xref: /openbmc/linux/kernel/kprobes.c (revision e8e0929d)
1 /*
2  *  Kernel Probes (KProbes)
3  *  kernel/kprobes.c
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License for more details.
14  *
15  * You should have received a copy of the GNU General Public License
16  * along with this program; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18  *
19  * Copyright (C) IBM Corporation, 2002, 2004
20  *
21  * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22  *		Probes initial implementation (includes suggestions from
23  *		Rusty Russell).
24  * 2004-Aug	Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
25  *		hlists and exceptions notifier as suggested by Andi Kleen.
26  * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
27  *		interface to access function arguments.
28  * 2004-Sep	Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
29  *		exceptions notifier to be first on the priority list.
30  * 2005-May	Hien Nguyen <hien@us.ibm.com>, Jim Keniston
31  *		<jkenisto@us.ibm.com> and Prasanna S Panchamukhi
32  *		<prasanna@in.ibm.com> added function-return probes.
33  */
34 #include <linux/kprobes.h>
35 #include <linux/hash.h>
36 #include <linux/init.h>
37 #include <linux/slab.h>
38 #include <linux/stddef.h>
39 #include <linux/module.h>
40 #include <linux/moduleloader.h>
41 #include <linux/kallsyms.h>
42 #include <linux/freezer.h>
43 #include <linux/seq_file.h>
44 #include <linux/debugfs.h>
45 #include <linux/kdebug.h>
46 #include <linux/memory.h>
47 
48 #include <asm-generic/sections.h>
49 #include <asm/cacheflush.h>
50 #include <asm/errno.h>
51 #include <asm/uaccess.h>
52 
53 #define KPROBE_HASH_BITS 6
54 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
55 
56 
57 /*
58  * Some oddball architectures like 64bit powerpc have function descriptors
59  * so this must be overridable.
60  */
61 #ifndef kprobe_lookup_name
62 #define kprobe_lookup_name(name, addr) \
63 	addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
64 #endif
65 
66 static int kprobes_initialized;
67 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
68 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
69 
70 /* NOTE: change this value only with kprobe_mutex held */
71 static bool kprobes_all_disarmed;
72 
73 static DEFINE_MUTEX(kprobe_mutex);	/* Protects kprobe_table */
74 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
75 static struct {
76 	spinlock_t lock ____cacheline_aligned_in_smp;
77 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
78 
79 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
80 {
81 	return &(kretprobe_table_locks[hash].lock);
82 }
83 
84 /*
85  * Normally, functions that we'd want to prohibit kprobes in, are marked
86  * __kprobes. But, there are cases where such functions already belong to
87  * a different section (__sched for preempt_schedule)
88  *
89  * For such cases, we now have a blacklist
90  */
91 static struct kprobe_blackpoint kprobe_blacklist[] = {
92 	{"preempt_schedule",},
93 	{NULL}    /* Terminator */
94 };
95 
96 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
97 /*
98  * kprobe->ainsn.insn points to the copy of the instruction to be
99  * single-stepped. x86_64, POWER4 and above have no-exec support and
100  * stepping on the instruction on a vmalloced/kmalloced/data page
101  * is a recipe for disaster
102  */
103 #define INSNS_PER_PAGE	(PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
104 
105 struct kprobe_insn_page {
106 	struct list_head list;
107 	kprobe_opcode_t *insns;		/* Page of instruction slots */
108 	char slot_used[INSNS_PER_PAGE];
109 	int nused;
110 	int ngarbage;
111 };
112 
113 enum kprobe_slot_state {
114 	SLOT_CLEAN = 0,
115 	SLOT_DIRTY = 1,
116 	SLOT_USED = 2,
117 };
118 
119 static DEFINE_MUTEX(kprobe_insn_mutex);	/* Protects kprobe_insn_pages */
120 static LIST_HEAD(kprobe_insn_pages);
121 static int kprobe_garbage_slots;
122 static int collect_garbage_slots(void);
123 
124 static int __kprobes check_safety(void)
125 {
126 	int ret = 0;
127 #if defined(CONFIG_PREEMPT) && defined(CONFIG_FREEZER)
128 	ret = freeze_processes();
129 	if (ret == 0) {
130 		struct task_struct *p, *q;
131 		do_each_thread(p, q) {
132 			if (p != current && p->state == TASK_RUNNING &&
133 			    p->pid != 0) {
134 				printk("Check failed: %s is running\n",p->comm);
135 				ret = -1;
136 				goto loop_end;
137 			}
138 		} while_each_thread(p, q);
139 	}
140 loop_end:
141 	thaw_processes();
142 #else
143 	synchronize_sched();
144 #endif
145 	return ret;
146 }
147 
148 /**
149  * __get_insn_slot() - Find a slot on an executable page for an instruction.
150  * We allocate an executable page if there's no room on existing ones.
151  */
152 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
153 {
154 	struct kprobe_insn_page *kip;
155 
156  retry:
157 	list_for_each_entry(kip, &kprobe_insn_pages, list) {
158 		if (kip->nused < INSNS_PER_PAGE) {
159 			int i;
160 			for (i = 0; i < INSNS_PER_PAGE; i++) {
161 				if (kip->slot_used[i] == SLOT_CLEAN) {
162 					kip->slot_used[i] = SLOT_USED;
163 					kip->nused++;
164 					return kip->insns + (i * MAX_INSN_SIZE);
165 				}
166 			}
167 			/* Surprise!  No unused slots.  Fix kip->nused. */
168 			kip->nused = INSNS_PER_PAGE;
169 		}
170 	}
171 
172 	/* If there are any garbage slots, collect it and try again. */
173 	if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
174 		goto retry;
175 	}
176 	/* All out of space.  Need to allocate a new page. Use slot 0. */
177 	kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
178 	if (!kip)
179 		return NULL;
180 
181 	/*
182 	 * Use module_alloc so this page is within +/- 2GB of where the
183 	 * kernel image and loaded module images reside. This is required
184 	 * so x86_64 can correctly handle the %rip-relative fixups.
