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

TOMOYO Linux Cross Reference
Linux/kernel/kprobes.c

Version: ~ [ linux-6.11-rc3 ] ~ [ linux-6.10.4 ] ~ [ linux-6.9.12 ] ~ [ linux-6.8.12 ] ~ [ linux-6.7.12 ] ~ [ linux-6.6.45 ] ~ [ linux-6.5.13 ] ~ [ linux-6.4.16 ] ~ [ linux-6.3.13 ] ~ [ linux-6.2.16 ] ~ [ linux-6.1.104 ] ~ [ linux-6.0.19 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.164 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.223 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.281 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.319 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.336 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.337 ] ~ [ linux-4.4.302 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-or-later
  2 /*
  3  *  Kernel Probes (KProbes)
  4  *
  5  * Copyright (C) IBM Corporation, 2002, 2004
  6  *
  7  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  8  *              Probes initial implementation (includes suggestions from
  9  *              Rusty Russell).
 10  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
 11  *              hlists and exceptions notifier as suggested by Andi Kleen.
 12  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
 13  *              interface to access function arguments.
 14  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
 15  *              exceptions notifier to be first on the priority list.
 16  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
 17  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
 18  *              <prasanna@in.ibm.com> added function-return probes.
 19  */
 20 
 21 #define pr_fmt(fmt) "kprobes: " fmt
 22 
 23 #include <linux/kprobes.h>
 24 #include <linux/hash.h>
 25 #include <linux/init.h>
 26 #include <linux/slab.h>
 27 #include <linux/stddef.h>
 28 #include <linux/export.h>
 29 #include <linux/kallsyms.h>
 30 #include <linux/freezer.h>
 31 #include <linux/seq_file.h>
 32 #include <linux/debugfs.h>
 33 #include <linux/sysctl.h>
 34 #include <linux/kdebug.h>
 35 #include <linux/memory.h>
 36 #include <linux/ftrace.h>
 37 #include <linux/cpu.h>
 38 #include <linux/jump_label.h>
 39 #include <linux/static_call.h>
 40 #include <linux/perf_event.h>
 41 #include <linux/execmem.h>
 42 
 43 #include <asm/sections.h>
 44 #include <asm/cacheflush.h>
 45 #include <asm/errno.h>
 46 #include <linux/uaccess.h>
 47 
 48 #define KPROBE_HASH_BITS 6
 49 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
 50 
 51 #if !defined(CONFIG_OPTPROBES) || !defined(CONFIG_SYSCTL)
 52 #define kprobe_sysctls_init() do { } while (0)
 53 #endif
 54 
 55 static int kprobes_initialized;
 56 /* kprobe_table can be accessed by
 57  * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
 58  * Or
 59  * - RCU hlist traversal under disabling preempt (breakpoint handlers)
 60  */
 61 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
 62 
 63 /* NOTE: change this value only with 'kprobe_mutex' held */
 64 static bool kprobes_all_disarmed;
 65 
 66 /* This protects 'kprobe_table' and 'optimizing_list' */
 67 static DEFINE_MUTEX(kprobe_mutex);
 68 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
 69 
 70 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
 71                                         unsigned int __unused)
 72 {
 73         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
 74 }
 75 
 76 /*
 77  * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
 78  * kprobes can not probe.
 79  */
 80 static LIST_HEAD(kprobe_blacklist);
 81 
 82 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
 83 /*
 84  * 'kprobe::ainsn.insn' points to the copy of the instruction to be
 85  * single-stepped. x86_64, POWER4 and above have no-exec support and
 86  * stepping on the instruction on a vmalloced/kmalloced/data page
 87  * is a recipe for disaster
 88  */
 89 struct kprobe_insn_page {
 90         struct list_head list;
 91         kprobe_opcode_t *insns;         /* Page of instruction slots */
 92         struct kprobe_insn_cache *cache;
 93         int nused;
 94         int ngarbage;
 95         char slot_used[];
 96 };
 97 
 98 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
 99         (offsetof(struct kprobe_insn_page, slot_used) + \
100          (sizeof(char) * (slots)))
101 
102 static int slots_per_page(struct kprobe_insn_cache *c)
103 {
104         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
105 }
106 
107 enum kprobe_slot_state {
108         SLOT_CLEAN = 0,
109         SLOT_DIRTY = 1,
110         SLOT_USED = 2,
111 };
112 
113 void __weak *alloc_insn_page(void)
114 {
115         /*
116          * Use execmem_alloc() so this page is within +/- 2GB of where the
117          * kernel image and loaded module images reside. This is required
118          * for most of the architectures.
119          * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
120          */
121         return execmem_alloc(EXECMEM_KPROBES, PAGE_SIZE);
122 }
123 
124 static void free_insn_page(void *page)
125 {
126         execmem_free(page);
127 }
128 
129 struct kprobe_insn_cache kprobe_insn_slots = {
130         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
131         .alloc = alloc_insn_page,
132         .free = free_insn_page,
133         .sym = KPROBE_INSN_PAGE_SYM,
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 
157                         for (i = 0; i < slots_per_page(c); i++) {
158                                 if (kip->slot_used[i] == SLOT_CLEAN) {
159                                         kip->slot_used[i] = SLOT_USED;
160                                         kip->nused++;
161                                         slot = kip->insns + (i * c->insn_size);
162                                         rcu_read_unlock();
163                                         goto out;
164                                 }
165                         }
166                         /* kip->nused is broken. Fix it. */
167                         kip->nused = slots_per_page(c);
168                         WARN_ON(1);
169                 }
170         }
171         rcu_read_unlock();
172 
173         /* If there are any garbage slots, collect it and try again. */
174         if (c->nr_garbage && collect_garbage_slots(c) == 0)
175                 goto retry;
176 
177         /* All out of space.  Need to allocate a new page. */
178         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
179         if (!kip)
180                 goto out;
181 
182         kip->insns = c->alloc();
183         if (!kip->insns) {
184                 kfree(kip);
185                 goto out;
186         }
187         INIT_LIST_HEAD(&kip->list);
188         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
189         kip->slot_used[0] = SLOT_USED;
190         kip->nused = 1;
191         kip->ngarbage = 0;
192         kip->cache = c;
193         list_add_rcu(&kip->list, &c->pages);
194         slot = kip->insns;
195 
196         /* Record the perf ksymbol register event after adding the page */
197         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
198                            PAGE_SIZE, false, c->sym);
199 out:
200         mutex_unlock(&c->mutex);
201         return slot;
202 }
203 
204 /* Return true if all garbages are collected, otherwise false. */
205 static bool 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                         /*
218                          * Record perf ksymbol unregister event before removing
219                          * the page.
220                          */
221                         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
222                                            (unsigned long)kip->insns, PAGE_SIZE, true,
223                                            kip->cache->sym);
224                         list_del_rcu(&kip->list);
225                         synchronize_rcu();
226                         kip->cache->free(kip->insns);
227                         kfree(kip);
228                 }
229                 return true;
230         }
231         return false;
232 }
233 
234 static int collect_garbage_slots(struct kprobe_insn_cache *c)
235 {
236         struct kprobe_insn_page *kip, *next;
237 
238         /* Ensure no-one is interrupted on the garbages */
239         synchronize_rcu();
240 
241         list_for_each_entry_safe(kip, next, &c->pages, list) {
242                 int i;
243 
244                 if (kip->ngarbage == 0)
245                         continue;
246                 kip->ngarbage = 0;      /* we will collect all garbages */
247                 for (i = 0; i < slots_per_page(c); i++) {
248                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
249                                 break;
250                 }
251         }
252         c->nr_garbage = 0;
253         return 0;
254 }
255 
256 void __free_insn_slot(struct kprobe_insn_cache *c,
257                       kprobe_opcode_t *slot, int dirty)
258 {
259         struct kprobe_insn_page *kip;
260         long idx;
261 
262         mutex_lock(&c->mutex);
263         rcu_read_lock();
264         list_for_each_entry_rcu(kip, &c->pages, list) {
265                 idx = ((long)slot - (long)kip->insns) /
266                         (c->insn_size * sizeof(kprobe_opcode_t));
267                 if (idx >= 0 && idx < slots_per_page(c))
268                         goto out;
269         }
270         /* Could not find this slot. */
271         WARN_ON(1);
272         kip = NULL;
273 out:
274         rcu_read_unlock();
275         /* Mark and sweep: this may sleep */
276         if (kip) {
277                 /* Check double free */
278                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
279                 if (dirty) {
280                         kip->slot_used[idx] = SLOT_DIRTY;
281                         kip->ngarbage++;
282                         if (++c->nr_garbage > slots_per_page(c))
283                                 collect_garbage_slots(c);
284                 } else {
285                         collect_one_slot(kip, idx);
286                 }
287         }
288         mutex_unlock(&c->mutex);
289 }
290 
291 /*
292  * Check given address is on the page of kprobe instruction slots.
293  * This will be used for checking whether the address on a stack
294  * is on a text area or not.
295  */
296 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
297 {
298         struct kprobe_insn_page *kip;
299         bool ret = false;
300 
301         rcu_read_lock();
302         list_for_each_entry_rcu(kip, &c->pages, list) {
303                 if (addr >= (unsigned long)kip->insns &&
304                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
305                         ret = true;
306                         break;
307                 }
308         }
309         rcu_read_unlock();
310 
311         return ret;
312 }
313 
314 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
315                              unsigned long *value, char *type, char *sym)
316 {
317         struct kprobe_insn_page *kip;
318         int ret = -ERANGE;
319 
320         rcu_read_lock();
321         list_for_each_entry_rcu(kip, &c->pages, list) {
322                 if ((*symnum)--)
323                         continue;
324                 strscpy(sym, c->sym, KSYM_NAME_LEN);
325                 *type = 't';
326                 *value = (unsigned long)kip->insns;
327                 ret = 0;
328                 break;
329         }
330         rcu_read_unlock();
331 
332         return ret;
333 }
334 
335 #ifdef CONFIG_OPTPROBES
336 void __weak *alloc_optinsn_page(void)
337 {
338         return alloc_insn_page();
339 }
340 
341 void __weak free_optinsn_page(void *page)
342 {
343         free_insn_page(page);
344 }
345 
346 /* For optimized_kprobe buffer */
347 struct kprobe_insn_cache kprobe_optinsn_slots = {
348         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
349         .alloc = alloc_optinsn_page,
350         .free = free_optinsn_page,
351         .sym = KPROBE_OPTINSN_PAGE_SYM,
352         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
353         /* .insn_size is initialized later */
354         .nr_garbage = 0,
355 };
356 #endif
357 #endif
358 
359 /* We have preemption disabled.. so it is safe to use __ versions */
360 static inline void set_kprobe_instance(struct kprobe *kp)
361 {
362         __this_cpu_write(kprobe_instance, kp);
363 }
364 
365 static inline void reset_kprobe_instance(void)
366 {
367         __this_cpu_write(kprobe_instance, NULL);
368 }
369 
370 /*
371  * This routine is called either:
372  *      - under the 'kprobe_mutex' - during kprobe_[un]register().
373  *                              OR
374  *      - with preemption disabled - from architecture specific code.
375  */
376 struct kprobe *get_kprobe(void *addr)
377 {
378         struct hlist_head *head;
379         struct kprobe *p;
380 
381         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
382         hlist_for_each_entry_rcu(p, head, hlist,
383                                  lockdep_is_held(&kprobe_mutex)) {
384                 if (p->addr == addr)
385                         return p;
386         }
387 
388         return NULL;
389 }
390 NOKPROBE_SYMBOL(get_kprobe);
391 
392 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
393 
394 /* Return true if 'p' is an aggregator */
395 static inline bool kprobe_aggrprobe(struct kprobe *p)
396 {
397         return p->pre_handler == aggr_pre_handler;
398 }
399 
400 /* Return true if 'p' is unused */
401 static inline bool kprobe_unused(struct kprobe *p)
402 {
403         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
404                list_empty(&p->list);
405 }
406 
407 /* Keep all fields in the kprobe consistent. */
408 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
409 {
410         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
411         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
412 }
413 
414 #ifdef CONFIG_OPTPROBES
415 /* NOTE: This is protected by 'kprobe_mutex'. */
416 static bool kprobes_allow_optimization;
417 
418 /*
419  * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
420  * This must be called from arch-dep optimized caller.
