kernel/kprobes.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Kernel Probes (KProbes)
   4  *  kernel/kprobes.c
   5  *
   6  * Copyright (C) IBM Corporation, 2002, 2004
   7  *
   8  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
   9  *              Probes initial implementation (includes suggestions from
  10  *              Rusty Russell).
  11  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  12  *              hlists and exceptions notifier as suggested by Andi Kleen.
  13  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  14  *              interface to access function arguments.
  15  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  16  *              exceptions notifier to be first on the priority list.
  17  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  18  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  19  *              <prasanna@in.ibm.com> added function-return probes.
  20  */
  21 #include <linux/kprobes.h>
  22 #include <linux/hash.h>
  23 #include <linux/init.h>
  24 #include <linux/slab.h>
  25 #include <linux/stddef.h>
  26 #include <linux/export.h>
  27 #include <linux/moduleloader.h>
  28 #include <linux/kallsyms.h>
  29 #include <linux/freezer.h>
  30 #include <linux/seq_file.h>
  31 #include <linux/debugfs.h>
  32 #include <linux/sysctl.h>
  33 #include <linux/kdebug.h>
  34 #include <linux/memory.h>
  35 #include <linux/ftrace.h>
  36 #include <linux/cpu.h>
  37 #include <linux/jump_label.h>
  38 #include <linux/perf_event.h>
  39
  40 #include <asm/sections.h>
  41 #include <asm/cacheflush.h>
  42 #include <asm/errno.h>
  43 #include <linux/uaccess.h>
  44
  45 #define KPROBE_HASH_BITS 6
  46 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  47
  48
  49 static int kprobes_initialized;
  50 /* kprobe_table can be accessed by
  51  * - Normal hlist traversal and RCU add/del under kprobe_mutex is held.
  52  * Or
  53  * - RCU hlist traversal under disabling preempt (breakpoint handlers)
  54  */
  55 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  56 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  57
  58 /* NOTE: change this value only with kprobe_mutex held */
  59 static bool kprobes_all_disarmed;
  60
  61 /* This protects kprobe_table and optimizing_list */
  62 static DEFINE_MUTEX(kprobe_mutex);
  63 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  64 static struct {
  65         raw_spinlock_t lock ____cacheline_aligned_in_smp;
  66 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  67
  68 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  69                                         unsigned int __unused)
  70 {
  71         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  72 }
  73
  74 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  75 {
  76         return &(kretprobe_table_locks[hash].lock);
  77 }
  78
  79 /* Blacklist -- list of struct kprobe_blacklist_entry */
  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         return module_alloc(PAGE_SIZE);
 116 }
 117
 118 void __weak free_insn_page(void *page)
 119 {
 120         module_memfree(page);
 121 }
 122
 123 struct kprobe_insn_cache kprobe_insn_slots = {
 124         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 125         .alloc = alloc_insn_page,
 126         .free = free_insn_page,
 127         .sym = KPROBE_INSN_PAGE_SYM,
 128         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 129         .insn_size = MAX_INSN_SIZE,
 130         .nr_garbage = 0,
 131 };
 132 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 133
 134 /**
 135  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 136  * We allocate an executable page if there's no room on existing ones.
 137  */
 138 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 139 {
 140         struct kprobe_insn_page *kip;
 141         kprobe_opcode_t *slot = NULL;
 142
 143         /* Since the slot array is not protected by rcu, we need a mutex */
 144         mutex_lock(&c->mutex);
 145  retry:
 146         rcu_read_lock();
 147         list_for_each_entry_rcu(kip, &c->pages, list) {
 148                 if (kip->nused < slots_per_page(c)) {
 149                         int i;
 150                         for (i = 0; i < slots_per_page(c); i++) {
 151                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 152                                         kip->slot_used[i] = SLOT_USED;
 153                                         kip->nused++;
 154                                         slot = kip->insns + (i * c->insn_size);
 155                                         rcu_read_unlock();
 156                                         goto out;
 157                                 }
 158                         }
 159                         /* kip->nused is broken. Fix it. */
 160                         kip->nused = slots_per_page(c);
 161                         WARN_ON(1);
 162                 }
 163         }
 164         rcu_read_unlock();
 165
 166         /* If there are any garbage slots, collect it and try again. */
 167         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 168                 goto retry;
 169
 170         /* All out of space.  Need to allocate a new page. */
 171         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 172         if (!kip)
 173                 goto out;
 174
 175         /*
 176          * Use module_alloc so this page is within +/- 2GB of where the
 177          * kernel image and loaded module images reside. This is required
 178          * so x86_64 can correctly handle the %rip-relative fixups.