185 	 */
186 	kip->insns = module_alloc(PAGE_SIZE);
187 	if (!kip->insns) {
188 		kfree(kip);
189 		return NULL;
190 	}
191 	INIT_LIST_HEAD(&kip->list);
192 	list_add(&kip->list, &kprobe_insn_pages);
193 	memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
194 	kip->slot_used[0] = SLOT_USED;
195 	kip->nused = 1;
196 	kip->ngarbage = 0;
197 	return kip->insns;
198 }
199 
200 kprobe_opcode_t __kprobes *get_insn_slot(void)
201 {
202 	kprobe_opcode_t *ret;
203 	mutex_lock(&kprobe_insn_mutex);
204 	ret = __get_insn_slot();
205 	mutex_unlock(&kprobe_insn_mutex);
206 	return ret;
207 }
208 
209 /* Return 1 if all garbages are collected, otherwise 0. */
210 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
211 {
212 	kip->slot_used[idx] = SLOT_CLEAN;
213 	kip->nused--;
214 	if (kip->nused == 0) {
215 		/*
216 		 * Page is no longer in use.  Free it unless
217 		 * it's the last one.  We keep the last one
218 		 * so as not to have to set it up again the
219 		 * next time somebody inserts a probe.
220 		 */
221 		if (!list_is_singular(&kprobe_insn_pages)) {
222 			list_del(&kip->list);
223 			module_free(NULL, kip->insns);
224 			kfree(kip);
225 		}
226 		return 1;
227 	}
228 	return 0;
229 }
230 
231 static int __kprobes collect_garbage_slots(void)
232 {
233 	struct kprobe_insn_page *kip, *next;
234 
235 	/* Ensure no-one is preepmted on the garbages */
236 	if (check_safety())
237 		return -EAGAIN;
238 
239 	list_for_each_entry_safe(kip, next, &kprobe_insn_pages, list) {
240 		int i;
241 		if (kip->ngarbage == 0)
242 			continue;
243 		kip->ngarbage = 0;	/* we will collect all garbages */
244 		for (i = 0; i < INSNS_PER_PAGE; i++) {
245 			if (kip->slot_used[i] == SLOT_DIRTY &&
246 			    collect_one_slot(kip, i))
247 				break;
248 		}
249 	}
250 	kprobe_garbage_slots = 0;
251 	return 0;
252 }
253 
254 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
255 {
256 	struct kprobe_insn_page *kip;
257 
258 	mutex_lock(&kprobe_insn_mutex);
259 	list_for_each_entry(kip, &kprobe_insn_pages, list) {
260 		if (kip->insns <= slot &&
261 		    slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
262 			int i = (slot - kip->insns) / MAX_INSN_SIZE;
263 			if (dirty) {
264 				kip->slot_used[i] = SLOT_DIRTY;
265 				kip->ngarbage++;
266 			} else
267 				collect_one_slot(kip, i);
268 			break;
269 		}
270 	}
271 
272 	if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
273 		collect_garbage_slots();
274 
275 	mutex_unlock(&kprobe_insn_mutex);
276 }
277 #endif
278 
279 /* We have preemption disabled.. so it is safe to use __ versions */
280 static inline void set_kprobe_instance(struct kprobe *kp)
281 {
282 	__get_cpu_var(kprobe_instance) = kp;
283 }
284 
285 static inline void reset_kprobe_instance(void)
286 {
287 	__get_cpu_var(kprobe_instance) = NULL;
288 }
289 
290 /*
291  * This routine is called either:
292  * 	- under the kprobe_mutex - during kprobe_[un]register()
293  * 				OR
294  * 	- with preemption disabled - from arch/xxx/kernel/kprobes.c
295  */
296 struct kprobe __kprobes *get_kprobe(void *addr)
297 {
298 	struct hlist_head *head;
299 	struct hlist_node *node;
300 	struct kprobe *p;
301 
302 	head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
303 	hlist_for_each_entry_rcu(p, node, head, hlist) {
304 		if (p->addr == addr)
305 			return p;
306 	}
307 	return NULL;
308 }
309 
310 /* Arm a kprobe with text_mutex */
311 static void __kprobes arm_kprobe(struct kprobe *kp)
312 {
313 	mutex_lock(&text_mutex);
314 	arch_arm_kprobe(kp);
315 	mutex_unlock(&text_mutex);
316 }
317 
318 /* Disarm a kprobe with text_mutex */
319 static void __kprobes disarm_kprobe(struct kprobe *kp)
320 {
321 	mutex_lock(&text_mutex);
322 	arch_disarm_kprobe(kp);
323 	mutex_unlock(&text_mutex);
324 }
325 
326 /*
327  * Aggregate handlers for multiple kprobes support - these handlers
328  * take care of invoking the individual kprobe handlers on p->list
329  */
330 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
331 {
332 	struct kprobe *kp;
333 
334 	list_for_each_entry_rcu(kp, &p->list, list) {
335 		if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
336 			set_kprobe_instance(kp);
337 			if (kp->pre_handler(kp, regs))
338 				return 1;
339 		}
340 		reset_kprobe_instance();
341 	}
342 	return 0;
343 }
344 
345 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
346 					unsigned long flags)
347 {
348 	struct kprobe *kp;
349 
350 	list_for_each_entry_rcu(kp, &p->list, list) {
351 		if (kp->post_handler && likely(!kprobe_disabled(kp))) {
352 			set_kprobe_instance(kp);
353 			kp->post_handler(kp, regs, flags);
354 			reset_kprobe_instance();
355 		}
356 	}
357 }
358 
359 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
360 					int trapnr)
361 {
362 	struct kprobe *cur = __get_cpu_var(kprobe_instance);
363 
364 	/*
365 	 * if we faulted "during" the execution of a user specified
366 	 * probe handler, invoke just that probe's fault handler
367 	 */
368 	if (cur && cur->fault_handler) {
369 		if (cur->fault_handler(cur, regs, trapnr))
370 			return 1;
371 	}
372 	return 0;
373 }
374 
375 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
376 {
377 	struct kprobe *cur = __get_cpu_var(kprobe_instance);
378 	int ret = 0;
379 
380 	if (cur && cur->break_handler) {
381 		if (cur->break_handler(cur, regs))
382 			ret = 1;
383 	}
384 	reset_kprobe_instance();
385 	return ret;
386 }
387 
388 /* Walks the list and increments nmissed count for multiprobe case */
389 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
390 {
391 	struct kprobe *kp;
392 	if (p->pre_handler != aggr_pre_handler) {
393 		p->nmissed++;
394 	} else {
395 		list_for_each_entry_rcu(kp, &p->list, list)
396 			kp->nmissed++;
397 	}
398 	return;
399 }
400 
401 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
402 				struct hlist_head *head)
403 {
404 	struct kretprobe *rp = ri->rp;
405 
406 	/* remove rp inst off the rprobe_inst_table */
407 	hlist_del(&ri->hlist);
408 	INIT_HLIST_NODE(&ri->hlist);
409 	if (likely(rp)) {
410 		spin_lock(&rp->lock);
411 		hlist_add_head(&ri->hlist, &rp->free_instances);
412 		spin_unlock(&rp->lock);
413 	} else
414 		/* Unregistering */
415 		hlist_add_head(&ri->hlist, head);
416 }
417 
418 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
419 			 struct hlist_head **head, unsigned long *flags)
420 {
421 	unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
422 	spinlock_t *hlist_lock;
423 
424 	*head = &kretprobe_inst_table[hash];
425 	hlist_lock = kretprobe_table_lock_ptr(hash);
426 	spin_lock_irqsave(hlist_lock, *flags);
427 }
428 
429 static void __kprobes kretprobe_table_lock(unsigned long hash,
430 	unsigned long *flags)
431 {
432 	spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
433 	spin_lock_irqsave(hlist_lock, *flags);
434 }
435 
436 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
437 	unsigned long *flags)
438 {
439 	unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
440 	spinlock_t *hlist_lock;
441 
442 	hlist_lock = kretprobe_table_lock_ptr(hash);
443 	spin_unlock_irqrestore(hlist_lock, *flags);
444 }
445 
446 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
447 {
448 	spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
449 	spin_unlock_irqrestore(hlist_lock, *flags);
450 }
451 
452 /*
453  * This function is called from finish_task_switch when task tk becomes dead,
454  * so that we can recycle any function-return probe instances associated
455  * with this task. These left over instances represent probed functions
456  * that have been called but will never return.
