kprobes: add __kprobes to kprobe internal functions
[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 int kprobes_initialized;
66 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
67 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
68
69 /* NOTE: change this value only with kprobe_mutex held */
70 static bool kprobe_enabled;
71
72 static DEFINE_MUTEX(kprobe_mutex);      /* Protects kprobe_table */
73 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
74 static struct {
75         spinlock_t lock ____cacheline_aligned_in_smp;
76 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
77
78 static spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
79 {
80         return &(kretprobe_table_locks[hash].lock);
81 }
82
83 /*
84  * Normally, functions that we'd want to prohibit kprobes in, are marked
85  * __kprobes. But, there are cases where such functions already belong to
86  * a different section (__sched for preempt_schedule)
87  *
88  * For such cases, we now have a blacklist
89  */
90 static struct kprobe_blackpoint kprobe_blacklist[] = {
91         {"preempt_schedule",},
92         {NULL}    /* Terminator */
93 };
94
95 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
96 /*
97  * kprobe->ainsn.insn points to the copy of the instruction to be
98  * single-stepped. x86_64, POWER4 and above have no-exec support and
99  * stepping on the instruction on a vmalloced/kmalloced/data page
100  * is a recipe for disaster
101  */
102 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
103
104 struct kprobe_insn_page {
105         struct hlist_node hlist;
106         kprobe_opcode_t *insns;         /* Page of instruction slots */
107         char slot_used[INSNS_PER_PAGE];
108         int nused;
109         int ngarbage;
110 };
111
112 enum kprobe_slot_state {
113         SLOT_CLEAN = 0,
114         SLOT_DIRTY = 1,
115         SLOT_USED = 2,
116 };
117
118 static DEFINE_MUTEX(kprobe_insn_mutex); /* Protects kprobe_insn_pages */
119 static struct hlist_head kprobe_insn_pages;
120 static int kprobe_garbage_slots;
121 static int collect_garbage_slots(void);
122
123 static int __kprobes check_safety(void)
124 {
125         int ret = 0;
126 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
127         ret = freeze_processes();
128         if (ret == 0) {
129                 struct task_struct *p, *q;
130                 do_each_thread(p, q) {
131                         if (p != current && p->state == TASK_RUNNING &&
132                             p->pid != 0) {
133                                 printk("Check failed: %s is running\n",p->comm);
134                                 ret = -1;
135                                 goto loop_end;
136                         }
137                 } while_each_thread(p, q);
138         }
139 loop_end:
140         thaw_processes();
141 #else
142         synchronize_sched();
143 #endif
144         return ret;
145 }
146
147 /**
148  * __get_insn_slot() - Find a slot on an executable page for an instruction.
149  * We allocate an executable page if there's no room on existing ones.
150  */
151 static kprobe_opcode_t __kprobes *__get_insn_slot(void)
152 {
153         struct kprobe_insn_page *kip;
154         struct hlist_node *pos;
155
156  retry:
157         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
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_HLIST_NODE(&kip->hlist);
192         hlist_add_head(&kip->hlist, &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                 hlist_del(&kip->hlist);
222                 if (hlist_empty(&kprobe_insn_pages)) {
223                         INIT_HLIST_NODE(&kip->hlist);
224                         hlist_add_head(&kip->hlist,
225                                        &kprobe_insn_pages);
226                 } else {
227                         module_free(NULL, kip->insns);
228                         kfree(kip);
229                 }
230                 return 1;
231         }
232         return 0;
233 }
234
235 static int __kprobes collect_garbage_slots(void)
236 {
237         struct kprobe_insn_page *kip;
238         struct hlist_node *pos, *next;
239         int safety;
240
241         /* Ensure no-one is preepmted on the garbages */
242         mutex_unlock(&kprobe_insn_mutex);
243         safety = check_safety();
244         mutex_lock(&kprobe_insn_mutex);
245         if (safety != 0)
246                 return -EAGAIN;
247
248         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
249                 int i;
250                 if (kip->ngarbage == 0)
251                         continue;
252                 kip->ngarbage = 0;      /* we will collect all garbages */
253                 for (i = 0; i < INSNS_PER_PAGE; i++) {
254                         if (kip->slot_used[i] == SLOT_DIRTY &&
255                             collect_one_slot(kip, i))
256                                 break;
257                 }
258         }
259         kprobe_garbage_slots = 0;
260         return 0;
261 }
262
263 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
264 {
265         struct kprobe_insn_page *kip;
266         struct hlist_node *pos;
267
268         mutex_lock(&kprobe_insn_mutex);
269         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
270                 if (kip->insns <= slot &&
271                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
272                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
273                         if (dirty) {
274                                 kip->slot_used[i] = SLOT_DIRTY;
275                                 kip->ngarbage++;
276                         } else {
277                                 collect_one_slot(kip, i);
278                         }
