~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/kernel/kprobes.c

Version: ~ [ linux-4.15-rc3 ] ~ [ linux-4.14.5 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.68 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.105 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.47 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.87 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.51 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.96 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/export.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/sysctl.h>
 46 #include <linux/kdebug.h>
 47 #include <linux/memory.h>
 48 #include <linux/ftrace.h>
 49 #include <linux/cpu.h>
 50 #include <linux/jump_label.h>
 51 
 52 #include <asm/sections.h>
 53 #include <asm/cacheflush.h>
 54 #include <asm/errno.h>
 55 #include <linux/uaccess.h>
 56 
 57 #define KPROBE_HASH_BITS 6
 58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
 59 
 60 
 61 static int kprobes_initialized;
 62 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
 63 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
 64 
 65 /* NOTE: change this value only with kprobe_mutex held */
 66 static bool kprobes_all_disarmed;
 67 
 68 /* This protects kprobe_table and optimizing_list */
 69 static DEFINE_MUTEX(kprobe_mutex);
 70 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
 71 static struct {
 72         raw_spinlock_t lock ____cacheline_aligned_in_smp;
 73 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
 74 
 75 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
 76                                         unsigned int __unused)
 77 {
 78         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
 79 }
 80 
 81 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
 82 {
 83         return &(kretprobe_table_locks[hash].lock);
 84 }
 85 
 86 /* Blacklist -- list of struct kprobe_blacklist_entry */
 87 static LIST_HEAD(kprobe_blacklist);
 88 
 89 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
 90 /*
 91  * kprobe->ainsn.insn points to the copy of the instruction to be
 92  * single-stepped. x86_64, POWER4 and above have no-exec support and
 93  * stepping on the instruction on a vmalloced/kmalloced/data page
 94  * is a recipe for disaster
 95  */
 96 struct kprobe_insn_page {
 97         struct list_head list;
 98         kprobe_opcode_t *insns;         /* Page of instruction slots */
 99         struct kprobe_insn_cache *cache;
100         int nused;
101         int ngarbage;
102         char slot_used[];
103 };
104 
105 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
106         (offsetof(struct kprobe_insn_page, slot_used) + \
107          (sizeof(char) * (slots)))
108 
109 static int slots_per_page(struct kprobe_insn_cache *c)
110 {
111         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
112 }
113 
114 enum kprobe_slot_state {
115         SLOT_CLEAN = 0,
116         SLOT_DIRTY = 1,
117         SLOT_USED = 2,
118 };
119 
120 void __weak *alloc_insn_page(void)
121 {
122         return module_alloc(PAGE_SIZE);
123 }
124 
125 void __weak free_insn_page(void *page)
126 {
127         module_memfree(page);
128 }
129 
130 struct kprobe_insn_cache kprobe_insn_slots = {
131         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
132         .alloc = alloc_insn_page,
133         .free = free_insn_page,
134         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
135         .insn_size = MAX_INSN_SIZE,
136         .nr_garbage = 0,
137 };
138 static int collect_garbage_slots(struct kprobe_insn_cache *c);
139 
140 /**
141  * __get_insn_slot() - Find a slot on an executable page for an instruction.
142  * We allocate an executable page if there's no room on existing ones.
143  */
144 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
145 {
146         struct kprobe_insn_page *kip;
147         kprobe_opcode_t *slot = NULL;
148 
149         /* Since the slot array is not protected by rcu, we need a mutex */
150         mutex_lock(&c->mutex);
151  retry:
152         rcu_read_lock();
153         list_for_each_entry_rcu(kip, &c->pages, list) {
154                 if (kip->nused < slots_per_page(c)) {
155                         int i;
156                         for (i = 0; i < slots_per_page(c); i++) {
157                                 if (kip->slot_used[i] == SLOT_CLEAN) {
158                                         kip->slot_used[i] = SLOT_USED;
159                                         kip->nused++;
160                                         slot = kip->insns + (i * c->insn_size);
161                                         rcu_read_unlock();
162                                         goto out;
163                                 }
164                         }
165                         /* kip->nused is broken. Fix it. */
166                         kip->nused = slots_per_page(c);
167                         WARN_ON(1);
168                 }
169         }
170         rcu_read_unlock();
171 
172         /* If there are any garbage slots, collect it and try again. */
173         if (c->nr_garbage && collect_garbage_slots(c) == 0)
174                 goto retry;
175 
176         /* All out of space.  Need to allocate a new page. */
177         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
178         if (!kip)
179                 goto out;
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 = c->alloc();
187         if (!kip->insns) {
188                 kfree(kip);
189                 goto out;
190         }
191         INIT_LIST_HEAD(&kip->list);
192         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
193         kip->slot_used[0] = SLOT_USED;
194         kip->nused = 1;
195         kip->ngarbage = 0;
196         kip->cache = c;
197         list_add_rcu(&kip->list, &c->pages);
198         slot = kip->insns;
199 out:
200         mutex_unlock(&c->mutex);
201         return slot;
202 }
203 
204 /* Return 1 if all garbages are collected, otherwise 0. */
205 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
206 {
207         kip->slot_used[idx] = SLOT_CLEAN;
208         kip->nused--;
209         if (kip->nused == 0) {
210                 /*
211                  * Page is no longer in use.  Free it unless
212                  * it's the last one.  We keep the last one
213                  * so as not to have to set it up again the
214                  * next time somebody inserts a probe.
215                  */
216                 if (!list_is_singular(&kip->list)) {
217                         list_del_rcu(&kip->list);
218                         synchronize_rcu();
219                         kip->cache->free(kip->insns);
220                         kfree(kip);
221                 }
222                 return 1;
223         }
224         return 0;
225 }
226 
227 static int collect_garbage_slots(struct kprobe_insn_cache *c)
228 {
229         struct kprobe_insn_page *kip, *next;
230 
231         /* Ensure no-one is interrupted on the garbages */
232         synchronize_sched();
233 
234         list_for_each_entry_safe(kip, next, &c->pages, list) {
235                 int i;
236                 if (kip->ngarbage == 0)
237                         continue;
238                 kip->ngarbage = 0;      /* we will collect all garbages */
239                 for (i = 0; i < slots_per_page(c); i++) {
240                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
241                                 break;
242                 }
243         }
244         c->nr_garbage = 0;
245         return 0;
246 }
247 
248 void __free_insn_slot(struct kprobe_insn_cache *c,
249                       kprobe_opcode_t *slot, int dirty)
250 {
251         struct kprobe_insn_page *kip;
252         long idx;
253 
254         mutex_lock(&c->mutex);
255         rcu_read_lock();
256         list_for_each_entry_rcu(kip, &c->pages, list) {
257                 idx = ((long)slot - (long)kip->insns) /
258                         (c->insn_size * sizeof(kprobe_opcode_t));
259                 if (idx >= 0 && idx < slots_per_page(c))
260                         goto out;
261         }
262         /* Could not find this slot. */
263         WARN_ON(1);
264         kip = NULL;
265 out:
266         rcu_read_unlock();
267         /* Mark and sweep: this may sleep */
268         if (kip) {
269                 /* Check double free */
270                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
271                 if (dirty) {
272                         kip->slot_used[idx] = SLOT_DIRTY;
273                         kip->ngarbage++;
274                         if (++c->nr_garbage > slots_per_page(c))
275                                 collect_garbage_slots(c);
276                 } else {
277                         collect_one_slot(kip, idx);
278                 }
279         }
280         mutex_unlock(&c->mutex);
281 }
282 
283 /*
284  * Check given address is on the page of kprobe instruction slots.
285  * This will be used for checking whether the address on a stack
286  * is on a text area or not.
287  */
288 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
289 {
290         struct kprobe_insn_page *kip;
291         bool ret = false;
292 
293         rcu_read_lock();
294         list_for_each_entry_rcu(kip, &c->pages, list) {
295                 if (addr >= (unsigned long)kip->insns &&
296                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
297                         ret = true;
298                         break;
299                 }
300         }
301         rcu_read_unlock();
302 
303         return ret;
304 }
305 
306 #ifdef CONFIG_OPTPROBES
307 /* For optimized_kprobe buffer */
308 struct kprobe_insn_cache kprobe_optinsn_slots = {
309         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
310         .alloc = alloc_insn_page,
311         .free = free_insn_page,
312         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
313         /* .insn_size is initialized later */
314         .nr_garbage = 0,
315 };
316 #endif
317 #endif
318 
319 /* We have preemption disabled.. so it is safe to use __ versions */
320 static inline void set_kprobe_instance(struct kprobe *kp)
321 {
322         __this_cpu_write(kprobe_instance, kp);
323 }
324 
325 static inline void reset_kprobe_instance(void)
326 {
327         __this_cpu_write(kprobe_instance, NULL);
328 }
329 
330 /*
331  * This routine is called either:
332  *      - under the kprobe_mutex - during kprobe_[un]register()
333  *                              OR
334  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
335  */
336 struct kprobe *get_kprobe(void *addr)
337 {
338         struct hlist_head *head;
339         struct kprobe *p;
340 
341         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
342         hlist_for_each_entry_rcu(p, head, hlist) {
343                 if (p->addr == addr)
344                         return p;
345         }
346 
347         return NULL;
348 }
349 NOKPROBE_SYMBOL(get_kprobe);
350 
351 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
352 
353 /* Return true if the kprobe is an aggregator */
354 static inline int kprobe_aggrprobe(struct kprobe *p)
355 {
356         return p->pre_handler == aggr_pre_handler;
357 }
358 
359 /* Return true(!0) if the kprobe is unused */
360 static inline int kprobe_unused(struct kprobe *p)
361 {
362         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
363                list_empty(&p->list);
364 }
365 
366 /*
367  * Keep all fields in the kprobe consistent
368  */
369 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
370 {
371         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
372         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
373 }
374 
375 #ifdef CONFIG_OPTPROBES
376 /* NOTE: change this value only with kprobe_mutex held */
377 static bool kprobes_allow_optimization;
378 
379 /*
380  * Call all pre_handler on the list, but ignores its return value.