457  */
458 void __kprobes kprobe_flush_task(struct task_struct *tk)
459 {
460 	struct kretprobe_instance *ri;
461 	struct hlist_head *head, empty_rp;
462 	struct hlist_node *node, *tmp;
463 	unsigned long hash, flags = 0;
464 
465 	if (unlikely(!kprobes_initialized))
466 		/* Early boot.  kretprobe_table_locks not yet initialized. */
467 		return;
468 
469 	hash = hash_ptr(tk, KPROBE_HASH_BITS);
470 	head = &kretprobe_inst_table[hash];
471 	kretprobe_table_lock(hash, &flags);
472 	hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
473 		if (ri->task == tk)
474 			recycle_rp_inst(ri, &empty_rp);
475 	}
476 	kretprobe_table_unlock(hash, &flags);
477 	INIT_HLIST_HEAD(&empty_rp);
478 	hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
479 		hlist_del(&ri->hlist);
480 		kfree(ri);
481 	}
482 }
483 
484 static inline void free_rp_inst(struct kretprobe *rp)
485 {
486 	struct kretprobe_instance *ri;
487 	struct hlist_node *pos, *next;
488 
489 	hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
490 		hlist_del(&ri->hlist);
491 		kfree(ri);
492 	}
493 }
494 
495 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
496 {
497 	unsigned long flags, hash;
498 	struct kretprobe_instance *ri;
499 	struct hlist_node *pos, *next;
500 	struct hlist_head *head;
501 
502 	/* No race here */
503 	for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
504 		kretprobe_table_lock(hash, &flags);
505 		head = &kretprobe_inst_table[hash];
506 		hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
507 			if (ri->rp == rp)
508 				ri->rp = NULL;
509 		}
510 		kretprobe_table_unlock(hash, &flags);
511 	}
512 	free_rp_inst(rp);
513 }
514 
515 /*
516  * Keep all fields in the kprobe consistent
517  */
518 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
519 {
520 	memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
521 	memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
522 }
523 
524 /*
525 * Add the new probe to ap->list. Fail if this is the
526 * second jprobe at the address - two jprobes can't coexist
527 */
528 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
529 {
530 	BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
531 	if (p->break_handler) {
532 		if (ap->break_handler)
533 			return -EEXIST;
534 		list_add_tail_rcu(&p->list, &ap->list);
535 		ap->break_handler = aggr_break_handler;
536 	} else
537 		list_add_rcu(&p->list, &ap->list);
538 	if (p->post_handler && !ap->post_handler)
539 		ap->post_handler = aggr_post_handler;
540 
541 	if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
542 		ap->flags &= ~KPROBE_FLAG_DISABLED;
543 		if (!kprobes_all_disarmed)
544 			/* Arm the breakpoint again. */
545 			arm_kprobe(ap);
546 	}
547 	return 0;
548 }
549 
550 /*
551  * Fill in the required fields of the "manager kprobe". Replace the
552  * earlier kprobe in the hlist with the manager kprobe
553  */
554 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
555 {
556 	copy_kprobe(p, ap);
557 	flush_insn_slot(ap);
558 	ap->addr = p->addr;
559 	ap->flags = p->flags;
560 	ap->pre_handler = aggr_pre_handler;
561 	ap->fault_handler = aggr_fault_handler;
562 	/* We don't care the kprobe which has gone. */
563 	if (p->post_handler && !kprobe_gone(p))
564 		ap->post_handler = aggr_post_handler;
565 	if (p->break_handler && !kprobe_gone(p))
566 		ap->break_handler = aggr_break_handler;
567 
568 	INIT_LIST_HEAD(&ap->list);
569 	list_add_rcu(&p->list, &ap->list);
570 
571 	hlist_replace_rcu(&p->hlist, &ap->hlist);
572 }
573 
574 /*
575  * This is the second or subsequent kprobe at the address - handle
576  * the intricacies
577  */
578 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
579 					  struct kprobe *p)
580 {
581 	int ret = 0;
582 	struct kprobe *ap = old_p;
583 
584 	if (old_p->pre_handler != aggr_pre_handler) {
585 		/* If old_p is not an aggr_probe, create new aggr_kprobe. */
586 		ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
587 		if (!ap)
588 			return -ENOMEM;
589 		add_aggr_kprobe(ap, old_p);
590 	}
591 
592 	if (kprobe_gone(ap)) {
593 		/*
594 		 * Attempting to insert new probe at the same location that
595 		 * had a probe in the module vaddr area which already
596 		 * freed. So, the instruction slot has already been
597 		 * released. We need a new slot for the new probe.
598 		 */
599 		ret = arch_prepare_kprobe(ap);
600 		if (ret)
601 			/*
602 			 * Even if fail to allocate new slot, don't need to
603 			 * free aggr_probe. It will be used next time, or
604 			 * freed by unregister_kprobe.