279                         break;
280                 }
281         }
282
283         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
284                 collect_garbage_slots();
285
286         mutex_unlock(&kprobe_insn_mutex);
287 }
288 #endif
289
290 /* We have preemption disabled.. so it is safe to use __ versions */
291 static inline void set_kprobe_instance(struct kprobe *kp)
292 {
293         __get_cpu_var(kprobe_instance) = kp;
294 }
295
296 static inline void reset_kprobe_instance(void)
297 {
298         __get_cpu_var(kprobe_instance) = NULL;
299 }
300
301 /*
302  * This routine is called either:
303  *      - under the kprobe_mutex - during kprobe_[un]register()
304  *                              OR
305  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
306  */
307 struct kprobe __kprobes *get_kprobe(void *addr)
308 {
309         struct hlist_head *head;
310         struct hlist_node *node;
311         struct kprobe *p;
312
313         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
314         hlist_for_each_entry_rcu(p, node, head, hlist) {
315                 if (p->addr == addr)
316                         return p;
317         }
318         return NULL;
319 }
320
321 /*
322  * Aggregate handlers for multiple kprobes support - these handlers
323  * take care of invoking the individual kprobe handlers on p->list
324  */
325 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
326 {
327         struct kprobe *kp;
328
329         list_for_each_entry_rcu(kp, &p->list, list) {
330                 if (kp->pre_handler) {
331                         set_kprobe_instance(kp);
332                         if (kp->pre_handler(kp, regs))
333                                 return 1;
334                 }
335                 reset_kprobe_instance();
336         }
337         return 0;
338 }
339
340 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
341                                         unsigned long flags)
342 {
343         struct kprobe *kp;
344
345         list_for_each_entry_rcu(kp, &p->list, list) {
346                 if (kp->post_handler) {
347                         set_kprobe_instance(kp);
348                         kp->post_handler(kp, regs, flags);
349                         reset_kprobe_instance();
350                 }
351         }
352 }
353
354 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
355                                         int trapnr)
356 {
357         struct kprobe *cur = __get_cpu_var(kprobe_instance);
358
359         /*
360          * if we faulted "during" the execution of a user specified
361          * probe handler, invoke just that probe's fault handler
362          */
363         if (cur && cur->fault_handler) {
364                 if (cur->fault_handler(cur, regs, trapnr))
365                         return 1;
366         }
367         return 0;
368 }
369
370 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
371 {
372         struct kprobe *cur = __get_cpu_var(kprobe_instance);
373         int ret = 0;
374
375         if (cur && cur->break_handler) {
376                 if (cur->break_handler(cur, regs))
377                         ret = 1;
378         }
379         reset_kprobe_instance();
380         return ret;
381 }
382
383 /* Walks the list and increments nmissed count for multiprobe case */
384 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
385 {
386         struct kprobe *kp;
387         if (p->pre_handler != aggr_pre_handler) {
388                 p->nmissed++;
389         } else {
390                 list_for_each_entry_rcu(kp, &p->list, list)
391                         kp->nmissed++;
392         }
393         return;
394 }
395
396 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
397                                 struct hlist_head *head)
398 {
399         struct kretprobe *rp = ri->rp;
400
401         /* remove rp inst off the rprobe_inst_table */
402         hlist_del(&ri->hlist);
403         INIT_HLIST_NODE(&ri->hlist);
404         if (likely(rp)) {
405                 spin_lock(&rp->lock);
406                 hlist_add_head(&ri->hlist, &rp->free_instances);
407                 spin_unlock(&rp->lock);
408         } else
409                 /* Unregistering */
410                 hlist_add_head(&ri->hlist, head);
411 }
412
413 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
414                          struct hlist_head **head, unsigned long *flags)
415 {
416         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
417         spinlock_t *hlist_lock;
418
419         *head = &kretprobe_inst_table[hash];
420         hlist_lock = kretprobe_table_lock_ptr(hash);
421         spin_lock_irqsave(hlist_lock, *flags);
422 }
423
424 static void __kprobes kretprobe_table_lock(unsigned long hash,
425         unsigned long *flags)
426 {
427         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
428         spin_lock_irqsave(hlist_lock, *flags);
429 }
430
431 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
432         unsigned long *flags)
433 {
434         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
435         spinlock_t *hlist_lock;
436
437         hlist_lock = kretprobe_table_lock_ptr(hash);
438         spin_unlock_irqrestore(hlist_lock, *flags);
439 }
440
441 void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
442 {
443         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
444         spin_unlock_irqrestore(hlist_lock, *flags);
445 }
446
447 /*
448  * This function is called from finish_task_switch when task tk becomes dead,
449  * so that we can recycle any function-return probe instances associated
450  * with this task. These left over instances represent probed functions
451  * that have been called but will never return.
