes
1739 			 * on the best effort basis.
1740 			 */
1741 			if (!kprobes_all_disarmed && !kprobe_disabled(orig_p)) {
1742 				ret = disarm_kprobe(orig_p, true);
1743 				if (ret) {
1744 					p->flags &= ~KPROBE_FLAG_DISABLED;
1745 					return ERR_PTR(ret);
1746 				}
1747 			}
1748 			orig_p->flags |= KPROBE_FLAG_DISABLED;
1749 		}
1750 	}
1751 
1752 	return orig_p;
1753 }
1754 
1755 /*
1756  * Unregister a kprobe without a scheduler synchronization.
1757  */
__unregister_kprobe_top(struct kprobe * p)1758 static int __unregister_kprobe_top(struct kprobe *p)
1759 {
1760 	struct kprobe *ap, *list_p;
1761 
1762 	/* Disable kprobe. This will disarm it if needed. */
1763 	ap = __disable_kprobe(p);
1764 	if (IS_ERR(ap))
1765 		return PTR_ERR(ap);
1766 
1767 	if (ap == p)
1768 		/*
1769 		 * This probe is an independent(and non-optimized) kprobe
1770 		 * (not an aggrprobe). Remove from the hash list.
1771 		 */
1772 		goto disarmed;
1773 
1774 	/* Following process expects this probe is an aggrprobe */
1775 	WARN_ON(!kprobe_aggrprobe(ap));
1776 
1777 	if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1778 		/*
1779 		 * !disarmed could be happen if the probe is under delayed
1780 		 * unoptimizing.
1781 		 */
1782 		goto disarmed;
1783 	else {
1784 		/* If disabling probe has special handlers, update aggrprobe */
1785 		if (p->post_handler && !kprobe_gone(p)) {
1786 			list_for_each_entry(list_p, &ap->list, list) {
1787 				if ((list_p != p) && (list_p->post_handler))
1788 					goto noclean;
1789 			}
1790 			/*
1791 			 * For the kprobe-on-ftrace case, we keep the
1792 			 * post_handler setting to identify this aggrprobe
1793 			 * armed with kprobe_ipmodify_ops.
1794 			 */
1795 			if (!kprobe_ftrace(ap))
1796 				ap->post_handler = NULL;
1797 		}
1798 noclean:
1799 		/*
1800 		 * Remove from the aggrprobe: this path will do nothing in
1801 		 * __unregister_kprobe_bottom().
1802 		 */
1803 		list_del_rcu(&p->list);
1804 		if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1805 			/*
1806 			 * Try to optimize this probe again, because post
1807 			 * handler may have been changed.
1808 			 */
1809 			optimize_kprobe(ap);
1810 	}
1811 	return 0;
1812 
1813 disarmed:
1814 	hlist_del_rcu(&ap->hlist);
1815 	return 0;
1816 }
1817 
__unregister_kprobe_bottom(struct kprobe * p)1818 static void __unregister_kprobe_bottom(struct kprobe *p)
1819 {
1820 	struct kprobe *ap;
1821 
1822 	if (list_empty(&p->list))
1823 		/* This is an independent kprobe */
1824 		arch_remove_kprobe(p);
1825 	else if (list_is_singular(&p->list)) {
1826 		/* This is the last child of an aggrprobe */
1827 		ap = list_entry(p->list.next, struct kprobe, list);
1828 		list_del(&p->list);
1829 		free_aggr_kprobe(ap);
1830 	}
1831 	/* Otherwise, do nothing. */
1832 }
1833 
register_kprobes(struct kprobe ** kps,int num)1834 int register_kprobes(struct kprobe **kps, int num)
1835 {
1836 	int i, ret = 0;
1837 
1838 	if (num <= 0)
1839 		return -EINVAL;
1840 	for (i = 0; i < num; i++) {
1841 		ret = register_kprobe(kps[i]);
1842 		if (ret < 0) {
1843 			if (i > 0)
1844 				unregister_kprobes(kps, i);
1845 			break;
1846 		}
1847 	}
1848 	return ret;
1849 }
1850 EXPORT_SYMBOL_GPL(register_kprobes);
1851 
unregister_kprobe(struct kprobe * p)1852 void unregister_kprobe(struct kprobe *p)
1853 {
1854 	unregister_kprobes(&p, 1);
1855 }
1856 EXPORT_SYMBOL_GPL(unregister_kprobe);
1857 
unregister_kprobes(struct kprobe ** kps,int num)1858 void unregister_kprobes(struct kprobe **kps, int num)
1859 {
1860 	int i;
1861 
1862 	if (num <= 0)
1863 		return;
1864 	mutex_lock(&kprobe_mutex);
1865 	for (i = 0; i < num; i++)
1866 		if (__unregister_kprobe_top(kps[i]) < 0)
1867 			kps[i]->addr = NULL;
1868 	mutex_unlock(&kprobe_mutex);
1869 
1870 	synchronize_rcu();
1871 	for (i = 0; i < num; i++)
1872 		if (kps[i]->addr)
1873 			__unregister_kprobe_bottom(kps[i]);
1874 }
1875 EXPORT_SYMBOL_GPL(unregister_kprobes);
1876 
kprobe_exceptions_notify(struct notifier_block * self,unsigned long val,void * data)1877 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1878 					unsigned long val, void *data)
1879 {
1880 	return NOTIFY_DONE;
1881 }
1882 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1883 
1884 static struct notifier_block kprobe_exceptions_nb = {
1885 	.notifier_call = kprobe_exceptions_notify,
1886 	.priority = 0x7fffffff /* we need to be notified first */
1887 };
1888 
1889 #ifdef CONFIG_KRETPROBES
1890 
1891 #if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
free_rp_inst_rcu(struct rcu_head * head)1892 static void free_rp_inst_rcu(struct rcu_head *head)
1893 {
1894 	struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1895 
1896 	if (refcount_dec_and_test(&ri->rph->ref))
1897 		kfree(ri->rph);
1898 	kfree(ri);
1899 }
1900 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1901 
recycle_rp_inst(struct kretprobe_instance * ri)1902 static void recycle_rp_inst(struct kretprobe_instance *ri)
1903 {
1904 	struct kretprobe *rp = get_kretprobe(ri);
1905 
1906 	if (likely(rp))
1907 		freelist_add(&ri->freelist, &rp->freelist);
1908 	else
1909 		call_rcu(&ri->rcu, free_rp_inst_rcu);
1910 }
1911 NOKPROBE_SYMBOL(recycle_rp_inst);
1912 
1913 /*
1914  * This function is called from delayed_put_task_struct() when a task is
1915  * dead and cleaned up to recycle any kretprobe instances associated with
1916  * this task. These left over instances represent probed functions that
1917  * have been called but will never return.