605 			 */
606 			return ret;
607 
608 		/*
609 		 * Clear gone flag to prevent allocating new slot again, and
610 		 * set disabled flag because it is not armed yet.
611 		 */
612 		ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
613 			    | KPROBE_FLAG_DISABLED;
614 	}
615 
616 	copy_kprobe(ap, p);
617 	return add_new_kprobe(ap, p);
618 }
619 
620 /* Try to disable aggr_kprobe, and return 1 if succeeded.*/
621 static int __kprobes try_to_disable_aggr_kprobe(struct kprobe *p)
622 {
623 	struct kprobe *kp;
624 
625 	list_for_each_entry_rcu(kp, &p->list, list) {
626 		if (!kprobe_disabled(kp))
627 			/*
628 			 * There is an active probe on the list.
629 			 * We can't disable aggr_kprobe.
630 			 */
631 			return 0;
632 	}
633 	p->flags |= KPROBE_FLAG_DISABLED;
634 	return 1;
635 }
636 
637 static int __kprobes in_kprobes_functions(unsigned long addr)
638 {
639 	struct kprobe_blackpoint *kb;
640 
641 	if (addr >= (unsigned long)__kprobes_text_start &&
642 	    addr < (unsigned long)__kprobes_text_end)
643 		return -EINVAL;
644 	/*
645 	 * If there exists a kprobe_blacklist, verify and
646 	 * fail any probe registration in the prohibited area
647 	 */
648 	for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
649 		if (kb->start_addr) {
650 			if (addr >= kb->start_addr &&
651 			    addr < (kb->start_addr + kb->range))
652 				return -EINVAL;
653 		}
654 	}
655 	return 0;
656 }
657 
658 /*
659  * If we have a symbol_name argument, look it up and add the offset field
660  * to it. This way, we can specify a relative address to a symbol.
661  */
662 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
663 {
664 	kprobe_opcode_t *addr = p->addr;
665 	if (p->symbol_name) {
666 		if (addr)
667 			return NULL;
668 		kprobe_lookup_name(p->symbol_name, addr);
669 	}
670 
671 	if (!addr)
672 		return NULL;
673 	return (kprobe_opcode_t *)(((char *)addr) + p->offset);
674 }
675 
676 int __kprobes register_kprobe(struct kprobe *p)
677 {
678 	int ret = 0;
679 	struct kprobe *old_p;
680 	struct module *probed_mod;
681 	kprobe_opcode_t *addr;
682 
683 	addr = kprobe_addr(p);
684 	if (!addr)
685 		return -EINVAL;
686 	p->addr = addr;
687 
688 	preempt_disable();
689 	if (!kernel_text_address((unsigned long) p->addr) ||
690 	    in_kprobes_functions((unsigned long) p->addr)) {
691 		preempt_enable();
692 		return -EINVAL;
693 	}
694 
695 	/* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
696 	p->flags &= KPROBE_FLAG_DISABLED;
697 
698 	/*
699 	 * Check if are we probing a module.
700 	 */
701 	probed_mod = __module_text_address((unsigned long) p->addr);
702 	if (probed_mod) {
703 		/*
704 		 * We must hold a refcount of the probed module while updating
705 		 * its code to prohibit unexpected unloading.
706 		 */
707 		if (unlikely(!try_module_get(probed_mod))) {
708 			preempt_enable();
709 			return -EINVAL;
710 		}
711 		/*
712 		 * If the module freed .init.text, we couldn't insert
713 		 * kprobes in there.
714 		 */
715 		if (within_module_init((unsigned long)p->addr, probed_mod) &&
716 		    probed_mod->state != MODULE_STATE_COMING) {
717 			module_put(probed_mod);
718 			preempt_enable();
719 			return -EINVAL;
720 		}
721 	}
722 	preempt_enable();
723 
724 	p->nmissed = 0;
725 	INIT_LIST_HEAD(&p->list);
726 	mutex_lock(&kprobe_mutex);
727 	old_p = get_kprobe(p->addr);
728 	if (old_p) {
729 		ret = register_aggr_kprobe(old_p, p);
730 		goto out;
731 	}
732 
733 	mutex_lock(&text_mutex);
734 	ret = arch_prepare_kprobe(p);
735 	if (ret)
736 		goto out_unlock_text;
737 
738 	INIT_HLIST_NODE(&p->hlist);
739 	hlist_add_head_rcu(&p->hlist,
740 		       &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
741 
742 	if (!kprobes_all_disarmed && !kprobe_disabled(p))
743 		arch_arm_kprobe(p);
744 
745 out_unlock_text:
746 	mutex_unlock(&text_mutex);
747 out:
748 	mutex_unlock(&kprobe_mutex);
749 
750 	if (probed_mod)
751 		module_put(probed_mod);
752 
753 	return ret;
754 }
755 EXPORT_SYMBOL_GPL(register_kprobe);
756 
757 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
758 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
759 {
760 	struct kprobe *old_p, *list_p;
761 
762 	old_p = get_kprobe(p->addr);
763 	if (unlikely(!old_p))
764 		return NULL;
765 
766 	if (p != old_p) {
767 		list_for_each_entry_rcu(list_p, &old_p->list, list)
768 			if (list_p == p)
769 			/* kprobe p is a valid probe */
770 				goto valid;
771 		return NULL;
772 	}
773 valid:
774 	return old_p;
775 }
776 
777 /*
778  * Unregister a kprobe without a scheduler synchronization.
779  */
780 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
781 {
782 	struct kprobe *old_p, *list_p;
783 
784 	old_p = __get_valid_kprobe(p);
785 	if (old_p == NULL)
786 		return -EINVAL;
787 
788 	if (old_p == p ||
789 	    (old_p->pre_handler == aggr_pre_handler &&
790 	     list_is_singular(&old_p->list))) {
791 		/*
792 		 * Only probe on the hash list. Disarm only if kprobes are
793 		 * enabled and not gone - otherwise, the breakpoint would
794 		 * already have been removed. We save on flushing icache.
