kprobes: add (un)register_kretprobes for batch registration
[linux-2.6.git] / kernel / kprobes.c
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
47 #include <asm-generic/sections.h>
48 #include <asm/cacheflush.h>
49 #include <asm/errno.h>
50 #include <asm/uaccess.h>
51
52 #define KPROBE_HASH_BITS 6
53 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
54
55
56 /*
57  * Some oddball architectures like 64bit powerpc have function descriptors
58  * so this must be overridable.
59  */
60 #ifndef kprobe_lookup_name
61 #define kprobe_lookup_name(name, addr) \
62         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
63 #endif
64
65 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
66 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
67
68 /* NOTE: change this value only with kprobe_mutex held */
69 static bool kprobe_enabled;
70
71 DEFINE_MUTEX(kprobe_mutex);             /* Protects kprobe_table */
72 DEFINE_SPINLOCK(kretprobe_lock);        /* Protects kretprobe_inst_table */
73 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
74
75 /*
76  * Normally, functions that we'd want to prohibit kprobes in, are marked
77  * __kprobes. But, there are cases where such functions already belong to
78  * a different section (__sched for preempt_schedule)
79  *
80  * For such cases, we now have a blacklist
81  */
82 struct kprobe_blackpoint kprobe_blacklist[] = {
83         {"preempt_schedule",},
84         {NULL}    /* Terminator */
85 };
86
87 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
88 /*
89  * kprobe->ainsn.insn points to the copy of the instruction to be
90  * single-stepped. x86_64, POWER4 and above have no-exec support and
91  * stepping on the instruction on a vmalloced/kmalloced/data page
92  * is a recipe for disaster
93  */
94 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
95
96 struct kprobe_insn_page {
97         struct hlist_node hlist;
98         kprobe_opcode_t *insns;         /* Page of instruction slots */
99         char slot_used[INSNS_PER_PAGE];
100         int nused;
101         int ngarbage;
102 };
103
104 enum kprobe_slot_state {
105         SLOT_CLEAN = 0,
106         SLOT_DIRTY = 1,
107         SLOT_USED = 2,
108 };
109
110 static struct hlist_head kprobe_insn_pages;
111 static int kprobe_garbage_slots;
112 static int collect_garbage_slots(void);
113
114 static int __kprobes check_safety(void)
115 {
116         int ret = 0;
117 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
118         ret = freeze_processes();
119         if (ret == 0) {
120                 struct task_struct *p, *q;
121                 do_each_thread(p, q) {
122                         if (p != current && p->state == TASK_RUNNING &&
123                             p->pid != 0) {
124                                 printk("Check failed: %s is running\n",p->comm);
125                                 ret = -1;
126                                 goto loop_end;
127                         }
128                 } while_each_thread(p, q);
129         }
130 loop_end:
131         thaw_processes();
132 #else
133         synchronize_sched();
134 #endif
135         return ret;
136 }
137
138 /**
139  * get_insn_slot() - Find a slot on an executable page for an instruction.
140  * We allocate an executable page if there's no room on existing ones.
141  */
142 kprobe_opcode_t __kprobes *get_insn_slot(void)
143 {
144         struct kprobe_insn_page *kip;
145         struct hlist_node *pos;
146
147  retry:
148         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
149                 if (kip->nused < INSNS_PER_PAGE) {
150                         int i;
151                         for (i = 0; i < INSNS_PER_PAGE; i++) {
152                                 if (kip->slot_used[i] == SLOT_CLEAN) {
153                                         kip->slot_used[i] = SLOT_USED;
154                                         kip->nused++;
155                                         return kip->insns + (i * MAX_INSN_SIZE);
156                                 }
157                         }
158                         /* Surprise!  No unused slots.  Fix kip->nused. */
159                         kip->nused = INSNS_PER_PAGE;
160                 }
161         }
162
163         /* If there are any garbage slots, collect it and try again. */
164         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
165                 goto retry;
166         }
167         /* All out of space.  Need to allocate a new page. Use slot 0. */
168         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
169         if (!kip)
170                 return NULL;
171
172         /*
173          * Use module_alloc so this page is within +/- 2GB of where the
174          * kernel image and loaded module images reside. This is required
175          * so x86_64 can correctly handle the %rip-relative fixups.
