kprobes: bugfix: try_module_get even if calling_mod is NULL
[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 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 struct hlist_head kprobe_insn_pages;
119 static int kprobe_garbage_slots;
120 static int collect_garbage_slots(void);
121
122 static int __kprobes check_safety(void)
123 {
124         int ret = 0;
125 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
126         ret = freeze_processes();
127         if (ret == 0) {
128                 struct task_struct *p, *q;
129                 do_each_thread(p, q) {
130                         if (p != current && p->state == TASK_RUNNING &&
131                             p->pid != 0) {
132                                 printk("Check failed: %s is running\n",p->comm);
133                                 ret = -1;
134                                 goto loop_end;
135                         }
136                 } while_each_thread(p, q);
137         }
138 loop_end:
139         thaw_processes();
140 #else
141         synchronize_sched();
142 #endif
143         return ret;
144 }
145
146 /**
147  * get_insn_slot() - Find a slot on an executable page for an instruction.
148  * We allocate an executable page if there's no room on existing ones.
149  */
150 kprobe_opcode_t __kprobes *get_insn_slot(void)
151 {
152         struct kprobe_insn_page *kip;
153         struct hlist_node *pos;
154
155  retry:
156         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
157                 if (kip->nused < INSNS_PER_PAGE) {
158                         int i;
159                         for (i = 0; i < INSNS_PER_PAGE; i++) {
160                                 if (kip->slot_used[i] == SLOT_CLEAN) {
161                                         kip->slot_used[i] = SLOT_USED;
162                                         kip->nused++;
163                                         return kip->insns + (i * MAX_INSN_SIZE);
164                                 }
165                         }
166                         /* Surprise!  No unused slots.  Fix kip->nused. */
167                         kip->nused = INSNS_PER_PAGE;
168                 }
169         }
170
171         /* If there are any garbage slots, collect it and try again. */
172         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
173                 goto retry;
174         }
175         /* All out of space.  Need to allocate a new page. Use slot 0. */
176         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
177         if (!kip)
178                 return NULL;
179
180         /*
181          * Use module_alloc so this page is within +/- 2GB of where the
182          * kernel image and loaded module images reside. This is required
183          * so x86_64 can correctly handle the %rip-relative fixups.
184          */
185         kip->insns = module_alloc(PAGE_SIZE);
186         if (!kip->insns) {
187                 kfree(kip);
188                 return NULL;
189         }
190         INIT_HLIST_NODE(&kip->hlist);
191         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
192         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
193         kip->slot_used[0] = SLOT_USED;
194         kip->nused = 1;
195         kip->ngarbage = 0;
196         return kip->insns;
197 }
198
199 /* Return 1 if all garbages are collected, otherwise 0. */
200 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
201 {
202         kip->slot_used[idx] = SLOT_CLEAN;
203         kip->nused--;
204         if (kip->nused == 0) {
205                 /*
206                  * Page is no longer in use.  Free it unless
207                  * it's the last one.  We keep the last one
208                  * so as not to have to set it up again the
209                  * next time somebody inserts a probe.
210                  */
211                 hlist_del(&kip->hlist);
212                 if (hlist_empty(&kprobe_insn_pages)) {
213                         INIT_HLIST_NODE(&kip->hlist);
214                         hlist_add_head(&kip->hlist,
215                                        &kprobe_insn_pages);
216                 } else {
217                         module_free(NULL, kip->insns);
218                         kfree(kip);
219                 }
220                 return 1;
221         }
222         return 0;
223 }
224
225 static int __kprobes collect_garbage_slots(void)
226 {
227         struct kprobe_insn_page *kip;
228         struct hlist_node *pos, *next;
229
230         /* Ensure no-one is preepmted on the garbages */
231         if (check_safety() != 0)
232                 return -EAGAIN;
233
234         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
235                 int i;
236                 if (kip->ngarbage == 0)
237                         continue;
238                 kip->ngarbage = 0;      /* we will collect all garbages */
239                 for (i = 0; i < INSNS_PER_PAGE; i++) {
240                         if (kip->slot_used[i] == SLOT_DIRTY &&
241                             collect_one_slot(kip, i))
242                                 break;
243                 }
244         }
245         kprobe_garbage_slots = 0;
246         return 0;
247 }
248
249 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
250 {
251         struct kprobe_insn_page *kip;
252         struct hlist_node *pos;
253
254         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
255                 if (kip->insns <= slot &&
256                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
257                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
258                         if (dirty) {
259                                 kip->slot_used[i] = SLOT_DIRTY;
260                                 kip->ngarbage++;