795 		 */
796 		if (!kprobes_all_disarmed && !kprobe_disabled(old_p))
797 			disarm_kprobe(p);
798 		hlist_del_rcu(&old_p->hlist);
799 	} else {
800 		if (p->break_handler && !kprobe_gone(p))
801 			old_p->break_handler = NULL;
802 		if (p->post_handler && !kprobe_gone(p)) {
803 			list_for_each_entry_rcu(list_p, &old_p->list, list) {
804 				if ((list_p != p) && (list_p->post_handler))
805 					goto noclean;
806 			}
807 			old_p->post_handler = NULL;
808 		}
809 noclean:
810 		list_del_rcu(&p->list);
811 		if (!kprobe_disabled(old_p)) {
812 			try_to_disable_aggr_kprobe(old_p);
813 			if (!kprobes_all_disarmed && kprobe_disabled(old_p))
814 				disarm_kprobe(old_p);
815 		}
816 	}
817 	return 0;
818 }
819 
820 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
821 {
822 	struct kprobe *old_p;
823 
824 	if (list_empty(&p->list))
825 		arch_remove_kprobe(p);
826 	else if (list_is_singular(&p->list)) {
827 		/* "p" is the last child of an aggr_kprobe */
828 		old_p = list_entry(p->list.next, struct kprobe, list);
829 		list_del(&p->list);
830 		arch_remove_kprobe(old_p);
831 		kfree(old_p);
832 	}
833 }
834 
835 int __kprobes register_kprobes(struct kprobe **kps, int num)
836 {
837 	int i, ret = 0;
838 
839 	if (num <= 0)
840 		return -EINVAL;
841 	for (i = 0; i < num; i++) {
842 		ret = register_kprobe(kps[i]);
843 		if (ret < 0) {
844 			if (i > 0)
845 				unregister_kprobes(kps, i);
846 			break;
847 		}
848 	}
849 	return ret;
850 }
851 EXPORT_SYMBOL_GPL(register_kprobes);
852 
853 void __kprobes unregister_kprobe(struct kprobe *p)
854 {
855 	unregister_kprobes(&p, 1);
856 }
857 EXPORT_SYMBOL_GPL(unregister_kprobe);
858 
859 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
860 {
861 	int i;
862 
863 	if (num <= 0)
864 		return;
865 	mutex_lock(&kprobe_mutex);
866 	for (i = 0; i < num; i++)
867 		if (__unregister_kprobe_top(kps[i]) < 0)
868 			kps[i]->addr = NULL;
869 	mutex_unlock(&kprobe_mutex);
870 
871 	synchronize_sched();
872 	for (i = 0; i < num; i++)
873 		if (kps[i]->addr)
874 			__unregister_kprobe_bottom(kps[i]);
875 }
876 EXPORT_SYMBOL_GPL(unregister_kprobes);
877 
878 static struct notifier_block kprobe_exceptions_nb = {
879 	.notifier_call = kprobe_exceptions_notify,
880 	.priority = 0x7fffffff /* we need to be notified first */
881 };
882 
883 unsigned long __weak arch_deref_entry_point(void *entry)
884 {
885 	return (unsigned long)entry;
886 }
887 
888 int __kprobes register_jprobes(struct jprobe **jps, int num)
889 {
890 	struct jprobe *jp;
891 	int ret = 0, i;
892 
893 	if (num <= 0)
894 		return -EINVAL;
895 	for (i = 0; i < num; i++) {
896 		unsigned long addr;
897 		jp = jps[i];
898 		addr = arch_deref_entry_point(jp->entry);
899 
900 		if (!kernel_text_address(addr))
901 			ret = -EINVAL;
902 		else {
903 			/* Todo: Verify probepoint is a function entry point */
904 			jp->kp.pre_handler = setjmp_pre_handler;
905 			jp->kp.break_handler = longjmp_break_handler;
906 			ret = register_kprobe(&jp->kp);
907 		}
908 		if (ret < 0) {
909 			if (i > 0)
910 				unregister_jprobes(jps, i);
911 			break;
912 		}
913 	}
914 	return ret;
915 }
916 EXPORT_SYMBOL_GPL(register_jprobes);
917 
918 int __kprobes register_jprobe(struct jprobe *jp)
919 {
920 	return register_jprobes(&jp, 1);
921 }
922 EXPORT_SYMBOL_GPL(register_jprobe);
923 
924 void __kprobes unregister_jprobe(struct jprobe *jp)
925 {
926 	unregister_jprobes(&jp, 1);
927 }
928 EXPORT_SYMBOL_GPL(unregister_jprobe);
929 
930 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
931 {
932 	int i;
933 
934 	if (num <= 0)
935 		return;
936 	mutex_lock(&kprobe_mutex);
937 	for (i = 0; i < num; i++)
938 		if (__unregister_kprobe_top(&jps[i]->kp) < 0)
939 			jps[i]->kp.addr = NULL;
940 	mutex_unlock(&kprobe_mutex);
941 
942 	synchronize_sched();
943 	for (i = 0; i < num; i++) {
944 		if (jps[i]->kp.addr)
945 			__unregister_kprobe_bottom(&jps[i]->kp);
946 	}
947 }
948 EXPORT_SYMBOL_GPL(unregister_jprobes);
949 
950 #ifdef CONFIG_KRETPROBES
951 /*
952  * This kprobe pre_handler is registered with every kretprobe. When probe
953  * hits it will set up the return probe.