 179          */
 180         kip->insns = c->alloc();
 181         if (!kip->insns) {
 182                 kfree(kip);
 183                 goto out;
 184         }
 185         INIT_LIST_HEAD(&kip->list);
 186         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 187         kip->slot_used[0] = SLOT_USED;
 188         kip->nused = 1;
 189         kip->ngarbage = 0;
 190         kip->cache = c;
 191         list_add_rcu(&kip->list, &c->pages);
 192         slot = kip->insns;
 193
 194         /* Record the perf ksymbol register event after adding the page */
 195         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
 196                            PAGE_SIZE, false, c->sym);
 197 out:
 198         mutex_unlock(&c->mutex);
 199         return slot;
 200 }
 201
 202 /* Return 1 if all garbages are collected, otherwise 0. */
 203 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 204 {
 205         kip->slot_used[idx] = SLOT_CLEAN;
 206         kip->nused--;
 207         if (kip->nused == 0) {
 208                 /*
 209                  * Page is no longer in use.  Free it unless
 210                  * it's the last one.  We keep the last one
 211                  * so as not to have to set it up again the
 212                  * next time somebody inserts a probe.
 213                  */
 214                 if (!list_is_singular(&kip->list)) {
 215                         /*
 216                          * Record perf ksymbol unregister event before removing
 217                          * the page.
 218                          */
 219                         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
 220                                            (unsigned long)kip->insns, PAGE_SIZE, true,
 221                                            kip->cache->sym);
 222                         list_del_rcu(&kip->list);
 223                         synchronize_rcu();
 224                         kip->cache->free(kip->insns);
 225                         kfree(kip);
 226                 }
 227                 return 1;
 228         }
 229         return 0;
 230 }
 231
 232 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 233 {
 234         struct kprobe_insn_page *kip, *next;
 235
 236         /* Ensure no-one is interrupted on the garbages */
 237         synchronize_rcu();
 238
 239         list_for_each_entry_safe(kip, next, &c->pages, list) {
 240                 int i;
 241                 if (kip->ngarbage == 0)
 242                         continue;
 243                 kip->ngarbage = 0;      /* we will collect all garbages */
 244                 for (i = 0; i < slots_per_page(c); i++) {
 245                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 246                                 break;
 247                 }
 248         }
 249         c->nr_garbage = 0;
 250         return 0;
 251 }
 252
 253 void __free_insn_slot(struct kprobe_insn_cache *c,
 254                       kprobe_opcode_t *slot, int dirty)
 255 {
 256         struct kprobe_insn_page *kip;
 257         long idx;
 258
 259         mutex_lock(&c->mutex);
 260         rcu_read_lock();
 261         list_for_each_entry_rcu(kip, &c->pages, list) {
 262                 idx = ((long)slot - (long)kip->insns) /
 263                         (c->insn_size * sizeof(kprobe_opcode_t));
 264                 if (idx >= 0 && idx < slots_per_page(c))
 265                         goto out;
 266         }
 267         /* Could not find this slot. */
 268         WARN_ON(1);
 269         kip = NULL;
 270 out:
 271         rcu_read_unlock();
 272         /* Mark and sweep: this may sleep */
 273         if (kip) {
 274                 /* Check double free */
 275                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 276                 if (dirty) {
 277                         kip->slot_used[idx] = SLOT_DIRTY;
 278                         kip->ngarbage++;
 279                         if (++c->nr_garbage > slots_per_page(c))
 280                                 collect_garbage_slots(c);
 281                 } else {
 282                         collect_one_slot(kip, idx);
 283                 }
 284         }
 285         mutex_unlock(&c->mutex);
 286 }
 287
 288 /*
 289  * Check given address is on the page of kprobe instruction slots.
 290  * This will be used for checking whether the address on a stack
 291  * is on a text area or not.