176          */
177         kip->insns = module_alloc(PAGE_SIZE);
178         if (!kip->insns) {
179                 kfree(kip);
180                 return NULL;
181         }
182         INIT_HLIST_NODE(&kip->hlist);
183         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
184         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
185         kip->slot_used[0] = SLOT_USED;
186         kip->nused = 1;
187         kip->ngarbage = 0;
188         return kip->insns;
189 }
190
191 /* Return 1 if all garbages are collected, otherwise 0. */
192 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
193 {
194         kip->slot_used[idx] = SLOT_CLEAN;
195         kip->nused--;
196         if (kip->nused == 0) {
197                 /*
198                  * Page is no longer in use.  Free it unless
199                  * it's the last one.  We keep the last one
200                  * so as not to have to set it up again the
201                  * next time somebody inserts a probe.
202                  */
203                 hlist_del(&kip->hlist);
204                 if (hlist_empty(&kprobe_insn_pages)) {
205                         INIT_HLIST_NODE(&kip->hlist);
206                         hlist_add_head(&kip->hlist,
207                                        &kprobe_insn_pages);
208                 } else {
209                         module_free(NULL, kip->insns);
210                         kfree(kip);
211                 }
212                 return 1;
213         }
214         return 0;
215 }
216
217 static int __kprobes collect_garbage_slots(void)
218 {
219         struct kprobe_insn_page *kip;
220         struct hlist_node *pos, *next;
221
222         /* Ensure no-one is preepmted on the garbages */
223         if (check_safety() != 0)
224                 return -EAGAIN;
225
226         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
227                 int i;
228                 if (kip->ngarbage == 0)
229                         continue;
230                 kip->ngarbage = 0;      /* we will collect all garbages */
231                 for (i = 0; i < INSNS_PER_PAGE; i++) {
232                         if (kip->slot_used[i] == SLOT_DIRTY &&
233                             collect_one_slot(kip, i))
234                                 break;
235                 }
236         }
237         kprobe_garbage_slots = 0;
238         return 0;
239 }
240
241 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
242 {
243         struct kprobe_insn_page *kip;
244         struct hlist_node *pos;
245
246         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
247                 if (kip->insns <= slot &&
248                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
249                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
250                         if (dirty) {
251                                 kip->slot_used[i] = SLOT_DIRTY;
252                                 kip->ngarbage++;
253                         } else {
254                                 collect_one_slot(kip, i);
255                         }
256                         break;
257                 }
258         }
259
260         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
261                 collect_garbage_slots();
262 }
263 #endif
264
265 /* We have preemption disabled.. so it is safe to use __ versions */
266 static inline void set_kprobe_instance(struct kprobe *kp)
267 {
268         __get_cpu_var(kprobe_instance) = kp;
269 }
270
271 static inline void reset_kprobe_instance(void)
272 {
273         __get_cpu_var(kprobe_instance) = NULL;
274 }
275
276 /*
277  * This routine is called either:
278  *      - under the kprobe_mutex - during kprobe_[un]register()
279  *                              OR
280  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
281  */
282 struct kprobe __kprobes *get_kprobe(void *addr)
283 {
284         struct hlist_head *head;
285         struct hlist_node *node;
286         struct kprobe *p;
287
288         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
289         hlist_for_each_entry_rcu(p, node, head, hlist) {
290                 if (p->addr == addr)
291                         return p;
292         }
293         return NULL;
294 }
295
296 /*
297  * Aggregate handlers for multiple kprobes support - these handlers
298  * take care of invoking the individual kprobe handlers on p->list
299  */
300 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
301 {
302         struct kprobe *kp;
303
304         list_for_each_entry_rcu(kp, &p->list, list) {
305                 if (kp->pre_handler) {
306                         set_kprobe_instance(kp);
307                         if (kp->pre_handler(kp, regs))
308                                 return 1;
309                 }
310                 reset_kprobe_instance();
311         }
312         return 0;
313 }
314
315 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
316                                         unsigned long flags)
317 {
318         struct kprobe *kp;
319
320         list_for_each_entry_rcu(kp, &p->list, list) {
321                 if (kp->post_handler) {
322                         set_kprobe_instance(kp);
323                         kp->post_handler(kp, regs, flags);
324                         reset_kprobe_instance();
325                 }
326         }
327 }
328
329 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
330                                         int trapnr)
331 {
332         struct kprobe *cur = __get_cpu_var(kprobe_instance);
333
334         /*
335          * if we faulted "during" the execution of a user specified
336          * probe handler, invoke just that probe's fault handler
337          */
338         if (cur && cur->fault_handler) {
339                 if (cur->fault_handler(cur, regs, trapnr))
340                         return 1;
341         }
342         return 0;
343 }
344
345 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
346 {
347         struct kprobe *cur = __get_cpu_var(kprobe_instance);
348         int ret = 0;
349
350         if (cur && cur->break_handler) {
351                 if (cur->break_handler(cur, regs))
352                         ret = 1;
353         }
354         reset_kprobe_instance();
355         return ret;
356 }
357
358 /* Walks the list and increments nmissed count for multiprobe case */
359 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
360 {
361         struct kprobe *kp;
362         if (p->pre_handler != aggr_pre_handler) {
363                 p->nmissed++;
364         } else {
365                 list_for_each_entry_rcu(kp, &p->list, list)
366                         kp->nmissed++;
367         }
368         return;
369 }
370
371 /* Called with kretprobe_lock held */
372 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
373                                 struct hlist_head *head)
374 {
375         /* remove rp inst off the rprobe_inst_table */
376         hlist_del(&ri->hlist);
377         if (ri->rp) {
378                 /* remove rp inst off the used list */
379                 hlist_del(&ri->uflist);
380                 /* put rp inst back onto the free list */
381                 INIT_HLIST_NODE(&ri->uflist);
382                 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
383         } else
384                 /* Unregistering */
385                 hlist_add_head(&ri->hlist, head);
386 }
387
388 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
389 {
390         return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
391 }
392
393 /*
394  * This function is called from finish_task_switch when task tk becomes dead,
395  * so that we can recycle any function-return probe instances associated
396  * with this task. These left over instances represent probed functions
397  * that have been called but will never return.