1918  */
kprobe_flush_task(struct task_struct * tk)1919 void kprobe_flush_task(struct task_struct *tk)
1920 {
1921 	struct kretprobe_instance *ri;
1922 	struct llist_node *node;
1923 
1924 	/* Early boot, not yet initialized. */
1925 	if (unlikely(!kprobes_initialized))
1926 		return;
1927 
1928 	kprobe_busy_begin();
1929 
1930 	node = __llist_del_all(&tk->kretprobe_instances);
1931 	while (node) {
1932 		ri = container_of(node, struct kretprobe_instance, llist);
1933 		node = node->next;
1934 
1935 		recycle_rp_inst(ri);
1936 	}
1937 
1938 	kprobe_busy_end();
1939 }
1940 NOKPROBE_SYMBOL(kprobe_flush_task);
1941 
free_rp_inst(struct kretprobe * rp)1942 static inline void free_rp_inst(struct kretprobe *rp)
1943 {
1944 	struct kretprobe_instance *ri;
1945 	struct freelist_node *node;
1946 	int count = 0;
1947 
1948 	node = rp->freelist.head;
1949 	while (node) {
1950 		ri = container_of(node, struct kretprobe_instance, freelist);
1951 		node = node->next;
1952 
1953 		kfree(ri);
1954 		count++;
1955 	}
1956 
1957 	if (refcount_sub_and_test(count, &rp->rph->ref)) {
1958 		kfree(rp->rph);
1959 		rp->rph = NULL;
1960 	}
1961 }
1962 
1963 /* This assumes the 'tsk' is the current task or the is not running. */
__kretprobe_find_ret_addr(struct task_struct * tsk,struct llist_node ** cur)1964 static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1965 						  struct llist_node **cur)
1966 {
1967 	struct kretprobe_instance *ri = NULL;
1968 	struct llist_node *node = *cur;
1969 
1970 	if (!node)
1971 		node = tsk->kretprobe_instances.first;
1972 	else
1973 		node = node->next;
1974 
1975 	while (node) {
1976 		ri = container_of(node, struct kretprobe_instance, llist);
1977 		if (ri->ret_addr != kretprobe_trampoline_addr()) {
1978 			*cur = node;
1979 			return ri->ret_addr;
1980 		}
1981 		node = node->next;
1982 	}
1983 	return NULL;
1984 }
1985 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1986 
1987 /**
1988  * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1989  * @tsk: Target task
1990  * @fp: A frame pointer
1991  * @cur: a storage of the loop cursor llist_node pointer for next call
1992  *
1993  * Find the correct return address modified by a kretprobe on @tsk in unsigned
1994  * long type. If it finds the return address, this returns that address value,
1995  * or this returns 0.
1996  * The @tsk must be 'current' or a task which is not running. @fp is a hint
1997  * to get the currect return address - which is compared with the
1998  * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1999  * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
2000  * first call, but '@cur' itself must NOT NULL.
2001  */
kretprobe_find_ret_addr(struct task_struct * tsk,void * fp,struct llist_node ** cur)2002 unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
2003 				      struct llist_node **cur)
2004 {
2005 	struct kretprobe_instance *ri = NULL;
2006 	kprobe_opcode_t *ret;
2007 
2008 	if (WARN_ON_ONCE(!cur))
2009 		return 0;
2010 
2011 	do {
2012 		ret = __kretprobe_find_ret_addr(tsk, cur);
2013 		if (!ret)
2014 			break;
2015 		ri = container_of(*cur, struct kretprobe_instance, llist);
2016 	} while (ri->fp != fp);
2017 
2018 	return (unsigned long)ret;
2019 }
2020 NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
2021 
arch_kretprobe_fixup_return(struct pt_regs * regs,kprobe_opcode_t * correct_ret_addr)2022 void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
2023 					kprobe_opcode_t *correct_ret_addr)
2024 {
2025 	/*
2026 	 * Do nothing by default. Please fill this to update the fake return
2027 	 * address on the stack with the correct one on each arch if possible.
2028 	 */
2029 }
2030 
__kretprobe_trampoline_handler(struct pt_regs * regs,void * frame_pointer)2031 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
2032 					     void *frame_pointer)
2033 {
2034 	struct kretprobe_instance *ri = NULL;
2035 	struct llist_node *first, *node = NULL;
2036 	kprobe_opcode_t *correct_ret_addr;
2037 	struct kretprobe *rp;
2038 
2039 	/* Find correct address and all nodes for this frame. */
2040 	correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
2041 	if (!correct_ret_addr) {
2042 		pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2043 		BUG_ON(1);
2044 	}
2045 
2046 	/*
2047 	 * Set the return address as the instruction pointer, because if the
2048 	 * user handler calls stack_trace_save_regs() with this 'regs',
2049 	 * the stack trace will start from the instruction pointer.