381  * This must be called from arch-dep optimized caller.
382  */
383 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
384 {
385         struct kprobe *kp;
386 
387         list_for_each_entry_rcu(kp, &p->list, list) {
388                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
389                         set_kprobe_instance(kp);
390                         kp->pre_handler(kp, regs);
391                 }
392                 reset_kprobe_instance();
393         }
394 }
395 NOKPROBE_SYMBOL(opt_pre_handler);
396 
397 /* Free optimized instructions and optimized_kprobe */
398 static void free_aggr_kprobe(struct kprobe *p)
399 {
400         struct optimized_kprobe *op;
401 
402         op = container_of(p, struct optimized_kprobe, kp);
403         arch_remove_optimized_kprobe(op);
404         arch_remove_kprobe(p);
405         kfree(op);
406 }
407 
408 /* Return true(!0) if the kprobe is ready for optimization. */
409 static inline int kprobe_optready(struct kprobe *p)
410 {
411         struct optimized_kprobe *op;
412 
413         if (kprobe_aggrprobe(p)) {
414                 op = container_of(p, struct optimized_kprobe, kp);
415                 return arch_prepared_optinsn(&op->optinsn);
416         }
417 
418         return 0;
419 }
420 
421 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
422 static inline int kprobe_disarmed(struct kprobe *p)
423 {
424         struct optimized_kprobe *op;
425 
426         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
427         if (!kprobe_aggrprobe(p))
428                 return kprobe_disabled(p);
429 
430         op = container_of(p, struct optimized_kprobe, kp);
431 
432         return kprobe_disabled(p) && list_empty(&op->list);
433 }
434 
435 /* Return true(!0) if the probe is queued on (un)optimizing lists */
436 static int kprobe_queued(struct kprobe *p)
437 {
438         struct optimized_kprobe *op;
439 
440         if (kprobe_aggrprobe(p)) {
441                 op = container_of(p, struct optimized_kprobe, kp);
442                 if (!list_empty(&op->list))
443                         return 1;
444         }
445         return 0;
446 }
447 
448 /*
449  * Return an optimized kprobe whose optimizing code replaces
450  * instructions including addr (exclude breakpoint).
451  */
452 static struct kprobe *get_optimized_kprobe(unsigned long addr)
453 {
454         int i;
455         struct kprobe *p = NULL;
456         struct optimized_kprobe *op;
457 
458         /* Don't check i == 0, since that is a breakpoint case. */
459         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
460                 p = get_kprobe((void *)(addr - i));
461 
462         if (p && kprobe_optready(p)) {
463                 op = container_of(p, struct optimized_kprobe, kp);
464                 if (arch_within_optimized_kprobe(op, addr))
465                         return p;
466         }
467 
468         return NULL;
469 }
470 
471 /* Optimization staging list, protected by kprobe_mutex */
472 static LIST_HEAD(optimizing_list);
473 static LIST_HEAD(unoptimizing_list);
474 static LIST_HEAD(freeing_list);
475 
476 static void kprobe_optimizer(struct work_struct *work);
477 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
478 #define OPTIMIZE_DELAY 5
479 
480 /*
481  * Optimize (replace a breakpoint with a jump) kprobes listed on
482  * optimizing_list.
483  */
484 static void do_optimize_kprobes(void)
485 {
486         /*
487          * The optimization/unoptimization refers online_cpus via
488          * stop_machine() and cpu-hotplug modifies online_cpus.
489          * And same time, text_mutex will be held in cpu-hotplug and here.
490          * This combination can cause a deadlock (cpu-hotplug try to lock
491          * text_mutex but stop_machine can not be done because online_cpus
492          * has been changed)
493          * To avoid this deadlock, caller must have locked cpu hotplug
494          * for preventing cpu-hotplug outside of text_mutex locking.
495          */
496         lockdep_assert_cpus_held();
497 
498         /* Optimization never be done when disarmed */
499         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
500             list_empty(&optimizing_list))
501                 return;
502 
503         mutex_lock(&text_mutex);
504         arch_optimize_kprobes(&optimizing_list);
505         mutex_unlock(&text_mutex);
506 }
507 
508 /*
509  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
510  * if need) kprobes listed on unoptimizing_list.
511  */
512 static void do_unoptimize_kprobes(void)
513 {
514         struct optimized_kprobe *op, *tmp;
515 
516         /* See comment in do_optimize_kprobes() */
517         lockdep_assert_cpus_held();
518 
519         /* Unoptimization must be done anytime */
520         if (list_empty(&unoptimizing_list))
521                 return;
522 
523         mutex_lock(&text_mutex);
524         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
525         /* Loop free_list for disarming */
526         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
527                 /* Disarm probes if marked disabled */
528                 if (kprobe_disabled(&op->kp))
529                         arch_disarm_kprobe(&op->kp);
530                 if (kprobe_unused(&op->kp)) {
531                         /*
532                          * Remove unused probes from hash list. After waiting
533                          * for synchronization, these probes are reclaimed.
534                          * (reclaiming is done by do_free_cleaned_kprobes.)
535                          */
536                         hlist_del_rcu(&op->kp.hlist);
537                 } else
538                         list_del_init(&op->list);
539         }
540         mutex_unlock(&text_mutex);
541 }
542 
543 /* Reclaim all kprobes on the free_list */
544 static void do_free_cleaned_kprobes(void)
545 {
546         struct optimized_kprobe *op, *tmp;
547 
548         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
549                 BUG_ON(!kprobe_unused(&op->kp));
550                 list_del_init(&op->list);
551                 free_aggr_kprobe(&op->kp);
552         }
553 }
554 
555 /* Start optimizer after OPTIMIZE_DELAY passed */
556 static void kick_kprobe_optimizer(void)
557 {
558         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
559 }
560 
561 /* Kprobe jump optimizer */
562 static void kprobe_optimizer(struct work_struct *work)
563 {
564         mutex_lock(&kprobe_mutex);
565         cpus_read_lock();
566         /* Lock modules while optimizing kprobes */
567         mutex_lock(&module_mutex);
568 
569         /*
570          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
571          * kprobes before waiting for quiesence period.
572          */
573         do_unoptimize_kprobes();
574 
575         /*
576          * Step 2: Wait for quiesence period to ensure all potentially
577          * preempted tasks to have normally scheduled. Because optprobe
578          * may modify multiple instructions, there is a chance that Nth
579          * instruction is preempted. In that case, such tasks can return
580          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
581          * Note that on non-preemptive kernel, this is transparently converted
582          * to synchronoze_sched() to wait for all interrupts to have completed.
583          */
584         synchronize_rcu_tasks();
585 
586         /* Step 3: Optimize kprobes after quiesence period */
587         do_optimize_kprobes();
588 
589         /* Step 4: Free cleaned kprobes after quiesence period */
590         do_free_cleaned_kprobes();
591 
592         mutex_unlock(&module_mutex);
593         cpus_read_unlock();
594         mutex_unlock(&kprobe_mutex);
595 
596         /* Step 5: Kick optimizer again if needed */
597         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
598                 kick_kprobe_optimizer();
599 }
600 
601 /* Wait for completing optimization and unoptimization */
602 void wait_for_kprobe_optimizer(void)
603 {
604         mutex_lock(&kprobe_mutex);
605 
606         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
607                 mutex_unlock(&kprobe_mutex);
608 
609                 /* this will also make optimizing_work execute immmediately */
610                 flush_delayed_work(&optimizing_work);
611                 /* @optimizing_work might not have been queued yet, relax */
612                 cpu_relax();
613 
614                 mutex_lock(&kprobe_mutex);
615         }
616 
617         mutex_unlock(&kprobe_mutex);
618 }
619 
620 /* Optimize kprobe if p is ready to be optimized */
621 static void optimize_kprobe(struct kprobe *p)
622 {
623         struct optimized_kprobe *op;
624 
625         /* Check if the kprobe is disabled or not ready for optimization. */
626         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
627             (kprobe_disabled(p) || kprobes_all_disarmed))
628                 return;
629 
630         /* Both of break_handler and post_handler are not supported. */
631         if (p->break_handler || p->post_handler)
632                 return;
633 
634         op = container_of(p, struct optimized_kprobe, kp);
635 
636         /* Check there is no other kprobes at the optimized instructions */
637         if (arch_check_optimized_kprobe(op) < 0)
638                 return;
639 
640         /* Check if it is already optimized. */
641         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
642                 return;
643         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
644 
645         if (!list_empty(&op->list))
646                 /* This is under unoptimizing. Just dequeue the probe */
647                 list_del_init(&op->list);
648         else {
649                 list_add(&op->list, &optimizing_list);
650                 kick_kprobe_optimizer();
651         }
652 }
653 
654 /* Short cut to direct unoptimizing */
655 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
656 {
657         lockdep_assert_cpus_held();
658         arch_unoptimize_kprobe(op);
659         if (kprobe_disabled(&op->kp))
660                 arch_disarm_kprobe(&op->kp);
661 }
662 
663 /* Unoptimize a kprobe if p is optimized */
664 static void unoptimize_kprobe(struct kprobe *p, bool force)
665 {
666         struct optimized_kprobe *op;
667 
668         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
669                 return; /* This is not an optprobe nor optimized */
670 
671         op = container_of(p, struct optimized_kprobe, kp);
672         if (!kprobe_optimized(p)) {
673                 /* Unoptimized or unoptimizing case */
674                 if (force && !list_empty(&op->list)) {
675                         /*
676                          * Only if this is unoptimizing kprobe and forced,
677                          * forcibly unoptimize it. (No need to unoptimize
678                          * unoptimized kprobe again :)
679                          */
680                         list_del_init(&op->list);
681                         force_unoptimize_kprobe(op);
682                 }
683                 return;
684         }
685 
686         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
687         if (!list_empty(&op->list)) {
688                 /* Dequeue from the optimization queue */
689                 list_del_init(&op->list);
690                 return;
691         }
692         /* Optimized kprobe case */
693         if (force)
694                 /* Forcibly update the code: this is a special case */
695                 force_unoptimize_kprobe(op);
696         else {
697                 list_add(&op->list, &unoptimizing_list);
698                 kick_kprobe_optimizer();
699         }
700 }
701 
702 /* Cancel unoptimizing for reusing */
703 static void reuse_unused_kprobe(struct kprobe *ap)
704 {
705         struct optimized_kprobe *op;
706 
707         BUG_ON(!kprobe_unused(ap));
708         /*
709          * Unused kprobe MUST be on the way of delayed unoptimizing (means
710          * there is still a relative jump) and disabled.