398  */
399 void __kprobes kprobe_flush_task(struct task_struct *tk)
400 {
401         struct kretprobe_instance *ri;
402         struct hlist_head *head, empty_rp;
403         struct hlist_node *node, *tmp;
404         unsigned long flags = 0;
405
406         INIT_HLIST_HEAD(&empty_rp);
407         spin_lock_irqsave(&kretprobe_lock, flags);
408         head = kretprobe_inst_table_head(tk);
409         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
410                 if (ri->task == tk)
411                         recycle_rp_inst(ri, &empty_rp);
412         }
413         spin_unlock_irqrestore(&kretprobe_lock, flags);
414
415         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
416                 hlist_del(&ri->hlist);
417                 kfree(ri);
418         }
419 }
420
421 static inline void free_rp_inst(struct kretprobe *rp)
422 {
423         struct kretprobe_instance *ri;
424         struct hlist_node *pos, *next;
425
426         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, uflist) {
427                 hlist_del(&ri->uflist);
428                 kfree(ri);
429         }
430 }
431
432 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
433 {
434         unsigned long flags;
435         struct kretprobe_instance *ri;
436         struct hlist_node *pos, *next;
437         /* No race here */
438         spin_lock_irqsave(&kretprobe_lock, flags);
439         hlist_for_each_entry_safe(ri, pos, next, &rp->used_instances, uflist) {
440                 ri->rp = NULL;
441                 hlist_del(&ri->uflist);
442         }
443         spin_unlock_irqrestore(&kretprobe_lock, flags);
444         free_rp_inst(rp);
445 }
446
447 /*
448  * Keep all fields in the kprobe consistent
449  */
450 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
451 {
452         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
453         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
454 }
455
456 /*
457 * Add the new probe to old_p->list. Fail if this is the
458 * second jprobe at the address - two jprobes can't coexist
459 */
460 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
461 {
462         if (p->break_handler) {
463                 if (old_p->break_handler)
464                         return -EEXIST;
465                 list_add_tail_rcu(&p->list, &old_p->list);
466                 old_p->break_handler = aggr_break_handler;
467         } else
468                 list_add_rcu(&p->list, &old_p->list);
469         if (p->post_handler && !old_p->post_handler)
470                 old_p->post_handler = aggr_post_handler;
471         return 0;
472 }
473
474 /*
475  * Fill in the required fields of the "manager kprobe". Replace the
476  * earlier kprobe in the hlist with the manager kprobe
477  */
478 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
479 {
480         copy_kprobe(p, ap);
481         flush_insn_slot(ap);
482         ap->addr = p->addr;
483         ap->pre_handler = aggr_pre_handler;
484         ap->fault_handler = aggr_fault_handler;
485         if (p->post_handler)
486                 ap->post_handler = aggr_post_handler;
487         if (p->break_handler)
488                 ap->break_handler = aggr_break_handler;
489
490         INIT_LIST_HEAD(&ap->list);
491         list_add_rcu(&p->list, &ap->list);
492
493         hlist_replace_rcu(&p->hlist, &ap->hlist);
494 }
495
496 /*
497  * This is the second or subsequent kprobe at the address - handle
498  * the intricacies
499  */
500 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
501                                           struct kprobe *p)
502 {
503         int ret = 0;
504         struct kprobe *ap;
505
506         if (old_p->pre_handler == aggr_pre_handler) {
507                 copy_kprobe(old_p, p);
508                 ret = add_new_kprobe(old_p, p);
509         } else {
510                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
511                 if (!ap)
512                         return -ENOMEM;
513                 add_aggr_kprobe(ap, old_p);
514                 copy_kprobe(ap, p);
515                 ret = add_new_kprobe(ap, p);
516         }
517         return ret;
518 }
519
520 static int __kprobes in_kprobes_functions(unsigned long addr)
521 {
522         struct kprobe_blackpoint *kb;
523
524         if (addr >= (unsigned long)__kprobes_text_start &&
525             addr < (unsigned long)__kprobes_text_end)
526                 return -EINVAL;
527         /*
528          * If there exists a kprobe_blacklist, verify and
529          * fail any probe registration in the prohibited area
530          */
531         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
532                 if (kb->start_addr) {
533                         if (addr >= kb->start_addr &&
534                             addr < (kb->start_addr + kb->range))
535                                 return -EINVAL;
536                 }
537         }
538         return 0;
539 }
540
541 /*
542  * If we have a symbol_name argument, look it up and add the offset field
543  * to it. This way, we can specify a relative address to a symbol.