421  */
422 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
423 {
424         struct kprobe *kp;
425 
426         list_for_each_entry_rcu(kp, &p->list, list) {
427                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
428                         set_kprobe_instance(kp);
429                         kp->pre_handler(kp, regs);
430                 }
431                 reset_kprobe_instance();
432         }
433 }
434 NOKPROBE_SYMBOL(opt_pre_handler);
435 
436 /* Free optimized instructions and optimized_kprobe */
437 static void free_aggr_kprobe(struct kprobe *p)
438 {
439         struct optimized_kprobe *op;
440 
441         op = container_of(p, struct optimized_kprobe, kp);
442         arch_remove_optimized_kprobe(op);
443         arch_remove_kprobe(p);
444         kfree(op);
445 }
446 
447 /* Return true if the kprobe is ready for optimization. */
448 static inline int kprobe_optready(struct kprobe *p)
449 {
450         struct optimized_kprobe *op;
451 
452         if (kprobe_aggrprobe(p)) {
453                 op = container_of(p, struct optimized_kprobe, kp);
454                 return arch_prepared_optinsn(&op->optinsn);
455         }
456 
457         return 0;
458 }
459 
460 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
461 bool kprobe_disarmed(struct kprobe *p)
462 {
463         struct optimized_kprobe *op;
464 
465         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
466         if (!kprobe_aggrprobe(p))
467                 return kprobe_disabled(p);
468 
469         op = container_of(p, struct optimized_kprobe, kp);
470 
471         return kprobe_disabled(p) && list_empty(&op->list);
472 }
473 
474 /* Return true if the probe is queued on (un)optimizing lists */
475 static bool kprobe_queued(struct kprobe *p)
476 {
477         struct optimized_kprobe *op;
478 
479         if (kprobe_aggrprobe(p)) {
480                 op = container_of(p, struct optimized_kprobe, kp);
481                 if (!list_empty(&op->list))
482                         return true;
483         }
484         return false;
485 }
486 
487 /*
488  * Return an optimized kprobe whose optimizing code replaces
489  * instructions including 'addr' (exclude breakpoint).
490  */
491 static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
492 {
493         int i;
494         struct kprobe *p = NULL;
495         struct optimized_kprobe *op;
496 
497         /* Don't check i == 0, since that is a breakpoint case. */
498         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
499                 p = get_kprobe(addr - i);
500 
501         if (p && kprobe_optready(p)) {
502                 op = container_of(p, struct optimized_kprobe, kp);
503                 if (arch_within_optimized_kprobe(op, addr))
504                         return p;
505         }
506 
507         return NULL;
508 }
509 
510 /* Optimization staging list, protected by 'kprobe_mutex' */
511 static LIST_HEAD(optimizing_list);
512 static LIST_HEAD(unoptimizing_list);
513 static LIST_HEAD(freeing_list);
514 
515 static void kprobe_optimizer(struct work_struct *work);
516 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
517 #define OPTIMIZE_DELAY 5
518 
519 /*
520  * Optimize (replace a breakpoint with a jump) kprobes listed on
521  * 'optimizing_list'.
522  */
523 static void do_optimize_kprobes(void)
524 {
525         lockdep_assert_held(&text_mutex);
526         /*
527          * The optimization/unoptimization refers 'online_cpus' via
528          * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
529          * And same time, 'text_mutex' will be held in cpu-hotplug and here.
530          * This combination can cause a deadlock (cpu-hotplug tries to lock
531          * 'text_mutex' but stop_machine() can not be done because
532          * the 'online_cpus' has been changed)
533          * To avoid this deadlock, caller must have locked cpu-hotplug
534          * for preventing cpu-hotplug outside of 'text_mutex' locking.
535          */
536         lockdep_assert_cpus_held();
537 
538         /* Optimization never be done when disarmed */
539         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
540             list_empty(&optimizing_list))
541                 return;
542 
543         arch_optimize_kprobes(&optimizing_list);
544 }
545 
546 /*
547  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
548  * if need) kprobes listed on 'unoptimizing_list'.
549  */
550 static void do_unoptimize_kprobes(void)
551 {
552         struct optimized_kprobe *op, *tmp;
553 
554         lockdep_assert_held(&text_mutex);
555         /* See comment in do_optimize_kprobes() */
556         lockdep_assert_cpus_held();
557 
558         if (!list_empty(&unoptimizing_list))
559                 arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
560 
561         /* Loop on 'freeing_list' for disarming and removing from kprobe hash list */
562         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
563                 /* Switching from detour code to origin */
564                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
565                 /* Disarm probes if marked disabled and not gone */
566                 if (kprobe_disabled(&op->kp) && !kprobe_gone(&op->kp))
567                         arch_disarm_kprobe(&op->kp);
568                 if (kprobe_unused(&op->kp)) {
569                         /*
570                          * Remove unused probes from hash list. After waiting
571                          * for synchronization, these probes are reclaimed.
572                          * (reclaiming is done by do_free_cleaned_kprobes().)
573                          */
574                         hlist_del_rcu(&op->kp.hlist);
575                 } else
576                         list_del_init(&op->list);
577         }
578 }
579 
580 /* Reclaim all kprobes on the 'freeing_list' */
581 static void do_free_cleaned_kprobes(void)
582 {
583         struct optimized_kprobe *op, *tmp;
584 
585         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
586                 list_del_init(&op->list);
587                 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
588                         /*
589                          * This must not happen, but if there is a kprobe
590                          * still in use, keep it on kprobes hash list.
591                          */
592                         continue;
593                 }
594                 free_aggr_kprobe(&op->kp);
595         }
596 }
597 
598 /* Start optimizer after OPTIMIZE_DELAY passed */
599 static void kick_kprobe_optimizer(void)
600 {
601         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
602 }
603 
604 /* Kprobe jump optimizer */
605 static void kprobe_optimizer(struct work_struct *work)
606 {
607         mutex_lock(&kprobe_mutex);
608         cpus_read_lock();
609         mutex_lock(&text_mutex);
610 
611         /*
612          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
613          * kprobes before waiting for quiesence period.
614          */
615         do_unoptimize_kprobes();
616 
617         /*
618          * Step 2: Wait for quiesence period to ensure all potentially
619          * preempted tasks to have normally scheduled. Because optprobe
620          * may modify multiple instructions, there is a chance that Nth
621          * instruction is preempted. In that case, such tasks can return
622          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
623          * Note that on non-preemptive kernel, this is transparently converted
624          * to synchronoze_sched() to wait for all interrupts to have completed.
625          */
626         synchronize_rcu_tasks();
627 
628         /* Step 3: Optimize kprobes after quiesence period */
629         do_optimize_kprobes();
630 
631         /* Step 4: Free cleaned kprobes after quiesence period */
632         do_free_cleaned_kprobes();
633 
634         mutex_unlock(&text_mutex);
635         cpus_read_unlock();
636 
637         /* Step 5: Kick optimizer again if needed */
638         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
639                 kick_kprobe_optimizer();
640 
641         mutex_unlock(&kprobe_mutex);
642 }
643 
644 /* Wait for completing optimization and unoptimization */
645 void wait_for_kprobe_optimizer(void)
646 {
647         mutex_lock(&kprobe_mutex);
648 
649         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
650                 mutex_unlock(&kprobe_mutex);
651 
652                 /* This will also make 'optimizing_work' execute immmediately */
653                 flush_delayed_work(&optimizing_work);
654                 /* 'optimizing_work' might not have been queued yet, relax */
655                 cpu_relax();
656 
657                 mutex_lock(&kprobe_mutex);
658         }
659 
660         mutex_unlock(&kprobe_mutex);
661 }
662 
663 bool optprobe_queued_unopt(struct optimized_kprobe *op)
664 {
665         struct optimized_kprobe *_op;
666 
667         list_for_each_entry(_op, &unoptimizing_list, list) {
668                 if (op == _op)
669                         return true;
670         }
671 
672         return false;
673 }
674 
675 /* Optimize kprobe if p is ready to be optimized */
676 static void optimize_kprobe(struct kprobe *p)
677 {
678         struct optimized_kprobe *op;
679 
680         /* Check if the kprobe is disabled or not ready for optimization. */
681         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
682             (kprobe_disabled(p) || kprobes_all_disarmed))
683                 return;
684 
685         /* kprobes with 'post_handler' can not be optimized */
686         if (p->post_handler)
687                 return;
688 
689         op = container_of(p, struct optimized_kprobe, kp);
690 
691         /* Check there is no other kprobes at the optimized instructions */
692         if (arch_check_optimized_kprobe(op) < 0)
693                 return;
694 
695         /* Check if it is already optimized. */
696         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
697                 if (optprobe_queued_unopt(op)) {
698                         /* This is under unoptimizing. Just dequeue the probe */
699                         list_del_init(&op->list);
700                 }
701                 return;
702         }
703         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
704 
705         /*
706          * On the 'unoptimizing_list' and 'optimizing_list',
707          * 'op' must have OPTIMIZED flag
708          */
709         if (WARN_ON_ONCE(!list_empty(&op->list)))
710                 return;
711 
712         list_add(&op->list, &optimizing_list);
713         kick_kprobe_optimizer();
714 }
715 
716 /* Short cut to direct unoptimizing */
717 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
718 {
719         lockdep_assert_cpus_held();
720         arch_unoptimize_kprobe(op);
721         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
722 }
723 
724 /* Unoptimize a kprobe if p is optimized */
725 static void unoptimize_kprobe(struct kprobe *p, bool force)
726 {
727         struct optimized_kprobe *op;
728 
729         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
730                 return; /* This is not an optprobe nor optimized */
731 
732         op = container_of(p, struct optimized_kprobe, kp);
733         if (!kprobe_optimized(p))
734                 return;
735 
736         if (!list_empty(&op->list)) {
737                 if (optprobe_queued_unopt(op)) {
738                         /* Queued in unoptimizing queue */
739                         if (force) {
740                                 /*
741                                  * Forcibly unoptimize the kprobe here, and queue it
742                                  * in the freeing list for release afterwards.
743                                  */
744                                 force_unoptimize_kprobe(op);
745                                 list_move(&op->list, &freeing_list);
746                         }
747                 } else {
748                         /* Dequeue from the optimizing queue */
749                         list_del_init(&op->list);
750                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
751                 }
752                 return;
753         }
754 
755         /* Optimized kprobe case */
756         if (force) {
757                 /* Forcibly update the code: this is a special case */
758                 force_unoptimize_kprobe(op);
759         } else {
760                 list_add(&op->list, &unoptimizing_list);
761                 kick_kprobe_optimizer();
762         }
763 }
764 
765 /* Cancel unoptimizing for reusing */
766 static int reuse_unused_kprobe(struct kprobe *ap)
767 {
768         struct optimized_kprobe *op;
769 
770         /*
771          * Unused kprobe MUST be on the way of delayed unoptimizing (means
772          * there is still a relative jump) and disabled.