261                         } else {
262                                 collect_one_slot(kip, i);
263                         }
264                         break;
265                 }
266         }
267
268         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
269                 collect_garbage_slots();
270 }
271 #endif
272
273 /* We have preemption disabled.. so it is safe to use __ versions */
274 static inline void set_kprobe_instance(struct kprobe *kp)
275 {
276         __get_cpu_var(kprobe_instance) = kp;
277 }
278
279 static inline void reset_kprobe_instance(void)
280 {
281         __get_cpu_var(kprobe_instance) = NULL;
282 }
283
284 /*
285  * This routine is called either:
286  *      - under the kprobe_mutex - during kprobe_[un]register()
287  *                              OR
288  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
289  */
290 struct kprobe __kprobes *get_kprobe(void *addr)
291 {
292         struct hlist_head *head;
293         struct hlist_node *node;
294         struct kprobe *p;
295
296         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
297         hlist_for_each_entry_rcu(p, node, head, hlist) {
298                 if (p->addr == addr)
299                         return p;
300         }
301         return NULL;
302 }
303
304 /*
305  * Aggregate handlers for multiple kprobes support - these handlers
306  * take care of invoking the individual kprobe handlers on p->list
307  */
308 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
309 {
310         struct kprobe *kp;
311
312         list_for_each_entry_rcu(kp, &p->list, list) {
313                 if (kp->pre_handler) {
314                         set_kprobe_instance(kp);
315                         if (kp->pre_handler(kp, regs))
316                                 return 1;
317                 }
318                 reset_kprobe_instance();
319         }
320         return 0;
321 }
322
323 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
324                                         unsigned long flags)
325 {
326         struct kprobe *kp;
327
328         list_for_each_entry_rcu(kp, &p->list, list) {
329                 if (kp->post_handler) {
330                         set_kprobe_instance(kp);
331                         kp->post_handler(kp, regs, flags);
332                         reset_kprobe_instance();
333                 }
334         }
335 }
336
337 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
338                                         int trapnr)
339 {
340         struct kprobe *cur = __get_cpu_var(kprobe_instance);
341
342         /*
343          * if we faulted "during" the execution of a user specified
344          * probe handler, invoke just that probe's fault handler
345          */
346         if (cur && cur->fault_handler) {
347                 if (cur->fault_handler(cur, regs, trapnr))
348                         return 1;
349         }
350         return 0;
351 }
352
353 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
354 {
355         struct kprobe *cur = __get_cpu_var(kprobe_instance);
356         int ret = 0;
357
358         if (cur && cur->break_handler) {
359                 if (cur->break_handler(cur, regs))
360                         ret = 1;
361         }
362         reset_kprobe_instance();
363         return ret;
364 }
365
366 /* Walks the list and increments nmissed count for multiprobe case */
367 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
368 {
369         struct kprobe *kp;
370         if (p->pre_handler != aggr_pre_handler) {
371                 p->nmissed++;
372         } else {
373                 list_for_each_entry_rcu(kp, &p->list, list)
374                         kp->nmissed++;
375         }
376         return;
377 }
378
379 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
380                                 struct hlist_head *head)
381 {
382         struct kretprobe *rp = ri->rp;
383
384         /* remove rp inst off the rprobe_inst_table */
385         hlist_del(&ri->hlist);
386         INIT_HLIST_NODE(&ri->hlist);
387         if (likely(rp)) {
388                 spin_lock(&rp->lock);
389                 hlist_add_head(&ri->hlist, &rp->free_instances);
390                 spin_unlock(&rp->lock);
391         } else
392                 /* Unregistering */
393                 hlist_add_head(&ri->hlist, head);
394 }
395
396 void kretprobe_hash_lock(struct task_struct *tsk,
397                          struct hlist_head **head, unsigned long *flags)
398 {
399         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
400         spinlock_t *hlist_lock;
401
402         *head = &kretprobe_inst_table[hash];
403         hlist_lock = kretprobe_table_lock_ptr(hash);
404         spin_lock_irqsave(hlist_lock, *flags);
405 }
406
407 static void kretprobe_table_lock(unsigned long hash, unsigned long *flags)
408 {
409         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
410         spin_lock_irqsave(hlist_lock, *flags);
411 }
412
413 void kretprobe_hash_unlock(struct task_struct *tsk, unsigned long *flags)
414 {
415         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
416         spinlock_t *hlist_lock;
417
418         hlist_lock = kretprobe_table_lock_ptr(hash);
419         spin_unlock_irqrestore(hlist_lock, *flags);
420 }
421
422 void kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
423 {
424         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
425         spin_unlock_irqrestore(hlist_lock, *flags);
426 }
427
428 /*
429  * This function is called from finish_task_switch when task tk becomes dead,
430  * so that we can recycle any function-return probe instances associated
431  * with this task. These left over instances represent probed functions
432  * that have been called but will never return.