544  */
545 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
546 {
547         kprobe_opcode_t *addr = p->addr;
548         if (p->symbol_name) {
549                 if (addr)
550                         return NULL;
551                 kprobe_lookup_name(p->symbol_name, addr);
552         }
553
554         if (!addr)
555                 return NULL;
556         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
557 }
558
559 static int __kprobes __register_kprobe(struct kprobe *p,
560         unsigned long called_from)
561 {
562         int ret = 0;
563         struct kprobe *old_p;
564         struct module *probed_mod;
565         kprobe_opcode_t *addr;
566
567         addr = kprobe_addr(p);
568         if (!addr)
569                 return -EINVAL;
570         p->addr = addr;
571
572         if (!kernel_text_address((unsigned long) p->addr) ||
573             in_kprobes_functions((unsigned long) p->addr))
574                 return -EINVAL;
575
576         p->mod_refcounted = 0;
577
578         /*
579          * Check if are we probing a module.
580          */
581         probed_mod = module_text_address((unsigned long) p->addr);
582         if (probed_mod) {
583                 struct module *calling_mod = module_text_address(called_from);
584                 /*
585                  * We must allow modules to probe themself and in this case
586                  * avoid incrementing the module refcount, so as to allow
587                  * unloading of self probing modules.
588                  */
589                 if (calling_mod && calling_mod != probed_mod) {
590                         if (unlikely(!try_module_get(probed_mod)))
591                                 return -EINVAL;
592                         p->mod_refcounted = 1;
593                 } else
594                         probed_mod = NULL;
595         }
596
597         p->nmissed = 0;
598         INIT_LIST_HEAD(&p->list);
599         mutex_lock(&kprobe_mutex);
600         old_p = get_kprobe(p->addr);
601         if (old_p) {
602                 ret = register_aggr_kprobe(old_p, p);
603                 goto out;
604         }
605
606         ret = arch_prepare_kprobe(p);
607         if (ret)
608                 goto out;
609
610         INIT_HLIST_NODE(&p->hlist);
611         hlist_add_head_rcu(&p->hlist,
612                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
613
614         if (kprobe_enabled)
615                 arch_arm_kprobe(p);
616
617 out:
618         mutex_unlock(&kprobe_mutex);
619
620         if (ret && probed_mod)
621                 module_put(probed_mod);
622         return ret;
623 }
624
625 /*
626  * Unregister a kprobe without a scheduler synchronization.
627  */
628 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
629 {
630         struct kprobe *old_p, *list_p;
631
632         old_p = get_kprobe(p->addr);
633         if (unlikely(!old_p))
634                 return -EINVAL;
635
636         if (p != old_p) {
637                 list_for_each_entry_rcu(list_p, &old_p->list, list)
638                         if (list_p == p)
639                         /* kprobe p is a valid probe */
640                                 goto valid_p;
641                 return -EINVAL;
642         }
643 valid_p:
644         if (old_p == p ||
645             (old_p->pre_handler == aggr_pre_handler &&
646              list_is_singular(&old_p->list))) {
647                 /*
648                  * Only probe on the hash list. Disarm only if kprobes are
649                  * enabled - otherwise, the breakpoint would already have
650                  * been removed. We save on flushing icache.