773          */
774         op = container_of(ap, struct optimized_kprobe, kp);
775         WARN_ON_ONCE(list_empty(&op->list));
776         /* Enable the probe again */
777         ap->flags &= ~KPROBE_FLAG_DISABLED;
778         /* Optimize it again. (remove from 'op->list') */
779         if (!kprobe_optready(ap))
780                 return -EINVAL;
781 
782         optimize_kprobe(ap);
783         return 0;
784 }
785 
786 /* Remove optimized instructions */
787 static void kill_optimized_kprobe(struct kprobe *p)
788 {
789         struct optimized_kprobe *op;
790 
791         op = container_of(p, struct optimized_kprobe, kp);
792         if (!list_empty(&op->list))
793                 /* Dequeue from the (un)optimization queue */
794                 list_del_init(&op->list);
795         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
796 
797         if (kprobe_unused(p)) {
798                 /*
799                  * Unused kprobe is on unoptimizing or freeing list. We move it
800                  * to freeing_list and let the kprobe_optimizer() remove it from
801                  * the kprobe hash list and free it.
802                  */
803                 if (optprobe_queued_unopt(op))
804                         list_move(&op->list, &freeing_list);
805         }
806 
807         /* Don't touch the code, because it is already freed. */
808         arch_remove_optimized_kprobe(op);
809 }
810 
811 static inline
812 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
813 {
814         if (!kprobe_ftrace(p))
815                 arch_prepare_optimized_kprobe(op, p);
816 }
817 
818 /* Try to prepare optimized instructions */
819 static void prepare_optimized_kprobe(struct kprobe *p)
820 {
821         struct optimized_kprobe *op;
822 
823         op = container_of(p, struct optimized_kprobe, kp);
824         __prepare_optimized_kprobe(op, p);
825 }
826 
827 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
828 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
829 {
830         struct optimized_kprobe *op;
831 
832         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
833         if (!op)
834                 return NULL;
835 
836         INIT_LIST_HEAD(&op->list);
837         op->kp.addr = p->addr;
838         __prepare_optimized_kprobe(op, p);
839 
840         return &op->kp;
841 }
842 
843 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
844 
845 /*
846  * Prepare an optimized_kprobe and optimize it.
847  * NOTE: 'p' must be a normal registered kprobe.
848  */
849 static void try_to_optimize_kprobe(struct kprobe *p)
850 {
851         struct kprobe *ap;
852         struct optimized_kprobe *op;
853 
854         /* Impossible to optimize ftrace-based kprobe. */
855         if (kprobe_ftrace(p))
856                 return;
857 
858         /* For preparing optimization, jump_label_text_reserved() is called. */
859         cpus_read_lock();
860         jump_label_lock();
861         mutex_lock(&text_mutex);
862 
863         ap = alloc_aggr_kprobe(p);
864         if (!ap)
865                 goto out;
866 
867         op = container_of(ap, struct optimized_kprobe, kp);
868         if (!arch_prepared_optinsn(&op->optinsn)) {
869                 /* If failed to setup optimizing, fallback to kprobe. */
870                 arch_remove_optimized_kprobe(op);
871                 kfree(op);
872                 goto out;
873         }
874 
875         init_aggr_kprobe(ap, p);
876         optimize_kprobe(ap);    /* This just kicks optimizer thread. */
877 
878 out:
879         mutex_unlock(&text_mutex);
880         jump_label_unlock();
881         cpus_read_unlock();
882 }
883 
884 static void optimize_all_kprobes(void)
885 {
886         struct hlist_head *head;
887         struct kprobe *p;
888         unsigned int i;
889 
890         mutex_lock(&kprobe_mutex);
891         /* If optimization is already allowed, just return. */
892         if (kprobes_allow_optimization)
893                 goto out;
894 
895         cpus_read_lock();
896         kprobes_allow_optimization = true;
897         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
898                 head = &kprobe_table[i];
899                 hlist_for_each_entry(p, head, hlist)
900                         if (!kprobe_disabled(p))
901                                 optimize_kprobe(p);
902         }
903         cpus_read_unlock();
904         pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
905 out:
906         mutex_unlock(&kprobe_mutex);
907 }
908 
909 #ifdef CONFIG_SYSCTL
910 static void unoptimize_all_kprobes(void)
911 {
912         struct hlist_head *head;
913         struct kprobe *p;
914         unsigned int i;
915 
916         mutex_lock(&kprobe_mutex);
917         /* If optimization is already prohibited, just return. */
918         if (!kprobes_allow_optimization) {
919                 mutex_unlock(&kprobe_mutex);
920                 return;
921         }
922 
923         cpus_read_lock();
924         kprobes_allow_optimization = false;
925         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
926                 head = &kprobe_table[i];
927                 hlist_for_each_entry(p, head, hlist) {
928                         if (!kprobe_disabled(p))
929                                 unoptimize_kprobe(p, false);
930                 }
931         }
932         cpus_read_unlock();
933         mutex_unlock(&kprobe_mutex);
934 
935         /* Wait for unoptimizing completion. */
936         wait_for_kprobe_optimizer();
937         pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
938 }
939 
940 static DEFINE_MUTEX(kprobe_sysctl_mutex);
941 static int sysctl_kprobes_optimization;
942 static int proc_kprobes_optimization_handler(const struct ctl_table *table,
943                                              int write, void *buffer,
944                                              size_t *length, loff_t *ppos)
945 {
946         int ret;
947 
948         mutex_lock(&kprobe_sysctl_mutex);
949         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
950         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
951 
952         if (sysctl_kprobes_optimization)
953                 optimize_all_kprobes();
954         else
955                 unoptimize_all_kprobes();
956         mutex_unlock(&kprobe_sysctl_mutex);
957 
958         return ret;
959 }
960 
961 static struct ctl_table kprobe_sysctls[] = {
962         {
963                 .procname       = "kprobes-optimization",
964                 .data           = &sysctl_kprobes_optimization,
965                 .maxlen         = sizeof(int),
966                 .mode           = 0644,
967                 .proc_handler   = proc_kprobes_optimization_handler,
968                 .extra1         = SYSCTL_ZERO,
969                 .extra2         = SYSCTL_ONE,
970         },
971 };
972 
973 static void __init kprobe_sysctls_init(void)
974 {
975         register_sysctl_init("debug", kprobe_sysctls);
976 }
977 #endif /* CONFIG_SYSCTL */
978 
979 /* Put a breakpoint for a probe. */
980 static void __arm_kprobe(struct kprobe *p)
981 {
982         struct kprobe *_p;
983 
984         lockdep_assert_held(&text_mutex);
985 
986         /* Find the overlapping optimized kprobes. */
987         _p = get_optimized_kprobe(p->addr);
988         if (unlikely(_p))
989                 /* Fallback to unoptimized kprobe */
990                 unoptimize_kprobe(_p, true);
991 
992         arch_arm_kprobe(p);
993         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
994 }
995 
996 /* Remove the breakpoint of a probe. */
997 static void __disarm_kprobe(struct kprobe *p, bool reopt)
998 {
999         struct kprobe *_p;
1000 
1001         lockdep_assert_held(&text_mutex);
1002 
1003         /* Try to unoptimize */
1004         unoptimize_kprobe(p, kprobes_all_disarmed);
1005 
1006         if (!kprobe_queued(p)) {
1007                 arch_disarm_kprobe(p);
1008                 /* If another kprobe was blocked, re-optimize it. */
1009                 _p = get_optimized_kprobe(p->addr);
1010                 if (unlikely(_p) && reopt)
1011                         optimize_kprobe(_p);
1012         }
1013         /*
1014          * TODO: Since unoptimization and real disarming will be done by
1015          * the worker thread, we can not check whether another probe are
1016          * unoptimized because of this probe here. It should be re-optimized
1017          * by the worker thread.
1018          */
1019 }
1020 
1021 #else /* !CONFIG_OPTPROBES */
1022 
1023 #define optimize_kprobe(p)                      do {} while (0)
1024 #define unoptimize_kprobe(p, f)                 do {} while (0)
1025 #define kill_optimized_kprobe(p)                do {} while (0)
1026 #define prepare_optimized_kprobe(p)             do {} while (0)
1027 #define try_to_optimize_kprobe(p)               do {} while (0)
1028 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
1029 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
1030 #define kprobe_disarmed(p)                      kprobe_disabled(p)
1031 #define wait_for_kprobe_optimizer()             do {} while (0)
1032 
1033 static int reuse_unused_kprobe(struct kprobe *ap)
1034 {
1035         /*
1036          * If the optimized kprobe is NOT supported, the aggr kprobe is
1037          * released at the same time that the last aggregated kprobe is
1038          * unregistered.
1039          * Thus there should be no chance to reuse unused kprobe.
1040          */
1041         WARN_ON_ONCE(1);
1042         return -EINVAL;
1043 }
1044 
1045 static void free_aggr_kprobe(struct kprobe *p)
1046 {
1047         arch_remove_kprobe(p);
1048         kfree(p);
1049 }
1050 
1051 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1052 {
1053         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1054 }
1055 #endif /* CONFIG_OPTPROBES */
1056 
1057 #ifdef CONFIG_KPROBES_ON_FTRACE
1058 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1059         .func = kprobe_ftrace_handler,
1060         .flags = FTRACE_OPS_FL_SAVE_REGS,
1061 };
1062 
1063 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1064         .func = kprobe_ftrace_handler,
1065         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1066 };
1067 
1068 static int kprobe_ipmodify_enabled;
1069 static int kprobe_ftrace_enabled;
1070 bool kprobe_ftrace_disabled;
1071 
1072 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1073                                int *cnt)
1074 {
1075         int ret;
1076 
1077         lockdep_assert_held(&kprobe_mutex);
1078 
1079         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1080         if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1081                 return ret;
1082 
1083         if (*cnt == 0) {
1084                 ret = register_ftrace_function(ops);
1085                 if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1086                         goto err_ftrace;
1087         }
1088 
1089         (*cnt)++;
1090         return ret;
1091 
1092 err_ftrace:
1093         /*
1094          * At this point, sinec ops is not registered, we should be sefe from
1095          * registering empty filter.