452  */
453 void __kprobes kprobe_flush_task(struct task_struct *tk)
454 {
455         struct kretprobe_instance *ri;
456         struct hlist_head *head, empty_rp;
457         struct hlist_node *node, *tmp;
458         unsigned long hash, flags = 0;
459
460         if (unlikely(!kprobes_initialized))
461                 /* Early boot.  kretprobe_table_locks not yet initialized. */
462                 return;
463
464         hash = hash_ptr(tk, KPROBE_HASH_BITS);
465         head = &kretprobe_inst_table[hash];
466         kretprobe_table_lock(hash, &flags);
467         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
468                 if (ri->task == tk)
469                         recycle_rp_inst(ri, &empty_rp);
470         }
471         kretprobe_table_unlock(hash, &flags);
472         INIT_HLIST_HEAD(&empty_rp);
473         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
474                 hlist_del(&ri->hlist);
475                 kfree(ri);
476         }
477 }
478
479 static inline void free_rp_inst(struct kretprobe *rp)
480 {
481         struct kretprobe_instance *ri;
482         struct hlist_node *pos, *next;
483
484         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
485                 hlist_del(&ri->hlist);
486                 kfree(ri);
487         }
488 }
489
490 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
491 {
492         unsigned long flags, hash;
493         struct kretprobe_instance *ri;
494         struct hlist_node *pos, *next;
495         struct hlist_head *head;
496
497         /* No race here */
498         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
499                 kretprobe_table_lock(hash, &flags);
500                 head = &kretprobe_inst_table[hash];
501                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
502                         if (ri->rp == rp)
503                                 ri->rp = NULL;
504                 }
505                 kretprobe_table_unlock(hash, &flags);
506         }
507         free_rp_inst(rp);
508 }
509
510 /*
511  * Keep all fields in the kprobe consistent
512  */
513 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
514 {
515         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
516         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
517 }
518
519 /*
520 * Add the new probe to old_p->list. Fail if this is the
521 * second jprobe at the address - two jprobes can't coexist
522 */
523 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
524 {
525         if (p->break_handler) {
526                 if (old_p->break_handler)
527                         return -EEXIST;
528                 list_add_tail_rcu(&p->list, &old_p->list);
529                 old_p->break_handler = aggr_break_handler;
530         } else
531                 list_add_rcu(&p->list, &old_p->list);
532         if (p->post_handler && !old_p->post_handler)
533                 old_p->post_handler = aggr_post_handler;
534         return 0;
535 }
536
537 /*
538  * Fill in the required fields of the "manager kprobe". Replace the
539  * earlier kprobe in the hlist with the manager kprobe
540  */
541 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
542 {
543         copy_kprobe(p, ap);
544         flush_insn_slot(ap);
545         ap->addr = p->addr;
546         ap->pre_handler = aggr_pre_handler;
547         ap->fault_handler = aggr_fault_handler;
548         if (p->post_handler)
549                 ap->post_handler = aggr_post_handler;
550         if (p->break_handler)
551                 ap->break_handler = aggr_break_handler;
552
553         INIT_LIST_HEAD(&ap->list);
554         list_add_rcu(&p->list, &ap->list);
555
556         hlist_replace_rcu(&p->hlist, &ap->hlist);
557 }
558
559 /*
560  * This is the second or subsequent kprobe at the address - handle
561  * the intricacies
562  */
563 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
564                                           struct kprobe *p)
565 {
566         int ret = 0;
567         struct kprobe *ap;
568
569         if (old_p->pre_handler == aggr_pre_handler) {
570                 copy_kprobe(old_p, p);
571                 ret = add_new_kprobe(old_p, p);
572         } else {
573                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
574                 if (!ap)
575                         return -ENOMEM;
576                 add_aggr_kprobe(ap, old_p);
577                 copy_kprobe(ap, p);
578                 ret = add_new_kprobe(ap, p);
579         }
580         return ret;
581 }
582
583 static int __kprobes in_kprobes_functions(unsigned long addr)
584 {
585         struct kprobe_blackpoint *kb;
586
587         if (addr >= (unsigned long)__kprobes_text_start &&
588             addr < (unsigned long)__kprobes_text_end)
589                 return -EINVAL;
590         /*
591          * If there exists a kprobe_blacklist, verify and
592          * fail any probe registration in the prohibited area
593          */
594         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
595                 if (kb->start_addr) {
596                         if (addr >= kb->start_addr &&
597                             addr < (kb->start_addr + kb->range))
598                                 return -EINVAL;
599                 }
600         }
601         return 0;
602 }
603
604 /*
605  * If we have a symbol_name argument, look it up and add the offset field
606  * to it. This way, we can specify a relative address to a symbol.