651                  */
652                 if (kprobe_enabled)
653                         arch_disarm_kprobe(p);
654                 hlist_del_rcu(&old_p->hlist);
655         } else {
656                 if (p->break_handler)
657                         old_p->break_handler = NULL;
658                 if (p->post_handler) {
659                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
660                                 if ((list_p != p) && (list_p->post_handler))
661                                         goto noclean;
662                         }
663                         old_p->post_handler = NULL;
664                 }
665 noclean:
666                 list_del_rcu(&p->list);
667         }
668         return 0;
669 }
670
671 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
672 {
673         struct module *mod;
674         struct kprobe *old_p;
675
676         if (p->mod_refcounted) {
677                 mod = module_text_address((unsigned long)p->addr);
678                 if (mod)
679                         module_put(mod);
680         }
681
682         if (list_empty(&p->list) || list_is_singular(&p->list)) {
683                 if (!list_empty(&p->list)) {
684                         /* "p" is the last child of an aggr_kprobe */
685                         old_p = list_entry(p->list.next, struct kprobe, list);
686                         list_del(&p->list);
687                         kfree(old_p);
688                 }
689                 arch_remove_kprobe(p);
690         }
691 }
692
693 static int __register_kprobes(struct kprobe **kps, int num,
694         unsigned long called_from)
695 {
696         int i, ret = 0;
697
698         if (num <= 0)
699                 return -EINVAL;
700         for (i = 0; i < num; i++) {
701                 ret = __register_kprobe(kps[i], called_from);
702                 if (ret < 0 && i > 0) {
703                         unregister_kprobes(kps, i);
704                         break;
705                 }
706         }
707         return ret;
708 }
709
710 /*
711  * Registration and unregistration functions for kprobe.
712  */
713 int __kprobes register_kprobe(struct kprobe *p)
714 {
715         return __register_kprobes(&p, 1,
716                                   (unsigned long)__builtin_return_address(0));
717 }
718
719 void __kprobes unregister_kprobe(struct kprobe *p)
720 {
721         unregister_kprobes(&p, 1);
722 }
723
724 int __kprobes register_kprobes(struct kprobe **kps, int num)
725 {
726         return __register_kprobes(kps, num,
727                                   (unsigned long)__builtin_return_address(0));
728 }
729
730 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
731 {
732         int i;
733
734         if (num <= 0)
735                 return;
736         mutex_lock(&kprobe_mutex);
737         for (i = 0; i < num; i++)
738                 if (__unregister_kprobe_top(kps[i]) < 0)
739                         kps[i]->addr = NULL;
740         mutex_unlock(&kprobe_mutex);
741
742         synchronize_sched();
743         for (i = 0; i < num; i++)
744                 if (kps[i]->addr)
745                         __unregister_kprobe_bottom(kps[i]);
746 }
747
748 static struct notifier_block kprobe_exceptions_nb = {
749         .notifier_call = kprobe_exceptions_notify,
750         .priority = 0x7fffffff /* we need to be notified first */
751 };
752
753 unsigned long __weak arch_deref_entry_point(void *entry)
754 {
755         return (unsigned long)entry;
756 }
757
758 int __kprobes register_jprobe(struct jprobe *jp)
759 {
760         unsigned long addr = arch_deref_entry_point(jp->entry);
761
762         if (!kernel_text_address(addr))
763                 return -EINVAL;
764
765         /* Todo: Verify probepoint is a function entry point */
766         jp->kp.pre_handler = setjmp_pre_handler;
767         jp->kp.break_handler = longjmp_break_handler;
768
769         return __register_kprobe(&jp->kp,
770                 (unsigned long)__builtin_return_address(0));
771 }
772
773 void __kprobes unregister_jprobe(struct jprobe *jp)
774 {
775         unregister_kprobe(&jp->kp);
776 }
777
778 #ifdef CONFIG_KRETPROBES
779 /*
780  * This kprobe pre_handler is registered with every kretprobe. When probe
781  * hits it will set up the return probe.