1096          */
1097         ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1098         return ret;
1099 }
1100 
1101 static int arm_kprobe_ftrace(struct kprobe *p)
1102 {
1103         bool ipmodify = (p->post_handler != NULL);
1104 
1105         return __arm_kprobe_ftrace(p,
1106                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1107                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1108 }
1109 
1110 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1111                                   int *cnt)
1112 {
1113         int ret;
1114 
1115         lockdep_assert_held(&kprobe_mutex);
1116 
1117         if (*cnt == 1) {
1118                 ret = unregister_ftrace_function(ops);
1119                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1120                         return ret;
1121         }
1122 
1123         (*cnt)--;
1124 
1125         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1126         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1127                   p->addr, ret);
1128         return ret;
1129 }
1130 
1131 static int disarm_kprobe_ftrace(struct kprobe *p)
1132 {
1133         bool ipmodify = (p->post_handler != NULL);
1134 
1135         return __disarm_kprobe_ftrace(p,
1136                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1137                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1138 }
1139 
1140 void kprobe_ftrace_kill(void)
1141 {
1142         kprobe_ftrace_disabled = true;
1143 }
1144 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1145 static inline int arm_kprobe_ftrace(struct kprobe *p)
1146 {
1147         return -ENODEV;
1148 }
1149 
1150 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1151 {
1152         return -ENODEV;
1153 }
1154 #endif
1155 
1156 static int prepare_kprobe(struct kprobe *p)
1157 {
1158         /* Must ensure p->addr is really on ftrace */
1159         if (kprobe_ftrace(p))
1160                 return arch_prepare_kprobe_ftrace(p);
1161 
1162         return arch_prepare_kprobe(p);
1163 }
1164 
1165 static int arm_kprobe(struct kprobe *kp)
1166 {
1167         if (unlikely(kprobe_ftrace(kp)))
1168                 return arm_kprobe_ftrace(kp);
1169 
1170         cpus_read_lock();
1171         mutex_lock(&text_mutex);
1172         __arm_kprobe(kp);
1173         mutex_unlock(&text_mutex);
1174         cpus_read_unlock();
1175 
1176         return 0;
1177 }
1178 
1179 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1180 {
1181         if (unlikely(kprobe_ftrace(kp)))
1182                 return disarm_kprobe_ftrace(kp);
1183 
1184         cpus_read_lock();
1185         mutex_lock(&text_mutex);
1186         __disarm_kprobe(kp, reopt);
1187         mutex_unlock(&text_mutex);
1188         cpus_read_unlock();
1189 
1190         return 0;
1191 }
1192 
1193 /*
1194  * Aggregate handlers for multiple kprobes support - these handlers
1195  * take care of invoking the individual kprobe handlers on p->list
1196  */
1197 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1198 {
1199         struct kprobe *kp;
1200 
1201         list_for_each_entry_rcu(kp, &p->list, list) {
1202                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1203                         set_kprobe_instance(kp);
1204                         if (kp->pre_handler(kp, regs))
1205                                 return 1;
1206                 }
1207                 reset_kprobe_instance();
1208         }
1209         return 0;
1210 }
1211 NOKPROBE_SYMBOL(aggr_pre_handler);
1212 
1213 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1214                               unsigned long flags)
1215 {
1216         struct kprobe *kp;
1217 
1218         list_for_each_entry_rcu(kp, &p->list, list) {
1219                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1220                         set_kprobe_instance(kp);
1221                         kp->post_handler(kp, regs, flags);
1222                         reset_kprobe_instance();
1223                 }
1224         }
1225 }
1226 NOKPROBE_SYMBOL(aggr_post_handler);
1227 
1228 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
1229 void kprobes_inc_nmissed_count(struct kprobe *p)
1230 {
1231         struct kprobe *kp;
1232 
1233         if (!kprobe_aggrprobe(p)) {
1234                 p->nmissed++;
1235         } else {
1236                 list_for_each_entry_rcu(kp, &p->list, list)
1237                         kp->nmissed++;
1238         }
1239 }
1240 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1241 
1242 static struct kprobe kprobe_busy = {
1243         .addr = (void *) get_kprobe,
1244 };
1245 
1246 void kprobe_busy_begin(void)
1247 {
1248         struct kprobe_ctlblk *kcb;
1249 
1250         preempt_disable();
1251         __this_cpu_write(current_kprobe, &kprobe_busy);
1252         kcb = get_kprobe_ctlblk();
1253         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1254 }
1255 
1256 void kprobe_busy_end(void)
1257 {
1258         __this_cpu_write(current_kprobe, NULL);
1259         preempt_enable();
1260 }
1261 
1262 /* Add the new probe to 'ap->list'. */
1263 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1264 {
1265         if (p->post_handler)
1266                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1267 
1268         list_add_rcu(&p->list, &ap->list);
1269         if (p->post_handler && !ap->post_handler)
1270                 ap->post_handler = aggr_post_handler;
1271 
1272         return 0;
1273 }
1274 
1275 /*
1276  * Fill in the required fields of the aggregator kprobe. Replace the
1277  * earlier kprobe in the hlist with the aggregator kprobe.
1278  */
1279 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1280 {
1281         /* Copy the insn slot of 'p' to 'ap'. */
1282         copy_kprobe(p, ap);
1283         flush_insn_slot(ap);
1284         ap->addr = p->addr;
1285         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1286         ap->pre_handler = aggr_pre_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 
1291         INIT_LIST_HEAD(&ap->list);
1292         INIT_HLIST_NODE(&ap->hlist);
1293 
1294         list_add_rcu(&p->list, &ap->list);
1295         hlist_replace_rcu(&p->hlist, &ap->hlist);
1296 }
1297 
1298 /*
1299  * This registers the second or subsequent kprobe at the same address.
1300  */
1301 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1302 {
1303         int ret = 0;
1304         struct kprobe *ap = orig_p;
1305 
1306         cpus_read_lock();
1307 
1308         /* For preparing optimization, jump_label_text_reserved() is called */
1309         jump_label_lock();
1310         mutex_lock(&text_mutex);
1311 
1312         if (!kprobe_aggrprobe(orig_p)) {
1313                 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1314                 ap = alloc_aggr_kprobe(orig_p);
1315                 if (!ap) {
1316                         ret = -ENOMEM;
1317                         goto out;
1318                 }
1319                 init_aggr_kprobe(ap, orig_p);
1320         } else if (kprobe_unused(ap)) {
1321                 /* This probe is going to die. Rescue it */
1322                 ret = reuse_unused_kprobe(ap);
1323                 if (ret)
1324                         goto out;
1325         }
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 the 'ap'. 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 the insn slot of 'p' to 'ap'. */
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                         ret = arm_kprobe(ap);
1368                         if (ret) {
1369                                 ap->flags |= KPROBE_FLAG_DISABLED;
1370                                 list_del_rcu(&p->list);
1371                                 synchronize_rcu();
1372                         }
1373                 }
1374         }
1375         return ret;
1376 }
1377 
1378 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1379 {
1380         /* The '__kprobes' functions and entry code must not be probed. */
1381         return addr >= (unsigned long)__kprobes_text_start &&
1382                addr < (unsigned long)__kprobes_text_end;
1383 }
1384 
1385 static bool __within_kprobe_blacklist(unsigned long addr)
1386 {
1387         struct kprobe_blacklist_entry *ent;
1388 
1389         if (arch_within_kprobe_blacklist(addr))
1390                 return true;
1391         /*
1392          * If 'kprobe_blacklist' is defined, check the address and
1393          * reject any probe registration in the prohibited area.
1394          */
1395         list_for_each_entry(ent, &kprobe_blacklist, list) {
1396                 if (addr >= ent->start_addr && addr < ent->end_addr)
1397                         return true;
1398         }
1399         return false;
1400 }
1401 
1402 bool within_kprobe_blacklist(unsigned long addr)
1403 {
1404         char symname[KSYM_NAME_LEN], *p;
1405 
1406         if (__within_kprobe_blacklist(addr))
1407                 return true;
1408 
1409         /* Check if the address is on a suffixed-symbol */
1410         if (!lookup_symbol_name(addr, symname)) {
1411                 p = strchr(symname, '.');
1412                 if (!p)
1413                         return false;
1414                 *p = '\0';
1415                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1416                 if (addr)
1417                         return __within_kprobe_blacklist(addr);
1418         }
1419         return false;
1420 }
1421 
1422 /*
1423  * arch_adjust_kprobe_addr - adjust the address
1424  * @addr: symbol base address
1425  * @offset: offset within the symbol
1426  * @on_func_entry: was this @addr+@offset on the function entry
1427  *
1428  * Typically returns @addr + @offset, except for special cases where the
1429  * function might be prefixed by a CFI landing pad, in that case any offset
1430  * inside the landing pad is mapped to the first 'real' instruction of the
1431  * symbol.
1432  *
1433  * Specifically, for things like IBT/BTI, skip the resp. ENDBR/BTI.C
1434  * instruction at +0.
1435  */
1436 kprobe_opcode_t *__weak arch_adjust_kprobe_addr(unsigned long addr,
1437                                                 unsigned long offset,
1438                                                 bool *on_func_entry)
1439 {
1440         *on_func_entry = !offset;
1441         return (kprobe_opcode_t *)(addr + offset);
1442 }
1443 
1444 /*
1445  * If 'symbol_name' is specified, look it up and add the 'offset'
1446  * to it. This way, we can specify a relative address to a symbol.
1447  * This returns encoded errors if it fails to look up symbol or invalid
1448  * combination of parameters.
1449  */
1450 static kprobe_opcode_t *
1451 _kprobe_addr(kprobe_opcode_t *addr, const char *symbol_name,
1452              unsigned long offset, bool *on_func_entry)
1453 {
1454         if ((symbol_name && addr) || (!symbol_name && !addr))
1455                 goto invalid;
1456 
1457         if (symbol_name) {
1458                 /*
1459                  * Input: @sym + @offset
1460                  * Output: @addr + @offset
1461                  *
1462                  * NOTE: kprobe_lookup_name() does *NOT* fold the offset
1463                  *       argument into it's output!
1464                  */
1465                 addr = kprobe_lookup_name(symbol_name, offset);
1466                 if (!addr)
1467                         return ERR_PTR(-ENOENT);
1468         }
1469 
1470         /*
1471          * So here we have @addr + @offset, displace it into a new
1472          * @addr' + @offset' where @addr' is the symbol start address.
1473          */
1474         addr = (void *)addr + offset;
1475         if (!kallsyms_lookup_size_offset((unsigned long)addr, NULL, &offset))
1476                 return ERR_PTR(-ENOENT);
1477         addr = (void *)addr - offset;
1478 
1479         /*
1480          * Then ask the architecture to re-combine them, taking care of
1481          * magical function entry details while telling us if this was indeed
1482          * at the start of the function.
1483          */
1484         addr = arch_adjust_kprobe_addr((unsigned long)addr, offset, on_func_entry);
1485         if (addr)
1486                 return addr;
1487 
1488 invalid:
1489         return ERR_PTR(-EINVAL);
1490 }
1491 
1492 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1493 {
1494         bool on_func_entry;
1495         return _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1496 }
1497 
1498 /*
1499  * Check the 'p' is valid and return the aggregator kprobe
1500  * at the same address.
1501  */
1502 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1503 {
1504         struct kprobe *ap, *list_p;
1505 
1506         lockdep_assert_held(&kprobe_mutex);
1507 
1508         ap = get_kprobe(p->addr);
1509         if (unlikely(!ap))
1510                 return NULL;
1511 
1512         if (p != ap) {
1513                 list_for_each_entry(list_p, &ap->list, list)
1514                         if (list_p == p)
1515                         /* kprobe p is a valid probe */
1516                                 goto valid;
1517                 return NULL;
1518         }
1519 valid:
1520         return ap;
1521 }
1522 
1523 /*
1524  * Warn and return error if the kprobe is being re-registered since
1525  * there must be a software bug.
1526  */
1527 static inline int warn_kprobe_rereg(struct kprobe *p)
1528 {
1529         int ret = 0;
1530 
1531         mutex_lock(&kprobe_mutex);
1532         if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1533                 ret = -EINVAL;
1534         mutex_unlock(&kprobe_mutex);
1535 
1536         return ret;
1537 }
1538 
1539 static int check_ftrace_location(struct kprobe *p)
1540 {
1541         unsigned long addr = (unsigned long)p->addr;
1542 
1543         if (ftrace_location(addr) == addr) {
1544 #ifdef CONFIG_KPROBES_ON_FTRACE
1545                 p->flags |= KPROBE_FLAG_FTRACE;
1546 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1547                 return -EINVAL;
1548 #endif
1549         }
1550         return 0;
1551 }
1552 
1553 static bool is_cfi_preamble_symbol(unsigned long addr)
1554 {
1555         char symbuf[KSYM_NAME_LEN];
1556 
1557         if (lookup_symbol_name(addr, symbuf))
1558                 return false;
1559 
1560         return str_has_prefix(symbuf, "__cfi_") ||
1561                 str_has_prefix(symbuf, "__pfx_");
1562 }
1563 
1564 static int check_kprobe_address_safe(struct kprobe *p,
1565                                      struct module **probed_mod)
1566 {
1567         int ret;
1568 
1569         ret = check_ftrace_location(p);
1570         if (ret)
1571                 return ret;
1572         jump_label_lock();
1573         preempt_disable();
1574 
1575         /* Ensure the address is in a text area, and find a module if exists. */
1576         *probed_mod = NULL;
1577         if (!core_kernel_text((unsigned long) p->addr)) {
1578                 *probed_mod = __module_text_address((unsigned long) p->addr);
1579                 if (!(*probed_mod)) {
1580                         ret = -EINVAL;
1581                         goto out;
1582                 }
1583         }
1584         /* Ensure it is not in reserved area. */
1585         if (in_gate_area_no_mm((unsigned long) p->addr) ||
1586             within_kprobe_blacklist((unsigned long) p->addr) ||
1587             jump_label_text_reserved(p->addr, p->addr) ||
1588             static_call_text_reserved(p->addr, p->addr) ||
1589             find_bug((unsigned long)p->addr) ||
1590             is_cfi_preamble_symbol((unsigned long)p->addr)) {
1591                 ret = -EINVAL;
1592                 goto out;
1593         }
1594 
1595         /* Get module refcount and reject __init functions for loaded modules. */
1596         if (IS_ENABLED(CONFIG_MODULES) && *probed_mod) {
1597                 /*
1598                  * We must hold a refcount of the probed module while updating
1599                  * its code to prohibit unexpected unloading.