711          */
712         op = container_of(ap, struct optimized_kprobe, kp);
713         if (unlikely(list_empty(&op->list)))
714                 printk(KERN_WARNING "Warning: found a stray unused "
715                         "aggrprobe@%p\n", ap->addr);
716         /* Enable the probe again */
717         ap->flags &= ~KPROBE_FLAG_DISABLED;
718         /* Optimize it again (remove from op->list) */
719         BUG_ON(!kprobe_optready(ap));
720         optimize_kprobe(ap);
721 }
722 
723 /* Remove optimized instructions */
724 static void kill_optimized_kprobe(struct kprobe *p)
725 {
726         struct optimized_kprobe *op;
727 
728         op = container_of(p, struct optimized_kprobe, kp);
729         if (!list_empty(&op->list))
730                 /* Dequeue from the (un)optimization queue */
731                 list_del_init(&op->list);
732         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
733 
734         if (kprobe_unused(p)) {
735                 /* Enqueue if it is unused */
736                 list_add(&op->list, &freeing_list);
737                 /*
738                  * Remove unused probes from the hash list. After waiting
739                  * for synchronization, this probe is reclaimed.
740                  * (reclaiming is done by do_free_cleaned_kprobes().)
741                  */
742                 hlist_del_rcu(&op->kp.hlist);
743         }
744 
745         /* Don't touch the code, because it is already freed. */
746         arch_remove_optimized_kprobe(op);
747 }
748 
749 static inline
750 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
751 {
752         if (!kprobe_ftrace(p))
753                 arch_prepare_optimized_kprobe(op, p);
754 }
755 
756 /* Try to prepare optimized instructions */
757 static void prepare_optimized_kprobe(struct kprobe *p)
758 {
759         struct optimized_kprobe *op;
760 
761         op = container_of(p, struct optimized_kprobe, kp);
762         __prepare_optimized_kprobe(op, p);
763 }
764 
765 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
766 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
767 {
768         struct optimized_kprobe *op;
769 
770         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
771         if (!op)
772                 return NULL;
773 
774         INIT_LIST_HEAD(&op->list);
775         op->kp.addr = p->addr;
776         __prepare_optimized_kprobe(op, p);
777 
778         return &op->kp;
779 }
780 
781 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
782 
783 /*
784  * Prepare an optimized_kprobe and optimize it
785  * NOTE: p must be a normal registered kprobe
786  */
787 static void try_to_optimize_kprobe(struct kprobe *p)
788 {
789         struct kprobe *ap;
790         struct optimized_kprobe *op;
791 
792         /* Impossible to optimize ftrace-based kprobe */
793         if (kprobe_ftrace(p))
794                 return;
795 
796         /* For preparing optimization, jump_label_text_reserved() is called */
797         cpus_read_lock();
798         jump_label_lock();
799         mutex_lock(&text_mutex);
800 
801         ap = alloc_aggr_kprobe(p);
802         if (!ap)
803                 goto out;
804 
805         op = container_of(ap, struct optimized_kprobe, kp);
806         if (!arch_prepared_optinsn(&op->optinsn)) {
807                 /* If failed to setup optimizing, fallback to kprobe */
808                 arch_remove_optimized_kprobe(op);
809                 kfree(op);
810                 goto out;
811         }
812 
813         init_aggr_kprobe(ap, p);
814         optimize_kprobe(ap);    /* This just kicks optimizer thread */
815 
816 out:
817         mutex_unlock(&text_mutex);
818         jump_label_unlock();
819         cpus_read_unlock();
820 }
821 
822 #ifdef CONFIG_SYSCTL
823 static void optimize_all_kprobes(void)
824 {
825         struct hlist_head *head;
826         struct kprobe *p;
827         unsigned int i;
828 
829         mutex_lock(&kprobe_mutex);
830         /* If optimization is already allowed, just return */
831         if (kprobes_allow_optimization)
832                 goto out;
833 
834         cpus_read_lock();
835         kprobes_allow_optimization = true;
836         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
837                 head = &kprobe_table[i];
838                 hlist_for_each_entry_rcu(p, head, hlist)
839                         if (!kprobe_disabled(p))
840                                 optimize_kprobe(p);
841         }
842         cpus_read_unlock();
843         printk(KERN_INFO "Kprobes globally optimized\n");
844 out:
845         mutex_unlock(&kprobe_mutex);
846 }
847 
848 static void unoptimize_all_kprobes(void)
849 {
850         struct hlist_head *head;
851         struct kprobe *p;
852         unsigned int i;
853 
854         mutex_lock(&kprobe_mutex);
855         /* If optimization is already prohibited, just return */
856         if (!kprobes_allow_optimization) {
857                 mutex_unlock(&kprobe_mutex);
858                 return;
859         }
860 
861         cpus_read_lock();
862         kprobes_allow_optimization = false;
863         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
864                 head = &kprobe_table[i];
865                 hlist_for_each_entry_rcu(p, head, hlist) {
866                         if (!kprobe_disabled(p))
867                                 unoptimize_kprobe(p, false);
868                 }
869         }
870         cpus_read_unlock();
871         mutex_unlock(&kprobe_mutex);
872 
873         /* Wait for unoptimizing completion */
874         wait_for_kprobe_optimizer();
875         printk(KERN_INFO "Kprobes globally unoptimized\n");
876 }
877 
878 static DEFINE_MUTEX(kprobe_sysctl_mutex);
879 int sysctl_kprobes_optimization;
880 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
881                                       void __user *buffer, size_t *length,
882                                       loff_t *ppos)
883 {
884         int ret;
885 
886         mutex_lock(&kprobe_sysctl_mutex);
887         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
888         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
889 
890         if (sysctl_kprobes_optimization)
891                 optimize_all_kprobes();
892         else
893                 unoptimize_all_kprobes();
894         mutex_unlock(&kprobe_sysctl_mutex);
895 
896         return ret;
897 }
898 #endif /* CONFIG_SYSCTL */
899 
900 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
901 static void __arm_kprobe(struct kprobe *p)
902 {
903         struct kprobe *_p;
904 
905         /* Check collision with other optimized kprobes */
906         _p = get_optimized_kprobe((unsigned long)p->addr);
907         if (unlikely(_p))
908                 /* Fallback to unoptimized kprobe */
909                 unoptimize_kprobe(_p, true);
910 
911         arch_arm_kprobe(p);
912         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
913 }
914 
915 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
916 static void __disarm_kprobe(struct kprobe *p, bool reopt)
917 {
918         struct kprobe *_p;
919 
920         /* Try to unoptimize */
921         unoptimize_kprobe(p, kprobes_all_disarmed);
922 
923         if (!kprobe_queued(p)) {
924                 arch_disarm_kprobe(p);
925                 /* If another kprobe was blocked, optimize it. */
926                 _p = get_optimized_kprobe((unsigned long)p->addr);
927                 if (unlikely(_p) && reopt)
928                         optimize_kprobe(_p);
929         }
930         /* TODO: reoptimize others after unoptimized this probe */
931 }
932 
933 #else /* !CONFIG_OPTPROBES */
934 
935 #define optimize_kprobe(p)                      do {} while (0)
936 #define unoptimize_kprobe(p, f)                 do {} while (0)
937 #define kill_optimized_kprobe(p)                do {} while (0)
938 #define prepare_optimized_kprobe(p)             do {} while (0)
939 #define try_to_optimize_kprobe(p)               do {} while (0)
940 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
941 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
942 #define kprobe_disarmed(p)                      kprobe_disabled(p)
943 #define wait_for_kprobe_optimizer()             do {} while (0)
944 
945 /* There should be no unused kprobes can be reused without optimization */
946 static void reuse_unused_kprobe(struct kprobe *ap)
947 {
948         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
949         BUG_ON(kprobe_unused(ap));
950 }
951 
952 static void free_aggr_kprobe(struct kprobe *p)
953 {
954         arch_remove_kprobe(p);
955         kfree(p);
956 }
957 
958 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
959 {
960         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
961 }
962 #endif /* CONFIG_OPTPROBES */
963 
964 #ifdef CONFIG_KPROBES_ON_FTRACE
965 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
966         .func = kprobe_ftrace_handler,
967         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
968 };
969 static int kprobe_ftrace_enabled;
970 
971 /* Must ensure p->addr is really on ftrace */
972 static int prepare_kprobe(struct kprobe *p)
973 {
974         if (!kprobe_ftrace(p))
975                 return arch_prepare_kprobe(p);
976 
977         return arch_prepare_kprobe_ftrace(p);
978 }
979 
980 /* Caller must lock kprobe_mutex */
981 static void arm_kprobe_ftrace(struct kprobe *p)
982 {
983         int ret;
984 
985         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
986                                    (unsigned long)p->addr, 0, 0);
987         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
988         kprobe_ftrace_enabled++;
989         if (kprobe_ftrace_enabled == 1) {
990                 ret = register_ftrace_function(&kprobe_ftrace_ops);
991                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
992         }
993 }
994 
995 /* Caller must lock kprobe_mutex */
996 static void disarm_kprobe_ftrace(struct kprobe *p)
997 {
998         int ret;
999 
1000         kprobe_ftrace_enabled--;
1001         if (kprobe_ftrace_enabled == 0) {
1002                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
1003                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
1004         }
1005         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
1006                            (unsigned long)p->addr, 1, 0);
1007         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
1008 }
1009 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1010 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
1011 #define arm_kprobe_ftrace(p)    do {} while (0)
1012 #define disarm_kprobe_ftrace(p) do {} while (0)
1013 #endif
1014 
1015 /* Arm a kprobe with text_mutex */
1016 static void arm_kprobe(struct kprobe *kp)
1017 {
1018         if (unlikely(kprobe_ftrace(kp))) {
1019                 arm_kprobe_ftrace(kp);
1020                 return;
1021         }
1022         cpus_read_lock();
1023         mutex_lock(&text_mutex);
1024         __arm_kprobe(kp);
1025         mutex_unlock(&text_mutex);
1026         cpus_read_unlock();
1027 }
1028 
1029 /* Disarm a kprobe with text_mutex */