433  */
434 void __kprobes kprobe_flush_task(struct task_struct *tk)
435 {
436         struct kretprobe_instance *ri;
437         struct hlist_head *head, empty_rp;
438         struct hlist_node *node, *tmp;
439         unsigned long hash, flags = 0;
440
441         if (unlikely(!kprobes_initialized))
442                 /* Early boot.  kretprobe_table_locks not yet initialized. */
443                 return;
444
445         hash = hash_ptr(tk, KPROBE_HASH_BITS);
446         head = &kretprobe_inst_table[hash];
447         kretprobe_table_lock(hash, &flags);
448         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
449                 if (ri->task == tk)
450                         recycle_rp_inst(ri, &empty_rp);
451         }
452         kretprobe_table_unlock(hash, &flags);
453         INIT_HLIST_HEAD(&empty_rp);
454         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
455                 hlist_del(&ri->hlist);
456                 kfree(ri);
457         }
458 }
459
460 static inline void free_rp_inst(struct kretprobe *rp)
461 {
462         struct kretprobe_instance *ri;
463         struct hlist_node *pos, *next;
464
465         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
466                 hlist_del(&ri->hlist);
467                 kfree(ri);
468         }
469 }
470
471 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
472 {
473         unsigned long flags, hash;
474         struct kretprobe_instance *ri;
475         struct hlist_node *pos, *next;
476         struct hlist_head *head;
477
478         /* No race here */
479         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
480                 kretprobe_table_lock(hash, &flags);
481                 head = &kretprobe_inst_table[hash];
482                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
483                         if (ri->rp == rp)
484                                 ri->rp = NULL;
485                 }
486                 kretprobe_table_unlock(hash, &flags);
487         }
488         free_rp_inst(rp);
489 }
490
491 /*
492  * Keep all fields in the kprobe consistent
493  */
494 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
495 {
496         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
497         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
498 }
499
500 /*
501 * Add the new probe to old_p->list. Fail if this is the
502 * second jprobe at the address - two jprobes can't coexist
503 */
504 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
505 {
506         if (p->break_handler) {
507                 if (old_p->break_handler)
508                         return -EEXIST;
509                 list_add_tail_rcu(&p->list, &old_p->list);
510                 old_p->break_handler = aggr_break_handler;
511         } else
512                 list_add_rcu(&p->list, &old_p->list);
513         if (p->post_handler && !old_p->post_handler)
514                 old_p->post_handler = aggr_post_handler;
515         return 0;
516 }
517
518 /*
519  * Fill in the required fields of the "manager kprobe". Replace the
520  * earlier kprobe in the hlist with the manager kprobe
521  */
522 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
523 {
524         copy_kprobe(p, ap);
525         flush_insn_slot(ap);
526         ap->addr = p->addr;
527         ap->pre_handler = aggr_pre_handler;
528         ap->fault_handler = aggr_fault_handler;
529         if (p->post_handler)
530                 ap->post_handler = aggr_post_handler;
531         if (p->break_handler)
532                 ap->break_handler = aggr_break_handler;
533
534         INIT_LIST_HEAD(&ap->list);
535         list_add_rcu(&p->list, &ap->list);
536
537         hlist_replace_rcu(&p->hlist, &ap->hlist);
538 }
539
540 /*
541  * This is the second or subsequent kprobe at the address - handle
542  * the intricacies
543  */
544 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
545                                           struct kprobe *p)
546 {
547         int ret = 0;
548         struct kprobe *ap;
549
550         if (old_p->pre_handler == aggr_pre_handler) {
551                 copy_kprobe(old_p, p);
552                 ret = add_new_kprobe(old_p, p);
553         } else {
554                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
555                 if (!ap)
556                         return -ENOMEM;
557                 add_aggr_kprobe(ap, old_p);
558                 copy_kprobe(ap, p);
559                 ret = add_new_kprobe(ap, p);
560         }
561         return ret;
562 }
563
564 static int __kprobes in_kprobes_functions(unsigned long addr)
565 {
566         struct kprobe_blackpoint *kb;
567
568         if (addr >= (unsigned long)__kprobes_text_start &&
569             addr < (unsigned long)__kprobes_text_end)
570                 return -EINVAL;
571         /*
572          * If there exists a kprobe_blacklist, verify and
573          * fail any probe registration in the prohibited area
574          */
575         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
576                 if (kb->start_addr) {
577                         if (addr >= kb->start_addr &&
578                             addr < (kb->start_addr + kb->range))
579                                 return -EINVAL;
580                 }
581         }
582         return 0;
583 }
584
585 /*
586  * If we have a symbol_name argument, look it up and add the offset field
587  * to it. This way, we can specify a relative address to a symbol.