1600                  */
1601                 if (unlikely(!try_module_get(*probed_mod))) {
1602                         ret = -ENOENT;
1603                         goto out;
1604                 }
1605 
1606                 /*
1607                  * If the module freed '.init.text', we couldn't insert
1608                  * kprobes in there.
1609                  */
1610                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1611                     !module_is_coming(*probed_mod)) {
1612                         module_put(*probed_mod);
1613                         *probed_mod = NULL;
1614                         ret = -ENOENT;
1615                 }
1616         }
1617 
1618 out:
1619         preempt_enable();
1620         jump_label_unlock();
1621 
1622         return ret;
1623 }
1624 
1625 int register_kprobe(struct kprobe *p)
1626 {
1627         int ret;
1628         struct kprobe *old_p;
1629         struct module *probed_mod;
1630         kprobe_opcode_t *addr;
1631         bool on_func_entry;
1632 
1633         /* Adjust probe address from symbol */
1634         addr = _kprobe_addr(p->addr, p->symbol_name, p->offset, &on_func_entry);
1635         if (IS_ERR(addr))
1636                 return PTR_ERR(addr);
1637         p->addr = addr;
1638 
1639         ret = warn_kprobe_rereg(p);
1640         if (ret)
1641                 return ret;
1642 
1643         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1644         p->flags &= KPROBE_FLAG_DISABLED;
1645         p->nmissed = 0;
1646         INIT_LIST_HEAD(&p->list);
1647 
1648         ret = check_kprobe_address_safe(p, &probed_mod);
1649         if (ret)
1650                 return ret;
1651 
1652         mutex_lock(&kprobe_mutex);
1653 
1654         if (on_func_entry)
1655                 p->flags |= KPROBE_FLAG_ON_FUNC_ENTRY;
1656 
1657         old_p = get_kprobe(p->addr);
1658         if (old_p) {
1659                 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1660                 ret = register_aggr_kprobe(old_p, p);
1661                 goto out;
1662         }
1663 
1664         cpus_read_lock();
1665         /* Prevent text modification */
1666         mutex_lock(&text_mutex);
1667         ret = prepare_kprobe(p);
1668         mutex_unlock(&text_mutex);
1669         cpus_read_unlock();
1670         if (ret)
1671                 goto out;
1672 
1673         INIT_HLIST_NODE(&p->hlist);
1674         hlist_add_head_rcu(&p->hlist,
1675                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1676 
1677         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1678                 ret = arm_kprobe(p);
1679                 if (ret) {
1680                         hlist_del_rcu(&p->hlist);
1681                         synchronize_rcu();
1682                         goto out;
1683                 }
1684         }
1685 
1686         /* Try to optimize kprobe */
1687         try_to_optimize_kprobe(p);
1688 out:
1689         mutex_unlock(&kprobe_mutex);
1690 
1691         if (probed_mod)
1692                 module_put(probed_mod);
1693 
1694         return ret;
1695 }
1696 EXPORT_SYMBOL_GPL(register_kprobe);
1697 
1698 /* Check if all probes on the 'ap' are disabled. */
1699 static bool aggr_kprobe_disabled(struct kprobe *ap)
1700 {
1701         struct kprobe *kp;
1702 
1703         lockdep_assert_held(&kprobe_mutex);
1704 
1705         list_for_each_entry(kp, &ap->list, list)
1706                 if (!kprobe_disabled(kp))
1707                         /*
1708                          * Since there is an active probe on the list,
1709                          * we can't disable this 'ap'.
1710                          */
1711                         return false;
1712 
1713         return true;
1714 }
1715 
1716 static struct kprobe *__disable_kprobe(struct kprobe *p)
1717 {
1718         struct kprobe *orig_p;
1719         int ret;
1720 
1721         lockdep_assert_held(&kprobe_mutex);
1722 
1723         /* Get an original kprobe for return */
1724         orig_p = __get_valid_kprobe(p);
1725         if (unlikely(orig_p == NULL))
1726                 return ERR_PTR(-EINVAL);
1727 
1728         if (!kprobe_disabled(p)) {
1729                 /* Disable probe if it is a child probe */
1730                 if (p != orig_p)
1731                         p->flags |= KPROBE_FLAG_DISABLED;
1732 
1733                 /* Try to disarm and disable this/parent probe */
1734                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1735                         /*
1736                          * Don't be lazy here.  Even if 'kprobes_all_disarmed'
1737                          * is false, 'orig_p' might not have been armed yet.
1738                          * Note arm_all_kprobes() __tries__ to arm all kprobes
1739                          * on the best effort basis.
1740                          */
1741                         if (!kprobes_all_disarmed && !kprobe_disabled(orig_p)) {
1742                                 ret = disarm_kprobe(orig_p, true);
1743                                 if (ret) {
1744                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1745                                         return ERR_PTR(ret);
1746                                 }
1747                         }
1748                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1749                 }
1750         }
1751 
1752         return orig_p;
1753 }
1754 
1755 /*
1756  * Unregister a kprobe without a scheduler synchronization.
1757  */
1758 static int __unregister_kprobe_top(struct kprobe *p)
1759 {
1760         struct kprobe *ap, *list_p;
1761 
1762         /* Disable kprobe. This will disarm it if needed. */
1763         ap = __disable_kprobe(p);
1764         if (IS_ERR(ap))
1765                 return PTR_ERR(ap);
1766 
1767         if (ap == p)
1768                 /*
1769                  * This probe is an independent(and non-optimized) kprobe
1770                  * (not an aggrprobe). Remove from the hash list.
1771                  */
1772                 goto disarmed;
1773 
1774         /* Following process expects this probe is an aggrprobe */
1775         WARN_ON(!kprobe_aggrprobe(ap));
1776 
1777         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1778                 /*
1779                  * !disarmed could be happen if the probe is under delayed
1780                  * unoptimizing.
1781                  */
1782                 goto disarmed;
1783         else {
1784                 /* If disabling probe has special handlers, update aggrprobe */
1785                 if (p->post_handler && !kprobe_gone(p)) {
1786                         list_for_each_entry(list_p, &ap->list, list) {
1787                                 if ((list_p != p) && (list_p->post_handler))
1788                                         goto noclean;
1789                         }
1790                         /*
1791                          * For the kprobe-on-ftrace case, we keep the
1792                          * post_handler setting to identify this aggrprobe
1793                          * armed with kprobe_ipmodify_ops.
1794                          */
1795                         if (!kprobe_ftrace(ap))
1796                                 ap->post_handler = NULL;
1797                 }
1798 noclean:
1799                 /*
1800                  * Remove from the aggrprobe: this path will do nothing in
1801                  * __unregister_kprobe_bottom().
1802                  */
1803                 list_del_rcu(&p->list);
1804                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1805                         /*
1806                          * Try to optimize this probe again, because post
1807                          * handler may have been changed.
1808                          */
1809                         optimize_kprobe(ap);
1810         }
1811         return 0;
1812 
1813 disarmed:
1814         hlist_del_rcu(&ap->hlist);
1815         return 0;
1816 }
1817 
1818 static void __unregister_kprobe_bottom(struct kprobe *p)
1819 {
1820         struct kprobe *ap;
1821 
1822         if (list_empty(&p->list))
1823                 /* This is an independent kprobe */
1824                 arch_remove_kprobe(p);
1825         else if (list_is_singular(&p->list)) {
1826                 /* This is the last child of an aggrprobe */
1827                 ap = list_entry(p->list.next, struct kprobe, list);
1828                 list_del(&p->list);
1829                 free_aggr_kprobe(ap);
1830         }
1831         /* Otherwise, do nothing. */
1832 }
1833 
1834 int register_kprobes(struct kprobe **kps, int num)
1835 {
1836         int i, ret = 0;
1837 
1838         if (num <= 0)
1839                 return -EINVAL;
1840         for (i = 0; i < num; i++) {
1841                 ret = register_kprobe(kps[i]);
1842                 if (ret < 0) {
1843                         if (i > 0)
1844                                 unregister_kprobes(kps, i);
1845                         break;
1846                 }
1847         }
1848         return ret;
1849 }
1850 EXPORT_SYMBOL_GPL(register_kprobes);
1851 
1852 void unregister_kprobe(struct kprobe *p)
1853 {
1854         unregister_kprobes(&p, 1);
1855 }
1856 EXPORT_SYMBOL_GPL(unregister_kprobe);
1857 
1858 void unregister_kprobes(struct kprobe **kps, int num)
1859 {
1860         int i;
1861 
1862         if (num <= 0)
1863                 return;
1864         mutex_lock(&kprobe_mutex);
1865         for (i = 0; i < num; i++)
1866                 if (__unregister_kprobe_top(kps[i]) < 0)
1867                         kps[i]->addr = NULL;
1868         mutex_unlock(&kprobe_mutex);
1869 
1870         synchronize_rcu();
1871         for (i = 0; i < num; i++)
1872                 if (kps[i]->addr)
1873                         __unregister_kprobe_bottom(kps[i]);
1874 }
1875 EXPORT_SYMBOL_GPL(unregister_kprobes);
1876 
1877 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1878                                         unsigned long val, void *data)
1879 {
1880         return NOTIFY_DONE;
1881 }
1882 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1883 
1884 static struct notifier_block kprobe_exceptions_nb = {
1885         .notifier_call = kprobe_exceptions_notify,
1886         .priority = 0x7fffffff /* we need to be notified first */
1887 };
1888 
1889 #ifdef CONFIG_KRETPROBES
1890 
1891 #if !defined(CONFIG_KRETPROBE_ON_RETHOOK)
1892 
1893 /* callbacks for objpool of kretprobe instances */
1894 static int kretprobe_init_inst(void *nod, void *context)
1895 {
1896         struct kretprobe_instance *ri = nod;
1897 
1898         ri->rph = context;
1899         return 0;
1900 }
1901 static int kretprobe_fini_pool(struct objpool_head *head, void *context)
1902 {
1903         kfree(context);
1904         return 0;
1905 }
1906 
1907 static void free_rp_inst_rcu(struct rcu_head *head)
1908 {
1909         struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1910         struct kretprobe_holder *rph = ri->rph;
1911 
1912         objpool_drop(ri, &rph->pool);
1913 }
1914 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1915 
1916 static void recycle_rp_inst(struct kretprobe_instance *ri)
1917 {
1918         struct kretprobe *rp = get_kretprobe(ri);
1919 
1920         if (likely(rp))
1921                 objpool_push(ri, &rp->rph->pool);
1922         else
1923                 call_rcu(&ri->rcu, free_rp_inst_rcu);
1924 }
1925 NOKPROBE_SYMBOL(recycle_rp_inst);
1926 
1927 /*
1928  * This function is called from delayed_put_task_struct() when a task is
1929  * dead and cleaned up to recycle any kretprobe instances associated with
1930  * this task. These left over instances represent probed functions that
1931  * have been called but will never return.