954  */
955 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
956 					   struct pt_regs *regs)
957 {
958 	struct kretprobe *rp = container_of(p, struct kretprobe, kp);
959 	unsigned long hash, flags = 0;
960 	struct kretprobe_instance *ri;
961 
962 	/*TODO: consider to only swap the RA after the last pre_handler fired */
963 	hash = hash_ptr(current, KPROBE_HASH_BITS);
964 	spin_lock_irqsave(&rp->lock, flags);
965 	if (!hlist_empty(&rp->free_instances)) {
966 		ri = hlist_entry(rp->free_instances.first,
967 				struct kretprobe_instance, hlist);
968 		hlist_del(&ri->hlist);
969 		spin_unlock_irqrestore(&rp->lock, flags);
970 
971 		ri->rp = rp;
972 		ri->task = current;
973 
974 		if (rp->entry_handler && rp->entry_handler(ri, regs))
975 			return 0;
976 
977 		arch_prepare_kretprobe(ri, regs);
978 
979 		/* XXX(hch): why is there no hlist_move_head? */
980 		INIT_HLIST_NODE(&ri->hlist);
981 		kretprobe_table_lock(hash, &flags);
982 		hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
983 		kretprobe_table_unlock(hash, &flags);
984 	} else {
985 		rp->nmissed++;
986 		spin_unlock_irqrestore(&rp->lock, flags);
987 	}
988 	return 0;
989 }
990 
991 int __kprobes register_kretprobe(struct kretprobe *rp)
992 {
993 	int ret = 0;
994 	struct kretprobe_instance *inst;
995 	int i;
996 	void *addr;
997 
998 	if (kretprobe_blacklist_size) {
999 		addr = kprobe_addr(&rp->kp);
1000 		if (!addr)
1001 			return -EINVAL;
1002 
1003 		for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1004 			if (kretprobe_blacklist[i].addr == addr)
1005 				return -EINVAL;
1006 		}
1007 	}
1008 
1009 	rp->kp.pre_handler = pre_handler_kretprobe;
1010 	rp->kp.post_handler = NULL;
1011 	rp->kp.fault_handler = NULL;
1012 	rp->kp.break_handler = NULL;
1013 
1014 	/* Pre-allocate memory for max kretprobe instances */
1015 	if (rp->maxactive <= 0) {
1016 #ifdef CONFIG_PREEMPT
1017 		rp->maxactive = max(10, 2 * NR_CPUS);
1018 #else
1019 		rp->maxactive = NR_CPUS;
1020 #endif
1021 	}
1022 	spin_lock_init(&rp->lock);
1023 	INIT_HLIST_HEAD(&rp->free_instances);
1024 	for (i = 0; i < rp->maxactive; i++) {
1025 		inst = kmalloc(sizeof(struct kretprobe_instance) +
1026 			       rp->data_size, GFP_KERNEL);
1027 		if (inst == NULL) {
1028 			free_rp_inst(rp);
1029 			return -ENOMEM;
1030 		}
1031 		INIT_HLIST_NODE(&inst->hlist);
1032 		hlist_add_head(&inst->hlist, &rp->free_instances);
1033 	}
1034 
1035 	rp->nmissed = 0;
1036 	/* Establish function entry probe point */
1037 	ret = register_kprobe(&rp->kp);
1038 	if (ret != 0)
1039 		free_rp_inst(rp);
1040 	return ret;
1041 }
1042 EXPORT_SYMBOL_GPL(register_kretprobe);
1043 
1044 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1045 {
1046 	int ret = 0, i;
1047 
1048 	if (num <= 0)
1049 		return -EINVAL;
1050 	for (i = 0; i < num; i++) {
1051 		ret = register_kretprobe(rps[i]);
1052 		if (ret < 0) {
1053 			if (i > 0)
1054 				unregister_kretprobes(rps, i);
1055 			break;
1056 		}
1057 	}
1058 	return ret;
1059 }
1060 EXPORT_SYMBOL_GPL(register_kretprobes);
1061 
1062 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1063 {
1064 	unregister_kretprobes(&rp, 1);
1065 }
1066 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1067 
1068 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1069 {
1070 	int i;
1071 
1072 	if (num <= 0)
1073 		return;
1074 	mutex_lock(&kprobe_mutex);
1075 	for (i = 0; i < num; i++)
1076 		if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1077 			rps[i]->kp.addr = NULL;
1078 	mutex_unlock(&kprobe_mutex);
1079 
1080 	synchronize_sched();
1081 	for (i = 0; i < num; i++) {
1082 		if (rps[i]->kp.addr) {
1083 			__unregister_kprobe_bottom(&rps[i]->kp);
1084 			cleanup_rp_inst(rps[i]);
1085 		}
1086 	}
1087 }
1088 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1089 
1090 #else /* CONFIG_KRETPROBES */
1091 int __kprobes register_kretprobe(struct kretprobe *rp)
1092 {
1093 	return -ENOSYS;
1094 }
1095 EXPORT_SYMBOL_GPL(register_kretprobe);
1096 
1097 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1098 {
1099 	return -ENOSYS;
1100 }
1101 EXPORT_SYMBOL_GPL(register_kretprobes);
1102 
1103 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1104 {
1105 }
1106 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1107 
1108 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1109 {
1110 }
1111 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1112 
1113 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1114 					   struct pt_regs *regs)
1115 {
1116 	return 0;
1117 }
1118 
1119 #endif /* CONFIG_KRETPROBES */
1120 
1121 /* Set the kprobe gone and remove its instruction buffer. */
1122 static void __kprobes kill_kprobe(struct kprobe *p)
1123 {
1124 	struct kprobe *kp;
1125 
1126 	p->flags |= KPROBE_FLAG_GONE;
1127 	if (p->pre_handler == aggr_pre_handler) {
1128 		/*
1129 		 * If this is an aggr_kprobe, we have to list all the
1130 		 * chained probes and mark them GONE.
1131 		 */
1132 		list_for_each_entry_rcu(kp, &p->list, list)
1133 			kp->flags |= KPROBE_FLAG_GONE;
1134 		p->post_handler = NULL;
1135 		p->break_handler = NULL;
1136 	}
1137 	/*
1138 	 * Here, we can remove insn_slot safely, because no thread calls
1139 	 * the original probed function (which will be freed soon) any more.
1140 	 */
1141 	arch_remove_kprobe(p);
1142 }
1143 
1144 /* Module notifier call back, checking kprobes on the module */
1145 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
1146 					     unsigned long val, void *data)
1147 {
1148 	struct module *mod = data;
1149 	struct hlist_head *head;
1150 	struct hlist_node *node;
1151 	struct kprobe *p;
1152 	unsigned int i;
1153 	int checkcore = (val == MODULE_STATE_GOING);
1154 
1155 	if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
1156 		return NOTIFY_DONE;
1157 
1158 	/*
1159 	 * When MODULE_STATE_GOING was notified, both of module .text and
1160 	 * .init.text sections would be freed. When MODULE_STATE_LIVE was
1161 	 * notified, only .init.text section would be freed. We need to
1162 	 * disable kprobes which have been inserted in the sections.
1163 	 */
1164 	mutex_lock(&kprobe_mutex);
1165 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1166 		head = &kprobe_table[i];
1167 		hlist_for_each_entry_rcu(p, node, head, hlist)
1168 			if (within_module_init((unsigned long)p->addr, mod) ||
1169 			    (checkcore &&
1170 			     within_module_core((unsigned long)p->addr, mod))) {
1171 				/*
1172 				 * The vaddr this probe is installed will soon
1173 				 * be vfreed buy not synced to disk. Hence,
1174 				 * disarming the breakpoint isn't needed.