1030 static void disarm_kprobe(struct kprobe *kp, bool reopt)
1031 {
1032         if (unlikely(kprobe_ftrace(kp))) {
1033                 disarm_kprobe_ftrace(kp);
1034                 return;
1035         }
1036 
1037         cpus_read_lock();
1038         mutex_lock(&text_mutex);
1039         __disarm_kprobe(kp, reopt);
1040         mutex_unlock(&text_mutex);
1041         cpus_read_unlock();
1042 }
1043 
1044 /*
1045  * Aggregate handlers for multiple kprobes support - these handlers
1046  * take care of invoking the individual kprobe handlers on p->list
1047  */
1048 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1049 {
1050         struct kprobe *kp;
1051 
1052         list_for_each_entry_rcu(kp, &p->list, list) {
1053                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1054                         set_kprobe_instance(kp);
1055                         if (kp->pre_handler(kp, regs))
1056                                 return 1;
1057                 }
1058                 reset_kprobe_instance();
1059         }
1060         return 0;
1061 }
1062 NOKPROBE_SYMBOL(aggr_pre_handler);
1063 
1064 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1065                               unsigned long flags)
1066 {
1067         struct kprobe *kp;
1068 
1069         list_for_each_entry_rcu(kp, &p->list, list) {
1070                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1071                         set_kprobe_instance(kp);
1072                         kp->post_handler(kp, regs, flags);
1073                         reset_kprobe_instance();
1074                 }
1075         }
1076 }
1077 NOKPROBE_SYMBOL(aggr_post_handler);
1078 
1079 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1080                               int trapnr)
1081 {
1082         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1083 
1084         /*
1085          * if we faulted "during" the execution of a user specified
1086          * probe handler, invoke just that probe's fault handler
1087          */
1088         if (cur && cur->fault_handler) {
1089                 if (cur->fault_handler(cur, regs, trapnr))
1090                         return 1;
1091         }
1092         return 0;
1093 }
1094 NOKPROBE_SYMBOL(aggr_fault_handler);
1095 
1096 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1097 {
1098         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1099         int ret = 0;
1100 
1101         if (cur && cur->break_handler) {
1102                 if (cur->break_handler(cur, regs))
1103                         ret = 1;
1104         }
1105         reset_kprobe_instance();
1106         return ret;
1107 }
1108 NOKPROBE_SYMBOL(aggr_break_handler);
1109 
1110 /* Walks the list and increments nmissed count for multiprobe case */
1111 void kprobes_inc_nmissed_count(struct kprobe *p)
1112 {
1113         struct kprobe *kp;
1114         if (!kprobe_aggrprobe(p)) {
1115                 p->nmissed++;
1116         } else {
1117                 list_for_each_entry_rcu(kp, &p->list, list)
1118                         kp->nmissed++;
1119         }
1120         return;
1121 }
1122 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1123 
1124 void recycle_rp_inst(struct kretprobe_instance *ri,
1125                      struct hlist_head *head)
1126 {
1127         struct kretprobe *rp = ri->rp;
1128 
1129         /* remove rp inst off the rprobe_inst_table */
1130         hlist_del(&ri->hlist);
1131         INIT_HLIST_NODE(&ri->hlist);
1132         if (likely(rp)) {
1133                 raw_spin_lock(&rp->lock);
1134                 hlist_add_head(&ri->hlist, &rp->free_instances);
1135                 raw_spin_unlock(&rp->lock);
1136         } else
1137                 /* Unregistering */
1138                 hlist_add_head(&ri->hlist, head);
1139 }
1140 NOKPROBE_SYMBOL(recycle_rp_inst);
1141 
1142 void kretprobe_hash_lock(struct task_struct *tsk,
1143                          struct hlist_head **head, unsigned long *flags)
1144 __acquires(hlist_lock)
1145 {
1146         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1147         raw_spinlock_t *hlist_lock;
1148 
1149         *head = &kretprobe_inst_table[hash];
1150         hlist_lock = kretprobe_table_lock_ptr(hash);
1151         raw_spin_lock_irqsave(hlist_lock, *flags);
1152 }
1153 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1154 
1155 static void kretprobe_table_lock(unsigned long hash,
1156                                  unsigned long *flags)
1157 __acquires(hlist_lock)
1158 {
1159         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1160         raw_spin_lock_irqsave(hlist_lock, *flags);
1161 }
1162 NOKPROBE_SYMBOL(kretprobe_table_lock);
1163 
1164 void kretprobe_hash_unlock(struct task_struct *tsk,
1165                            unsigned long *flags)
1166 __releases(hlist_lock)
1167 {
1168         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1169         raw_spinlock_t *hlist_lock;
1170 
1171         hlist_lock = kretprobe_table_lock_ptr(hash);
1172         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1173 }
1174 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1175 
1176 static void kretprobe_table_unlock(unsigned long hash,
1177                                    unsigned long *flags)
1178 __releases(hlist_lock)
1179 {
1180         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1181         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1182 }
1183 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1184 
1185 /*
1186  * This function is called from finish_task_switch when task tk becomes dead,
1187  * so that we can recycle any function-return probe instances associated
1188  * with this task. These left over instances represent probed functions
1189  * that have been called but will never return.
1190  */
1191 void kprobe_flush_task(struct task_struct *tk)
1192 {
1193         struct kretprobe_instance *ri;
1194         struct hlist_head *head, empty_rp;
1195         struct hlist_node *tmp;
1196         unsigned long hash, flags = 0;
1197 
1198         if (unlikely(!kprobes_initialized))
1199                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1200                 return;
1201 
1202         INIT_HLIST_HEAD(&empty_rp);
1203         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1204         head = &kretprobe_inst_table[hash];
1205         kretprobe_table_lock(hash, &flags);
1206         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1207                 if (ri->task == tk)
1208                         recycle_rp_inst(ri, &empty_rp);
1209         }
1210         kretprobe_table_unlock(hash, &flags);
1211         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1212                 hlist_del(&ri->hlist);
1213                 kfree(ri);
1214         }
1215 }
1216 NOKPROBE_SYMBOL(kprobe_flush_task);
1217 
1218 static inline void free_rp_inst(struct kretprobe *rp)
1219 {
1220         struct kretprobe_instance *ri;
1221         struct hlist_node *next;
1222 
1223         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1224                 hlist_del(&ri->hlist);
1225                 kfree(ri);
1226         }
1227 }
1228 
1229 static void cleanup_rp_inst(struct kretprobe *rp)
1230 {
1231         unsigned long flags, hash;
1232         struct kretprobe_instance *ri;
1233         struct hlist_node *next;
1234         struct hlist_head *head;
1235 
1236         /* No race here */
1237         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1238                 kretprobe_table_lock(hash, &flags);
1239                 head = &kretprobe_inst_table[hash];
1240                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1241                         if (ri->rp == rp)
1242                                 ri->rp = NULL;
1243                 }
1244                 kretprobe_table_unlock(hash, &flags);
1245         }
1246         free_rp_inst(rp);
1247 }
1248 NOKPROBE_SYMBOL(cleanup_rp_inst);
1249 
1250 /*
1251 * Add the new probe to ap->list. Fail if this is the
1252 * second jprobe at the address - two jprobes can't coexist
1253 */
1254 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1255 {
1256         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1257 
1258         if (p->break_handler || p->post_handler)
1259                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1260 
1261         if (p->break_handler) {
1262                 if (ap->break_handler)
1263                         return -EEXIST;
1264                 list_add_tail_rcu(&p->list, &ap->list);
1265                 ap->break_handler = aggr_break_handler;
1266         } else
1267                 list_add_rcu(&p->list, &ap->list);
1268         if (p->post_handler && !ap->post_handler)
1269                 ap->post_handler = aggr_post_handler;
1270 
1271         return 0;
1272 }
1273 
1274 /*
1275  * Fill in the required fields of the "manager kprobe". Replace the
1276  * earlier kprobe in the hlist with the manager kprobe
1277  */
1278 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1279 {
1280         /* Copy p's insn slot to ap */
1281         copy_kprobe(p, ap);
1282         flush_insn_slot(ap);
1283         ap->addr = p->addr;
1284         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1285         ap->pre_handler = aggr_pre_handler;
1286         ap->fault_handler = aggr_fault_handler;
1287         /* We don't care the kprobe which has gone. */
1288         if (p->post_handler && !kprobe_gone(p))
1289                 ap->post_handler = aggr_post_handler;
1290         if (p->break_handler && !kprobe_gone(p))
1291                 ap->break_handler = aggr_break_handler;
1292 
1293         INIT_LIST_HEAD(&ap->list);
1294         INIT_HLIST_NODE(&ap->hlist);
1295 
1296         list_add_rcu(&p->list, &ap->list);
1297         hlist_replace_rcu(&p->hlist, &ap->hlist);
1298 }
1299 
1300 /*
1301  * This is the second or subsequent kprobe at the address - handle
1302  * the intricacies
1303  */
1304 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1305 {
1306         int ret = 0;
1307         struct kprobe *ap = orig_p;
1308 
1309         cpus_read_lock();
1310 
1311         /* For preparing optimization, jump_label_text_reserved() is called */
1312         jump_label_lock();
1313         mutex_lock(&text_mutex);
1314 
1315         if (!kprobe_aggrprobe(orig_p)) {
1316                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1317                 ap = alloc_aggr_kprobe(orig_p);
1318                 if (!ap) {
1319                         ret = -ENOMEM;
1320                         goto out;
1321                 }
1322                 init_aggr_kprobe(ap, orig_p);
1323         } else if (kprobe_unused(ap))
1324                 /* This probe is going to die. Rescue it */
1325                 reuse_unused_kprobe(ap);
1326 
1327         if (kprobe_gone(ap)) {
1328                 /*
1329                  * Attempting to insert new probe at the same location that
1330                  * had a probe in the module vaddr area which already
1331                  * freed. So, the instruction slot has already been
1332                  * released. We need a new slot for the new probe.