1932  */
1933 void kprobe_flush_task(struct task_struct *tk)
1934 {
1935         struct kretprobe_instance *ri;
1936         struct llist_node *node;
1937 
1938         /* Early boot, not yet initialized. */
1939         if (unlikely(!kprobes_initialized))
1940                 return;
1941 
1942         kprobe_busy_begin();
1943 
1944         node = __llist_del_all(&tk->kretprobe_instances);
1945         while (node) {
1946                 ri = container_of(node, struct kretprobe_instance, llist);
1947                 node = node->next;
1948 
1949                 recycle_rp_inst(ri);
1950         }
1951 
1952         kprobe_busy_end();
1953 }
1954 NOKPROBE_SYMBOL(kprobe_flush_task);
1955 
1956 static inline void free_rp_inst(struct kretprobe *rp)
1957 {
1958         struct kretprobe_holder *rph = rp->rph;
1959 
1960         if (!rph)
1961                 return;
1962         rp->rph = NULL;
1963         objpool_fini(&rph->pool);
1964 }
1965 
1966 /* This assumes the 'tsk' is the current task or the is not running. */
1967 static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1968                                                   struct llist_node **cur)
1969 {
1970         struct kretprobe_instance *ri = NULL;
1971         struct llist_node *node = *cur;
1972 
1973         if (!node)
1974                 node = tsk->kretprobe_instances.first;
1975         else
1976                 node = node->next;
1977 
1978         while (node) {
1979                 ri = container_of(node, struct kretprobe_instance, llist);
1980                 if (ri->ret_addr != kretprobe_trampoline_addr()) {
1981                         *cur = node;
1982                         return ri->ret_addr;
1983                 }
1984                 node = node->next;
1985         }
1986         return NULL;
1987 }
1988 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1989 
1990 /**
1991  * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1992  * @tsk: Target task
1993  * @fp: A frame pointer
1994  * @cur: a storage of the loop cursor llist_node pointer for next call
1995  *
1996  * Find the correct return address modified by a kretprobe on @tsk in unsigned
1997  * long type. If it finds the return address, this returns that address value,
1998  * or this returns 0.
1999  * The @tsk must be 'current' or a task which is not running. @fp is a hint
2000  * to get the currect return address - which is compared with the
2001  * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
2002  * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
2003  * first call, but '@cur' itself must NOT NULL.
2004  */
2005 unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
2006                                       struct llist_node **cur)
2007 {
2008         struct kretprobe_instance *ri;
2009         kprobe_opcode_t *ret;
2010 
2011         if (WARN_ON_ONCE(!cur))
2012                 return 0;
2013 
2014         do {
2015                 ret = __kretprobe_find_ret_addr(tsk, cur);
2016                 if (!ret)
2017                         break;
2018                 ri = container_of(*cur, struct kretprobe_instance, llist);
2019         } while (ri->fp != fp);
2020 
2021         return (unsigned long)ret;
2022 }
2023 NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
2024 
2025 void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
2026                                         kprobe_opcode_t *correct_ret_addr)
2027 {
2028         /*
2029          * Do nothing by default. Please fill this to update the fake return
2030          * address on the stack with the correct one on each arch if possible.
2031          */
2032 }
2033 
2034 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
2035                                              void *frame_pointer)
2036 {
2037         struct kretprobe_instance *ri = NULL;
2038         struct llist_node *first, *node = NULL;
2039         kprobe_opcode_t *correct_ret_addr;
2040         struct kretprobe *rp;
2041 
2042         /* Find correct address and all nodes for this frame. */
2043         correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
2044         if (!correct_ret_addr) {
2045                 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
2046                 BUG_ON(1);
2047         }
2048 
2049         /*
2050          * Set the return address as the instruction pointer, because if the
2051          * user handler calls stack_trace_save_regs() with this 'regs',
2052          * the stack trace will start from the instruction pointer.
2053          */
2054         instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
2055 
2056         /* Run the user handler of the nodes. */
2057         first = current->kretprobe_instances.first;
2058         while (first) {
2059                 ri = container_of(first, struct kretprobe_instance, llist);
2060 
2061                 if (WARN_ON_ONCE(ri->fp != frame_pointer))
2062                         break;
2063 
2064                 rp = get_kretprobe(ri);
2065                 if (rp && rp->handler) {
2066                         struct kprobe *prev = kprobe_running();
2067 
2068                         __this_cpu_write(current_kprobe, &rp->kp);
2069                         ri->ret_addr = correct_ret_addr;
2070                         rp->handler(ri, regs);
2071                         __this_cpu_write(current_kprobe, prev);
2072                 }
2073                 if (first == node)
2074                         break;
2075 
2076                 first = first->next;
2077         }
2078 
2079         arch_kretprobe_fixup_return(regs, correct_ret_addr);
2080 
2081         /* Unlink all nodes for this frame. */
2082         first = current->kretprobe_instances.first;
2083         current->kretprobe_instances.first = node->next;
2084         node->next = NULL;
2085 
2086         /* Recycle free instances. */
2087         while (first) {
2088                 ri = container_of(first, struct kretprobe_instance, llist);
2089                 first = first->next;
2090 
2091                 recycle_rp_inst(ri);
2092         }
2093 
2094         return (unsigned long)correct_ret_addr;
2095 }
2096 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
2097 
2098 /*
2099  * This kprobe pre_handler is registered with every kretprobe. When probe
2100  * hits it will set up the return probe.
2101  */
2102 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2103 {
2104         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2105         struct kretprobe_holder *rph = rp->rph;
2106         struct kretprobe_instance *ri;
2107 
2108         ri = objpool_pop(&rph->pool);
2109         if (!ri) {
2110                 rp->nmissed++;
2111                 return 0;
2112         }
2113 
2114         if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2115                 objpool_push(ri, &rph->pool);
2116                 return 0;
2117         }
2118 
2119         arch_prepare_kretprobe(ri, regs);
2120 
2121         __llist_add(&ri->llist, &current->kretprobe_instances);
2122 
2123         return 0;
2124 }
2125 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2126 #else /* CONFIG_KRETPROBE_ON_RETHOOK */
2127 /*
2128  * This kprobe pre_handler is registered with every kretprobe. When probe
2129  * hits it will set up the return probe.
2130  */
2131 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2132 {
2133         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2134         struct kretprobe_instance *ri;
2135         struct rethook_node *rhn;
2136 
2137         rhn = rethook_try_get(rp->rh);
2138         if (!rhn) {
2139                 rp->nmissed++;
2140                 return 0;
2141         }
2142 
2143         ri = container_of(rhn, struct kretprobe_instance, node);
2144 
2145         if (rp->entry_handler && rp->entry_handler(ri, regs))
2146                 rethook_recycle(rhn);
2147         else
2148                 rethook_hook(rhn, regs, kprobe_ftrace(p));
2149 
2150         return 0;
2151 }
2152 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2153 
2154 static void kretprobe_rethook_handler(struct rethook_node *rh, void *data,
2155                                       unsigned long ret_addr,
2156                                       struct pt_regs *regs)
2157 {
2158         struct kretprobe *rp = (struct kretprobe *)data;
2159         struct kretprobe_instance *ri;
2160         struct kprobe_ctlblk *kcb;
2161 
2162         /* The data must NOT be null. This means rethook data structure is broken. */
2163         if (WARN_ON_ONCE(!data) || !rp->handler)
2164                 return;
2165 
2166         __this_cpu_write(current_kprobe, &rp->kp);
2167         kcb = get_kprobe_ctlblk();
2168         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
2169 
2170         ri = container_of(rh, struct kretprobe_instance, node);
2171         rp->handler(ri, regs);
2172 
2173         __this_cpu_write(current_kprobe, NULL);
2174 }
2175 NOKPROBE_SYMBOL(kretprobe_rethook_handler);
2176 
2177 #endif /* !CONFIG_KRETPROBE_ON_RETHOOK */
2178 
2179 /**
2180  * kprobe_on_func_entry() -- check whether given address is function entry
2181  * @addr: Target address
2182  * @sym:  Target symbol name
2183  * @offset: The offset from the symbol or the address
2184  *
2185  * This checks whether the given @addr+@offset or @sym+@offset is on the
2186  * function entry address or not.
2187  * This returns 0 if it is the function entry, or -EINVAL if it is not.
2188  * And also it returns -ENOENT if it fails the symbol or address lookup.
2189  * Caller must pass @addr or @sym (either one must be NULL), or this
2190  * returns -EINVAL.
2191  */
2192 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2193 {
2194         bool on_func_entry;
2195         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset, &on_func_entry);
2196 
2197         if (IS_ERR(kp_addr))
2198                 return PTR_ERR(kp_addr);
2199 
2200         if (!on_func_entry)
2201                 return -EINVAL;
2202 
2203         return 0;
2204 }
2205 
2206 int register_kretprobe(struct kretprobe *rp)
2207 {
2208         int ret;
2209         int i;
2210         void *addr;
2211 
2212         ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2213         if (ret)
2214                 return ret;
2215 
2216         /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2217         if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2218                 return -EINVAL;
2219 
2220         if (kretprobe_blacklist_size) {
2221                 addr = kprobe_addr(&rp->kp);
2222                 if (IS_ERR(addr))
2223                         return PTR_ERR(addr);
2224 
2225                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2226                         if (kretprobe_blacklist[i].addr == addr)
2227                                 return -EINVAL;
2228                 }
2229         }
2230 
2231         if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2232                 return -E2BIG;
2233 
2234         rp->kp.pre_handler = pre_handler_kretprobe;
2235         rp->kp.post_handler = NULL;
2236 
2237         /* Pre-allocate memory for max kretprobe instances */
2238         if (rp->maxactive <= 0)
2239                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2240 
2241 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2242         rp->rh = rethook_alloc((void *)rp, kretprobe_rethook_handler,
2243                                 sizeof(struct kretprobe_instance) +
2244                                 rp->data_size, rp->maxactive);
2245         if (IS_ERR(rp->rh))
2246                 return PTR_ERR(rp->rh);
2247 
2248         rp->nmissed = 0;
2249         /* Establish function entry probe point */
2250         ret = register_kprobe(&rp->kp);
2251         if (ret != 0) {
2252                 rethook_free(rp->rh);
2253                 rp->rh = NULL;
2254         }
2255 #else   /* !CONFIG_KRETPROBE_ON_RETHOOK */
2256         rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2257         if (!rp->rph)
2258                 return -ENOMEM;
2259 
2260         if (objpool_init(&rp->rph->pool, rp->maxactive, rp->data_size +
2261                         sizeof(struct kretprobe_instance), GFP_KERNEL,
2262                         rp->rph, kretprobe_init_inst, kretprobe_fini_pool)) {
2263                 kfree(rp->rph);
2264                 rp->rph = NULL;
2265                 return -ENOMEM;
2266         }
2267         rcu_assign_pointer(rp->rph->rp, rp);
2268         rp->nmissed = 0;
2269         /* Establish function entry probe point */
2270         ret = register_kprobe(&rp->kp);
2271         if (ret != 0)
2272                 free_rp_inst(rp);
2273 #endif
2274         return ret;
2275 }
2276 EXPORT_SYMBOL_GPL(register_kretprobe);
2277 
2278 int register_kretprobes(struct kretprobe **rps, int num)
2279 {
2280         int ret = 0, i;
2281 
2282         if (num <= 0)
2283                 return -EINVAL;
2284         for (i = 0; i < num; i++) {
2285                 ret = register_kretprobe(rps[i]);
2286                 if (ret < 0) {
2287                         if (i > 0)
2288                                 unregister_kretprobes(rps, i);
2289                         break;
2290                 }
2291         }
2292         return ret;
2293 }
2294 EXPORT_SYMBOL_GPL(register_kretprobes);
2295 
2296 void unregister_kretprobe(struct kretprobe *rp)
2297 {
2298         unregister_kretprobes(&rp, 1);
2299 }
2300 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2301 
2302 void unregister_kretprobes(struct kretprobe **rps, int num)
2303 {
2304         int i;
2305 
2306         if (num <= 0)
2307                 return;
2308         mutex_lock(&kprobe_mutex);
2309         for (i = 0; i < num; i++) {
2310                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2311                         rps[i]->kp.addr = NULL;
2312 #ifdef CONFIG_KRETPROBE_ON_RETHOOK
2313                 rethook_free(rps[i]->rh);
2314 #else
2315                 rcu_assign_pointer(rps[i]->rph->rp, NULL);
2316 #endif
2317         }
2318         mutex_unlock(&kprobe_mutex);
2319 
2320         synchronize_rcu();
2321         for (i = 0; i < num; i++) {
2322                 if (rps[i]->kp.addr) {
2323                         __unregister_kprobe_bottom(&rps[i]->kp);
2324 #ifndef CONFIG_KRETPROBE_ON_RETHOOK
2325                         free_rp_inst(rps[i]);
2326 #endif
2327                 }
2328         }
2329 }
2330 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2331 
2332 #else /* CONFIG_KRETPROBES */
2333 int register_kretprobe(struct kretprobe *rp)
2334 {
2335         return -EOPNOTSUPP;
2336 }
2337 EXPORT_SYMBOL_GPL(register_kretprobe);
2338 
2339 int register_kretprobes(struct kretprobe **rps, int num)
2340 {
2341         return -EOPNOTSUPP;
2342 }
2343 EXPORT_SYMBOL_GPL(register_kretprobes);
2344 
2345 void unregister_kretprobe(struct kretprobe *rp)
2346 {
2347 }
2348 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2349 
2350 void unregister_kretprobes(struct kretprobe **rps, int num)
2351 {
2352 }
2353 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2354 
2355 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2356 {
2357         return 0;
2358 }
2359 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2360 
2361 #endif /* CONFIG_KRETPROBES */
2362 
2363 /* Set the kprobe gone and remove its instruction buffer. */
2364 static void kill_kprobe(struct kprobe *p)
2365 {
2366         struct kprobe *kp;
2367 
2368         lockdep_assert_held(&kprobe_mutex);
2369 
2370         /*
2371          * The module is going away. We should disarm the kprobe which
2372          * is using ftrace, because ftrace framework is still available at
2373          * 'MODULE_STATE_GOING' notification.