607  */
608 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
609 {
610         kprobe_opcode_t *addr = p->addr;
611         if (p->symbol_name) {
612                 if (addr)
613                         return NULL;
614                 kprobe_lookup_name(p->symbol_name, addr);
615         }
616
617         if (!addr)
618                 return NULL;
619         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
620 }
621
622 static int __kprobes __register_kprobe(struct kprobe *p,
623         unsigned long called_from)
624 {
625         int ret = 0;
626         struct kprobe *old_p;
627         struct module *probed_mod;
628         kprobe_opcode_t *addr;
629
630         addr = kprobe_addr(p);
631         if (!addr)
632                 return -EINVAL;
633         p->addr = addr;
634
635         preempt_disable();
636         if (!__kernel_text_address((unsigned long) p->addr) ||
637             in_kprobes_functions((unsigned long) p->addr)) {
638                 preempt_enable();
639                 return -EINVAL;
640         }
641
642         p->mod_refcounted = 0;
643
644         /*
645          * Check if are we probing a module.
646          */
647         probed_mod = __module_text_address((unsigned long) p->addr);
648         if (probed_mod) {
649                 struct module *calling_mod;
650                 calling_mod = __module_text_address(called_from);
651                 /*
652                  * We must allow modules to probe themself and in this case
653                  * avoid incrementing the module refcount, so as to allow
654                  * unloading of self probing modules.
655                  */
656                 if (calling_mod != probed_mod) {
657                         if (unlikely(!try_module_get(probed_mod))) {
658                                 preempt_enable();
659                                 return -EINVAL;
660                         }
661                         p->mod_refcounted = 1;
662                 } else
663                         probed_mod = NULL;
664         }
665         preempt_enable();
666
667         p->nmissed = 0;
668         INIT_LIST_HEAD(&p->list);
669         mutex_lock(&kprobe_mutex);
670         old_p = get_kprobe(p->addr);
671         if (old_p) {
672                 ret = register_aggr_kprobe(old_p, p);
673                 goto out;
674         }
675
676         ret = arch_prepare_kprobe(p);
677         if (ret)
678                 goto out;
679
680         INIT_HLIST_NODE(&p->hlist);
681         hlist_add_head_rcu(&p->hlist,
682                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
683
684         if (kprobe_enabled)
685                 arch_arm_kprobe(p);
686
687 out:
688         mutex_unlock(&kprobe_mutex);
689
690         if (ret && probed_mod)
691                 module_put(probed_mod);
692         return ret;
693 }
694
695 /*
696  * Unregister a kprobe without a scheduler synchronization.
697  */
698 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
699 {
700         struct kprobe *old_p, *list_p;
701
702         old_p = get_kprobe(p->addr);
703         if (unlikely(!old_p))
704                 return -EINVAL;
705
706         if (p != old_p) {
707                 list_for_each_entry_rcu(list_p, &old_p->list, list)
708                         if (list_p == p)
709                         /* kprobe p is a valid probe */
710                                 goto valid_p;
711                 return -EINVAL;
712         }
713 valid_p:
714         if (old_p == p ||
715             (old_p->pre_handler == aggr_pre_handler &&
716              list_is_singular(&old_p->list))) {
717                 /*
718                  * Only probe on the hash list. Disarm only if kprobes are
719                  * enabled - otherwise, the breakpoint would already have
720                  * been removed. We save on flushing icache.
721                  */
722                 if (kprobe_enabled)
723                         arch_disarm_kprobe(p);
724                 hlist_del_rcu(&old_p->hlist);
725         } else {
726                 if (p->break_handler)
727                         old_p->break_handler = NULL;
728                 if (p->post_handler) {
729                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
730                                 if ((list_p != p) && (list_p->post_handler))
731                                         goto noclean;
732                         }
733                         old_p->post_handler = NULL;
734                 }
735 noclean:
736                 list_del_rcu(&p->list);
737         }
738         return 0;
739 }
740
741 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
742 {
743         struct module *mod;
744         struct kprobe *old_p;
745
746         if (p->mod_refcounted) {
747                 /*
748                  * Since we've already incremented refcount,
749                  * we don't need to disable preemption.