1333                  */
1334                 ret = arch_prepare_kprobe(ap);
1335                 if (ret)
1336                         /*
1337                          * Even if fail to allocate new slot, don't need to
1338                          * free aggr_probe. It will be used next time, or
1339                          * freed by unregister_kprobe.
1340                          */
1341                         goto out;
1342 
1343                 /* Prepare optimized instructions if possible. */
1344                 prepare_optimized_kprobe(ap);
1345 
1346                 /*
1347                  * Clear gone flag to prevent allocating new slot again, and
1348                  * set disabled flag because it is not armed yet.
1349                  */
1350                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1351                             | KPROBE_FLAG_DISABLED;
1352         }
1353 
1354         /* Copy ap's insn slot to p */
1355         copy_kprobe(ap, p);
1356         ret = add_new_kprobe(ap, p);
1357 
1358 out:
1359         mutex_unlock(&text_mutex);
1360         jump_label_unlock();
1361         cpus_read_unlock();
1362 
1363         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1364                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1365                 if (!kprobes_all_disarmed)
1366                         /* Arm the breakpoint again. */
1367                         arm_kprobe(ap);
1368         }
1369         return ret;
1370 }
1371 
1372 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1373 {
1374         /* The __kprobes marked functions and entry code must not be probed */
1375         return addr >= (unsigned long)__kprobes_text_start &&
1376                addr < (unsigned long)__kprobes_text_end;
1377 }
1378 
1379 bool within_kprobe_blacklist(unsigned long addr)
1380 {
1381         struct kprobe_blacklist_entry *ent;
1382 
1383         if (arch_within_kprobe_blacklist(addr))
1384                 return true;
1385         /*
1386          * If there exists a kprobe_blacklist, verify and
1387          * fail any probe registration in the prohibited area
1388          */
1389         list_for_each_entry(ent, &kprobe_blacklist, list) {
1390                 if (addr >= ent->start_addr && addr < ent->end_addr)
1391                         return true;
1392         }
1393 
1394         return false;
1395 }
1396 
1397 /*
1398  * If we have a symbol_name argument, look it up and add the offset field
1399  * to it. This way, we can specify a relative address to a symbol.
1400  * This returns encoded errors if it fails to look up symbol or invalid
1401  * combination of parameters.
1402  */
1403 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1404                         const char *symbol_name, unsigned int offset)
1405 {
1406         if ((symbol_name && addr) || (!symbol_name && !addr))
1407                 goto invalid;
1408 
1409         if (symbol_name) {
1410                 addr = kprobe_lookup_name(symbol_name, offset);
1411                 if (!addr)
1412                         return ERR_PTR(-ENOENT);
1413         }
1414 
1415         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1416         if (addr)
1417                 return addr;
1418 
1419 invalid:
1420         return ERR_PTR(-EINVAL);
1421 }
1422 
1423 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1424 {
1425         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1426 }
1427 
1428 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1429 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1430 {
1431         struct kprobe *ap, *list_p;
1432 
1433         ap = get_kprobe(p->addr);
1434         if (unlikely(!ap))
1435                 return NULL;
1436 
1437         if (p != ap) {
1438                 list_for_each_entry_rcu(list_p, &ap->list, list)
1439                         if (list_p == p)
1440                         /* kprobe p is a valid probe */
1441                                 goto valid;
1442                 return NULL;
1443         }
1444 valid:
1445         return ap;
1446 }
1447 
1448 /* Return error if the kprobe is being re-registered */
1449 static inline int check_kprobe_rereg(struct kprobe *p)
1450 {
1451         int ret = 0;
1452 
1453         mutex_lock(&kprobe_mutex);
1454         if (__get_valid_kprobe(p))
1455                 ret = -EINVAL;
1456         mutex_unlock(&kprobe_mutex);
1457 
1458         return ret;
1459 }
1460 
1461 int __weak arch_check_ftrace_location(struct kprobe *p)
1462 {
1463         unsigned long ftrace_addr;
1464 
1465         ftrace_addr = ftrace_location((unsigned long)p->addr);
1466         if (ftrace_addr) {
1467 #ifdef CONFIG_KPROBES_ON_FTRACE
1468                 /* Given address is not on the instruction boundary */
1469                 if ((unsigned long)p->addr != ftrace_addr)
1470                         return -EILSEQ;
1471                 p->flags |= KPROBE_FLAG_FTRACE;
1472 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1473                 return -EINVAL;
1474 #endif
1475         }
1476         return 0;
1477 }
1478 
1479 static int check_kprobe_address_safe(struct kprobe *p,
1480                                      struct module **probed_mod)
1481 {
1482         int ret;
1483 
1484         ret = arch_check_ftrace_location(p);
1485         if (ret)
1486                 return ret;
1487         jump_label_lock();
1488         preempt_disable();
1489 
1490         /* Ensure it is not in reserved area nor out of text */
1491         if (!kernel_text_address((unsigned long) p->addr) ||
1492             within_kprobe_blacklist((unsigned long) p->addr) ||
1493             jump_label_text_reserved(p->addr, p->addr)) {
1494                 ret = -EINVAL;
1495                 goto out;
1496         }
1497 
1498         /* Check if are we probing a module */
1499         *probed_mod = __module_text_address((unsigned long) p->addr);
1500         if (*probed_mod) {
1501                 /*
1502                  * We must hold a refcount of the probed module while updating
1503                  * its code to prohibit unexpected unloading.
1504                  */
1505                 if (unlikely(!try_module_get(*probed_mod))) {
1506                         ret = -ENOENT;
1507                         goto out;
1508                 }
1509 
1510                 /*
1511                  * If the module freed .init.text, we couldn't insert
1512                  * kprobes in there.
1513                  */
1514                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1515                     (*probed_mod)->state != MODULE_STATE_COMING) {
1516                         module_put(*probed_mod);
1517                         *probed_mod = NULL;
1518                         ret = -ENOENT;
1519                 }
1520         }
1521 out:
1522         preempt_enable();
1523         jump_label_unlock();
1524 
1525         return ret;
1526 }
1527 
1528 int register_kprobe(struct kprobe *p)
1529 {
1530         int ret;
1531         struct kprobe *old_p;
1532         struct module *probed_mod;
1533         kprobe_opcode_t *addr;
1534 
1535         /* Adjust probe address from symbol */
1536         addr = kprobe_addr(p);
1537         if (IS_ERR(addr))
1538                 return PTR_ERR(addr);
1539         p->addr = addr;
1540 
1541         ret = check_kprobe_rereg(p);
1542         if (ret)
1543                 return ret;
1544 
1545         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1546         p->flags &= KPROBE_FLAG_DISABLED;
1547         p->nmissed = 0;
1548         INIT_LIST_HEAD(&p->list);
1549 
1550         ret = check_kprobe_address_safe(p, &probed_mod);
1551         if (ret)
1552                 return ret;
1553 
1554         mutex_lock(&kprobe_mutex);
1555 
1556         old_p = get_kprobe(p->addr);
1557         if (old_p) {
1558                 /* Since this may unoptimize old_p, locking text_mutex. */
1559                 ret = register_aggr_kprobe(old_p, p);
1560                 goto out;
1561         }
1562 
1563         cpus_read_lock();
1564         /* Prevent text modification */
1565         mutex_lock(&text_mutex);
1566         ret = prepare_kprobe(p);
1567         mutex_unlock(&text_mutex);
1568         cpus_read_unlock();
1569         if (ret)
1570                 goto out;
1571 
1572         INIT_HLIST_NODE(&p->hlist);
1573         hlist_add_head_rcu(&p->hlist,
1574                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1575 
1576         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1577                 arm_kprobe(p);
1578 
1579         /* Try to optimize kprobe */
1580         try_to_optimize_kprobe(p);
1581 out:
1582         mutex_unlock(&kprobe_mutex);
1583 
1584         if (probed_mod)
1585                 module_put(probed_mod);
1586 
1587         return ret;
1588 }
1589 EXPORT_SYMBOL_GPL(register_kprobe);
1590 
1591 /* Check if all probes on the aggrprobe are disabled */
1592 static int aggr_kprobe_disabled(struct kprobe *ap)
1593 {
1594         struct kprobe *kp;
1595 
1596         list_for_each_entry_rcu(kp, &ap->list, list)
1597                 if (!kprobe_disabled(kp))
1598                         /*
1599                          * There is an active probe on the list.
1600                          * We can't disable this ap.
1601                          */
1602                         return 0;
1603 
1604         return 1;
1605 }
1606 
1607 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1608 static struct kprobe *__disable_kprobe(struct kprobe *p)
1609 {
1610         struct kprobe *orig_p;
1611 
1612         /* Get an original kprobe for return */
1613         orig_p = __get_valid_kprobe(p);
1614         if (unlikely(orig_p == NULL))
1615                 return NULL;
1616 
1617         if (!kprobe_disabled(p)) {
1618                 /* Disable probe if it is a child probe */
1619                 if (p != orig_p)
1620                         p->flags |= KPROBE_FLAG_DISABLED;
1621 
1622                 /* Try to disarm and disable this/parent probe */
1623                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1624                         /*
1625                          * If kprobes_all_disarmed is set, orig_p
1626                          * should have already been disarmed, so
1627                          * skip unneed disarming process.