2050 	 */
2051 	instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
2052 
2053 	/* Run the user handler of the nodes. */
2054 	first = current->kretprobe_instances.first;
2055 	while (first) {
2056 		ri = container_of(first, struct kretprobe_instance, llist);
2057 
2058 		if (WARN_ON_ONCE(ri->fp != frame_pointer))
2059 			break;
2060 
2061 		rp = get_kretprobe(ri);
2062 		if (rp && rp->handler) {
2063 			struct kprobe *prev = kprobe_running();
2064 
2065 			__this_cpu_write(current_kprobe, &rp->kp);
2066 			ri->ret_addr = correct_ret_addr;
2067 			rp->handler(ri, regs);
2068 			__this_cpu_write(current_kprobe, prev);
2069 		}
2070 		if (first == node)
2071 			break;
2072 
2073 		first = first->next;
2074 	}
2075 
2076 	arch_kretprobe_fixup_return(regs, correct_ret_addr);
2077 
2078 	/* Unlink all nodes for this frame. */
2079 	first = current->kretprobe_instances.first;
2080 	current->kretprobe_instances.first = node->next;
2081 	node->next = NULL;
2082 
2083 	/* Recycle free instances. */
2084 	while (first) {
2085 		ri = container_of(first, struct kretprobe_instance, llist);
2086 		first = first->next;
2087 
2088 		recycle_rp_inst(ri);
2089 	}
2090 
2091 	return (unsigned long)correct_ret_addr;
2092 }
NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)2093 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2094 
2095 /*
2096  * This kprobe pre_handler is registered with every kretprobe. When probe
2097  * hits it will set up the return probe.
2098  */
2099 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2100 {
2101 	struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2102 	struct kretprobe_instance *ri;
2103 	struct freelist_node *fn;
2104 
2105 	fn = freelist_try_get(&rp->freelist);
2106 	if (!fn) {
2107 		rp->nmissed++;
2108 		return 0;
2109 	}
2110 
2111 	ri = container_of(fn, struct kretprobe_instance, freelist);
2112 
2113 	if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2114 		freelist_add(&ri->freelist, &rp->freelist);
2115 		return 0;
2116 	}
2117 
2118 	arch_prepare_kretprobe(ri, regs);
2119 
2120 	__llist_add(&ri->llist, &current->kretprobe_instances);
2121 
2122 	return 0;
2123 }
2124 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2125 #else /* CONFIG_KRETPROBE_ON_RETHOOK */
2126 /*
2127  * This kprobe pre_handler is registered with every kretprobe. When probe
2128  * hits it will set up the return probe.
2129  */
pre_handler_kretprobe(struct kprobe * p,struct pt_regs * regs)2130 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2131 {
2132 	struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2133 	struct kretprobe_instance *ri;
2134 	struct rethook_node *rhn;
2135 
2136 	rhn = rethook_try_get(rp->rh);
2137 	if (!rhn) {
2138 		rp->nmissed++;
2139 		return 0;
2140 	}
2141 
2142 	ri = container_of(rhn, struct kretprobe_instance, node);
2143 
2144 	if (rp->entry_handler && rp->entry_handler(ri, regs))
2145 		rethook_recycle(rhn);
2146 	else
2147 		rethook_hook(rhn, regs, kprobe_ftrace(p));
2148 
2149 	return 0;
2150 }
2151 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2152 
kretprobe_rethook_handler(struct rethook_node * rh,void * data,unsigned long ret_addr,struct pt_regs * regs)2153 static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
2154 				      unsigned long ret_addr,
2155 				      struct pt_regs *regs)
2156 {
2157 	struct kretprobe *rp = (struct kretprobe *)data;
2158 	struct kretprobe_instance *ri;
2159 	struct kprobe_ctlblk *kcb;
2160 
2161 	/* The data must NOT be null. This means rethook data structure is broken. */
2162 	if (WARN_ON_ONCE(!data) || !rp->handler)
2163 		return;
2164 
2165 	__this_cpu_write(current_kprobe, &rp->kp);
2166 	kcb = get_kprobe_ctlblk();
2167 	kcb->kprobe_status = KPROBE_HIT_ACTIVE;
2168 
2169 	ri = container_of(rh, struct kretprobe_instance, node);
2170 	rp->handler(ri, regs);
2171 
2172 	__this_cpu_write(current_kprobe, NULL);
2173 }
2174 NOKPROBE_SYMBOL(kretprobe_rethook_handler);
2175 
2176 #endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2177 
2178 /**
2179  * kprobe_on_func_entry() -- check whether given address is function entry
2180  * @addr: Target address
2181  * @sym:  Target symbol name
2182  * @offset: The offset from the symbol or the address
2183  *
2184  * This checks whether the given @addr+@offset or @sym+@offset is on the
2185  * function entry address or not.
2186  * This returns 0 if it is the function entry, or -EINVAL if it is not.
2187  * And also it returns -ENOENT if it fails the symbol or address lookup.
2188  * Caller must pass @addr or @sym (either one must be NULL), or this
2189  * returns -EINVAL.