782  */
783 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
784                                            struct pt_regs *regs)
785 {
786         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
787         unsigned long flags = 0;
788
789         /*TODO: consider to only swap the RA after the last pre_handler fired */
790         spin_lock_irqsave(&kretprobe_lock, flags);
791         if (!hlist_empty(&rp->free_instances)) {
792                 struct kretprobe_instance *ri;
793
794                 ri = hlist_entry(rp->free_instances.first,
795                                  struct kretprobe_instance, uflist);
796                 ri->rp = rp;
797                 ri->task = current;
798
799                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
800                         spin_unlock_irqrestore(&kretprobe_lock, flags);
801                         return 0;
802                 }
803
804                 arch_prepare_kretprobe(ri, regs);
805
806                 /* XXX(hch): why is there no hlist_move_head? */
807                 hlist_del(&ri->uflist);
808                 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
809                 hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
810         } else
811                 rp->nmissed++;
812         spin_unlock_irqrestore(&kretprobe_lock, flags);
813         return 0;
814 }
815
816 static int __kprobes __register_kretprobe(struct kretprobe *rp,
817                                           unsigned long called_from)
818 {
819         int ret = 0;
820         struct kretprobe_instance *inst;
821         int i;
822         void *addr;
823
824         if (kretprobe_blacklist_size) {
825                 addr = kprobe_addr(&rp->kp);
826                 if (!addr)
827                         return -EINVAL;
828
829                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
830                         if (kretprobe_blacklist[i].addr == addr)
831                                 return -EINVAL;
832                 }
833         }
834
835         rp->kp.pre_handler = pre_handler_kretprobe;
836         rp->kp.post_handler = NULL;
837         rp->kp.fault_handler = NULL;
838         rp->kp.break_handler = NULL;
839
840         /* Pre-allocate memory for max kretprobe instances */
841         if (rp->maxactive <= 0) {
842 #ifdef CONFIG_PREEMPT
843                 rp->maxactive = max(10, 2 * NR_CPUS);
844 #else
845                 rp->maxactive = NR_CPUS;
846 #endif
847         }
848         INIT_HLIST_HEAD(&rp->used_instances);
849         INIT_HLIST_HEAD(&rp->free_instances);
850         for (i = 0; i < rp->maxactive; i++) {
851                 inst = kmalloc(sizeof(struct kretprobe_instance) +
852                                rp->data_size, GFP_KERNEL);
853                 if (inst == NULL) {
854                         free_rp_inst(rp);
855                         return -ENOMEM;
856                 }
857                 INIT_HLIST_NODE(&inst->uflist);
858                 hlist_add_head(&inst->uflist, &rp->free_instances);
859         }
860
861         rp->nmissed = 0;
862         /* Establish function entry probe point */
863         ret = __register_kprobe(&rp->kp, called_from);
864         if (ret != 0)
865                 free_rp_inst(rp);
866         return ret;
867 }
868
869 static int __register_kretprobes(struct kretprobe **rps, int num,
870         unsigned long called_from)
871 {
872         int ret = 0, i;
873
874         if (num <= 0)
875                 return -EINVAL;
876         for (i = 0; i < num; i++) {
877                 ret = __register_kretprobe(rps[i], called_from);
878                 if (ret < 0 && i > 0) {
879                         unregister_kretprobes(rps, i);
880                         break;
881                 }
882         }
883         return ret;
884 }
885
886 int __kprobes register_kretprobe(struct kretprobe *rp)
887 {
888         return __register_kretprobes(&rp, 1,
889                         (unsigned long)__builtin_return_address(0));
890 }
891
892 void __kprobes unregister_kretprobe(struct kretprobe *rp)
893 {
894         unregister_kretprobes(&rp, 1);
895 }
896
897 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
898 {
899         return __register_kretprobes(rps, num,
900                         (unsigned long)__builtin_return_address(0));
901 }
902
903 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
904 {
905         int i;
906
907         if (num <= 0)
908                 return;
909         mutex_lock(&kprobe_mutex);
910         for (i = 0; i < num; i++)
911                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
912                         rps[i]->kp.addr = NULL;
913         mutex_unlock(&kprobe_mutex);
914
915         synchronize_sched();
916         for (i = 0; i < num; i++) {
917                 if (rps[i]->kp.addr) {
918                         __unregister_kprobe_bottom(&rps[i]->kp);
919                         cleanup_rp_inst(rps[i]);
920                 }
921         }
922 }
923
924 #else /* CONFIG_KRETPROBES */
925 int __kprobes register_kretprobe(struct kretprobe *rp)
926 {
927         return -ENOSYS;
928 }
929
930 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
931 {
932         return -ENOSYS;
933 }
934 void __kprobes unregister_kretprobe(struct kretprobe *rp)
935 {
936 }
937
938 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
939 {
940 }
941
942 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
943                                            struct pt_regs *regs)
944 {
945         return 0;
946 }
947
948 #endif /* CONFIG_KRETPROBES */
949
950 static int __init init_kprobes(void)
951 {
952         int i, err = 0;
953         unsigned long offset = 0, size = 0;
954         char *modname, namebuf[128];
955         const char *symbol_name;
956         void *addr;
957         struct kprobe_blackpoint *kb;
958
959         /* FIXME allocate the probe table, currently defined statically */
960         /* initialize all list heads */
961         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
962                 INIT_HLIST_HEAD(&kprobe_table[i]);
963                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
964         }
965
966         /*
967          * Lookup and populate the kprobe_blacklist.