588  */
589 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
590 {
591         kprobe_opcode_t *addr = p->addr;
592         if (p->symbol_name) {
593                 if (addr)
594                         return NULL;
595                 kprobe_lookup_name(p->symbol_name, addr);
596         }
597
598         if (!addr)
599                 return NULL;
600         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
601 }
602
603 static int __kprobes __register_kprobe(struct kprobe *p,
604         unsigned long called_from)
605 {
606         int ret = 0;
607         struct kprobe *old_p;
608         struct module *probed_mod;
609         kprobe_opcode_t *addr;
610
611         addr = kprobe_addr(p);
612         if (!addr)
613                 return -EINVAL;
614         p->addr = addr;
615
616         preempt_disable();
617         if (!__kernel_text_address((unsigned long) p->addr) ||
618             in_kprobes_functions((unsigned long) p->addr)) {
619                 preempt_enable();
620                 return -EINVAL;
621         }
622
623         p->mod_refcounted = 0;
624
625         /*
626          * Check if are we probing a module.
627          */
628         probed_mod = __module_text_address((unsigned long) p->addr);
629         if (probed_mod) {
630                 struct module *calling_mod;
631                 calling_mod = __module_text_address(called_from);
632                 /*
633                  * We must allow modules to probe themself and in this case
634                  * avoid incrementing the module refcount, so as to allow
635                  * unloading of self probing modules.
636                  */
637                 if (calling_mod != probed_mod) {
638                         if (unlikely(!try_module_get(probed_mod))) {
639                                 preempt_enable();
640                                 return -EINVAL;
641                         }
642                         p->mod_refcounted = 1;
643                 } else
644                         probed_mod = NULL;
645         }
646         preempt_enable();
647
648         p->nmissed = 0;
649         INIT_LIST_HEAD(&p->list);
650         mutex_lock(&kprobe_mutex);
651         old_p = get_kprobe(p->addr);
652         if (old_p) {
653                 ret = register_aggr_kprobe(old_p, p);
654                 goto out;
655         }
656
657         ret = arch_prepare_kprobe(p);
658         if (ret)
659                 goto out;
660
661         INIT_HLIST_NODE(&p->hlist);
662         hlist_add_head_rcu(&p->hlist,
663                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
664
665         if (kprobe_enabled)
666                 arch_arm_kprobe(p);
667
668 out:
669         mutex_unlock(&kprobe_mutex);
670
671         if (ret && probed_mod)
672                 module_put(probed_mod);
673         return ret;
674 }
675
676 /*
677  * Unregister a kprobe without a scheduler synchronization.
678  */
679 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
680 {
681         struct kprobe *old_p, *list_p;
682
683         old_p = get_kprobe(p->addr);
684         if (unlikely(!old_p))
685                 return -EINVAL;
686
687         if (p != old_p) {
688                 list_for_each_entry_rcu(list_p, &old_p->list, list)
689                         if (list_p == p)
690                         /* kprobe p is a valid probe */
691                                 goto valid_p;
692                 return -EINVAL;
693         }
694 valid_p:
695         if (old_p == p ||
696             (old_p->pre_handler == aggr_pre_handler &&
697              list_is_singular(&old_p->list))) {
698                 /*
699                  * Only probe on the hash list. Disarm only if kprobes are
700                  * enabled - otherwise, the breakpoint would already have
701                  * been removed. We save on flushing icache.
702                  */
703                 if (kprobe_enabled)
704                         arch_disarm_kprobe(p);
705                 hlist_del_rcu(&old_p->hlist);
706         } else {
707                 if (p->break_handler)
708                         old_p->break_handler = NULL;
709                 if (p->post_handler) {
710                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
711                                 if ((list_p != p) && (list_p->post_handler))
712                                         goto noclean;
713                         }
714                         old_p->post_handler = NULL;
715                 }
716 noclean:
717                 list_del_rcu(&p->list);
718         }
719         return 0;
720 }
721
722 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
723 {
724         struct module *mod;
725         struct kprobe *old_p;
726
727         if (p->mod_refcounted) {
728                 /*
729                  * Since we've already incremented refcount,
730                  * we don't need to disable preemption.
731                  */
732                 mod = module_text_address((unsigned long)p->addr);
733                 if (mod)
734                         module_put(mod);
735         }
736
737         if (list_empty(&p->list) || list_is_singular(&p->list)) {
738                 if (!list_empty(&p->list)) {
739                         /* "p" is the last child of an aggr_kprobe */
740                         old_p = list_entry(p->list.next, struct kprobe, list);
741                         list_del(&p->list);
742                         kfree(old_p);
743                 }
744                 arch_remove_kprobe(p);
745         }
746 }
747
748 static int __register_kprobes(struct kprobe **kps, int num,
749         unsigned long called_from)
750 {
751         int i, ret = 0;
752
753         if (num <= 0)
754                 return -EINVAL;
755         for (i = 0; i < num; i++) {
756                 ret = __register_kprobe(kps[i], called_from);
757                 if (ret < 0) {
758                         if (i > 0)
759                                 unregister_kprobes(kps, i);
760                         break;
761                 }
762         }
763         return ret;
764 }
765
766 /*
767  * Registration and unregistration functions for kprobe.
768  */
769 int __kprobes register_kprobe(struct kprobe *p)
770 {
771         return __register_kprobes(&p, 1,
772                                   (unsigned long)__builtin_return_address(0));
773 }
774
775 void __kprobes unregister_kprobe(struct kprobe *p)
776 {
777         unregister_kprobes(&p, 1);
778 }
779
780 int __kprobes register_kprobes(struct kprobe **kps, int num)
781 {
782         return __register_kprobes(kps, num,
783                                   (unsigned long)__builtin_return_address(0));
784 }
785
786 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
787 {
788         int i;
789
790         if (num <= 0)
791                 return;
792         mutex_lock(&kprobe_mutex);
793         for (i = 0; i < num; i++)
794                 if (__unregister_kprobe_top(kps[i]) < 0)
795                         kps[i]->addr = NULL;
796         mutex_unlock(&kprobe_mutex);
797
798         synchronize_sched();
799         for (i = 0; i < num; i++)
800                 if (kps[i]->addr)
801                         __unregister_kprobe_bottom(kps[i]);
802 }
803
804 static struct notifier_block kprobe_exceptions_nb = {
805         .notifier_call = kprobe_exceptions_notify,
806         .priority = 0x7fffffff /* we need to be notified first */
807 };
808
809 unsigned long __weak arch_deref_entry_point(void *entry)
810 {
811         return (unsigned long)entry;
812 }
813
814 static int __register_jprobes(struct jprobe **jps, int num,
815         unsigned long called_from)
816 {
817         struct jprobe *jp;
818         int ret = 0, i;
819
820         if (num <= 0)
821                 return -EINVAL;
822         for (i = 0; i < num; i++) {
823                 unsigned long addr;
824                 jp = jps[i];
825                 addr = arch_deref_entry_point(jp->entry);
826
827                 if (!kernel_text_address(addr))
828                         ret = -EINVAL;
829                 else {
830                         /* Todo: Verify probepoint is a function entry point */
831                         jp->kp.pre_handler = setjmp_pre_handler;
832                         jp->kp.break_handler = longjmp_break_handler;
833                         ret = __register_kprobe(&jp->kp, called_from);
834                 }
835                 if (ret < 0) {
836                         if (i > 0)
837                                 unregister_jprobes(jps, i);
838                         break;
839                 }
840         }
841         return ret;
842 }
843
844 int __kprobes register_jprobe(struct jprobe *jp)
845 {
846         return __register_jprobes(&jp, 1,
847                 (unsigned long)__builtin_return_address(0));
848 }
849
850 void __kprobes unregister_jprobe(struct jprobe *jp)
851 {
852         unregister_jprobes(&jp, 1);
853 }
854
855 int __kprobes register_jprobes(struct jprobe **jps, int num)
856 {
857         return __register_jprobes(jps, num,
858                 (unsigned long)__builtin_return_address(0));
859 }
860
861 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
862 {
863         int i;
864
865         if (num <= 0)
866                 return;
867         mutex_lock(&kprobe_mutex);
868         for (i = 0; i < num; i++)
869                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
870                         jps[i]->kp.addr = NULL;
871         mutex_unlock(&kprobe_mutex);
872
873         synchronize_sched();
874         for (i = 0; i < num; i++) {
875                 if (jps[i]->kp.addr)
876                         __unregister_kprobe_bottom(&jps[i]->kp);
877         }
878 }
879
880 #ifdef CONFIG_KRETPROBES
881 /*
882  * This kprobe pre_handler is registered with every kretprobe. When probe
883  * hits it will set up the return probe.
884  */
885 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
886                                            struct pt_regs *regs)
887 {
888         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
889         unsigned long hash, flags = 0;
890         struct kretprobe_instance *ri;
891
892         /*TODO: consider to only swap the RA after the last pre_handler fired */
893         hash = hash_ptr(current, KPROBE_HASH_BITS);
894         spin_lock_irqsave(&rp->lock, flags);
895         if (!hlist_empty(&rp->free_instances)) {
896                 ri = hlist_entry(rp->free_instances.first,
897                                 struct kretprobe_instance, hlist);
898                 hlist_del(&ri->hlist);
899                 spin_unlock_irqrestore(&rp->lock, flags);
900
901                 ri->rp = rp;
902                 ri->task = current;
903
904                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
905                         spin_unlock_irqrestore(&rp->lock, flags);
906                         return 0;
907                 }
908
909                 arch_prepare_kretprobe(ri, regs);
910
911                 /* XXX(hch): why is there no hlist_move_head? */
912                 INIT_HLIST_NODE(&ri->hlist);
913                 kretprobe_table_lock(hash, &flags);
914                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
915                 kretprobe_table_unlock(hash, &flags);
916         } else {
917                 rp->nmissed++;
918                 spin_unlock_irqrestore(&rp->lock, flags);
919         }
920         return 0;
921 }
922
923 static int __kprobes __register_kretprobe(struct kretprobe *rp,
924                                           unsigned long called_from)
925 {
926         int ret = 0;
927         struct kretprobe_instance *inst;
928         int i;
929         void *addr;
930
931         if (kretprobe_blacklist_size) {
932                 addr = kprobe_addr(&rp->kp);
933                 if (!addr)
934                         return -EINVAL;
935
936                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
937                         if (kretprobe_blacklist[i].addr == addr)
938                                 return -EINVAL;
939                 }
940         }
941
942         rp->kp.pre_handler = pre_handler_kretprobe;
943         rp->kp.post_handler = NULL;
944         rp->kp.fault_handler = NULL;
945         rp->kp.break_handler = NULL;
946
947         /* Pre-allocate memory for max kretprobe instances */
948         if (rp->maxactive <= 0) {
949 #ifdef CONFIG_PREEMPT
950                 rp->maxactive = max(10, 2 * NR_CPUS);
951 #else
952                 rp->maxactive = NR_CPUS;
953 #endif
954         }
955         spin_lock_init(&rp->lock);
956         INIT_HLIST_HEAD(&rp->free_instances);
957         for (i = 0; i < rp->maxactive; i++) {
958                 inst = kmalloc(sizeof(struct kretprobe_instance) +
959                                rp->data_size, GFP_KERNEL);
960                 if (inst == NULL) {
961                         free_rp_inst(rp);
962                         return -ENOMEM;
963                 }
964                 INIT_HLIST_NODE(&inst->hlist);
965                 hlist_add_head(&inst->hlist, &rp->free_instances);
966         }
967
968         rp->nmissed = 0;
969         /* Establish function entry probe point */
970         ret = __register_kprobe(&rp->kp, called_from);
971         if (ret != 0)
972                 free_rp_inst(rp);
973         return ret;
974 }
975
976 static int __register_kretprobes(struct kretprobe **rps, int num,
977         unsigned long called_from)
978 {
979         int ret = 0, i;
980
981         if (num <= 0)
982                 return -EINVAL;
983         for (i = 0; i < num; i++) {
984                 ret = __register_kretprobe(rps[i], called_from);
985                 if (ret < 0) {
986                         if (i > 0)
987                                 unregister_kretprobes(rps, i);
988                         break;
989                 }
990         }
991         return ret;
992 }
993
994 int __kprobes register_kretprobe(struct kretprobe *rp)
995 {
996         return __register_kretprobes(&rp, 1,
997                         (unsigned long)__builtin_return_address(0));
998 }
999
1000 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1001 {
1002         unregister_kretprobes(&rp, 1);
1003 }
1004
1005 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1006 {
1007         return __register_kretprobes(rps, num,
1008                         (unsigned long)__builtin_return_address(0));
1009 }
1010
1011 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1012 {
1013         int i;
1014
1015         if (num <= 0)
1016                 return;
1017         mutex_lock(&kprobe_mutex);
1018         for (i = 0; i < num; i++)
1019                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1020                         rps[i]->kp.addr = NULL;
1021         mutex_unlock(&kprobe_mutex);
1022
1023         synchronize_sched();
1024         for (i = 0; i < num; i++) {
1025                 if (rps[i]->kp.addr) {
1026                         __unregister_kprobe_bottom(&rps[i]->kp);
1027                         cleanup_rp_inst(rps[i]);
1028                 }
1029         }
1030 }
1031
1032 #else /* CONFIG_KRETPROBES */
1033 int __kprobes register_kretprobe(struct kretprobe *rp)
1034 {
1035         return -ENOSYS;
1036 }
1037
1038 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1039 {
1040         return -ENOSYS;
1041 }
1042 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1043 {
1044 }
1045
1046 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1047 {
1048 }
1049
1050 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1051                                            struct pt_regs *regs)
1052 {
1053         return 0;
1054 }
1055
1056 #endif /* CONFIG_KRETPROBES */
1057
1058 static int __init init_kprobes(void)
1059 {
1060         int i, err = 0;
1061         unsigned long offset = 0, size = 0;
1062         char *modname, namebuf[128];
1063         const char *symbol_name;
1064         void *addr;
1065         struct kprobe_blackpoint *kb;
1066
1067         /* FIXME allocate the probe table, currently defined statically */
1068         /* initialize all list heads */
1069         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1070                 INIT_HLIST_HEAD(&kprobe_table[i]);
1071                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1072                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1073         }
1074
1075         /*
1076          * Lookup and populate the kprobe_blacklist.
1077          *
1078          * Unlike the kretprobe blacklist, we'll need to determine
1079          * the range of addresses that belong to the said functions,
1080          * since a kprobe need not necessarily be at the beginning
1081          * of a function.
1082          */
1083         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1084                 kprobe_lookup_name(kb->name, addr);
1085                 if (!addr)
1086                         continue;
1087
1088                 kb->start_addr = (unsigned long)addr;
1089                 symbol_name = kallsyms_lookup(kb->start_addr,
1090                                 &size, &offset, &modname, namebuf);
1091                 if (!symbol_name)
1092                         kb->range = 0;
1093                 else
1094                         kb->range = size;
1095         }
1096
1097         if (kretprobe_blacklist_size) {
1098                 /* lookup the function address from its name */
1099                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1100                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1101                                            kretprobe_blacklist[i].addr);
1102                         if (!kretprobe_blacklist[i].addr)
1103                                 printk("kretprobe: lookup failed: %s\n",
1104                                        kretprobe_blacklist[i].name);
1105                 }
1106         }
1107
1108         /* By default, kprobes are enabled */
1109         kprobe_enabled = true;
1110
1111         err = arch_init_kprobes();
1112         if (!err)
1113                 err = register_die_notifier(&kprobe_exceptions_nb);
1114         kprobes_initialized = (err == 0);
1115
1116         if (!err)
1117                 init_test_probes();
1118         return err;
1119 }
1120
1121 #ifdef CONFIG_DEBUG_FS
1122 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1123                 const char *sym, int offset,char *modname)
1124 {
1125         char *kprobe_type;
1126
1127         if (p->pre_handler == pre_handler_kretprobe)
1128                 kprobe_type = "r";
1129         else if (p->pre_handler == setjmp_pre_handler)
1130                 kprobe_type = "j";
1131         else
1132                 kprobe_type = "k";
1133         if (sym)
1134                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1135                         sym, offset, (modname ? modname : " "));
1136         else
1137                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1138 }
1139
1140 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1141 {
1142         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1143 }
1144
1145 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1146 {
1147         (*pos)++;
1148         if (*pos >= KPROBE_TABLE_SIZE)
1149                 return NULL;
1150         return pos;
1151 }
1152
1153 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1154 {
1155         /* Nothing to do */
1156 }
1157
1158 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1159 {
1160         struct hlist_head *head;
1161         struct hlist_node *node;
1162         struct kprobe *p, *kp;
1163         const char *sym = NULL;
1164         unsigned int i = *(loff_t *) v;
1165         unsigned long offset = 0;
1166         char *modname, namebuf[128];
1167
1168         head = &kprobe_table[i];
1169         preempt_disable();
1170         hlist_for_each_entry_rcu(p, node, head, hlist) {
1171                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1172                                         &offset, &modname, namebuf);
1173                 if (p->pre_handler == aggr_pre_handler) {
1174                         list_for_each_entry_rcu(kp, &p->list, list)
1175                                 report_probe(pi, kp, sym, offset, modname);
1176                 } else
1177                         report_probe(pi, p, sym, offset, modname);
1178         }
1179         preempt_enable();
1180         return 0;
1181 }
1182
1183 static struct seq_operations kprobes_seq_ops = {
1184         .start = kprobe_seq_start,
1185         .next  = kprobe_seq_next,
1186         .stop  = kprobe_seq_stop,
1187         .show  = show_kprobe_addr
1188 };
1189
1190 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1191 {
1192         return seq_open(filp, &kprobes_seq_ops);
1193 }
1194
1195 static struct file_operations debugfs_kprobes_operations = {
1196         .open           = kprobes_open,
1197         .read           = seq_read,
1198         .llseek         = seq_lseek,
1199         .release        = seq_release,
1200 };
1201
1202 static void __kprobes enable_all_kprobes(void)
1203 {
1204         struct hlist_head *head;
1205         struct hlist_node *node;
1206         struct kprobe *p;