2374          */
2375         if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2376                 disarm_kprobe_ftrace(p);
2377 
2378         p->flags |= KPROBE_FLAG_GONE;
2379         if (kprobe_aggrprobe(p)) {
2380                 /*
2381                  * If this is an aggr_kprobe, we have to list all the
2382                  * chained probes and mark them GONE.
2383                  */
2384                 list_for_each_entry(kp, &p->list, list)
2385                         kp->flags |= KPROBE_FLAG_GONE;
2386                 p->post_handler = NULL;
2387                 kill_optimized_kprobe(p);
2388         }
2389         /*
2390          * Here, we can remove insn_slot safely, because no thread calls
2391          * the original probed function (which will be freed soon) any more.
2392          */
2393         arch_remove_kprobe(p);
2394 }
2395 
2396 /* Disable one kprobe */
2397 int disable_kprobe(struct kprobe *kp)
2398 {
2399         int ret = 0;
2400         struct kprobe *p;
2401 
2402         mutex_lock(&kprobe_mutex);
2403 
2404         /* Disable this kprobe */
2405         p = __disable_kprobe(kp);
2406         if (IS_ERR(p))
2407                 ret = PTR_ERR(p);
2408 
2409         mutex_unlock(&kprobe_mutex);
2410         return ret;
2411 }
2412 EXPORT_SYMBOL_GPL(disable_kprobe);
2413 
2414 /* Enable one kprobe */
2415 int enable_kprobe(struct kprobe *kp)
2416 {
2417         int ret = 0;
2418         struct kprobe *p;
2419 
2420         mutex_lock(&kprobe_mutex);
2421 
2422         /* Check whether specified probe is valid. */
2423         p = __get_valid_kprobe(kp);
2424         if (unlikely(p == NULL)) {
2425                 ret = -EINVAL;
2426                 goto out;
2427         }
2428 
2429         if (kprobe_gone(kp)) {
2430                 /* This kprobe has gone, we couldn't enable it. */
2431                 ret = -EINVAL;
2432                 goto out;
2433         }
2434 
2435         if (p != kp)
2436                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2437 
2438         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2439                 p->flags &= ~KPROBE_FLAG_DISABLED;
2440                 ret = arm_kprobe(p);
2441                 if (ret) {
2442                         p->flags |= KPROBE_FLAG_DISABLED;
2443                         if (p != kp)
2444                                 kp->flags |= KPROBE_FLAG_DISABLED;
2445                 }
2446         }
2447 out:
2448         mutex_unlock(&kprobe_mutex);
2449         return ret;
2450 }
2451 EXPORT_SYMBOL_GPL(enable_kprobe);
2452 
2453 /* Caller must NOT call this in usual path. This is only for critical case */
2454 void dump_kprobe(struct kprobe *kp)
2455 {
2456         pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2457                kp->symbol_name, kp->offset, kp->addr);
2458 }
2459 NOKPROBE_SYMBOL(dump_kprobe);
2460 
2461 int kprobe_add_ksym_blacklist(unsigned long entry)
2462 {
2463         struct kprobe_blacklist_entry *ent;
2464         unsigned long offset = 0, size = 0;
2465 
2466         if (!kernel_text_address(entry) ||
2467             !kallsyms_lookup_size_offset(entry, &size, &offset))
2468                 return -EINVAL;
2469 
2470         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2471         if (!ent)
2472                 return -ENOMEM;
2473         ent->start_addr = entry;
2474         ent->end_addr = entry + size;
2475         INIT_LIST_HEAD(&ent->list);
2476         list_add_tail(&ent->list, &kprobe_blacklist);
2477 
2478         return (int)size;
2479 }
2480 
2481 /* Add all symbols in given area into kprobe blacklist */
2482 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2483 {
2484         unsigned long entry;
2485         int ret = 0;
2486 
2487         for (entry = start; entry < end; entry += ret) {
2488                 ret = kprobe_add_ksym_blacklist(entry);
2489                 if (ret < 0)
2490                         return ret;
2491                 if (ret == 0)   /* In case of alias symbol */
2492                         ret = 1;
2493         }
2494         return 0;
2495 }
2496 
2497 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2498                                    char *type, char *sym)
2499 {
2500         return -ERANGE;
2501 }
2502 
2503 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2504                        char *sym)
2505 {
2506 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2507         if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2508                 return 0;
2509 #ifdef CONFIG_OPTPROBES
2510         if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2511                 return 0;
2512 #endif
2513 #endif
2514         if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2515                 return 0;
2516         return -ERANGE;
2517 }
2518 
2519 int __init __weak arch_populate_kprobe_blacklist(void)
2520 {
2521         return 0;
2522 }
2523 
2524 /*
2525  * Lookup and populate the kprobe_blacklist.
2526  *
2527  * Unlike the kretprobe blacklist, we'll need to determine
2528  * the range of addresses that belong to the said functions,
2529  * since a kprobe need not necessarily be at the beginning
2530  * of a function.
2531  */
2532 static int __init populate_kprobe_blacklist(unsigned long *start,
2533                                              unsigned long *end)
2534 {
2535         unsigned long entry;
2536         unsigned long *iter;
2537         int ret;
2538 
2539         for (iter = start; iter < end; iter++) {
2540                 entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2541                 ret = kprobe_add_ksym_blacklist(entry);
2542                 if (ret == -EINVAL)
2543                         continue;
2544                 if (ret < 0)
2545                         return ret;
2546         }
2547 
2548         /* Symbols in '__kprobes_text' are blacklisted */
2549         ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2550                                         (unsigned long)__kprobes_text_end);
2551         if (ret)
2552                 return ret;
2553 
2554         /* Symbols in 'noinstr' section are blacklisted */
2555         ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2556                                         (unsigned long)__noinstr_text_end);
2557 
2558         return ret ? : arch_populate_kprobe_blacklist();
2559 }
2560 
2561 #ifdef CONFIG_MODULES
2562 /* Remove all symbols in given area from kprobe blacklist */
2563 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2564 {
2565         struct kprobe_blacklist_entry *ent, *n;
2566 
2567         list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2568                 if (ent->start_addr < start || ent->start_addr >= end)
2569                         continue;
2570                 list_del(&ent->list);
2571                 kfree(ent);
2572         }
2573 }
2574 
2575 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2576 {
2577         kprobe_remove_area_blacklist(entry, entry + 1);
2578 }
2579 
2580 static void add_module_kprobe_blacklist(struct module *mod)
2581 {
2582         unsigned long start, end;
2583         int i;
2584 
2585         if (mod->kprobe_blacklist) {
2586                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2587                         kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2588         }
2589 
2590         start = (unsigned long)mod->kprobes_text_start;
2591         if (start) {
2592                 end = start + mod->kprobes_text_size;
2593                 kprobe_add_area_blacklist(start, end);
2594         }
2595 
2596         start = (unsigned long)mod->noinstr_text_start;
2597         if (start) {
2598                 end = start + mod->noinstr_text_size;
2599                 kprobe_add_area_blacklist(start, end);
2600         }
2601 }
2602 
2603 static void remove_module_kprobe_blacklist(struct module *mod)
2604 {
2605         unsigned long start, end;
2606         int i;
2607 
2608         if (mod->kprobe_blacklist) {
2609                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2610                         kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2611         }
2612 
2613         start = (unsigned long)mod->kprobes_text_start;
2614         if (start) {
2615                 end = start + mod->kprobes_text_size;
2616                 kprobe_remove_area_blacklist(start, end);
2617         }
2618 
2619         start = (unsigned long)mod->noinstr_text_start;
2620         if (start) {
2621                 end = start + mod->noinstr_text_size;
2622                 kprobe_remove_area_blacklist(start, end);
2623         }
2624 }
2625 
2626 /* Module notifier call back, checking kprobes on the module */
2627 static int kprobes_module_callback(struct notifier_block *nb,
2628                                    unsigned long val, void *data)
2629 {
2630         struct module *mod = data;
2631         struct hlist_head *head;
2632         struct kprobe *p;
2633         unsigned int i;
2634         int checkcore = (val == MODULE_STATE_GOING);
2635 
2636         if (val == MODULE_STATE_COMING) {
2637                 mutex_lock(&kprobe_mutex);
2638                 add_module_kprobe_blacklist(mod);
2639                 mutex_unlock(&kprobe_mutex);
2640         }
2641         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2642                 return NOTIFY_DONE;
2643 
2644         /*
2645          * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2646          * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2647          * notified, only '.init.text' section would be freed. We need to
2648          * disable kprobes which have been inserted in the sections.
2649          */
2650         mutex_lock(&kprobe_mutex);
2651         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2652                 head = &kprobe_table[i];
2653                 hlist_for_each_entry(p, head, hlist)
2654                         if (within_module_init((unsigned long)p->addr, mod) ||
2655                             (checkcore &&
2656                              within_module_core((unsigned long)p->addr, mod))) {
2657                                 /*
2658                                  * The vaddr this probe is installed will soon
2659                                  * be vfreed buy not synced to disk. Hence,
2660                                  * disarming the breakpoint isn't needed.
2661                                  *
2662                                  * Note, this will also move any optimized probes
2663                                  * that are pending to be removed from their
2664                                  * corresponding lists to the 'freeing_list' and
2665                                  * will not be touched by the delayed
2666                                  * kprobe_optimizer() work handler.