750                  */
751                 mod = module_text_address((unsigned long)p->addr);
752                 if (mod)
753                         module_put(mod);
754         }
755
756         if (list_empty(&p->list) || list_is_singular(&p->list)) {
757                 if (!list_empty(&p->list)) {
758                         /* "p" is the last child of an aggr_kprobe */
759                         old_p = list_entry(p->list.next, struct kprobe, list);
760                         list_del(&p->list);
761                         kfree(old_p);
762                 }
763                 arch_remove_kprobe(p);
764         }
765 }
766
767 static int __kprobes __register_kprobes(struct kprobe **kps, int num,
768         unsigned long called_from)
769 {
770         int i, ret = 0;
771
772         if (num <= 0)
773                 return -EINVAL;
774         for (i = 0; i < num; i++) {
775                 ret = __register_kprobe(kps[i], called_from);
776                 if (ret < 0) {
777                         if (i > 0)
778                                 unregister_kprobes(kps, i);
779                         break;
780                 }
781         }
782         return ret;
783 }
784
785 /*
786  * Registration and unregistration functions for kprobe.
787  */
788 int __kprobes register_kprobe(struct kprobe *p)
789 {
790         return __register_kprobes(&p, 1,
791                                   (unsigned long)__builtin_return_address(0));
792 }
793
794 void __kprobes unregister_kprobe(struct kprobe *p)
795 {
796         unregister_kprobes(&p, 1);
797 }
798
799 int __kprobes register_kprobes(struct kprobe **kps, int num)
800 {
801         return __register_kprobes(kps, num,
802                                   (unsigned long)__builtin_return_address(0));
803 }
804
805 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
806 {
807         int i;
808
809         if (num <= 0)
810                 return;
811         mutex_lock(&kprobe_mutex);
812         for (i = 0; i < num; i++)
813                 if (__unregister_kprobe_top(kps[i]) < 0)
814                         kps[i]->addr = NULL;
815         mutex_unlock(&kprobe_mutex);
816
817         synchronize_sched();
818         for (i = 0; i < num; i++)
819                 if (kps[i]->addr)
820                         __unregister_kprobe_bottom(kps[i]);
821 }
822
823 static struct notifier_block kprobe_exceptions_nb = {
824         .notifier_call = kprobe_exceptions_notify,
825         .priority = 0x7fffffff /* we need to be notified first */
826 };
827
828 unsigned long __weak arch_deref_entry_point(void *entry)
829 {
830         return (unsigned long)entry;
831 }
832
833 static int __kprobes __register_jprobes(struct jprobe **jps, int num,
834         unsigned long called_from)
835 {
836         struct jprobe *jp;
837         int ret = 0, i;
838
839         if (num <= 0)
840                 return -EINVAL;
841         for (i = 0; i < num; i++) {
842                 unsigned long addr;
843                 jp = jps[i];
844                 addr = arch_deref_entry_point(jp->entry);
845
846                 if (!kernel_text_address(addr))
847                         ret = -EINVAL;
848                 else {
849                         /* Todo: Verify probepoint is a function entry point */
850                         jp->kp.pre_handler = setjmp_pre_handler;
851                         jp->kp.break_handler = longjmp_break_handler;
852                         ret = __register_kprobe(&jp->kp, called_from);
853                 }
854                 if (ret < 0) {
855                         if (i > 0)
856                                 unregister_jprobes(jps, i);
857                         break;
858                 }
859         }
860         return ret;
861 }
862
863 int __kprobes register_jprobe(struct jprobe *jp)
864 {
865         return __register_jprobes(&jp, 1,
866                 (unsigned long)__builtin_return_address(0));
867 }
868
869 void __kprobes unregister_jprobe(struct jprobe *jp)
870 {
871         unregister_jprobes(&jp, 1);
872 }
873
874 int __kprobes register_jprobes(struct jprobe **jps, int num)
875 {
876         return __register_jprobes(jps, num,
877                 (unsigned long)__builtin_return_address(0));
878 }
879
880 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
881 {
882         int i;
883
884         if (num <= 0)
885                 return;
886         mutex_lock(&kprobe_mutex);
887         for (i = 0; i < num; i++)
888                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
889                         jps[i]->kp.addr = NULL;
890         mutex_unlock(&kprobe_mutex);
891
892         synchronize_sched();
893         for (i = 0; i < num; i++) {
894                 if (jps[i]->kp.addr)
895                         __unregister_kprobe_bottom(&jps[i]->kp);
896         }
897 }
898
899 #ifdef CONFIG_KRETPROBES
900 /*
901  * This kprobe pre_handler is registered with every kretprobe. When probe
902  * hits it will set up the return probe.