2190  */
kprobe_on_func_entry(kprobe_opcode_t * addr,const char * sym,unsigned long offset)2191 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2192 {
2193 	bool on_func_entry;
2194 	kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset, &on_func_entry);
2195 
2196 	if (IS_ERR(kp_addr))
2197 		return PTR_ERR(kp_addr);
2198 
2199 	if (!on_func_entry)
2200 		return -EINVAL;
2201 
2202 	return 0;
2203 }
2204 
register_kretprobe(struct kretprobe * rp)2205 int register_kretprobe(struct kretprobe *rp)
2206 {
2207 	int ret;
2208 	struct kretprobe_instance *inst;
2209 	int i;
2210 	void *addr;
2211 
2212 	ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2213 	if (ret)
2214 		return ret;
2215 
2216 	/* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2217 	if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2218 		return -EINVAL;
2219 
2220 	if (kretprobe_blacklist_size) {
2221 		addr = kprobe_addr(&rp->kp);
2222 		if (IS_ERR(addr))
2223 			return PTR_ERR(addr);
2224 
2225 		for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2226 			if (kretprobe_blacklist[i].addr == addr)
2227 				return -EINVAL;
2228 		}
2229 	}
2230 
2231 	if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2232 		return -E2BIG;
2233 
2234 	rp->kp.pre_handler = pre_handler_kretprobe;
2235 	rp->kp.post_handler = NULL;
2236 
2237 	/* Pre-allocate memory for max kretprobe instances */
2238 	if (rp->maxactive <= 0)
2239 		rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2240 
2241 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2242 	rp->rh = rethook_alloc((void *)rp, kretprobe_rethook_handler);
2243 	if (!rp->rh)
2244 		return -ENOMEM;
2245 
2246 	for (i = 0; i < rp->maxactive; i++) {
2247 		inst = kzalloc(struct_size(inst, data, rp->data_size), GFP_KERNEL);
2248 		if (inst == NULL) {
2249 			rethook_free(rp->rh);
2250 			rp->rh = NULL;
2251 			return -ENOMEM;
2252 		}
2253 		rethook_add_node(rp->rh, &inst->node);
2254 	}
2255 	rp->nmissed = 0;
2256 	/* Establish function entry probe point */
2257 	ret = register_kprobe(&rp->kp);
2258 	if (ret != 0) {
2259 		rethook_free(rp->rh);
2260 		rp->rh = NULL;
2261 	}
2262 #else	/* !CONFIG_KRETPROBE_ON_RETHOOK */
2263 	rp->freelist.head = NULL;
2264 	rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2265 	if (!rp->rph)
2266 		return -ENOMEM;
2267 
2268 	rcu_assign_pointer(rp->rph->rp, rp);
2269 	for (i = 0; i < rp->maxactive; i++) {
2270 		inst = kzalloc(struct_size(inst, data, rp->data_size), GFP_KERNEL);
2271 		if (inst == NULL) {
2272 			refcount_set(&rp->rph->ref, i);
2273 			free_rp_inst(rp);
2274 			return -ENOMEM;
2275 		}
2276 		inst->rph = rp->rph;
2277 		freelist_add(&inst->freelist, &rp->freelist);
2278 	}
2279 	refcount_set(&rp->rph->ref, i);
2280 
2281 	rp->nmissed = 0;
2282 	/* Establish function entry probe point */
2283 	ret = register_kprobe(&rp->kp);
2284 	if (ret != 0)
2285 		free_rp_inst(rp);
2286 #endif
2287 	return ret;
2288 }
2289 EXPORT_SYMBOL_GPL(register_kretprobe);
2290 
register_kretprobes(struct kretprobe ** rps,int num)2291 int register_kretprobes(struct kretprobe **rps, int num)
2292 {
2293 	int ret = 0, i;
2294 
2295 	if (num <= 0)
2296 		return -EINVAL;
2297 	for (i = 0; i < num; i++) {
2298 		ret = register_kretprobe(rps[i]);
2299 		if (ret < 0) {
2300 			if (i > 0)
2301 				unregister_kretprobes(rps, i);
2302 			break;
2303 		}
2304 	}
2305 	return ret;
2306 }
2307 EXPORT_SYMBOL_GPL(register_kretprobes);
2308 
unregister_kretprobe(struct kretprobe * rp)2309 void unregister_kretprobe(struct kretprobe *rp)
2310 {
2311 	unregister_kretprobes(&rp, 1);
2312 }
2313 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2314 
unregister_kretprobes(struct kretprobe ** rps,int num)2315 void unregister_kretprobes(struct kretprobe **rps, int num)
2316 {
2317 	int i;
2318 
2319 	if (num <= 0)
2320 		return;
2321 	mutex_lock(&kprobe_mutex);
2322 	for (i = 0; i < num; i++) {
2323 		if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2324 			rps[i]->kp.addr = NULL;
2325 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2326 		rethook_free(rps[i]->rh);
2327 #else
2328 		rcu_assign_pointer(rps[i]->rph->rp, NULL);
2329 #endif
2330 	}
2331 	mutex_unlock(&kprobe_mutex);
2332 
2333 	synchronize_rcu();
2334 	for (i = 0; i < num; i++) {
2335 		if (rps[i]->kp.addr) {
2336 			__unregister_kprobe_bottom(&rps[i]->kp);
2337 #ifndef CONFIG_KRETPROBE_ON_RETHOOK
2338 			free_rp_inst(rps[i]);
2339 #endif
2340 		}
2341 	}
2342 }
2343 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2344 
2345 #else /* CONFIG_KRETPROBES */
register_kretprobe(struct kretprobe * rp)2346 int register_kretprobe(struct kretprobe *rp)
2347 {
2348 	return -EOPNOTSUPP;
2349 }
2350 EXPORT_SYMBOL_GPL(register_kretprobe);
2351 
register_kretprobes(struct kretprobe ** rps,int num)2352 int register_kretprobes(struct kretprobe **rps, int num)
2353 {
2354 	return -EOPNOTSUPP;
2355 }
2356 EXPORT_SYMBOL_GPL(register_kretprobes);
2357 
unregister_kretprobe(struct kretprobe * rp)2358 void unregister_kretprobe(struct kretprobe *rp)
2359 {
2360 }
2361 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2362 
unregister_kretprobes(struct kretprobe ** rps,int num)2363 void unregister_kretprobes(struct kretprobe **rps, int num)
2364 {
2365 }
2366 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2367 
pre_handler_kretprobe(struct kprobe * p,struct pt_regs * regs)2368 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2369 {
2370 	return 0;
2371 }
2372 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2373 
2374 #endif /* CONFIG_KRETPROBES */
2375 
2376 /* Set the kprobe gone and remove its instruction buffer. */
kill_kprobe(struct kprobe * p)2377 static void kill_kprobe(struct kprobe *p)
2378 {
2379 	struct kprobe *kp;
2380 
2381 	lockdep_assert_held(&kprobe_mutex);
2382 
2383 	/*
2384 	 * The module is going away. We should disarm the kprobe which
2385 	 * is using ftrace, because ftrace framework is still available at
2386 	 * 'MODULE_STATE_GOING' notification.