1207         unsigned int i;
1208
1209         mutex_lock(&kprobe_mutex);
1210
1211         /* If kprobes are already enabled, just return */
1212         if (kprobe_enabled)
1213                 goto already_enabled;
1214
1215         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1216                 head = &kprobe_table[i];
1217                 hlist_for_each_entry_rcu(p, node, head, hlist)
1218                         arch_arm_kprobe(p);
1219         }
1220
1221         kprobe_enabled = true;
1222         printk(KERN_INFO "Kprobes globally enabled\n");
1223
1224 already_enabled:
1225         mutex_unlock(&kprobe_mutex);
1226         return;
1227 }
1228
1229 static void __kprobes disable_all_kprobes(void)
1230 {
1231         struct hlist_head *head;
1232         struct hlist_node *node;
1233         struct kprobe *p;
1234         unsigned int i;
1235
1236         mutex_lock(&kprobe_mutex);
1237
1238         /* If kprobes are already disabled, just return */
1239         if (!kprobe_enabled)
1240                 goto already_disabled;
1241
1242         kprobe_enabled = false;
1243         printk(KERN_INFO "Kprobes globally disabled\n");
1244         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1245                 head = &kprobe_table[i];
1246                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1247                         if (!arch_trampoline_kprobe(p))
1248                                 arch_disarm_kprobe(p);
1249                 }
1250         }
1251
1252         mutex_unlock(&kprobe_mutex);
1253         /* Allow all currently running kprobes to complete */
1254         synchronize_sched();
1255         return;
1256
1257 already_disabled:
1258         mutex_unlock(&kprobe_mutex);
1259         return;
1260 }
1261
1262 /*
1263  * XXX: The debugfs bool file interface doesn't allow for callbacks
1264  * when the bool state is switched. We can reuse that facility when
1265  * available
1266  */
1267 static ssize_t read_enabled_file_bool(struct file *file,
1268                char __user *user_buf, size_t count, loff_t *ppos)
1269 {
1270         char buf[3];
1271
1272         if (kprobe_enabled)
1273                 buf[0] = '1';
1274         else
1275                 buf[0] = '0';
1276         buf[1] = '\n';
1277         buf[2] = 0x00;
1278         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1279 }
1280
1281 static ssize_t write_enabled_file_bool(struct file *file,
1282                const char __user *user_buf, size_t count, loff_t *ppos)
1283 {
1284         char buf[32];
1285         int buf_size;
1286
1287         buf_size = min(count, (sizeof(buf)-1));
1288         if (copy_from_user(buf, user_buf, buf_size))
1289                 return -EFAULT;
1290
1291         switch (buf[0]) {
1292         case 'y':
1293         case 'Y':
1294         case '1':
1295                 enable_all_kprobes();
1296                 break;
1297         case 'n':
1298         case 'N':
1299         case '0':
1300                 disable_all_kprobes();
1301                 break;
1302         }
1303
1304         return count;
1305 }
1306
1307 static struct file_operations fops_kp = {
1308         .read =         read_enabled_file_bool,
1309         .write =        write_enabled_file_bool,
1310 };
1311
1312 static int __kprobes debugfs_kprobe_init(void)
1313 {
1314         struct dentry *dir, *file;
1315         unsigned int value = 1;
1316
1317         dir = debugfs_create_dir("kprobes", NULL);
1318         if (!dir)
1319                 return -ENOMEM;
1320
1321         file = debugfs_create_file("list", 0444, dir, NULL,
1322                                 &debugfs_kprobes_operations);
1323         if (!file) {
1324                 debugfs_remove(dir);
1325                 return -ENOMEM;
1326         }
1327
1328         file = debugfs_create_file("enabled", 0600, dir,
1329                                         &value, &fops_kp);
1330         if (!file) {
1331                 debugfs_remove(dir);
1332                 return -ENOMEM;
1333         }
1334
1335         return 0;
1336 }
1337
1338 late_initcall(debugfs_kprobe_init);
1339 #endif /* CONFIG_DEBUG_FS */
1340
1341 module_init(init_kprobes);
1342
1343 EXPORT_SYMBOL_GPL(register_kprobe);
1344 EXPORT_SYMBOL_GPL(unregister_kprobe);
1345 EXPORT_SYMBOL_GPL(register_kprobes);
1346 EXPORT_SYMBOL_GPL(unregister_kprobes);
1347 EXPORT_SYMBOL_GPL(register_jprobe);
1348 EXPORT_SYMBOL_GPL(unregister_jprobe);
1349 EXPORT_SYMBOL_GPL(register_jprobes);
1350 EXPORT_SYMBOL_GPL(unregister_jprobes);
1351 EXPORT_SYMBOL_GPL(jprobe_return);
1352 EXPORT_SYMBOL_GPL(register_kretprobe);
1353 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1354 EXPORT_SYMBOL_GPL(register_kretprobes);
1355 EXPORT_SYMBOL_GPL(unregister_kretprobes);