1628                          */
1629                         if (!kprobes_all_disarmed)
1630                                 disarm_kprobe(orig_p, true);
1631                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1632                 }
1633         }
1634 
1635         return orig_p;
1636 }
1637 
1638 /*
1639  * Unregister a kprobe without a scheduler synchronization.
1640  */
1641 static int __unregister_kprobe_top(struct kprobe *p)
1642 {
1643         struct kprobe *ap, *list_p;
1644 
1645         /* Disable kprobe. This will disarm it if needed. */
1646         ap = __disable_kprobe(p);
1647         if (ap == NULL)
1648                 return -EINVAL;
1649 
1650         if (ap == p)
1651                 /*
1652                  * This probe is an independent(and non-optimized) kprobe
1653                  * (not an aggrprobe). Remove from the hash list.
1654                  */
1655                 goto disarmed;
1656 
1657         /* Following process expects this probe is an aggrprobe */
1658         WARN_ON(!kprobe_aggrprobe(ap));
1659 
1660         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1661                 /*
1662                  * !disarmed could be happen if the probe is under delayed
1663                  * unoptimizing.
1664                  */
1665                 goto disarmed;
1666         else {
1667                 /* If disabling probe has special handlers, update aggrprobe */
1668                 if (p->break_handler && !kprobe_gone(p))
1669                         ap->break_handler = NULL;
1670                 if (p->post_handler && !kprobe_gone(p)) {
1671                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1672                                 if ((list_p != p) && (list_p->post_handler))
1673                                         goto noclean;
1674                         }
1675                         ap->post_handler = NULL;
1676                 }
1677 noclean:
1678                 /*
1679                  * Remove from the aggrprobe: this path will do nothing in
1680                  * __unregister_kprobe_bottom().
1681                  */
1682                 list_del_rcu(&p->list);
1683                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1684                         /*
1685                          * Try to optimize this probe again, because post
1686                          * handler may have been changed.
1687                          */
1688                         optimize_kprobe(ap);
1689         }
1690         return 0;
1691 
1692 disarmed:
1693         BUG_ON(!kprobe_disarmed(ap));
1694         hlist_del_rcu(&ap->hlist);
1695         return 0;
1696 }
1697 
1698 static void __unregister_kprobe_bottom(struct kprobe *p)
1699 {
1700         struct kprobe *ap;
1701 
1702         if (list_empty(&p->list))
1703                 /* This is an independent kprobe */
1704                 arch_remove_kprobe(p);
1705         else if (list_is_singular(&p->list)) {
1706                 /* This is the last child of an aggrprobe */
1707                 ap = list_entry(p->list.next, struct kprobe, list);
1708                 list_del(&p->list);
1709                 free_aggr_kprobe(ap);
1710         }
1711         /* Otherwise, do nothing. */
1712 }
1713 
1714 int register_kprobes(struct kprobe **kps, int num)
1715 {
1716         int i, ret = 0;
1717 
1718         if (num <= 0)
1719                 return -EINVAL;
1720         for (i = 0; i < num; i++) {
1721                 ret = register_kprobe(kps[i]);
1722                 if (ret < 0) {
1723                         if (i > 0)
1724                                 unregister_kprobes(kps, i);
1725                         break;
1726                 }
1727         }
1728         return ret;
1729 }
1730 EXPORT_SYMBOL_GPL(register_kprobes);
1731 
1732 void unregister_kprobe(struct kprobe *p)
1733 {
1734         unregister_kprobes(&p, 1);
1735 }
1736 EXPORT_SYMBOL_GPL(unregister_kprobe);
1737 
1738 void unregister_kprobes(struct kprobe **kps, int num)
1739 {
1740         int i;
1741 
1742         if (num <= 0)
1743                 return;
1744         mutex_lock(&kprobe_mutex);
1745         for (i = 0; i < num; i++)
1746                 if (__unregister_kprobe_top(kps[i]) < 0)
1747                         kps[i]->addr = NULL;
1748         mutex_unlock(&kprobe_mutex);
1749 
1750         synchronize_sched();
1751         for (i = 0; i < num; i++)
1752                 if (kps[i]->addr)
1753                         __unregister_kprobe_bottom(kps[i]);
1754 }
1755 EXPORT_SYMBOL_GPL(unregister_kprobes);
1756 
1757 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1758                                         unsigned long val, void *data)
1759 {
1760         return NOTIFY_DONE;
1761 }
1762 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1763 
1764 static struct notifier_block kprobe_exceptions_nb = {
1765         .notifier_call = kprobe_exceptions_notify,
1766         .priority = 0x7fffffff /* we need to be notified first */
1767 };
1768 
1769 unsigned long __weak arch_deref_entry_point(void *entry)
1770 {
1771         return (unsigned long)entry;
1772 }
1773 
1774 #if 0
1775 int register_jprobes(struct jprobe **jps, int num)
1776 {
1777         int ret = 0, i;
1778 
1779         if (num <= 0)
1780                 return -EINVAL;
1781 
1782         for (i = 0; i < num; i++) {
1783                 ret = register_jprobe(jps[i]);
1784 
1785                 if (ret < 0) {
1786                         if (i > 0)
1787                                 unregister_jprobes(jps, i);
1788                         break;
1789                 }
1790         }
1791 
1792         return ret;
1793 }
1794 EXPORT_SYMBOL_GPL(register_jprobes);
1795 
1796 int register_jprobe(struct jprobe *jp)
1797 {
1798         unsigned long addr, offset;
1799         struct kprobe *kp = &jp->kp;
1800 
1801         /*
1802          * Verify probepoint as well as the jprobe handler are
1803          * valid function entry points.
1804          */
1805         addr = arch_deref_entry_point(jp->entry);
1806 
1807         if (kallsyms_lookup_size_offset(addr, NULL, &offset) && offset == 0 &&
1808             kprobe_on_func_entry(kp->addr, kp->symbol_name, kp->offset)) {
1809                 kp->pre_handler = setjmp_pre_handler;
1810                 kp->break_handler = longjmp_break_handler;
1811                 return register_kprobe(kp);
1812         }
1813 
1814         return -EINVAL;
1815 }
1816 EXPORT_SYMBOL_GPL(register_jprobe);
1817 
1818 void unregister_jprobe(struct jprobe *jp)
1819 {
1820         unregister_jprobes(&jp, 1);
1821 }
1822 EXPORT_SYMBOL_GPL(unregister_jprobe);
1823 
1824 void unregister_jprobes(struct jprobe **jps, int num)
1825 {
1826         int i;
1827 
1828         if (num <= 0)
1829                 return;
1830         mutex_lock(&kprobe_mutex);
1831         for (i = 0; i < num; i++)
1832                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1833                         jps[i]->kp.addr = NULL;
1834         mutex_unlock(&kprobe_mutex);
1835 
1836         synchronize_sched();
1837         for (i = 0; i < num; i++) {
1838                 if (jps[i]->kp.addr)
1839                         __unregister_kprobe_bottom(&jps[i]->kp);
1840         }
1841 }
1842 EXPORT_SYMBOL_GPL(unregister_jprobes);
1843 #endif
1844 
1845 #ifdef CONFIG_KRETPROBES
1846 /*
1847  * This kprobe pre_handler is registered with every kretprobe. When probe
1848  * hits it will set up the return probe.