1175 				 */
1176 				kill_kprobe(p);
1177 			}
1178 	}
1179 	mutex_unlock(&kprobe_mutex);
1180 	return NOTIFY_DONE;
1181 }
1182 
1183 static struct notifier_block kprobe_module_nb = {
1184 	.notifier_call = kprobes_module_callback,
1185 	.priority = 0
1186 };
1187 
1188 static int __init init_kprobes(void)
1189 {
1190 	int i, err = 0;
1191 	unsigned long offset = 0, size = 0;
1192 	char *modname, namebuf[128];
1193 	const char *symbol_name;
1194 	void *addr;
1195 	struct kprobe_blackpoint *kb;
1196 
1197 	/* FIXME allocate the probe table, currently defined statically */
1198 	/* initialize all list heads */
1199 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1200 		INIT_HLIST_HEAD(&kprobe_table[i]);
1201 		INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1202 		spin_lock_init(&(kretprobe_table_locks[i].lock));
1203 	}
1204 
1205 	/*
1206 	 * Lookup and populate the kprobe_blacklist.
1207 	 *
1208 	 * Unlike the kretprobe blacklist, we'll need to determine
1209 	 * the range of addresses that belong to the said functions,
1210 	 * since a kprobe need not necessarily be at the beginning
1211 	 * of a function.
1212 	 */
1213 	for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1214 		kprobe_lookup_name(kb->name, addr);
1215 		if (!addr)
1216 			continue;
1217 
1218 		kb->start_addr = (unsigned long)addr;
1219 		symbol_name = kallsyms_lookup(kb->start_addr,
1220 				&size, &offset, &modname, namebuf);
1221 		if (!symbol_name)
1222 			kb->range = 0;
1223 		else
1224 			kb->range = size;
1225 	}
1226 
1227 	if (kretprobe_blacklist_size) {
1228 		/* lookup the function address from its name */
1229 		for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1230 			kprobe_lookup_name(kretprobe_blacklist[i].name,
1231 					   kretprobe_blacklist[i].addr);
1232 			if (!kretprobe_blacklist[i].addr)
1233 				printk("kretprobe: lookup failed: %s\n",
1234 				       kretprobe_blacklist[i].name);
1235 		}
1236 	}
1237 
1238 	/* By default, kprobes are armed */
1239 	kprobes_all_disarmed = false;
1240 
1241 	err = arch_init_kprobes();
1242 	if (!err)
1243 		err = register_die_notifier(&kprobe_exceptions_nb);
1244 	if (!err)
1245 		err = register_module_notifier(&kprobe_module_nb);
1246 
1247 	kprobes_initialized = (err == 0);
1248 
1249 	if (!err)
1250 		init_test_probes();
1251 	return err;
1252 }
1253 
1254 #ifdef CONFIG_DEBUG_FS
1255 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1256 		const char *sym, int offset,char *modname)
1257 {
1258 	char *kprobe_type;
1259 
1260 	if (p->pre_handler == pre_handler_kretprobe)
1261 		kprobe_type = "r";
1262 	else if (p->pre_handler == setjmp_pre_handler)
1263 		kprobe_type = "j";
1264 	else
1265 		kprobe_type = "k";
1266 	if (sym)
1267 		seq_printf(pi, "%p  %s  %s+0x%x  %s %s%s\n",
1268 			p->addr, kprobe_type, sym, offset,
1269 			(modname ? modname : " "),
1270 			(kprobe_gone(p) ? "[GONE]" : ""),
1271 			((kprobe_disabled(p) && !kprobe_gone(p)) ?
1272 			 "[DISABLED]" : ""));
1273 	else
1274 		seq_printf(pi, "%p  %s  %p %s%s\n",
1275 			p->addr, kprobe_type, p->addr,
1276 			(kprobe_gone(p) ? "[GONE]" : ""),
1277 			((kprobe_disabled(p) && !kprobe_gone(p)) ?
1278 			 "[DISABLED]" : ""));
1279 }
1280 
1281 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1282 {
1283 	return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1284 }
1285 
1286 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1287 {
1288 	(*pos)++;
1289 	if (*pos >= KPROBE_TABLE_SIZE)
1290 		return NULL;
1291 	return pos;
1292 }
1293 
1294 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1295 {
1296 	/* Nothing to do */
1297 }
1298 
1299 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1300 {
1301 	struct hlist_head *head;
1302 	struct hlist_node *node;
1303 	struct kprobe *p, *kp;
1304 	const char *sym = NULL;
1305 	unsigned int i = *(loff_t *) v;
1306 	unsigned long offset = 0;
1307 	char *modname, namebuf[128];
1308 
1309 	head = &kprobe_table[i];
1310 	preempt_disable();
1311 	hlist_for_each_entry_rcu(p, node, head, hlist) {
1312 		sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1313 					&offset, &modname, namebuf);
1314 		if (p->pre_handler == aggr_pre_handler) {
1315 			list_for_each_entry_rcu(kp, &p->list, list)
1316 				report_probe(pi, kp, sym, offset, modname);
1317 		} else
1318 			report_probe(pi, p, sym, offset, modname);
1319 	}
1320 	preempt_enable();
1321 	return 0;
1322 }
1323 
1324 static const struct seq_operations kprobes_seq_ops = {
1325 	.start = kprobe_seq_start,
1326 	.next  = kprobe_seq_next,
1327 	.stop  = kprobe_seq_stop,
1328 	.show  = show_kprobe_addr
1329 };
1330 
1331 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1332 {
1333 	return seq_open(filp, &kprobes_seq_ops);
1334 }
1335 
1336 static const struct file_operations debugfs_kprobes_operations = {
1337 	.open           = kprobes_open,
1338 	.read           = seq_read,
1339 	.llseek         = seq_lseek,
1340 	.release        = seq_release,
1341 };
1342 
1343 /* Disable one kprobe */
1344 int __kprobes disable_kprobe(struct kprobe *kp)
1345 {
1346 	int ret = 0;
1347 	struct kprobe *p;
1348 
1349 	mutex_lock(&kprobe_mutex);
1350 
1351 	/* Check whether specified probe is valid. */