2387 	 */
2388 	if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2389 		disarm_kprobe_ftrace(p);
2390 
2391 	p->flags |= KPROBE_FLAG_GONE;
2392 	if (kprobe_aggrprobe(p)) {
2393 		/*
2394 		 * If this is an aggr_kprobe, we have to list all the
2395 		 * chained probes and mark them GONE.
2396 		 */
2397 		list_for_each_entry(kp, &p->list, list)
2398 			kp->flags |= KPROBE_FLAG_GONE;
2399 		p->post_handler = NULL;
2400 		kill_optimized_kprobe(p);
2401 	}
2402 	/*
2403 	 * Here, we can remove insn_slot safely, because no thread calls
2404 	 * the original probed function (which will be freed soon) any more.
2405 	 */
2406 	arch_remove_kprobe(p);
2407 }
2408 
2409 /* Disable one kprobe */
disable_kprobe(struct kprobe * kp)2410 int disable_kprobe(struct kprobe *kp)
2411 {
2412 	int ret = 0;
2413 	struct kprobe *p;
2414 
2415 	mutex_lock(&kprobe_mutex);
2416 
2417 	/* Disable this kprobe */
2418 	p = __disable_kprobe(kp);
2419 	if (IS_ERR(p))
2420 		ret = PTR_ERR(p);
2421 
2422 	mutex_unlock(&kprobe_mutex);
2423 	return ret;
2424 }
2425 EXPORT_SYMBOL_GPL(disable_kprobe);
2426 
2427 /* Enable one kprobe */
enable_kprobe(struct kprobe * kp)2428 int enable_kprobe(struct kprobe *kp)
2429 {
2430 	int ret = 0;
2431 	struct kprobe *p;
2432 
2433 	mutex_lock(&kprobe_mutex);
2434 
2435 	/* Check whether specified probe is valid. */
2436 	p = __get_valid_kprobe(kp);
2437 	if (unlikely(p == NULL)) {
2438 		ret = -EINVAL;
2439 		goto out;
2440 	}
2441 
2442 	if (kprobe_gone(kp)) {
2443 		/* This kprobe has gone, we couldn't enable it. */
2444 		ret = -EINVAL;
2445 		goto out;
2446 	}
2447 
2448 	if (p != kp)
2449 		kp->flags &= ~KPROBE_FLAG_DISABLED;
2450 
2451 	if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2452 		p->flags &= ~KPROBE_FLAG_DISABLED;
2453 		ret = arm_kprobe(p);
2454 		if (ret) {
2455 			p->flags |= KPROBE_FLAG_DISABLED;
2456 			if (p != kp)
2457 				kp->flags |= KPROBE_FLAG_DISABLED;
2458 		}
2459 	}
2460 out:
2461 	mutex_unlock(&kprobe_mutex);
2462 	return ret;
2463 }
2464 EXPORT_SYMBOL_GPL(enable_kprobe);
2465 
2466 /* Caller must NOT call this in usual path. This is only for critical case */
dump_kprobe(struct kprobe * kp)2467 void dump_kprobe(struct kprobe *kp)
2468 {
2469 	pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2470 	       kp->symbol_name, kp->offset, kp->addr);
2471 }
2472 NOKPROBE_SYMBOL(dump_kprobe);
2473 
kprobe_add_ksym_blacklist(unsigned long entry)2474 int kprobe_add_ksym_blacklist(unsigned long entry)
2475 {
2476 	struct kprobe_blacklist_entry *ent;
2477 	unsigned long offset = 0, size = 0;
2478 
2479 	if (!kernel_text_address(entry) ||
2480 	    !kallsyms_lookup_size_offset(entry, &size, &offset))
2481 		return -EINVAL;
2482 
2483 	ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2484 	if (!ent)
2485 		return -ENOMEM;
2486 	ent->start_addr = entry;
2487 	ent->end_addr = entry + size;
2488 	INIT_LIST_HEAD(&ent->list);
2489 	list_add_tail(&ent->list, &kprobe_blacklist);
2490 
2491 	return (int)size;
2492 }
2493 
2494 /* Add all symbols in given area into kprobe blacklist */
kprobe_add_area_blacklist(unsigned long start,unsigned long end)2495 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2496 {
2497 	unsigned long entry;
2498 	int ret = 0;
2499 
2500 	for (entry = start; entry < end; entry += ret) {
2501 		ret = kprobe_add_ksym_blacklist(entry);
2502 		if (ret < 0)
2503 			return ret;
2504 		if (ret == 0)	/* In case of alias symbol */
2505 			ret = 1;
2506 	}
2507 	return 0;
2508 }
2509 
2510 /* Remove all symbols in given area from kprobe blacklist */
kprobe_remove_area_blacklist(unsigned long start,unsigned long end)2511 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2512 {
2513 	struct kprobe_blacklist_entry *ent, *n;
2514 
2515 	list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2516 		if (ent->start_addr < start || ent->start_addr >= end)
2517 			continue;
2518 		list_del(&ent->list);
2519 		kfree(ent);
2520 	}
2521 }
2522 
kprobe_remove_ksym_blacklist(unsigned long entry)2523 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2524 {
2525 	kprobe_remove_area_blacklist(entry, entry + 1);
2526 }
2527 
arch_kprobe_get_kallsym(unsigned int * symnum,unsigned long * value,char * type,char * sym)2528 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2529 				   char *type, char *sym)
2530 {
2531 	return -ERANGE;
2532 }
2533 
kprobe_get_kallsym(unsigned int symnum,unsigned long * value,char * type,char * sym)2534 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2535 		       char *sym)
2536 {
2537 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2538 	if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2539 		return 0;
2540 #ifdef CONFIG_OPTPROBES
2541 	if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2542 		return 0;
2543 #endif
2544 #endif
2545 	if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2546 		return 0;
2547 	return -ERANGE;
2548 }
2549 
arch_populate_kprobe_blacklist(void)2550 int __init __weak arch_populate_kprobe_blacklist(void)
2551 {
2552 	return 0;
2553 }
2554 
2555 /*
2556  * Lookup and populate the kprobe_blacklist.