968          *
969          * Unlike the kretprobe blacklist, we'll need to determine
970          * the range of addresses that belong to the said functions,
971          * since a kprobe need not necessarily be at the beginning
972          * of a function.
973          */
974         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
975                 kprobe_lookup_name(kb->name, addr);
976                 if (!addr)
977                         continue;
978
979                 kb->start_addr = (unsigned long)addr;
980                 symbol_name = kallsyms_lookup(kb->start_addr,
981                                 &size, &offset, &modname, namebuf);
982                 if (!symbol_name)
983                         kb->range = 0;
984                 else
985                         kb->range = size;
986         }
987
988         if (kretprobe_blacklist_size) {
989                 /* lookup the function address from its name */
990                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
991                         kprobe_lookup_name(kretprobe_blacklist[i].name,
992                                            kretprobe_blacklist[i].addr);
993                         if (!kretprobe_blacklist[i].addr)
994                                 printk("kretprobe: lookup failed: %s\n",
995                                        kretprobe_blacklist[i].name);
996                 }
997         }
998
999         /* By default, kprobes are enabled */
1000         kprobe_enabled = true;
1001
1002         err = arch_init_kprobes();
1003         if (!err)
1004                 err = register_die_notifier(&kprobe_exceptions_nb);
1005
1006         if (!err)
1007                 init_test_probes();
1008         return err;
1009 }
1010
1011 #ifdef CONFIG_DEBUG_FS
1012 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1013                 const char *sym, int offset,char *modname)
1014 {
1015         char *kprobe_type;
1016
1017         if (p->pre_handler == pre_handler_kretprobe)
1018                 kprobe_type = "r";
1019         else if (p->pre_handler == setjmp_pre_handler)
1020                 kprobe_type = "j";
1021         else
1022                 kprobe_type = "k";
1023         if (sym)
1024                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1025                         sym, offset, (modname ? modname : " "));
1026         else
1027                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1028 }
1029
1030 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1031 {
1032         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1033 }
1034
1035 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1036 {
1037         (*pos)++;
1038         if (*pos >= KPROBE_TABLE_SIZE)
1039                 return NULL;
1040         return pos;
1041 }
1042
1043 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1044 {
1045         /* Nothing to do */
1046 }
1047
1048 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1049 {
1050         struct hlist_head *head;
1051         struct hlist_node *node;
1052         struct kprobe *p, *kp;
1053         const char *sym = NULL;
1054         unsigned int i = *(loff_t *) v;
1055         unsigned long offset = 0;
1056         char *modname, namebuf[128];
1057
1058         head = &kprobe_table[i];
1059         preempt_disable();
1060         hlist_for_each_entry_rcu(p, node, head, hlist) {
1061                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1062                                         &offset, &modname, namebuf);
1063                 if (p->pre_handler == aggr_pre_handler) {
1064                         list_for_each_entry_rcu(kp, &p->list, list)
1065                                 report_probe(pi, kp, sym, offset, modname);
1066                 } else
1067                         report_probe(pi, p, sym, offset, modname);
1068         }
1069         preempt_enable();
1070         return 0;
1071 }
1072
1073 static struct seq_operations kprobes_seq_ops = {
1074         .start = kprobe_seq_start,
1075         .next  = kprobe_seq_next,
1076         .stop  = kprobe_seq_stop,
1077         .show  = show_kprobe_addr
1078 };
1079
1080 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1081 {
1082         return seq_open(filp, &kprobes_seq_ops);
1083 }
1084
1085 static struct file_operations debugfs_kprobes_operations = {
1086         .open           = kprobes_open,
1087         .read           = seq_read,
1088         .llseek         = seq_lseek,
1089         .release        = seq_release,
1090 };
1091
1092 static void __kprobes enable_all_kprobes(void)
1093 {
1094         struct hlist_head *head;
1095         struct hlist_node *node;
1096         struct kprobe *p;
1097         unsigned int i;