1352 	p = __get_valid_kprobe(kp);
1353 	if (unlikely(p == NULL)) {
1354 		ret = -EINVAL;
1355 		goto out;
1356 	}
1357 
1358 	/* If the probe is already disabled (or gone), just return */
1359 	if (kprobe_disabled(kp))
1360 		goto out;
1361 
1362 	kp->flags |= KPROBE_FLAG_DISABLED;
1363 	if (p != kp)
1364 		/* When kp != p, p is always enabled. */
1365 		try_to_disable_aggr_kprobe(p);
1366 
1367 	if (!kprobes_all_disarmed && kprobe_disabled(p))
1368 		disarm_kprobe(p);
1369 out:
1370 	mutex_unlock(&kprobe_mutex);
1371 	return ret;
1372 }
1373 EXPORT_SYMBOL_GPL(disable_kprobe);
1374 
1375 /* Enable one kprobe */
1376 int __kprobes enable_kprobe(struct kprobe *kp)
1377 {
1378 	int ret = 0;
1379 	struct kprobe *p;
1380 
1381 	mutex_lock(&kprobe_mutex);
1382 
1383 	/* Check whether specified probe is valid. */
1384 	p = __get_valid_kprobe(kp);
1385 	if (unlikely(p == NULL)) {
1386 		ret = -EINVAL;
1387 		goto out;
1388 	}
1389 
1390 	if (kprobe_gone(kp)) {
1391 		/* This kprobe has gone, we couldn't enable it. */
1392 		ret = -EINVAL;
1393 		goto out;
1394 	}
1395 
1396 	if (!kprobes_all_disarmed && kprobe_disabled(p))
1397 		arm_kprobe(p);
1398 
1399 	p->flags &= ~KPROBE_FLAG_DISABLED;
1400 	if (p != kp)
1401 		kp->flags &= ~KPROBE_FLAG_DISABLED;
1402 out:
1403 	mutex_unlock(&kprobe_mutex);
1404 	return ret;
1405 }
1406 EXPORT_SYMBOL_GPL(enable_kprobe);
1407 
1408 static void __kprobes arm_all_kprobes(void)
1409 {
1410 	struct hlist_head *head;
1411 	struct hlist_node *node;
1412 	struct kprobe *p;
1413 	unsigned int i;
1414 
1415 	mutex_lock(&kprobe_mutex);
1416 
1417 	/* If kprobes are armed, just return */
1418 	if (!kprobes_all_disarmed)
1419 		goto already_enabled;
1420 
1421 	mutex_lock(&text_mutex);
1422 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1423 		head = &kprobe_table[i];
1424 		hlist_for_each_entry_rcu(p, node, head, hlist)
1425 			if (!kprobe_disabled(p))
1426 				arch_arm_kprobe(p);
1427 	}
1428 	mutex_unlock(&text_mutex);
1429 
1430 	kprobes_all_disarmed = false;
1431 	printk(KERN_INFO "Kprobes globally enabled\n");
1432 
1433 already_enabled:
1434 	mutex_unlock(&kprobe_mutex);
1435 	return;
1436 }
1437 
1438 static void __kprobes disarm_all_kprobes(void)
1439 {
1440 	struct hlist_head *head;
1441 	struct hlist_node *node;
1442 	struct kprobe *p;
1443 	unsigned int i;
1444 
1445 	mutex_lock(&kprobe_mutex);
1446 
1447 	/* If kprobes are already disarmed, just return */
1448 	if (kprobes_all_disarmed)
1449 		goto already_disabled;
1450 
1451 	kprobes_all_disarmed = true;
1452 	printk(KERN_INFO "Kprobes globally disabled\n");
1453 	mutex_lock(&text_mutex);
1454 	for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1455 		head = &kprobe_table[i];
1456 		hlist_for_each_entry_rcu(p, node, head, hlist) {
1457 			if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
1458 				arch_disarm_kprobe(p);
1459 		}
1460 	}
1461 
1462 	mutex_unlock(&text_mutex);
1463 	mutex_unlock(&kprobe_mutex);
1464 	/* Allow all currently running kprobes to complete */
1465 	synchronize_sched();
1466 	return;
1467 
1468 already_disabled:
1469 	mutex_unlock(&kprobe_mutex);
1470 	return;
1471 }
1472 
1473 /*
1474  * XXX: The debugfs bool file interface doesn't allow for callbacks
1475  * when the bool state is switched. We can reuse that facility when
1476  * available
1477  */
1478 static ssize_t read_enabled_file_bool(struct file *file,
1479 	       char __user *user_buf, size_t count, loff_t *ppos)
1480 {
1481 	char buf[3];
1482 
1483 	if (!kprobes_all_disarmed)
1484 		buf[0] = '1';
1485 	else
1486 		buf[0] = '0';
1487 	buf[1] = '\n';
1488 	buf[2] = 0x00;
1489 	return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1490 }
1491 
1492 static ssize_t write_enabled_file_bool(struct file *file,
1493 	       const char __user *user_buf, size_t count, loff_t *ppos)
1494 {
1495 	char buf[32];
1496 	int buf_size;
1497 
1498 	buf_size = min(count, (sizeof(buf)-1));
1499 	if (copy_from_user(buf, user_buf, buf_size))
1500 		return -EFAULT;
1501 
1502 	switch (buf[0]) {
1503 	case 'y':
1504 	case 'Y':
1505 	case '1':
1506 		arm_all_kprobes();
1507 		break;
1508 	case 'n':
1509 	case 'N':
1510 	case '0':
1511 		disarm_all_kprobes();
1512 		break;
1513 	}
1514 
1515 	return count;
1516 }
1517 
1518 static const struct file_operations fops_kp = {
1519 	.read =         read_enabled_file_bool,
1520 	.write =        write_enabled_file_bool,
1521 };
1522 
1523 static int __kprobes debugfs_kprobe_init(void)
1524 {
1525 	struct dentry *dir, *file;
1526 	unsigned int value = 1;
1527 
1528 	dir = debugfs_create_dir("kprobes", NULL);
1529 	if (!dir)
1530 		return -ENOMEM;
1531 
1532 	file = debugfs_create_file("list", 0444, dir, NULL,
1533 				&debugfs_kprobes_operations);
1534 	if (!file) {
1535 		debugfs_remove(dir);
1536 		return -ENOMEM;
1537 	}
1538 
1539 	file = debugfs_create_file("enabled", 0600, dir,
1540 					&value, &fops_kp);
1541 	if (!file) {
1542 		debugfs_remove(dir);
1543 		return -ENOMEM;
1544 	}
1545 
1546 	return 0;
1547 }
1548 
1549 late_initcall(debugfs_kprobe_init);
1550 #endif /* CONFIG_DEBUG_FS */
1551 
1552 module_init(init_kprobes);
1553 
1554 /* defined in arch/.../kernel/kprobes.c */
1555 EXPORT_SYMBOL_GPL(jprobe_return);
1556