1849  */
1850 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1851 {
1852         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1853         unsigned long hash, flags = 0;
1854         struct kretprobe_instance *ri;
1855 
1856         /*
1857          * To avoid deadlocks, prohibit return probing in NMI contexts,
1858          * just skip the probe and increase the (inexact) 'nmissed'
1859          * statistical counter, so that the user is informed that
1860          * something happened:
1861          */
1862         if (unlikely(in_nmi())) {
1863                 rp->nmissed++;
1864                 return 0;
1865         }
1866 
1867         /* TODO: consider to only swap the RA after the last pre_handler fired */
1868         hash = hash_ptr(current, KPROBE_HASH_BITS);
1869         raw_spin_lock_irqsave(&rp->lock, flags);
1870         if (!hlist_empty(&rp->free_instances)) {
1871                 ri = hlist_entry(rp->free_instances.first,
1872                                 struct kretprobe_instance, hlist);
1873                 hlist_del(&ri->hlist);
1874                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1875 
1876                 ri->rp = rp;
1877                 ri->task = current;
1878 
1879                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1880                         raw_spin_lock_irqsave(&rp->lock, flags);
1881                         hlist_add_head(&ri->hlist, &rp->free_instances);
1882                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1883                         return 0;
1884                 }
1885 
1886                 arch_prepare_kretprobe(ri, regs);
1887 
1888                 /* XXX(hch): why is there no hlist_move_head? */
1889                 INIT_HLIST_NODE(&ri->hlist);
1890                 kretprobe_table_lock(hash, &flags);
1891                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1892                 kretprobe_table_unlock(hash, &flags);
1893         } else {
1894                 rp->nmissed++;
1895                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1896         }
1897         return 0;
1898 }
1899 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1900 
1901 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1902 {
1903         return !offset;
1904 }
1905 
1906 bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1907 {
1908         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1909 
1910         if (IS_ERR(kp_addr))
1911                 return false;
1912 
1913         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) ||
1914                                                 !arch_kprobe_on_func_entry(offset))
1915                 return false;
1916 
1917         return true;
1918 }
1919 
1920 int register_kretprobe(struct kretprobe *rp)
1921 {
1922         int ret = 0;
1923         struct kretprobe_instance *inst;
1924         int i;
1925         void *addr;
1926 
1927         if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset))
1928                 return -EINVAL;
1929 
1930         if (kretprobe_blacklist_size) {
1931                 addr = kprobe_addr(&rp->kp);
1932                 if (IS_ERR(addr))
1933                         return PTR_ERR(addr);
1934 
1935                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1936                         if (kretprobe_blacklist[i].addr == addr)
1937                                 return -EINVAL;
1938                 }
1939         }
1940 
1941         rp->kp.pre_handler = pre_handler_kretprobe;
1942         rp->kp.post_handler = NULL;
1943         rp->kp.fault_handler = NULL;
1944         rp->kp.break_handler = NULL;
1945 
1946         /* Pre-allocate memory for max kretprobe instances */
1947         if (rp->maxactive <= 0) {
1948 #ifdef CONFIG_PREEMPT
1949                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1950 #else
1951                 rp->maxactive = num_possible_cpus();
1952 #endif
1953         }
1954         raw_spin_lock_init(&rp->lock);
1955         INIT_HLIST_HEAD(&rp->free_instances);
1956         for (i = 0; i < rp->maxactive; i++) {
1957                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1958                                rp->data_size, GFP_KERNEL);
1959                 if (inst == NULL) {
1960                         free_rp_inst(rp);
1961                         return -ENOMEM;
1962                 }
1963                 INIT_HLIST_NODE(&inst->hlist);
1964                 hlist_add_head(&inst->hlist, &rp->free_instances);
1965         }
1966 
1967         rp->nmissed = 0;
1968         /* Establish function entry probe point */
1969         ret = register_kprobe(&rp->kp);
1970         if (ret != 0)
1971                 free_rp_inst(rp);
1972         return ret;
1973 }
1974 EXPORT_SYMBOL_GPL(register_kretprobe);
1975 
1976 int register_kretprobes(struct kretprobe **rps, int num)
1977 {
1978         int ret = 0, i;
1979 
1980         if (num <= 0)
1981                 return -EINVAL;
1982         for (i = 0; i < num; i++) {
1983                 ret = register_kretprobe(rps[i]);
1984                 if (ret < 0) {
1985                         if (i > 0)
1986                                 unregister_kretprobes(rps, i);
1987                         break;
1988                 }
1989         }
1990         return ret;
1991 }
1992 EXPORT_SYMBOL_GPL(register_kretprobes);
1993 
1994 void unregister_kretprobe(struct kretprobe *rp)
1995 {
1996         unregister_kretprobes(&rp, 1);
1997 }
1998 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1999 
2000 void unregister_kretprobes(struct kretprobe **rps, int num)
2001 {
2002         int i;
2003 
2004         if (num <= 0)
2005                 return;
2006         mutex_lock(&kprobe_mutex);
2007         for (i = 0; i < num; i++)
2008                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2009                         rps[i]->kp.addr = NULL;
2010         mutex_unlock(&kprobe_mutex);
2011 
2012         synchronize_sched();
2013         for (i = 0; i < num; i++) {
2014                 if (rps[i]->kp.addr) {
2015                         __unregister_kprobe_bottom(&rps[i]->kp);
2016                         cleanup_rp_inst(rps[i]);
2017                 }
2018         }
2019 }
2020 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2021 
2022 #else /* CONFIG_KRETPROBES */
2023 int register_kretprobe(struct kretprobe *rp)
2024 {
2025         return -ENOSYS;
2026 }
2027 EXPORT_SYMBOL_GPL(register_kretprobe);
2028 
2029 int register_kretprobes(struct kretprobe **rps, int num)
2030 {
2031         return -ENOSYS;
2032 }
2033 EXPORT_SYMBOL_GPL(register_kretprobes);
2034 
2035 void unregister_kretprobe(struct kretprobe *rp)
2036 {
2037 }
2038 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2039 
2040 void unregister_kretprobes(struct kretprobe **rps, int num)
2041 {
2042 }
2043 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2044 
2045 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2046 {
2047         return 0;
2048 }
2049 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2050 
2051 #endif /* CONFIG_KRETPROBES */
2052 
2053 /* Set the kprobe gone and remove its instruction buffer. */
2054 static void kill_kprobe(struct kprobe *p)
2055 {
2056         struct kprobe *kp;
2057 
2058         p->flags |= KPROBE_FLAG_GONE;
2059         if (kprobe_aggrprobe(p)) {
2060                 /*
2061                  * If this is an aggr_kprobe, we have to list all the
2062                  * chained probes and mark them GONE.
2063                  */
2064                 list_for_each_entry_rcu(kp, &p->list, list)
2065                         kp->flags |= KPROBE_FLAG_GONE;
2066                 p->post_handler = NULL;
2067                 p->break_handler = NULL;
2068                 kill_optimized_kprobe(p);
2069         }
2070         /*
2071          * Here, we can remove insn_slot safely, because no thread calls
2072          * the original probed function (which will be freed soon) any more.
2073          */
2074         arch_remove_kprobe(p);
2075 }
2076 
2077 /* Disable one kprobe */
2078 int disable_kprobe(struct kprobe *kp)
2079 {
2080         int ret = 0;
2081 
2082         mutex_lock(&kprobe_mutex);
2083 
2084         /* Disable this kprobe */
2085         if (__disable_kprobe(kp) == NULL)
2086                 ret = -EINVAL;
2087 
2088         mutex_unlock(&kprobe_mutex);
2089         return ret;
2090 }
2091 EXPORT_SYMBOL_GPL(disable_kprobe);
2092 
2093 /* Enable one kprobe */
2094 int enable_kprobe(struct kprobe *kp)
2095 {
2096         int ret = 0;
2097         struct kprobe *p;
2098 
2099         mutex_lock(&kprobe_mutex);
2100 
2101         /* Check whether specified probe is valid. */
2102         p = __get_valid_kprobe(kp);
2103         if (unlikely(p == NULL)) {
2104                 ret = -EINVAL;
2105                 goto out;
2106         }
2107 
2108         if (kprobe_gone(kp)) {
2109                 /* This kprobe has gone, we couldn't enable it. */
2110                 ret = -EINVAL;
2111                 goto out;
2112         }
2113 
2114         if (p != kp)
2115                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2116 
2117         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2118                 p->flags &= ~KPROBE_FLAG_DISABLED;
2119                 arm_kprobe(p);
2120         }
2121 out:
2122         mutex_unlock(&kprobe_mutex);
2123         return ret;
2124 }
2125 EXPORT_SYMBOL_GPL(enable_kprobe);
2126 
2127 void dump_kprobe(struct kprobe *kp)
2128 {
2129         printk(KERN_WARNING "Dumping kprobe:\n");
2130         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2131                kp->symbol_name, kp->addr, kp->offset);
2132 }
2133 NOKPROBE_SYMBOL(dump_kprobe);
2134 
2135 /*
2136  * Lookup and populate the kprobe_blacklist.
2137  *
2138  * Unlike the kretprobe blacklist, we'll need to determine
2139  * the range of addresses that belong to the said functions,
2140  * since a kprobe need not necessarily be at the beginning
2141  * of a function.
2142  */
2143 static int __init populate_kprobe_blacklist(unsigned long *start,
2144                                              unsigned long *end)
2145 {
2146         unsigned long *iter;
2147         struct kprobe_blacklist_entry *ent;
2148         unsigned long entry, offset = 0, size = 0;
2149 
2150         for (iter = start; iter < end; iter++) {
2151                 entry = arch_deref_entry_point((void *)*iter);
2152 
2153                 if (!kernel_text_address(entry) ||
2154                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2155                         pr_err("Failed to find blacklist at %p\n",
2156                                 (void *)entry);
2157                         continue;
2158                 }
2159 
2160                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2161                 if (!ent)
2162                         return -ENOMEM;
2163                 ent->start_addr = entry;
2164                 ent->end_addr = entry + size;
2165                 INIT_LIST_HEAD(&ent->list);
2166                 list_add_tail(&ent->list, &kprobe_blacklist);
2167         }
2168         return 0;
2169 }
2170 
2171 /* Module notifier call back, checking kprobes on the module */
2172 static int kprobes_module_callback(struct notifier_block *nb,
2173                                    unsigned long val, void *data)
2174 {
2175         struct module *mod = data;
2176         struct hlist_head *head;
2177         struct kprobe *p;
2178         unsigned int i;
2179         int checkcore = (val == MODULE_STATE_GOING);
2180 
2181         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2182                 return NOTIFY_DONE;
2183 
2184         /*
2185          * When MODULE_STATE_GOING was notified, both of module .text and
2186          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2187          * notified, only .init.text section would be freed. We need to
2188          * disable kprobes which have been inserted in the sections.
2189          */
2190         mutex_lock(&kprobe_mutex);
2191         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2192                 head = &kprobe_table[i];
2193                 hlist_for_each_entry_rcu(p, head, hlist)
2194                         if (within_module_init((unsigned long)p->addr, mod) ||
2195                             (checkcore &&
2196                              within_module_core((unsigned long)p->addr, mod))) {
2197                                 /*
2198                                  * The vaddr this probe is installed will soon
2199                                  * be vfreed buy not synced to disk. Hence,
2200                                  * disarming the breakpoint isn't needed.
2201                                  *
2202                                  * Note, this will also move any optimized probes
2203                                  * that are pending to be removed from their
2204                                  * corresponding lists to the freeing_list and
2205                                  * will not be touched by the delayed
2206                                  * kprobe_optimizer work handler.