903  */
904 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
905                                            struct pt_regs *regs)
906 {
907         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
908         unsigned long hash, flags = 0;
909         struct kretprobe_instance *ri;
910
911         /*TODO: consider to only swap the RA after the last pre_handler fired */
912         hash = hash_ptr(current, KPROBE_HASH_BITS);
913         spin_lock_irqsave(&rp->lock, flags);
914         if (!hlist_empty(&rp->free_instances)) {
915                 ri = hlist_entry(rp->free_instances.first,
916                                 struct kretprobe_instance, hlist);
917                 hlist_del(&ri->hlist);
918                 spin_unlock_irqrestore(&rp->lock, flags);
919
920                 ri->rp = rp;
921                 ri->task = current;
922
923                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
924                         spin_unlock_irqrestore(&rp->lock, flags);
925                         return 0;
926                 }
927
928                 arch_prepare_kretprobe(ri, regs);
929
930                 /* XXX(hch): why is there no hlist_move_head? */
931                 INIT_HLIST_NODE(&ri->hlist);
932                 kretprobe_table_lock(hash, &flags);
933                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
934                 kretprobe_table_unlock(hash, &flags);
935         } else {
936                 rp->nmissed++;
937                 spin_unlock_irqrestore(&rp->lock, flags);
938         }
939         return 0;
940 }
941
942 static int __kprobes __register_kretprobe(struct kretprobe *rp,
943                                           unsigned long called_from)
944 {
945         int ret = 0;
946         struct kretprobe_instance *inst;
947         int i;
948         void *addr;
949
950         if (kretprobe_blacklist_size) {
951                 addr = kprobe_addr(&rp->kp);
952                 if (!addr)
953                         return -EINVAL;
954
955                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
956                         if (kretprobe_blacklist[i].addr == addr)
957                                 return -EINVAL;
958                 }
959         }
960
961         rp->kp.pre_handler = pre_handler_kretprobe;
962         rp->kp.post_handler = NULL;
963         rp->kp.fault_handler = NULL;
964         rp->kp.break_handler = NULL;
965
966         /* Pre-allocate memory for max kretprobe instances */
967         if (rp->maxactive <= 0) {
968 #ifdef CONFIG_PREEMPT
969                 rp->maxactive = max(10, 2 * NR_CPUS);
970 #else
971                 rp->maxactive = NR_CPUS;
972 #endif
973         }
974         spin_lock_init(&rp->lock);
975         INIT_HLIST_HEAD(&rp->free_instances);
976         for (i = 0; i < rp->maxactive; i++) {
977                 inst = kmalloc(sizeof(struct kretprobe_instance) +
978                                rp->data_size, GFP_KERNEL);
979                 if (inst == NULL) {
980                         free_rp_inst(rp);
981                         return -ENOMEM;
982                 }
983                 INIT_HLIST_NODE(&inst->hlist);
984                 hlist_add_head(&inst->hlist, &rp->free_instances);
985         }
986
987         rp->nmissed = 0;
988         /* Establish function entry probe point */
989         ret = __register_kprobe(&rp->kp, called_from);
990         if (ret != 0)
991                 free_rp_inst(rp);
992         return ret;
993 }
994
995 static int __kprobes __register_kretprobes(struct kretprobe **rps, int num,
996         unsigned long called_from)
997 {
998         int ret = 0, i;
999
1000         if (num <= 0)
1001                 return -EINVAL;
1002         for (i = 0; i < num; i++) {
1003                 ret = __register_kretprobe(rps[i], called_from);
1004                 if (ret < 0) {
1005                         if (i > 0)
1006                                 unregister_kretprobes(rps, i);
1007                         break;
1008                 }
1009         }
1010         return ret;
1011 }
1012
1013 int __kprobes register_kretprobe(struct kretprobe *rp)
1014 {
1015         return __register_kretprobes(&rp, 1,
1016                         (unsigned long)__builtin_return_address(0));
1017 }
1018
1019 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1020 {
1021         unregister_kretprobes(&rp, 1);
1022 }
1023
1024 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1025 {
1026         return __register_kretprobes(rps, num,
1027                         (unsigned long)__builtin_return_address(0));
1028 }
1029
1030 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1031 {
1032         int i;
1033
1034         if (num <= 0)
1035                 return;
1036         mutex_lock(&kprobe_mutex);
1037         for (i = 0; i < num; i++)
1038                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1039                         rps[i]->kp.addr = NULL;
1040         mutex_unlock(&kprobe_mutex);
1041
1042         synchronize_sched();
1043         for (i = 0; i < num; i++) {
1044                 if (rps[i]->kp.addr) {
1045                         __unregister_kprobe_bottom(&rps[i]->kp);
1046                         cleanup_rp_inst(rps[i]);
1047                 }
1048         }
1049 }
1050
1051 #else /* CONFIG_KRETPROBES */
1052 int __kprobes register_kretprobe(struct kretprobe *rp)
1053 {
1054         return -ENOSYS;
1055 }
1056
1057 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1058 {
1059         return -ENOSYS;
1060 }
1061 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1062 {
1063 }
1064
1065 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1066 {
1067 }
1068
1069 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1070                                            struct pt_regs *regs)
1071 {
1072         return 0;
1073 }
1074
1075 #endif /* CONFIG_KRETPROBES */
1076
1077 static int __init init_kprobes(void)
1078 {
1079         int i, err = 0;
1080         unsigned long offset = 0, size = 0;
1081         char *modname, namebuf[128];
1082         const char *symbol_name;
1083         void *addr;
1084         struct kprobe_blackpoint *kb;
1085
1086         /* FIXME allocate the probe table, currently defined statically */
1087         /* initialize all list heads */
1088         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1089                 INIT_HLIST_HEAD(&kprobe_table[i]);
1090                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1091                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1092         }
1093
1094         /*
1095          * Lookup and populate the kprobe_blacklist.
1096          *
1097          * Unlike the kretprobe blacklist, we'll need to determine
1098          * the range of addresses that belong to the said functions,
1099          * since a kprobe need not necessarily be at the beginning
1100          * of a function.
1101          */
1102         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1103                 kprobe_lookup_name(kb->name, addr);
1104                 if (!addr)
1105                         continue;
1106
1107                 kb->start_addr = (unsigned long)addr;
1108                 symbol_name = kallsyms_lookup(kb->start_addr,
1109                                 &size, &offset, &modname, namebuf);
1110                 if (!symbol_name)
1111                         kb->range = 0;
1112                 else
1113                         kb->range = size;
1114         }
1115
1116         if (kretprobe_blacklist_size) {
1117                 /* lookup the function address from its name */
1118                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1119                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1120                                            kretprobe_blacklist[i].addr);
1121                         if (!kretprobe_blacklist[i].addr)
1122                                 printk("kretprobe: lookup failed: %s\n",
1123                                        kretprobe_blacklist[i].name);
1124                 }
1125         }
1126
1127         /* By default, kprobes are enabled */
1128         kprobe_enabled = true;
1129
1130         err = arch_init_kprobes();
1131         if (!err)
1132                 err = register_die_notifier(&kprobe_exceptions_nb);
1133         kprobes_initialized = (err == 0);
1134
1135         if (!err)
1136                 init_test_probes();
1137         return err;
1138 }
1139
1140 #ifdef CONFIG_DEBUG_FS
1141 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1142                 const char *sym, int offset,char *modname)
1143 {
1144         char *kprobe_type;
1145
1146         if (p->pre_handler == pre_handler_kretprobe)
1147                 kprobe_type = "r";
1148         else if (p->pre_handler == setjmp_pre_handler)
1149                 kprobe_type = "j";
1150         else
1151                 kprobe_type = "k";
1152         if (sym)
1153                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1154                         sym, offset, (modname ? modname : " "));
1155         else
1156                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1157 }
1158
1159 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1160 {
1161         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1162 }
1163
1164 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1165 {
1166         (*pos)++;
1167         if (*pos >= KPROBE_TABLE_SIZE)
1168                 return NULL;
1169         return pos;
1170 }
1171
1172 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1173 {
1174         /* Nothing to do */
1175 }
1176
1177 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1178 {
1179         struct hlist_head *head;
1180         struct hlist_node *node;
1181         struct kprobe *p, *kp;
1182         const char *sym = NULL;
1183         unsigned int i = *(loff_t *) v;
1184         unsigned long offset = 0;
1185         char *modname, namebuf[128];
1186
1187         head = &kprobe_table[i];
1188         preempt_disable();
1189         hlist_for_each_entry_rcu(p, node, head, hlist) {
1190                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1191                                         &offset, &modname, namebuf);
1192                 if (p->pre_handler == aggr_pre_handler) {
1193                         list_for_each_entry_rcu(kp, &p->list, list)
1194                                 report_probe(pi, kp, sym, offset, modname);
1195                 } else
1196                         report_probe(pi, p, sym, offset, modname);
1197         }
1198         preempt_enable();
1199         return 0;
1200 }
1201
1202 static struct seq_operations kprobes_seq_ops = {
1203         .start = kprobe_seq_start,
1204         .next  = kprobe_seq_next,
1205         .stop  = kprobe_seq_stop,
1206         .show  = show_kprobe_addr
1207 };
1208
1209 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1210 {
1211         return seq_open(filp, &kprobes_seq_ops);
1212 }
1213
1214 static struct file_operations debugfs_kprobes_operations = {
1215         .open           = kprobes_open,
1216         .read           = seq_read,
1217         .llseek         = seq_lseek,
1218         .release        = seq_release,
1219 };
1220
1221 static void __kprobes enable_all_kprobes(void)
1222 {
1223         struct hlist_head *head;
1224         struct hlist_node *node;
1225         struct kprobe *p;
1226         unsigned int i;
1227
1228         mutex_lock(&kprobe_mutex);
1229
1230         /* If kprobes are already enabled, just return */
1231         if (kprobe_enabled)
1232                 goto already_enabled;
1233
1234         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1235                 head = &kprobe_table[i];
1236                 hlist_for_each_entry_rcu(p, node, head, hlist)
1237                         arch_arm_kprobe(p);
1238         }
1239
1240         kprobe_enabled = true;
1241         printk(KERN_INFO "Kprobes globally enabled\n");
1242
1243 already_enabled:
1244         mutex_unlock(&kprobe_mutex);
1245         return;
1246 }
1247
1248 static void __kprobes disable_all_kprobes(void)
1249 {
1250         struct hlist_head *head;
1251         struct hlist_node *node;
1252         struct kprobe *p;
1253         unsigned int i;
1254
1255         mutex_lock(&kprobe_mutex);
1256
1257         /* If kprobes are already disabled, just return */
1258         if (!kprobe_enabled)
1259                 goto already_disabled;
1260
1261         kprobe_enabled = false;
1262         printk(KERN_INFO "Kprobes globally disabled\n");
1263         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1264                 head = &kprobe_table[i];
1265                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1266                         if (!arch_trampoline_kprobe(p))
1267                                 arch_disarm_kprobe(p);
1268                 }
1269         }
1270
1271         mutex_unlock(&kprobe_mutex);
1272         /* Allow all currently running kprobes to complete */
1273         synchronize_sched();
1274         return;
1275
1276 already_disabled:
1277         mutex_unlock(&kprobe_mutex);
1278         return;
1279 }
1280
1281 /*
1282  * XXX: The debugfs bool file interface doesn't allow for callbacks
1283  * when the bool state is switched. We can reuse that facility when
1284  * available
1285  */
1286 static ssize_t read_enabled_file_bool(struct file *file,
1287                char __user *user_buf, size_t count, loff_t *ppos)
1288 {
1289         char buf[3];
1290
1291         if (kprobe_enabled)
1292                 buf[0] = '1';
1293         else
1294                 buf[0] = '0';
1295         buf[1] = '\n';
1296         buf[2] = 0x00;
1297         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1298 }
1299
1300 static ssize_t write_enabled_file_bool(struct file *file,
1301                const char __user *user_buf, size_t count, loff_t *ppos)
1302 {
1303         char buf[32];
1304         int buf_size;
1305
1306         buf_size = min(count, (sizeof(buf)-1));
1307         if (copy_from_user(buf, user_buf, buf_size))
1308                 return -EFAULT;
1309
1310         switch (buf[0]) {
1311         case 'y':
1312         case 'Y':
1313         case '1':
1314                 enable_all_kprobes();
1315                 break;
1316         case 'n':
1317         case 'N':
1318         case '0':
1319                 disable_all_kprobes();
1320                 break;
1321         }
1322
1323         return count;
1324 }
1325
1326 static struct file_operations fops_kp = {
1327         .read =         read_enabled_file_bool,
1328         .write =        write_enabled_file_bool,
1329 };
1330
1331 static int __kprobes debugfs_kprobe_init(void)
1332 {
1333         struct dentry *dir, *file;
1334         unsigned int value = 1;
1335
1336         dir = debugfs_create_dir("kprobes", NULL);
1337         if (!dir)
1338                 return -ENOMEM;
1339
1340         file = debugfs_create_file("list", 0444, dir, NULL,
1341                                 &debugfs_kprobes_operations);
1342         if (!file) {
1343                 debugfs_remove(dir);
1344                 return -ENOMEM;
1345         }
1346
1347         file = debugfs_create_file("enabled", 0600, dir,
1348                                         &value, &fops_kp);
1349         if (!file) {
1350                 debugfs_remove(dir);
1351                 return -ENOMEM;
1352         }
1353
1354         return 0;
1355 }
1356
1357 late_initcall(debugfs_kprobe_init);
1358 #endif /* CONFIG_DEBUG_FS */
1359
1360 module_init(init_kprobes);
1361
1362 EXPORT_SYMBOL_GPL(register_kprobe);
1363 EXPORT_SYMBOL_GPL(unregister_kprobe);
1364 EXPORT_SYMBOL_GPL(register_kprobes);
1365 EXPORT_SYMBOL_GPL(unregister_kprobes);
1366 EXPORT_SYMBOL_GPL(register_jprobe);
1367 EXPORT_SYMBOL_GPL(unregister_jprobe);
1368 EXPORT_SYMBOL_GPL(register_jprobes);
1369 EXPORT_SYMBOL_GPL(unregister_jprobes);
1370 EXPORT_SYMBOL_GPL(jprobe_return);
1371 EXPORT_SYMBOL_GPL(register_kretprobe);
1372 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1373 EXPORT_SYMBOL_GPL(register_kretprobes);
1374 EXPORT_SYMBOL_GPL(unregister_kretprobes);