2557  *
2558  * Unlike the kretprobe blacklist, we'll need to determine
2559  * the range of addresses that belong to the said functions,
2560  * since a kprobe need not necessarily be at the beginning
2561  * of a function.
2562  */
populate_kprobe_blacklist(unsigned long * start,unsigned long * end)2563 static int __init populate_kprobe_blacklist(unsigned long *start,
2564 					     unsigned long *end)
2565 {
2566 	unsigned long entry;
2567 	unsigned long *iter;
2568 	int ret;
2569 
2570 	for (iter = start; iter < end; iter++) {
2571 		entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2572 		ret = kprobe_add_ksym_blacklist(entry);
2573 		if (ret == -EINVAL)
2574 			continue;
2575 		if (ret < 0)
2576 			return ret;
2577 	}
2578 
2579 	/* Symbols in '__kprobes_text' are blacklisted */
2580 	ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2581 					(unsigned long)__kprobes_text_end);
2582 	if (ret)
2583 		return ret;
2584 
2585 	/* Symbols in 'noinstr' section are blacklisted */
2586 	ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2587 					(unsigned long)__noinstr_text_end);
2588 
2589 	return ret ? : arch_populate_kprobe_blacklist();
2590 }
2591 
add_module_kprobe_blacklist(struct module * mod)2592 static void add_module_kprobe_blacklist(struct module *mod)
2593 {
2594 	unsigned long start, end;
2595 	int i;
2596 
2597 	if (mod->kprobe_blacklist) {
2598 		for (i = 0; i < mod->num_kprobe_blacklist; i++)
2599 			kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2600 	}
2601 
2602 	start = (unsigned long)mod->kprobes_text_start;
2603 	if (start) {
2604 		end = start + mod->kprobes_text_size;
2605 		kprobe_add_area_blacklist(start, end);
2606 	}
2607 
2608 	start = (unsigned long)mod->noinstr_text_start;
2609 	if (start) {
2610 		end = start + mod->noinstr_text_size;
2611 		kprobe_add_area_blacklist(start, end);
2612 	}
2613 }
2614 
remove_module_kprobe_blacklist(struct module * mod)2615 static void remove_module_kprobe_blacklist(struct module *mod)
2616 {
2617 	unsigned long start, end;
2618 	int i;
2619 
2620 	if (mod->kprobe_blacklist) {
2621 		for (i = 0; i < mod->num_kprobe_blacklist; i++)
2622 			kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2623 	}
2624 
2625 	start = (unsigned long)mod->kprobes_text_start;
2626 	if (start) {
2627 		end = start + mod->kprobes_text_size;
2628 		kprobe_remove_area_blacklist(start, end);
2629 	}
2630 
2631 	start = (unsigned long)mod->noinstr_text_start;
2632 	if (start) {
2633 		end = start + mod->noinstr_text_size;
2634 		kprobe_remove_area_blacklist(start, end);
2635 	}
2636 }
2637 
2638 /* Module notifier call back, checking kprobes on the module */
kprobes_module_callback(struct notifier_block * nb,unsigned long val,void * data)2639 static int kprobes_module_callback(struct notifier_block *nb,
2640 				   unsigned long val, void *data)
2641 {
2642 	struct module *mod = data;
2643 	struct hlist_head *head;
2644 	struct kprobe *p;
2645 	unsigned int i;
2646 	int checkcore = (val == MODULE_STATE_GOING);
2647 
2648 	if (val == MODULE_STATE_COMING) {
2649 		mutex_lock(&kprobe_mutex);
2650 		add_module_kprobe_blacklist(mod);
2651 		mutex_unlock(&kprobe_mutex);
2652 	}
2653 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2654 		return NOTIFY_DONE;
2655 
2656 	/*
2657 	 * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2658 	 * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2659 	 * notified, only '.init.text' section would be freed. We need to
2660 	 * disable kprobes which have been inserted in the sections.
2661 	 */
2662 	mutex_lock(&kprobe_mutex);
2663 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2664 		head = &kprobe_table[i];
2665 		hlist_for_each_entry(p, head, hlist)
2666 			if (within_module_init((unsigned long)p->addr, mod) ||
2667 			    (checkcore &&
2668 			     within_module_core((unsigned long)p->addr, mod))) {
2669 				/*
2670 				 * The vaddr this probe is installed will soon
2671 				 * be vfreed buy not synced to disk. Hence,
2672 				 * disarming the breakpoint isn't needed.
2673 				 *
2674 				 * Note, this will also move any optimized probes
2675 				 * that are pending to be removed from their
2676 				 * corresponding lists to the 'freeing_list' and
2677 				 * will not be touched by the delayed
2678 				 * kprobe_optimizer() work handler.