2667                                  */
2668                                 kill_kprobe(p);
2669                         }
2670         }
2671         if (val == MODULE_STATE_GOING)
2672                 remove_module_kprobe_blacklist(mod);
2673         mutex_unlock(&kprobe_mutex);
2674         return NOTIFY_DONE;
2675 }
2676 
2677 static struct notifier_block kprobe_module_nb = {
2678         .notifier_call = kprobes_module_callback,
2679         .priority = 0
2680 };
2681 
2682 static int kprobe_register_module_notifier(void)
2683 {
2684         return register_module_notifier(&kprobe_module_nb);
2685 }
2686 #else
2687 static int kprobe_register_module_notifier(void)
2688 {
2689         return 0;
2690 }
2691 #endif /* CONFIG_MODULES */
2692 
2693 void kprobe_free_init_mem(void)
2694 {
2695         void *start = (void *)(&__init_begin);
2696         void *end = (void *)(&__init_end);
2697         struct hlist_head *head;
2698         struct kprobe *p;
2699         int i;
2700 
2701         mutex_lock(&kprobe_mutex);
2702 
2703         /* Kill all kprobes on initmem because the target code has been freed. */
2704         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2705                 head = &kprobe_table[i];
2706                 hlist_for_each_entry(p, head, hlist) {
2707                         if (start <= (void *)p->addr && (void *)p->addr < end)
2708                                 kill_kprobe(p);
2709                 }
2710         }
2711 
2712         mutex_unlock(&kprobe_mutex);
2713 }
2714 
2715 static int __init init_kprobes(void)
2716 {
2717         int i, err;
2718 
2719         /* FIXME allocate the probe table, currently defined statically */
2720         /* initialize all list heads */
2721         for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2722                 INIT_HLIST_HEAD(&kprobe_table[i]);
2723 
2724         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2725                                         __stop_kprobe_blacklist);
2726         if (err)
2727                 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2728 
2729         if (kretprobe_blacklist_size) {
2730                 /* lookup the function address from its name */
2731                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2732                         kretprobe_blacklist[i].addr =
2733                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2734                         if (!kretprobe_blacklist[i].addr)
2735                                 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2736                                        kretprobe_blacklist[i].name);
2737                 }
2738         }
2739 
2740         /* By default, kprobes are armed */
2741         kprobes_all_disarmed = false;
2742 
2743 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2744         /* Init 'kprobe_optinsn_slots' for allocation */
2745         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2746 #endif
2747 
2748         err = arch_init_kprobes();
2749         if (!err)
2750                 err = register_die_notifier(&kprobe_exceptions_nb);
2751         if (!err)
2752                 err = kprobe_register_module_notifier();
2753 
2754         kprobes_initialized = (err == 0);
2755         kprobe_sysctls_init();
2756         return err;
2757 }
2758 early_initcall(init_kprobes);
2759 
2760 #if defined(CONFIG_OPTPROBES)
2761 static int __init init_optprobes(void)
2762 {
2763         /*
2764          * Enable kprobe optimization - this kicks the optimizer which
2765          * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2766          * not spawned in early initcall. So delay the optimization.
2767          */
2768         optimize_all_kprobes();
2769 
2770         return 0;
2771 }
2772 subsys_initcall(init_optprobes);
2773 #endif
2774 
2775 #ifdef CONFIG_DEBUG_FS
2776 static void report_probe(struct seq_file *pi, struct kprobe *p,
2777                 const char *sym, int offset, char *modname, struct kprobe *pp)
2778 {
2779         char *kprobe_type;
2780         void *addr = p->addr;
2781 
2782         if (p->pre_handler == pre_handler_kretprobe)
2783                 kprobe_type = "r";
2784         else
2785                 kprobe_type = "k";
2786 
2787         if (!kallsyms_show_value(pi->file->f_cred))
2788                 addr = NULL;
2789 
2790         if (sym)
2791                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2792                         addr, kprobe_type, sym, offset,
2793                         (modname ? modname : " "));
2794         else    /* try to use %pS */
2795                 seq_printf(pi, "%px  %s  %pS ",
2796                         addr, kprobe_type, p->addr);
2797 
2798         if (!pp)
2799                 pp = p;
2800         seq_printf(pi, "%s%s%s%s\n",
2801                 (kprobe_gone(p) ? "[GONE]" : ""),
2802                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2803                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2804                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2805 }
2806 
2807 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2808 {
2809         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2810 }
2811 
2812 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2813 {
2814         (*pos)++;
2815         if (*pos >= KPROBE_TABLE_SIZE)
2816                 return NULL;
2817         return pos;
2818 }
2819 
2820 static void kprobe_seq_stop(struct seq_file *f, void *v)
2821 {
2822         /* Nothing to do */
2823 }
2824 
2825 static int show_kprobe_addr(struct seq_file *pi, void *v)
2826 {
2827         struct hlist_head *head;
2828         struct kprobe *p, *kp;
2829         const char *sym;
2830         unsigned int i = *(loff_t *) v;
2831         unsigned long offset = 0;
2832         char *modname, namebuf[KSYM_NAME_LEN];
2833 
2834         head = &kprobe_table[i];
2835         preempt_disable();
2836         hlist_for_each_entry_rcu(p, head, hlist) {
2837                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2838                                         &offset, &modname, namebuf);
2839                 if (kprobe_aggrprobe(p)) {
2840                         list_for_each_entry_rcu(kp, &p->list, list)
2841                                 report_probe(pi, kp, sym, offset, modname, p);
2842                 } else
2843                         report_probe(pi, p, sym, offset, modname, NULL);
2844         }
2845         preempt_enable();
2846         return 0;
2847 }
2848 
2849 static const struct seq_operations kprobes_sops = {
2850         .start = kprobe_seq_start,
2851         .next  = kprobe_seq_next,
2852         .stop  = kprobe_seq_stop,
2853         .show  = show_kprobe_addr
2854 };
2855 
2856 DEFINE_SEQ_ATTRIBUTE(kprobes);
2857 
2858 /* kprobes/blacklist -- shows which functions can not be probed */
2859 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2860 {
2861         mutex_lock(&kprobe_mutex);
2862         return seq_list_start(&kprobe_blacklist, *pos);
2863 }
2864 
2865 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2866 {
2867         return seq_list_next(v, &kprobe_blacklist, pos);
2868 }
2869 
2870 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2871 {
2872         struct kprobe_blacklist_entry *ent =
2873                 list_entry(v, struct kprobe_blacklist_entry, list);
2874 
2875         /*
2876          * If '/proc/kallsyms' is not showing kernel address, we won't
2877          * show them here either.
2878          */
2879         if (!kallsyms_show_value(m->file->f_cred))
2880                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2881                            (void *)ent->start_addr);
2882         else
2883                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2884                            (void *)ent->end_addr, (void *)ent->start_addr);
2885         return 0;
2886 }
2887 
2888 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2889 {
2890         mutex_unlock(&kprobe_mutex);
2891 }
2892 
2893 static const struct seq_operations kprobe_blacklist_sops = {
2894         .start = kprobe_blacklist_seq_start,
2895         .next  = kprobe_blacklist_seq_next,
2896         .stop  = kprobe_blacklist_seq_stop,
2897         .show  = kprobe_blacklist_seq_show,
2898 };
2899 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2900 
2901 static int arm_all_kprobes(void)
2902 {
2903         struct hlist_head *head;
2904         struct kprobe *p;
2905         unsigned int i, total = 0, errors = 0;
2906         int err, ret = 0;
2907 
2908         mutex_lock(&kprobe_mutex);
2909 
2910         /* If kprobes are armed, just return */
2911         if (!kprobes_all_disarmed)
2912                 goto already_enabled;
2913 
2914         /*
2915          * optimize_kprobe() called by arm_kprobe() checks
2916          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2917          * arm_kprobe.
2918          */
2919         kprobes_all_disarmed = false;
2920         /* Arming kprobes doesn't optimize kprobe itself */
2921         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2922                 head = &kprobe_table[i];
2923                 /* Arm all kprobes on a best-effort basis */
2924                 hlist_for_each_entry(p, head, hlist) {
2925                         if (!kprobe_disabled(p)) {
2926                                 err = arm_kprobe(p);
2927                                 if (err)  {
2928                                         errors++;
2929                                         ret = err;
2930                                 }
2931                                 total++;
2932                         }
2933                 }
2934         }
2935 
2936         if (errors)
2937                 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2938                         errors, total);
2939         else
2940                 pr_info("Kprobes globally enabled\n");
2941 
2942 already_enabled:
2943         mutex_unlock(&kprobe_mutex);
2944         return ret;
2945 }
2946 
2947 static int disarm_all_kprobes(void)
2948 {
2949         struct hlist_head *head;
2950         struct kprobe *p;
2951         unsigned int i, total = 0, errors = 0;
2952         int err, ret = 0;
2953 
2954         mutex_lock(&kprobe_mutex);
2955 
2956         /* If kprobes are already disarmed, just return */
2957         if (kprobes_all_disarmed) {
2958                 mutex_unlock(&kprobe_mutex);
2959                 return 0;
2960         }
2961 
2962         kprobes_all_disarmed = true;
2963 
2964         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2965                 head = &kprobe_table[i];
2966                 /* Disarm all kprobes on a best-effort basis */
2967                 hlist_for_each_entry(p, head, hlist) {
2968                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2969                                 err = disarm_kprobe(p, false);
2970                                 if (err) {
2971                                         errors++;
2972                                         ret = err;
2973                                 }
2974                                 total++;
2975                         }
2976                 }
2977         }
2978 
2979         if (errors)
2980                 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2981                         errors, total);
2982         else
2983                 pr_info("Kprobes globally disabled\n");
2984 
2985         mutex_unlock(&kprobe_mutex);
2986 
2987         /* Wait for disarming all kprobes by optimizer */
2988         wait_for_kprobe_optimizer();
2989 
2990         return ret;
2991 }
2992 
2993 /*
2994  * XXX: The debugfs bool file interface doesn't allow for callbacks
2995  * when the bool state is switched. We can reuse that facility when
2996  * available
2997  */
2998 static ssize_t read_enabled_file_bool(struct file *file,
2999                char __user *user_buf, size_t count, loff_t *ppos)
3000 {
3001         char buf[3];
3002 
3003         if (!kprobes_all_disarmed)
3004                 buf[0] = '1';
3005         else
3006                 buf[0] = '';
3007         buf[1] = '\n';
3008         buf[2] = 0x00;
3009         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
3010 }
3011 
3012 static ssize_t write_enabled_file_bool(struct file *file,
3013                const char __user *user_buf, size_t count, loff_t *ppos)
3014 {
3015         bool enable;
3016         int ret;
3017 
3018         ret = kstrtobool_from_user(user_buf, count, &enable);
3019         if (ret)
3020                 return ret;
3021 
3022         ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
3023         if (ret)
3024                 return ret;
3025 
3026         return count;
3027 }
3028 
3029 static const struct file_operations fops_kp = {
3030         .read =         read_enabled_file_bool,
3031         .write =        write_enabled_file_bool,
3032         .llseek =       default_llseek,
3033 };
3034 
3035 static int __init debugfs_kprobe_init(void)
3036 {
3037         struct dentry *dir;
3038 
3039         dir = debugfs_create_dir("kprobes", NULL);
3040 
3041         debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
3042 
3043         debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
3044 
3045         debugfs_create_file("blacklist", 0400, dir, NULL,
3046                             &kprobe_blacklist_fops);
3047 
3048         return 0;
3049 }
3050 
3051 late_initcall(debugfs_kprobe_init);
3052 #endif /* CONFIG_DEBUG_FS */
3053 

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

kernel.org | git.kernel.org | LWN.net | Project Home | 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.

sflogo.php