2207                                  */
2208                                 kill_kprobe(p);
2209                         }
2210         }
2211         mutex_unlock(&kprobe_mutex);
2212         return NOTIFY_DONE;
2213 }
2214 
2215 static struct notifier_block kprobe_module_nb = {
2216         .notifier_call = kprobes_module_callback,
2217         .priority = 0
2218 };
2219 
2220 /* Markers of _kprobe_blacklist section */
2221 extern unsigned long __start_kprobe_blacklist[];
2222 extern unsigned long __stop_kprobe_blacklist[];
2223 
2224 static int __init init_kprobes(void)
2225 {
2226         int i, err = 0;
2227 
2228         /* FIXME allocate the probe table, currently defined statically */
2229         /* initialize all list heads */
2230         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2231                 INIT_HLIST_HEAD(&kprobe_table[i]);
2232                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2233                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2234         }
2235 
2236         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2237                                         __stop_kprobe_blacklist);
2238         if (err) {
2239                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2240                 pr_err("Please take care of using kprobes.\n");
2241         }
2242 
2243         if (kretprobe_blacklist_size) {
2244                 /* lookup the function address from its name */
2245                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2246                         kretprobe_blacklist[i].addr =
2247                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2248                         if (!kretprobe_blacklist[i].addr)
2249                                 printk("kretprobe: lookup failed: %s\n",
2250                                        kretprobe_blacklist[i].name);
2251                 }
2252         }
2253 
2254 #if defined(CONFIG_OPTPROBES)
2255 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2256         /* Init kprobe_optinsn_slots */
2257         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2258 #endif
2259         /* By default, kprobes can be optimized */
2260         kprobes_allow_optimization = true;
2261 #endif
2262 
2263         /* By default, kprobes are armed */
2264         kprobes_all_disarmed = false;
2265 
2266         err = arch_init_kprobes();
2267         if (!err)
2268                 err = register_die_notifier(&kprobe_exceptions_nb);
2269         if (!err)
2270                 err = register_module_notifier(&kprobe_module_nb);
2271 
2272         kprobes_initialized = (err == 0);
2273 
2274         if (!err)
2275                 init_test_probes();
2276         return err;
2277 }
2278 
2279 #ifdef CONFIG_DEBUG_FS
2280 static void report_probe(struct seq_file *pi, struct kprobe *p,
2281                 const char *sym, int offset, char *modname, struct kprobe *pp)
2282 {
2283         char *kprobe_type;
2284 
2285         if (p->pre_handler == pre_handler_kretprobe)
2286                 kprobe_type = "r";
2287         else if (p->pre_handler == setjmp_pre_handler)
2288                 kprobe_type = "j";
2289         else
2290                 kprobe_type = "k";
2291 
2292         if (sym)
2293                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2294                         p->addr, kprobe_type, sym, offset,
2295                         (modname ? modname : " "));
2296         else
2297                 seq_printf(pi, "%p  %s  %p ",
2298                         p->addr, kprobe_type, p->addr);
2299 
2300         if (!pp)
2301                 pp = p;
2302         seq_printf(pi, "%s%s%s%s\n",
2303                 (kprobe_gone(p) ? "[GONE]" : ""),
2304                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2305                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2306                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2307 }
2308 
2309 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2310 {
2311         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2312 }
2313 
2314 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2315 {
2316         (*pos)++;
2317         if (*pos >= KPROBE_TABLE_SIZE)
2318                 return NULL;
2319         return pos;
2320 }
2321 
2322 static void kprobe_seq_stop(struct seq_file *f, void *v)
2323 {
2324         /* Nothing to do */
2325 }
2326 
2327 static int show_kprobe_addr(struct seq_file *pi, void *v)
2328 {
2329         struct hlist_head *head;
2330         struct kprobe *p, *kp;
2331         const char *sym = NULL;
2332         unsigned int i = *(loff_t *) v;
2333         unsigned long offset = 0;
2334         char *modname, namebuf[KSYM_NAME_LEN];
2335 
2336         head = &kprobe_table[i];
2337         preempt_disable();
2338         hlist_for_each_entry_rcu(p, head, hlist) {
2339                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2340                                         &offset, &modname, namebuf);
2341                 if (kprobe_aggrprobe(p)) {
2342                         list_for_each_entry_rcu(kp, &p->list, list)
2343                                 report_probe(pi, kp, sym, offset, modname, p);
2344                 } else
2345                         report_probe(pi, p, sym, offset, modname, NULL);
2346         }
2347         preempt_enable();
2348         return 0;
2349 }
2350 
2351 static const struct seq_operations kprobes_seq_ops = {
2352         .start = kprobe_seq_start,
2353         .next  = kprobe_seq_next,
2354         .stop  = kprobe_seq_stop,
2355         .show  = show_kprobe_addr
2356 };
2357 
2358 static int kprobes_open(struct inode *inode, struct file *filp)
2359 {
2360         return seq_open(filp, &kprobes_seq_ops);
2361 }
2362 
2363 static const struct file_operations debugfs_kprobes_operations = {
2364         .open           = kprobes_open,
2365         .read           = seq_read,
2366         .llseek         = seq_lseek,
2367         .release        = seq_release,
2368 };
2369 
2370 /* kprobes/blacklist -- shows which functions can not be probed */
2371 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2372 {
2373         return seq_list_start(&kprobe_blacklist, *pos);
2374 }
2375 
2376 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2377 {
2378         return seq_list_next(v, &kprobe_blacklist, pos);
2379 }
2380 
2381 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2382 {
2383         struct kprobe_blacklist_entry *ent =
2384                 list_entry(v, struct kprobe_blacklist_entry, list);
2385 
2386         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2387                    (void *)ent->end_addr, (void *)ent->start_addr);
2388         return 0;
2389 }
2390 
2391 static const struct seq_operations kprobe_blacklist_seq_ops = {
2392         .start = kprobe_blacklist_seq_start,
2393         .next  = kprobe_blacklist_seq_next,
2394         .stop  = kprobe_seq_stop,       /* Reuse void function */
2395         .show  = kprobe_blacklist_seq_show,
2396 };
2397 
2398 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2399 {
2400         return seq_open(filp, &kprobe_blacklist_seq_ops);
2401 }
2402 
2403 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2404         .open           = kprobe_blacklist_open,
2405         .read           = seq_read,
2406         .llseek         = seq_lseek,
2407         .release        = seq_release,
2408 };
2409 
2410 static void arm_all_kprobes(void)
2411 {
2412         struct hlist_head *head;
2413         struct kprobe *p;
2414         unsigned int i;
2415 
2416         mutex_lock(&kprobe_mutex);
2417 
2418         /* If kprobes are armed, just return */
2419         if (!kprobes_all_disarmed)
2420                 goto already_enabled;
2421 
2422         /*
2423          * optimize_kprobe() called by arm_kprobe() checks
2424          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2425          * arm_kprobe.
2426          */
2427         kprobes_all_disarmed = false;
2428         /* Arming kprobes doesn't optimize kprobe itself */
2429         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2430                 head = &kprobe_table[i];
2431                 hlist_for_each_entry_rcu(p, head, hlist)
2432                         if (!kprobe_disabled(p))
2433                                 arm_kprobe(p);
2434         }
2435 
2436         printk(KERN_INFO "Kprobes globally enabled\n");
2437 
2438 already_enabled:
2439         mutex_unlock(&kprobe_mutex);
2440         return;
2441 }
2442 
2443 static void disarm_all_kprobes(void)
2444 {
2445         struct hlist_head *head;
2446         struct kprobe *p;
2447         unsigned int i;
2448 
2449         mutex_lock(&kprobe_mutex);
2450 
2451         /* If kprobes are already disarmed, just return */
2452         if (kprobes_all_disarmed) {
2453                 mutex_unlock(&kprobe_mutex);
2454                 return;
2455         }
2456 
2457         kprobes_all_disarmed = true;
2458         printk(KERN_INFO "Kprobes globally disabled\n");
2459 
2460         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2461                 head = &kprobe_table[i];
2462                 hlist_for_each_entry_rcu(p, head, hlist) {
2463                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2464                                 disarm_kprobe(p, false);
2465                 }
2466         }
2467         mutex_unlock(&kprobe_mutex);
2468 
2469         /* Wait for disarming all kprobes by optimizer */
2470         wait_for_kprobe_optimizer();
2471 }
2472 
2473 /*
2474  * XXX: The debugfs bool file interface doesn't allow for callbacks
2475  * when the bool state is switched. We can reuse that facility when
2476  * available
2477  */
2478 static ssize_t read_enabled_file_bool(struct file *file,
2479                char __user *user_buf, size_t count, loff_t *ppos)
2480 {
2481         char buf[3];
2482 
2483         if (!kprobes_all_disarmed)
2484                 buf[0] = '1';
2485         else
2486                 buf[0] = '';
2487         buf[1] = '\n';
2488         buf[2] = 0x00;
2489         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2490 }
2491 
2492 static ssize_t write_enabled_file_bool(struct file *file,
2493                const char __user *user_buf, size_t count, loff_t *ppos)
2494 {
2495         char buf[32];
2496         size_t buf_size;
2497 
2498         buf_size = min(count, (sizeof(buf)-1));
2499         if (copy_from_user(buf, user_buf, buf_size))
2500                 return -EFAULT;
2501 
2502         buf[buf_size] = '\0';
2503         switch (buf[0]) {
2504         case 'y':
2505         case 'Y':
2506         case '1':
2507                 arm_all_kprobes();
2508                 break;
2509         case 'n':
2510         case 'N':
2511         case '':
2512                 disarm_all_kprobes();
2513                 break;
2514         default:
2515                 return -EINVAL;
2516         }
2517 
2518         return count;
2519 }
2520 
2521 static const struct file_operations fops_kp = {
2522         .read =         read_enabled_file_bool,
2523         .write =        write_enabled_file_bool,
2524         .llseek =       default_llseek,
2525 };
2526 
2527 static int __init debugfs_kprobe_init(void)
2528 {
2529         struct dentry *dir, *file;
2530         unsigned int value = 1;
2531 
2532         dir = debugfs_create_dir("kprobes", NULL);
2533         if (!dir)
2534                 return -ENOMEM;
2535 
2536         file = debugfs_create_file("list", 0444, dir, NULL,
2537                                 &debugfs_kprobes_operations);
2538         if (!file)
2539                 goto error;
2540 
2541         file = debugfs_create_file("enabled", 0600, dir,
2542                                         &value, &fops_kp);
2543         if (!file)
2544                 goto error;
2545 
2546         file = debugfs_create_file("blacklist", 0444, dir, NULL,
2547                                 &debugfs_kprobe_blacklist_ops);
2548         if (!file)
2549                 goto error;
2550 
2551         return 0;
2552 
2553 error:
2554         debugfs_remove(dir);
2555         return -ENOMEM;
2556 }
2557 
2558 late_initcall(debugfs_kprobe_init);
2559 #endif /* CONFIG_DEBUG_FS */
2560 
2561 module_init(init_kprobes);
2562 
2563 /* defined in arch/.../kernel/kprobes.c */
2564 EXPORT_SYMBOL_GPL(jprobe_return);
2565 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp