]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - kernel/kprobes.c
kprobes: add kprobe_insn_mutex and cleanup arch_remove_kprobe()
[linux-2.6-omap-h63xx.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 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 kretprobe_table_lock(unsigned long hash, unsigned long *flags)
425 {
426         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
427         spin_lock_irqsave(hlist_lock, *flags);
428 }
429
430 void kretprobe_hash_unlock(struct task_struct *tsk, unsigned long *flags)
431 {
432         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
433         spinlock_t *hlist_lock;
434
435         hlist_lock = kretprobe_table_lock_ptr(hash);
436         spin_unlock_irqrestore(hlist_lock, *flags);
437 }
438
439 void kretprobe_table_unlock(unsigned long hash, unsigned long *flags)
440 {
441         spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
442         spin_unlock_irqrestore(hlist_lock, *flags);
443 }
444
445 /*
446  * This function is called from finish_task_switch when task tk becomes dead,
447  * so that we can recycle any function-return probe instances associated
448  * with this task. These left over instances represent probed functions
449  * that have been called but will never return.
450  */
451 void __kprobes kprobe_flush_task(struct task_struct *tk)
452 {
453         struct kretprobe_instance *ri;
454         struct hlist_head *head, empty_rp;
455         struct hlist_node *node, *tmp;
456         unsigned long hash, flags = 0;
457
458         if (unlikely(!kprobes_initialized))
459                 /* Early boot.  kretprobe_table_locks not yet initialized. */
460                 return;
461
462         hash = hash_ptr(tk, KPROBE_HASH_BITS);
463         head = &kretprobe_inst_table[hash];
464         kretprobe_table_lock(hash, &flags);
465         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
466                 if (ri->task == tk)
467                         recycle_rp_inst(ri, &empty_rp);
468         }
469         kretprobe_table_unlock(hash, &flags);
470         INIT_HLIST_HEAD(&empty_rp);
471         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
472                 hlist_del(&ri->hlist);
473                 kfree(ri);
474         }
475 }
476
477 static inline void free_rp_inst(struct kretprobe *rp)
478 {
479         struct kretprobe_instance *ri;
480         struct hlist_node *pos, *next;
481
482         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
483                 hlist_del(&ri->hlist);
484                 kfree(ri);
485         }
486 }
487
488 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
489 {
490         unsigned long flags, hash;
491         struct kretprobe_instance *ri;
492         struct hlist_node *pos, *next;
493         struct hlist_head *head;
494
495         /* No race here */
496         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
497                 kretprobe_table_lock(hash, &flags);
498                 head = &kretprobe_inst_table[hash];
499                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
500                         if (ri->rp == rp)
501                                 ri->rp = NULL;
502                 }
503                 kretprobe_table_unlock(hash, &flags);
504         }
505         free_rp_inst(rp);
506 }
507
508 /*
509  * Keep all fields in the kprobe consistent
510  */
511 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
512 {
513         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
514         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
515 }
516
517 /*
518 * Add the new probe to old_p->list. Fail if this is the
519 * second jprobe at the address - two jprobes can't coexist
520 */
521 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
522 {
523         if (p->break_handler) {
524                 if (old_p->break_handler)
525                         return -EEXIST;
526                 list_add_tail_rcu(&p->list, &old_p->list);
527                 old_p->break_handler = aggr_break_handler;
528         } else
529                 list_add_rcu(&p->list, &old_p->list);
530         if (p->post_handler && !old_p->post_handler)
531                 old_p->post_handler = aggr_post_handler;
532         return 0;
533 }
534
535 /*
536  * Fill in the required fields of the "manager kprobe". Replace the
537  * earlier kprobe in the hlist with the manager kprobe
538  */
539 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
540 {
541         copy_kprobe(p, ap);
542         flush_insn_slot(ap);
543         ap->addr = p->addr;
544         ap->pre_handler = aggr_pre_handler;
545         ap->fault_handler = aggr_fault_handler;
546         if (p->post_handler)
547                 ap->post_handler = aggr_post_handler;
548         if (p->break_handler)
549                 ap->break_handler = aggr_break_handler;
550
551         INIT_LIST_HEAD(&ap->list);
552         list_add_rcu(&p->list, &ap->list);
553
554         hlist_replace_rcu(&p->hlist, &ap->hlist);
555 }
556
557 /*
558  * This is the second or subsequent kprobe at the address - handle
559  * the intricacies
560  */
561 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
562                                           struct kprobe *p)
563 {
564         int ret = 0;
565         struct kprobe *ap;
566
567         if (old_p->pre_handler == aggr_pre_handler) {
568                 copy_kprobe(old_p, p);
569                 ret = add_new_kprobe(old_p, p);
570         } else {
571                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
572                 if (!ap)
573                         return -ENOMEM;
574                 add_aggr_kprobe(ap, old_p);
575                 copy_kprobe(ap, p);
576                 ret = add_new_kprobe(ap, p);
577         }
578         return ret;
579 }
580
581 static int __kprobes in_kprobes_functions(unsigned long addr)
582 {
583         struct kprobe_blackpoint *kb;
584
585         if (addr >= (unsigned long)__kprobes_text_start &&
586             addr < (unsigned long)__kprobes_text_end)
587                 return -EINVAL;
588         /*
589          * If there exists a kprobe_blacklist, verify and
590          * fail any probe registration in the prohibited area
591          */
592         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
593                 if (kb->start_addr) {
594                         if (addr >= kb->start_addr &&
595                             addr < (kb->start_addr + kb->range))
596                                 return -EINVAL;
597                 }
598         }
599         return 0;
600 }
601
602 /*
603  * If we have a symbol_name argument, look it up and add the offset field
604  * to it. This way, we can specify a relative address to a symbol.
605  */
606 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
607 {
608         kprobe_opcode_t *addr = p->addr;
609         if (p->symbol_name) {
610                 if (addr)
611                         return NULL;
612                 kprobe_lookup_name(p->symbol_name, addr);
613         }
614
615         if (!addr)
616                 return NULL;
617         return (kprobe_opcode_t *)(((char *)addr) + p->offset);
618 }
619
620 static int __kprobes __register_kprobe(struct kprobe *p,
621         unsigned long called_from)
622 {
623         int ret = 0;
624         struct kprobe *old_p;
625         struct module *probed_mod;
626         kprobe_opcode_t *addr;
627
628         addr = kprobe_addr(p);
629         if (!addr)
630                 return -EINVAL;
631         p->addr = addr;
632
633         preempt_disable();
634         if (!__kernel_text_address((unsigned long) p->addr) ||
635             in_kprobes_functions((unsigned long) p->addr)) {
636                 preempt_enable();
637                 return -EINVAL;
638         }
639
640         p->mod_refcounted = 0;
641
642         /*
643          * Check if are we probing a module.
644          */
645         probed_mod = __module_text_address((unsigned long) p->addr);
646         if (probed_mod) {
647                 struct module *calling_mod;
648                 calling_mod = __module_text_address(called_from);
649                 /*
650                  * We must allow modules to probe themself and in this case
651                  * avoid incrementing the module refcount, so as to allow
652                  * unloading of self probing modules.
653                  */
654                 if (calling_mod != probed_mod) {
655                         if (unlikely(!try_module_get(probed_mod))) {
656                                 preempt_enable();
657                                 return -EINVAL;
658                         }
659                         p->mod_refcounted = 1;
660                 } else
661                         probed_mod = NULL;
662         }
663         preempt_enable();
664
665         p->nmissed = 0;
666         INIT_LIST_HEAD(&p->list);
667         mutex_lock(&kprobe_mutex);
668         old_p = get_kprobe(p->addr);
669         if (old_p) {
670                 ret = register_aggr_kprobe(old_p, p);
671                 goto out;
672         }
673
674         ret = arch_prepare_kprobe(p);
675         if (ret)
676                 goto out;
677
678         INIT_HLIST_NODE(&p->hlist);
679         hlist_add_head_rcu(&p->hlist,
680                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
681
682         if (kprobe_enabled)
683                 arch_arm_kprobe(p);
684
685 out:
686         mutex_unlock(&kprobe_mutex);
687
688         if (ret && probed_mod)
689                 module_put(probed_mod);
690         return ret;
691 }
692
693 /*
694  * Unregister a kprobe without a scheduler synchronization.
695  */
696 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
697 {
698         struct kprobe *old_p, *list_p;
699
700         old_p = get_kprobe(p->addr);
701         if (unlikely(!old_p))
702                 return -EINVAL;
703
704         if (p != old_p) {
705                 list_for_each_entry_rcu(list_p, &old_p->list, list)
706                         if (list_p == p)
707                         /* kprobe p is a valid probe */
708                                 goto valid_p;
709                 return -EINVAL;
710         }
711 valid_p:
712         if (old_p == p ||
713             (old_p->pre_handler == aggr_pre_handler &&
714              list_is_singular(&old_p->list))) {
715                 /*
716                  * Only probe on the hash list. Disarm only if kprobes are
717                  * enabled - otherwise, the breakpoint would already have
718                  * been removed. We save on flushing icache.
719                  */
720                 if (kprobe_enabled)
721                         arch_disarm_kprobe(p);
722                 hlist_del_rcu(&old_p->hlist);
723         } else {
724                 if (p->break_handler)
725                         old_p->break_handler = NULL;
726                 if (p->post_handler) {
727                         list_for_each_entry_rcu(list_p, &old_p->list, list) {
728                                 if ((list_p != p) && (list_p->post_handler))
729                                         goto noclean;
730                         }
731                         old_p->post_handler = NULL;
732                 }
733 noclean:
734                 list_del_rcu(&p->list);
735         }
736         return 0;
737 }
738
739 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
740 {
741         struct module *mod;
742         struct kprobe *old_p;
743
744         if (p->mod_refcounted) {
745                 /*
746                  * Since we've already incremented refcount,
747                  * we don't need to disable preemption.
748                  */
749                 mod = module_text_address((unsigned long)p->addr);
750                 if (mod)
751                         module_put(mod);
752         }
753
754         if (list_empty(&p->list) || list_is_singular(&p->list)) {
755                 if (!list_empty(&p->list)) {
756                         /* "p" is the last child of an aggr_kprobe */
757                         old_p = list_entry(p->list.next, struct kprobe, list);
758                         list_del(&p->list);
759                         kfree(old_p);
760                 }
761                 arch_remove_kprobe(p);
762         }
763 }
764
765 static int __register_kprobes(struct kprobe **kps, int num,
766         unsigned long called_from)
767 {
768         int i, ret = 0;
769
770         if (num <= 0)
771                 return -EINVAL;
772         for (i = 0; i < num; i++) {
773                 ret = __register_kprobe(kps[i], called_from);
774                 if (ret < 0) {
775                         if (i > 0)
776                                 unregister_kprobes(kps, i);
777                         break;
778                 }
779         }
780         return ret;
781 }
782
783 /*
784  * Registration and unregistration functions for kprobe.
785  */
786 int __kprobes register_kprobe(struct kprobe *p)
787 {
788         return __register_kprobes(&p, 1,
789                                   (unsigned long)__builtin_return_address(0));
790 }
791
792 void __kprobes unregister_kprobe(struct kprobe *p)
793 {
794         unregister_kprobes(&p, 1);
795 }
796
797 int __kprobes register_kprobes(struct kprobe **kps, int num)
798 {
799         return __register_kprobes(kps, num,
800                                   (unsigned long)__builtin_return_address(0));
801 }
802
803 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
804 {
805         int i;
806
807         if (num <= 0)
808                 return;
809         mutex_lock(&kprobe_mutex);
810         for (i = 0; i < num; i++)
811                 if (__unregister_kprobe_top(kps[i]) < 0)
812                         kps[i]->addr = NULL;
813         mutex_unlock(&kprobe_mutex);
814
815         synchronize_sched();
816         for (i = 0; i < num; i++)
817                 if (kps[i]->addr)
818                         __unregister_kprobe_bottom(kps[i]);
819 }
820
821 static struct notifier_block kprobe_exceptions_nb = {
822         .notifier_call = kprobe_exceptions_notify,
823         .priority = 0x7fffffff /* we need to be notified first */
824 };
825
826 unsigned long __weak arch_deref_entry_point(void *entry)
827 {
828         return (unsigned long)entry;
829 }
830
831 static int __register_jprobes(struct jprobe **jps, int num,
832         unsigned long called_from)
833 {
834         struct jprobe *jp;
835         int ret = 0, i;
836
837         if (num <= 0)
838                 return -EINVAL;
839         for (i = 0; i < num; i++) {
840                 unsigned long addr;
841                 jp = jps[i];
842                 addr = arch_deref_entry_point(jp->entry);
843
844                 if (!kernel_text_address(addr))
845                         ret = -EINVAL;
846                 else {
847                         /* Todo: Verify probepoint is a function entry point */
848                         jp->kp.pre_handler = setjmp_pre_handler;
849                         jp->kp.break_handler = longjmp_break_handler;
850                         ret = __register_kprobe(&jp->kp, called_from);
851                 }
852                 if (ret < 0) {
853                         if (i > 0)
854                                 unregister_jprobes(jps, i);
855                         break;
856                 }
857         }
858         return ret;
859 }
860
861 int __kprobes register_jprobe(struct jprobe *jp)
862 {
863         return __register_jprobes(&jp, 1,
864                 (unsigned long)__builtin_return_address(0));
865 }
866
867 void __kprobes unregister_jprobe(struct jprobe *jp)
868 {
869         unregister_jprobes(&jp, 1);
870 }
871
872 int __kprobes register_jprobes(struct jprobe **jps, int num)
873 {
874         return __register_jprobes(jps, num,
875                 (unsigned long)__builtin_return_address(0));
876 }
877
878 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
879 {
880         int i;
881
882         if (num <= 0)
883                 return;
884         mutex_lock(&kprobe_mutex);
885         for (i = 0; i < num; i++)
886                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
887                         jps[i]->kp.addr = NULL;
888         mutex_unlock(&kprobe_mutex);
889
890         synchronize_sched();
891         for (i = 0; i < num; i++) {
892                 if (jps[i]->kp.addr)
893                         __unregister_kprobe_bottom(&jps[i]->kp);
894         }
895 }
896
897 #ifdef CONFIG_KRETPROBES
898 /*
899  * This kprobe pre_handler is registered with every kretprobe. When probe
900  * hits it will set up the return probe.
901  */
902 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
903                                            struct pt_regs *regs)
904 {
905         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
906         unsigned long hash, flags = 0;
907         struct kretprobe_instance *ri;
908
909         /*TODO: consider to only swap the RA after the last pre_handler fired */
910         hash = hash_ptr(current, KPROBE_HASH_BITS);
911         spin_lock_irqsave(&rp->lock, flags);
912         if (!hlist_empty(&rp->free_instances)) {
913                 ri = hlist_entry(rp->free_instances.first,
914                                 struct kretprobe_instance, hlist);
915                 hlist_del(&ri->hlist);
916                 spin_unlock_irqrestore(&rp->lock, flags);
917
918                 ri->rp = rp;
919                 ri->task = current;
920
921                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
922                         spin_unlock_irqrestore(&rp->lock, flags);
923                         return 0;
924                 }
925
926                 arch_prepare_kretprobe(ri, regs);
927
928                 /* XXX(hch): why is there no hlist_move_head? */
929                 INIT_HLIST_NODE(&ri->hlist);
930                 kretprobe_table_lock(hash, &flags);
931                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
932                 kretprobe_table_unlock(hash, &flags);
933         } else {
934                 rp->nmissed++;
935                 spin_unlock_irqrestore(&rp->lock, flags);
936         }
937         return 0;
938 }
939
940 static int __kprobes __register_kretprobe(struct kretprobe *rp,
941                                           unsigned long called_from)
942 {
943         int ret = 0;
944         struct kretprobe_instance *inst;
945         int i;
946         void *addr;
947
948         if (kretprobe_blacklist_size) {
949                 addr = kprobe_addr(&rp->kp);
950                 if (!addr)
951                         return -EINVAL;
952
953                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
954                         if (kretprobe_blacklist[i].addr == addr)
955                                 return -EINVAL;
956                 }
957         }
958
959         rp->kp.pre_handler = pre_handler_kretprobe;
960         rp->kp.post_handler = NULL;
961         rp->kp.fault_handler = NULL;
962         rp->kp.break_handler = NULL;
963
964         /* Pre-allocate memory for max kretprobe instances */
965         if (rp->maxactive <= 0) {
966 #ifdef CONFIG_PREEMPT
967                 rp->maxactive = max(10, 2 * NR_CPUS);
968 #else
969                 rp->maxactive = NR_CPUS;
970 #endif
971         }
972         spin_lock_init(&rp->lock);
973         INIT_HLIST_HEAD(&rp->free_instances);
974         for (i = 0; i < rp->maxactive; i++) {
975                 inst = kmalloc(sizeof(struct kretprobe_instance) +
976                                rp->data_size, GFP_KERNEL);
977                 if (inst == NULL) {
978                         free_rp_inst(rp);
979                         return -ENOMEM;
980                 }
981                 INIT_HLIST_NODE(&inst->hlist);
982                 hlist_add_head(&inst->hlist, &rp->free_instances);
983         }
984
985         rp->nmissed = 0;
986         /* Establish function entry probe point */
987         ret = __register_kprobe(&rp->kp, called_from);
988         if (ret != 0)
989                 free_rp_inst(rp);
990         return ret;
991 }
992
993 static int __register_kretprobes(struct kretprobe **rps, int num,
994         unsigned long called_from)
995 {
996         int ret = 0, i;
997
998         if (num <= 0)
999                 return -EINVAL;
1000         for (i = 0; i < num; i++) {
1001                 ret = __register_kretprobe(rps[i], called_from);
1002                 if (ret < 0) {
1003                         if (i > 0)
1004                                 unregister_kretprobes(rps, i);
1005                         break;
1006                 }
1007         }
1008         return ret;
1009 }
1010
1011 int __kprobes register_kretprobe(struct kretprobe *rp)
1012 {
1013         return __register_kretprobes(&rp, 1,
1014                         (unsigned long)__builtin_return_address(0));
1015 }
1016
1017 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1018 {
1019         unregister_kretprobes(&rp, 1);
1020 }
1021
1022 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1023 {
1024         return __register_kretprobes(rps, num,
1025                         (unsigned long)__builtin_return_address(0));
1026 }
1027
1028 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1029 {
1030         int i;
1031
1032         if (num <= 0)
1033                 return;
1034         mutex_lock(&kprobe_mutex);
1035         for (i = 0; i < num; i++)
1036                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1037                         rps[i]->kp.addr = NULL;
1038         mutex_unlock(&kprobe_mutex);
1039
1040         synchronize_sched();
1041         for (i = 0; i < num; i++) {
1042                 if (rps[i]->kp.addr) {
1043                         __unregister_kprobe_bottom(&rps[i]->kp);
1044                         cleanup_rp_inst(rps[i]);
1045                 }
1046         }
1047 }
1048
1049 #else /* CONFIG_KRETPROBES */
1050 int __kprobes register_kretprobe(struct kretprobe *rp)
1051 {
1052         return -ENOSYS;
1053 }
1054
1055 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1056 {
1057         return -ENOSYS;
1058 }
1059 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1060 {
1061 }
1062
1063 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1064 {
1065 }
1066
1067 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1068                                            struct pt_regs *regs)
1069 {
1070         return 0;
1071 }
1072
1073 #endif /* CONFIG_KRETPROBES */
1074
1075 static int __init init_kprobes(void)
1076 {
1077         int i, err = 0;
1078         unsigned long offset = 0, size = 0;
1079         char *modname, namebuf[128];
1080         const char *symbol_name;
1081         void *addr;
1082         struct kprobe_blackpoint *kb;
1083
1084         /* FIXME allocate the probe table, currently defined statically */
1085         /* initialize all list heads */
1086         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1087                 INIT_HLIST_HEAD(&kprobe_table[i]);
1088                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
1089                 spin_lock_init(&(kretprobe_table_locks[i].lock));
1090         }
1091
1092         /*
1093          * Lookup and populate the kprobe_blacklist.
1094          *
1095          * Unlike the kretprobe blacklist, we'll need to determine
1096          * the range of addresses that belong to the said functions,
1097          * since a kprobe need not necessarily be at the beginning
1098          * of a function.
1099          */
1100         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1101                 kprobe_lookup_name(kb->name, addr);
1102                 if (!addr)
1103                         continue;
1104
1105                 kb->start_addr = (unsigned long)addr;
1106                 symbol_name = kallsyms_lookup(kb->start_addr,
1107                                 &size, &offset, &modname, namebuf);
1108                 if (!symbol_name)
1109                         kb->range = 0;
1110                 else
1111                         kb->range = size;
1112         }
1113
1114         if (kretprobe_blacklist_size) {
1115                 /* lookup the function address from its name */
1116                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1117                         kprobe_lookup_name(kretprobe_blacklist[i].name,
1118                                            kretprobe_blacklist[i].addr);
1119                         if (!kretprobe_blacklist[i].addr)
1120                                 printk("kretprobe: lookup failed: %s\n",
1121                                        kretprobe_blacklist[i].name);
1122                 }
1123         }
1124
1125         /* By default, kprobes are enabled */
1126         kprobe_enabled = true;
1127
1128         err = arch_init_kprobes();
1129         if (!err)
1130                 err = register_die_notifier(&kprobe_exceptions_nb);
1131         kprobes_initialized = (err == 0);
1132
1133         if (!err)
1134                 init_test_probes();
1135         return err;
1136 }
1137
1138 #ifdef CONFIG_DEBUG_FS
1139 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
1140                 const char *sym, int offset,char *modname)
1141 {
1142         char *kprobe_type;
1143
1144         if (p->pre_handler == pre_handler_kretprobe)
1145                 kprobe_type = "r";
1146         else if (p->pre_handler == setjmp_pre_handler)
1147                 kprobe_type = "j";
1148         else
1149                 kprobe_type = "k";
1150         if (sym)
1151                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
1152                         sym, offset, (modname ? modname : " "));
1153         else
1154                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
1155 }
1156
1157 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
1158 {
1159         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
1160 }
1161
1162 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
1163 {
1164         (*pos)++;
1165         if (*pos >= KPROBE_TABLE_SIZE)
1166                 return NULL;
1167         return pos;
1168 }
1169
1170 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
1171 {
1172         /* Nothing to do */
1173 }
1174
1175 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
1176 {
1177         struct hlist_head *head;
1178         struct hlist_node *node;
1179         struct kprobe *p, *kp;
1180         const char *sym = NULL;
1181         unsigned int i = *(loff_t *) v;
1182         unsigned long offset = 0;
1183         char *modname, namebuf[128];
1184
1185         head = &kprobe_table[i];
1186         preempt_disable();
1187         hlist_for_each_entry_rcu(p, node, head, hlist) {
1188                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
1189                                         &offset, &modname, namebuf);
1190                 if (p->pre_handler == aggr_pre_handler) {
1191                         list_for_each_entry_rcu(kp, &p->list, list)
1192                                 report_probe(pi, kp, sym, offset, modname);
1193                 } else
1194                         report_probe(pi, p, sym, offset, modname);
1195         }
1196         preempt_enable();
1197         return 0;
1198 }
1199
1200 static struct seq_operations kprobes_seq_ops = {
1201         .start = kprobe_seq_start,
1202         .next  = kprobe_seq_next,
1203         .stop  = kprobe_seq_stop,
1204         .show  = show_kprobe_addr
1205 };
1206
1207 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
1208 {
1209         return seq_open(filp, &kprobes_seq_ops);
1210 }
1211
1212 static struct file_operations debugfs_kprobes_operations = {
1213         .open           = kprobes_open,
1214         .read           = seq_read,
1215         .llseek         = seq_lseek,
1216         .release        = seq_release,
1217 };
1218
1219 static void __kprobes enable_all_kprobes(void)
1220 {
1221         struct hlist_head *head;
1222         struct hlist_node *node;
1223         struct kprobe *p;
1224         unsigned int i;
1225
1226         mutex_lock(&kprobe_mutex);
1227
1228         /* If kprobes are already enabled, just return */
1229         if (kprobe_enabled)
1230                 goto already_enabled;
1231
1232         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1233                 head = &kprobe_table[i];
1234                 hlist_for_each_entry_rcu(p, node, head, hlist)
1235                         arch_arm_kprobe(p);
1236         }
1237
1238         kprobe_enabled = true;
1239         printk(KERN_INFO "Kprobes globally enabled\n");
1240
1241 already_enabled:
1242         mutex_unlock(&kprobe_mutex);
1243         return;
1244 }
1245
1246 static void __kprobes disable_all_kprobes(void)
1247 {
1248         struct hlist_head *head;
1249         struct hlist_node *node;
1250         struct kprobe *p;
1251         unsigned int i;
1252
1253         mutex_lock(&kprobe_mutex);
1254
1255         /* If kprobes are already disabled, just return */
1256         if (!kprobe_enabled)
1257                 goto already_disabled;
1258
1259         kprobe_enabled = false;
1260         printk(KERN_INFO "Kprobes globally disabled\n");
1261         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1262                 head = &kprobe_table[i];
1263                 hlist_for_each_entry_rcu(p, node, head, hlist) {
1264                         if (!arch_trampoline_kprobe(p))
1265                                 arch_disarm_kprobe(p);
1266                 }
1267         }
1268
1269         mutex_unlock(&kprobe_mutex);
1270         /* Allow all currently running kprobes to complete */
1271         synchronize_sched();
1272         return;
1273
1274 already_disabled:
1275         mutex_unlock(&kprobe_mutex);
1276         return;
1277 }
1278
1279 /*
1280  * XXX: The debugfs bool file interface doesn't allow for callbacks
1281  * when the bool state is switched. We can reuse that facility when
1282  * available
1283  */
1284 static ssize_t read_enabled_file_bool(struct file *file,
1285                char __user *user_buf, size_t count, loff_t *ppos)
1286 {
1287         char buf[3];
1288
1289         if (kprobe_enabled)
1290                 buf[0] = '1';
1291         else
1292                 buf[0] = '0';
1293         buf[1] = '\n';
1294         buf[2] = 0x00;
1295         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
1296 }
1297
1298 static ssize_t write_enabled_file_bool(struct file *file,
1299                const char __user *user_buf, size_t count, loff_t *ppos)
1300 {
1301         char buf[32];
1302         int buf_size;
1303
1304         buf_size = min(count, (sizeof(buf)-1));
1305         if (copy_from_user(buf, user_buf, buf_size))
1306                 return -EFAULT;
1307
1308         switch (buf[0]) {
1309         case 'y':
1310         case 'Y':
1311         case '1':
1312                 enable_all_kprobes();
1313                 break;
1314         case 'n':
1315         case 'N':
1316         case '0':
1317                 disable_all_kprobes();
1318                 break;
1319         }
1320
1321         return count;
1322 }
1323
1324 static struct file_operations fops_kp = {
1325         .read =         read_enabled_file_bool,
1326         .write =        write_enabled_file_bool,
1327 };
1328
1329 static int __kprobes debugfs_kprobe_init(void)
1330 {
1331         struct dentry *dir, *file;
1332         unsigned int value = 1;
1333
1334         dir = debugfs_create_dir("kprobes", NULL);
1335         if (!dir)
1336                 return -ENOMEM;
1337
1338         file = debugfs_create_file("list", 0444, dir, NULL,
1339                                 &debugfs_kprobes_operations);
1340         if (!file) {
1341                 debugfs_remove(dir);
1342                 return -ENOMEM;
1343         }
1344
1345         file = debugfs_create_file("enabled", 0600, dir,
1346                                         &value, &fops_kp);
1347         if (!file) {
1348                 debugfs_remove(dir);
1349                 return -ENOMEM;
1350         }
1351
1352         return 0;
1353 }
1354
1355 late_initcall(debugfs_kprobe_init);
1356 #endif /* CONFIG_DEBUG_FS */
1357
1358 module_init(init_kprobes);
1359
1360 EXPORT_SYMBOL_GPL(register_kprobe);
1361 EXPORT_SYMBOL_GPL(unregister_kprobe);
1362 EXPORT_SYMBOL_GPL(register_kprobes);
1363 EXPORT_SYMBOL_GPL(unregister_kprobes);
1364 EXPORT_SYMBOL_GPL(register_jprobe);
1365 EXPORT_SYMBOL_GPL(unregister_jprobe);
1366 EXPORT_SYMBOL_GPL(register_jprobes);
1367 EXPORT_SYMBOL_GPL(unregister_jprobes);
1368 EXPORT_SYMBOL_GPL(jprobe_return);
1369 EXPORT_SYMBOL_GPL(register_kretprobe);
1370 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1371 EXPORT_SYMBOL_GPL(register_kretprobes);
1372 EXPORT_SYMBOL_GPL(unregister_kretprobes);