1098
1099         mutex_lock(&kprobe_mutex);
1100
1101         /* If kprobes are already enabled, just return */
1102         if (kprobe_enabled)
1103                 goto already_enabled;
1104
1105         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1106                 head = &kprobe_table[i];
1107                 hlist_for_each_entry_rcu(p, node, head, hlist)
1108                         arch_arm_kprobe(p);
1109         }
1110
1111         kprobe_enabled = true;
1112         printk(KERN_INFO "Kprobes globally enabled\n");
1113
1114 already_enabled:
1115         mutex_unlock(&kprobe_mutex);
1116         return;
1117 }
1118
1119 static void __kprobes disable_all_kprobes(void)
1120 {
1121         struct hlist_head *head;
1122         struct hlist_node *node;
1123         struct kprobe *p;
1124         unsigned int i;
1125
1126         mutex_lock(&kprobe_mutex);
1127
1128         /* If kprobes are already disabled, just return */
1129         if (!kprobe_enabled)
1130                 goto already_disabled;
1131
1132         kprobe_enabled = false;
1133         printk(KERN_INFO "Kprobes globally disabled\n");
1134         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1135                 head = &kprobe_table[i];
1136                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1137                         if (!arch_trampoline_kprobe(p))
1138                                 arch_disarm_kprobe(p);
1139                 }
1140         }
1141
1142         mutex_unlock(&kprobe_mutex);
1143         /* Allow all currently running kprobes to complete */
1144         synchronize_sched();
1145         return;
1146
1147 already_disabled:
1148         mutex_unlock(&kprobe_mutex);
1149         return;
1150 }
1151
1152 /*
1153  * XXX: The debugfs bool file interface doesn't allow for callbacks
1154  * when the bool state is switched. We can reuse that facility when
1155  * available
1156  */
1157 static ssize_t read_enabled_file_bool(struct file *file,
1158                char __user *user_buf, size_t count, loff_t *ppos)
1159 {
1160         char buf[3];
1161
1162         if (kprobe_enabled)
1163                 buf[0] = '1';
1164         else
1165                 buf[0] = '0';
1166         buf[1] = '\n';
1167         buf[2] = 0x00;
1168         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1169 }
1170
1171 static ssize_t write_enabled_file_bool(struct file *file,
1172                const char __user *user_buf, size_t count, loff_t *ppos)
1173 {
1174         char buf[32];
1175         int buf_size;
1176
1177         buf_size = min(count, (sizeof(buf)-1));
1178         if (copy_from_user(buf, user_buf, buf_size))
1179                 return -EFAULT;
1180
1181         switch (buf[0]) {
1182         case 'y':
1183         case 'Y':
1184         case '1':
1185                 enable_all_kprobes();
1186                 break;
1187         case 'n':
1188         case 'N':
1189         case '0':
1190                 disable_all_kprobes();
1191                 break;
1192         }
1193
1194         return count;
1195 }
1196
1197 static struct file_operations fops_kp = {
1198         .read =         read_enabled_file_bool,
1199         .write =        write_enabled_file_bool,
1200 };
1201
1202 static int __kprobes debugfs_kprobe_init(void)
1203 {
1204         struct dentry *dir, *file;
1205         unsigned int value = 1;
1206
1207         dir = debugfs_create_dir("kprobes", NULL);
1208         if (!dir)
1209                 return -ENOMEM;
1210
1211         file = debugfs_create_file("list", 0444, dir, NULL,
1212                                 &debugfs_kprobes_operations);
1213         if (!file) {
1214                 debugfs_remove(dir);
1215                 return -ENOMEM;
1216         }
1217
1218         file = debugfs_create_file("enabled", 0600, dir,
1219                                         &value, &fops_kp);
1220         if (!file) {
1221                 debugfs_remove(dir);
1222                 return -ENOMEM;
1223         }
1224
1225         return 0;
1226 }
1227
1228 late_initcall(debugfs_kprobe_init);
1229 #endif /* CONFIG_DEBUG_FS */
1230
1231 module_init(init_kprobes);
1232
1233 EXPORT_SYMBOL_GPL(register_kprobe);
1234 EXPORT_SYMBOL_GPL(unregister_kprobe);
1235 EXPORT_SYMBOL_GPL(register_kprobes);
1236 EXPORT_SYMBOL_GPL(unregister_kprobes);
1237 EXPORT_SYMBOL_GPL(register_jprobe);
1238 EXPORT_SYMBOL_GPL(unregister_jprobe);
1239 #ifdef CONFIG_KPROBES
1240 EXPORT_SYMBOL_GPL(jprobe_return);
1241 #endif
1242
1243 #ifdef CONFIG_KPROBES
1244 EXPORT_SYMBOL_GPL(register_kretprobe);
1245 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1246 EXPORT_SYMBOL_GPL(register_kretprobes);
1247 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1248 #endif