2679 				 */
2680 				kill_kprobe(p);
2681 			}
2682 	}
2683 	if (val == MODULE_STATE_GOING)
2684 		remove_module_kprobe_blacklist(mod);
2685 	mutex_unlock(&kprobe_mutex);
2686 	return NOTIFY_DONE;
2687 }
2688 
2689 static struct notifier_block kprobe_module_nb = {
2690 	.notifier_call = kprobes_module_callback,
2691 	.priority = 0
2692 };
2693 
kprobe_free_init_mem(void)2694 void kprobe_free_init_mem(void)
2695 {
2696 	void *start = (void *)(&__init_begin);
2697 	void *end = (void *)(&__init_end);
2698 	struct hlist_head *head;
2699 	struct kprobe *p;
2700 	int i;
2701 
2702 	mutex_lock(&kprobe_mutex);
2703 
2704 	/* Kill all kprobes on initmem because the target code has been freed. */
2705 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2706 		head = &kprobe_table[i];
2707 		hlist_for_each_entry(p, head, hlist) {
2708 			if (start <= (void *)p->addr && (void *)p->addr < end)
2709 				kill_kprobe(p);
2710 		}
2711 	}
2712 
2713 	mutex_unlock(&kprobe_mutex);
2714 }
2715 
init_kprobes(void)2716 static int __init init_kprobes(void)
2717 {
2718 	int i, err;
2719 
2720 	/* FIXME allocate the probe table, currently defined statically */
2721 	/* initialize all list heads */
2722 	for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2723 		INIT_HLIST_HEAD(&kprobe_table[i]);
2724 
2725 	err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2726 					__stop_kprobe_blacklist);
2727 	if (err)
2728 		pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2729 
2730 	if (kretprobe_blacklist_size) {
2731 		/* lookup the function address from its name */
2732 		for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2733 			kretprobe_blacklist[i].addr =
2734 				kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2735 			if (!kretprobe_blacklist[i].addr)
2736 				pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2737 				       kretprobe_blacklist[i].name);
2738 		}
2739 	}
2740 
2741 	/* By default, kprobes are armed */
2742 	kprobes_all_disarmed = false;
2743 
2744 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2745 	/* Init 'kprobe_optinsn_slots' for allocation */
2746 	kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2747 #endif
2748 
2749 	err = arch_init_kprobes();
2750 	if (!err)
2751 		err = register_die_notifier(&kprobe_exceptions_nb);
2752 	if (!err)
2753 		err = register_module_notifier(&kprobe_module_nb);
2754 
2755 	kprobes_initialized = (err == 0);
2756 	kprobe_sysctls_init();
2757 	return err;
2758 }
2759 early_initcall(init_kprobes);
2760 
2761 #if defined(CONFIG_OPTPROBES)
init_optprobes(void)2762 static int __init init_optprobes(void)
2763 {
2764 	/*
2765 	 * Enable kprobe optimization - this kicks the optimizer which
2766 	 * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2767 	 * not spawned in early initcall. So delay the optimization.
2768 	 */
2769 	optimize_all_kprobes();
2770 
2771 	return 0;
2772 }
2773 subsys_initcall(init_optprobes);
2774 #endif
2775 
2776 #ifdef CONFIG_DEBUG_FS
report_probe(struct seq_file * pi,struct kprobe * p,const char * sym,int offset,char * modname,struct kprobe * pp)2777 static void report_probe(struct seq_file *pi, struct kprobe *p,
2778 		const char *sym, int offset, char *modname, struct kprobe *pp)
2779 {
2780 	char *kprobe_type;
2781 	void *addr = p->addr;
2782 
2783 	if (p->pre_handler == pre_handler_kretprobe)
2784 		kprobe_type = "r";
2785 	else
2786 		kprobe_type = "k";
2787 
2788 	if (!kallsyms_show_value(pi->file->f_cred))
2789 		addr = NULL;
2790 
2791 	if (sym)
2792 		seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2793 			addr, kprobe_type, sym, offset,
2794 			(modname ? modname : " "));
2795 	else	/* try to use %pS */
2796 		seq_printf(pi, "%px  %s  %pS ",
2797 			addr, kprobe_type, p->addr);
2798 
2799 	if (!pp)
2800 		pp = p;
2801 	seq_printf(pi, "%s%s%s%s\n",
2802 		(kprobe_gone(p) ? "[GONE]" : ""),
2803 		((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2804 		(kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2805 		(kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2806 }
2807 
kprobe_seq_start(struct seq_file * f,loff_t * pos)2808 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2809 {
2810 	return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2811 }
2812 
kprobe_seq_next(struct seq_file * f,void * v,loff_t * pos)2813 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2814 {
2815 	(*pos)++;
2816 	if (*pos >= KPROBE_TABLE_SIZE)
2817 		return NULL;
2818 	return pos;
2819 }
2820 
kprobe_seq_stop(struct seq_file * f,void * v)2821 static void kprobe_seq_stop(struct seq_file *f, void *v)
2822 {
2823 	/* Nothing to do */
2824 }
2825 
show_kprobe_addr(struct seq_file * pi,void * v)2826 static int show_kprobe_addr(struct seq_file *pi, void *v)
2827 {
2828 	struct hlist_head *head;
2829 	struct kprobe *p, *kp;
2830 	const char *sym = NULL;
2831 	unsigned int i = *(loff_t *) v;
2832 	unsigned long offset = 0;
2833 	char *modname, namebuf[KSYM_NAME_LEN];
2834 
2835 	head = &kprobe_table[i];
2836 	preempt_disable();
2837 	hlist_for_each_entry_rcu(p, head, hlist) {
2838 		sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2839 					&offset, &modname, namebuf);
2840 		if (kprobe_aggrprobe(p)) {
2841 			list_for_each_entry_rcu(kp, &p->list, list)
2842 				report_probe(pi, kp, sym, offset, modname, p);
2843 		} else
2844 			report_probe(pi, p, sym, offset, modname, NULL);
2845 	}
2846 	preempt_enable();
2847 	return 0;
2848 }
2849 
2850 static const struct seq_operations kprobes_sops = {
2851 	.start = kprobe_seq_start,
2852 	.next  = kprobe_seq_next,
2853 	.stop  = kprobe_seq_stop,
2854 	.show  = show_kprobe_addr
2855 };
2856 
2857 DEFINE_SEQ_ATTRIBUTE(kprobes);
2858 
2859 /* kprobes/blacklist -- shows which functions can not be probed */
kprobe_blacklist_seq_start(struct seq_file * m,loff_t * pos)2860 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2861 {
2862 	mutex_lock(&kprobe_mutex);
2863 	return seq_list_start(&kprobe_blacklist, *pos);
2864 }
2865 
kprobe_blacklist_seq_next(struct seq_file * m,void * v,loff_t * pos)2866 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2867 {
2868 	return seq_list_next(v, &kprobe_blacklist, pos);
2869 }
2870 
kprobe_blacklist_seq_show(struct seq_file * m,void * v)2871 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2872 {
2873 	struct kprobe_blacklist_entry *ent =
2874 		list_entry(v, struct kprobe_blacklist_entry, list);
2875 
2876 	/*
2877 	 * If '/proc/kallsyms' is not showing kernel address, we won't
2878 	 * show them here either.
2879 	 */
2880 	if (!kallsyms_show_value(m->file->f_cred))
2881 		seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2882 			   (void *)ent->start_addr);
2883 	else
2884 		seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2885 			   (void *)ent->end_addr, (void *)ent->start_addr);
2886 	return 0;
2887 }
2888 
kprobe_blacklist_seq_stop(struct seq_file * f,void * v)2889 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2890 {
2891 	mutex_unlock(&kprobe_mutex);
2892 }
2893 
2894 static const struct seq_operations kprobe_blacklist_sops = {
2895 	.start = kprobe_blacklist_seq_start,
2896 	.next  = kprobe_blacklist_seq_next,
2897 	.stop  = kprobe_blacklist_seq_stop,
2898 	.show  = kprobe_blacklist_seq_show,
2899 };
2900 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2901 
arm_all_kprobes(void)2902 static int arm_all_kprobes(void)
2903 {
2904 	struct hlist_head *head;
2905 	struct kprobe *p;
2906 	unsigned int i, total = 0, errors = 0;
2907 	int err, ret = 0;
2908 
2909 	mutex_lock(&kprobe_mutex);
2910 
2911 	/* If kprobes are armed, just return */
2912 	if (!kprobes_all_disarmed)
2913 		goto already_enabled;
2914 
2915 	/*
2916 	 * optimize_kprobe() called by arm_kprobe() checks
2917 	 * kprobes_all_disarmed, so set kprobes_all_disarmed before
2918 	 * arm_kprobe.
2919 	 */
2920 	kprobes_all_disarmed = false;
2921 	/* Arming kprobes doesn't optimize kprobe itself */
2922 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2923 		head = &kprobe_table[i];
2924 		/* Arm all kprobes on a best-effort basis */
2925 		hlist_for_each_entry(p, head, hlist) {
2926 			if (!kprobe_disabled(p)) {
2927 				err = arm_kprobe(p);
2928 				if (err)  {
2929 					errors++;
2930 					ret = err;
2931 				}
2932 				total++;
2933 			}
2934 		}
2935 	}
2936 
2937 	if (errors)
2938 		pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2939 			errors, total);
2940 	else
2941 		pr_info("Kprobes globally enabled\n");
2942 
2943 already_enabled:
2944 	mutex_unlock(&kprobe_mutex);
2945 	return ret;
2946 }
2947 
disarm_all_kprobes(void)2948 static int disarm_all_kprobes(void)
2949 {
2950 	struct hlist_head *head;
2951 	struct kprobe *p;
2952 	unsigned int i, total = 0, errors = 0;
2953 	int err, ret = 0;
2954 
2955 	mutex_lock(&kprobe_mutex);
2956 
2957 	/* If kprobes are already disarmed, just return */
2958 	if (kprobes_all_disarmed) {
2959 		mutex_unlock(&kprobe_mutex);
2960 		return 0;
2961 	}
2962 
2963 	kprobes_all_disarmed = true;
2964 
2965 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2966 		head = &kprobe_table[i];
2967 		/* Disarm all kprobes on a best-effort basis */
2968 		hlist_for_each_entry(p, head, hlist) {
2969 			if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2970 				err = disarm_kprobe(p, false);
2971 				if (err) {
2972 					errors++;
2973 					ret = err;
2974 				}
2975 				total++;
2976 			}
2977 		}
2978 	}
2979 
2980 	if (errors)
2981 		pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2982 			errors, total);
2983 	else
2984 		pr_info("Kprobes globally disabled\n");
2985 
2986 	mutex_unlock(&kprobe_mutex);
2987 
2988 	/* Wait for disarming all kprobes by optimizer */
2989 	wait_for_kprobe_optimizer();
2990 
2991 	return ret;
2992 }
2993 
2994 /*
2995  * XXX: The debugfs bool file interface doesn't allow for callbacks
2996  * when the bool state is switched. We can reuse that facility when
2997  * available
2998  */
read_enabled_file_bool(struct file * file,char __user * user_buf,size_t count,loff_t * ppos)2999 static ssize_t read_enabled_file_bool(struct file *file,
3000 	       char __user *user_buf, size_t count, loff_t *ppos)
3001 {
3002 	char buf[3];
3003 
3004 	if (!kprobes_all_disarmed)
3005 		buf[0] = '1';
3006 	else
3007 		buf[0] = '0';
3008 	buf[1] = '\n';
3009 	buf[2] = 0x00;
3010 	return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
3011 }
3012 
write_enabled_file_bool(struct file * file,const char __user * user_buf,size_t count,loff_t * ppos)3013 static ssize_t write_enabled_file_bool(struct file *file,
3014 	       const char __user *user_buf, size_t count, loff_t *ppos)
3015 {
3016 	bool enable;
3017 	int ret;
3018 
3019 	ret = kstrtobool_from_user(user_buf, count, &enable);
3020 	if (ret)
3021 		return ret;
3022 
3023 	ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
3024 	if (ret)
3025 		return ret;
3026 
3027 	return count;
3028 }
3029 
3030 static const struct file_operations fops_kp = {
3031 	.read =         read_enabled_file_bool,
3032 	.write =        write_enabled_file_bool,
3033 	.llseek =	default_llseek,
3034 };
3035 
debugfs_kprobe_init(void)3036 static int __init debugfs_kprobe_init(void)
3037 {
3038 	struct dentry *dir;
3039 
3040 	dir = debugfs_create_dir("kprobes", NULL);
3041 
3042 	debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
3043 
3044 	debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
3045 
3046 	debugfs_create_file("blacklist", 0400, dir, NULL,
3047 			    &kprobe_blacklist_fops);
3048 
3049 	return 0;
3050 }
3051 
3052 late_initcall(debugfs_kprobe_init);
3053 #endif /* CONFIG_DEBUG_FS */
3054