 292  */
 293 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 294 {
 295         struct kprobe_insn_page *kip;
 296         bool ret = false;
 297
 298         rcu_read_lock();
 299         list_for_each_entry_rcu(kip, &c->pages, list) {
 300                 if (addr >= (unsigned long)kip->insns &&
 301                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 302                         ret = true;
 303                         break;
 304                 }
 305         }
 306         rcu_read_unlock();
 307
 308         return ret;
 309 }
 310
 311 int kprobe_cache_get_kallsym(struct kprobe_insn_cache *c, unsigned int *symnum,
 312                              unsigned long *value, char *type, char *sym)
 313 {
 314         struct kprobe_insn_page *kip;
 315         int ret = -ERANGE;
 316
 317         rcu_read_lock();
 318         list_for_each_entry_rcu(kip, &c->pages, list) {
 319                 if ((*symnum)--)
 320                         continue;
 321                 strlcpy(sym, c->sym, KSYM_NAME_LEN);
 322                 *type = 't';
 323                 *value = (unsigned long)kip->insns;
 324                 ret = 0;
 325                 break;
 326         }
 327         rcu_read_unlock();
 328
 329         return ret;
 330 }
 331
 332 #ifdef CONFIG_OPTPROBES
 333 /* For optimized_kprobe buffer */
 334 struct kprobe_insn_cache kprobe_optinsn_slots = {
 335         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 336         .alloc = alloc_insn_page,
 337         .free = free_insn_page,
 338         .sym = KPROBE_OPTINSN_PAGE_SYM,
 339         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 340         /* .insn_size is initialized later */
 341         .nr_garbage = 0,
 342 };
 343 #endif
 344 #endif
 345
 346 /* We have preemption disabled.. so it is safe to use __ versions */
 347 static inline void set_kprobe_instance(struct kprobe *kp)
 348 {
 349         __this_cpu_write(kprobe_instance, kp);
 350 }
 351
 352 static inline void reset_kprobe_instance(void)
 353 {
 354         __this_cpu_write(kprobe_instance, NULL);
 355 }
 356
 357 /*
 358  * This routine is called either:
 359  *      - under the kprobe_mutex - during kprobe_[un]register()
 360  *                              OR
 361  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 362  */
 363 struct kprobe *get_kprobe(void *addr)
 364 {
 365         struct hlist_head *head;
 366         struct kprobe *p;
 367
 368         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 369         hlist_for_each_entry_rcu(p, head, hlist,
 370                                  lockdep_is_held(&kprobe_mutex)) {
 371                 if (p->addr == addr)
 372                         return p;
 373         }
 374
 375         return NULL;
 376 }
 377 NOKPROBE_SYMBOL(get_kprobe);
 378
 379 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 380
 381 /* Return true if the kprobe is an aggregator */
 382 static inline int kprobe_aggrprobe(struct kprobe *p)
 383 {
 384         return p->pre_handler == aggr_pre_handler;
 385 }
 386
 387 /* Return true(!0) if the kprobe is unused */
 388 static inline int kprobe_unused(struct kprobe *p)
 389 {
 390         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 391                list_empty(&p->list);
 392 }
 393
 394 /*
 395  * Keep all fields in the kprobe consistent
 396  */
 397 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 398 {
 399         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 400         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 401 }
 402
 403 #ifdef CONFIG_OPTPROBES
 404 /* NOTE: change this value only with kprobe_mutex held */
 405 static bool kprobes_allow_optimization;
 406
 407 /*
 408  * Call all pre_handler on the list, but ignores its return value.
 409  * This must be called from arch-dep optimized caller.
 410  */
 411 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 412 {
 413         struct kprobe *kp;
 414
 415         list_for_each_entry_rcu(kp, &p->list, list) {
 416                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 417                         set_kprobe_instance(kp);
 418                         kp->pre_handler(kp, regs);
 419                 }
 420                 reset_kprobe_instance();
 421         }
 422 }
 423 NOKPROBE_SYMBOL(opt_pre_handler);
 424
 425 /* Free optimized instructions and optimized_kprobe */
 426 static void free_aggr_kprobe(struct kprobe *p)
 427 {
 428         struct optimized_kprobe *op;
 429
 430         op = container_of(p, struct optimized_kprobe, kp);
 431         arch_remove_optimized_kprobe(op);
 432         arch_remove_kprobe(p);
 433         kfree(op);
 434 }
 435
 436 /* Return true(!0) if the kprobe is ready for optimization. */
 437 static inline int kprobe_optready(struct kprobe *p)
 438 {
 439         struct optimized_kprobe *op;
 440
 441         if (kprobe_aggrprobe(p)) {
 442                 op = container_of(p, struct optimized_kprobe, kp);
 443                 return arch_prepared_optinsn(&op->optinsn);
 444         }
 445
 446         return 0;
 447 }
 448
 449 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 450 static inline int kprobe_disarmed(struct kprobe *p)
 451 {
 452         struct optimized_kprobe *op;
 453
 454         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 455         if (!kprobe_aggrprobe(p))
 456                 return kprobe_disabled(p);
 457
 458         op = container_of(p, struct optimized_kprobe, kp);
 459
 460         return kprobe_disabled(p) && list_empty(&op->list);
 461 }
 462
 463 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 464 static int kprobe_queued(struct kprobe *p)
 465 {
 466         struct optimized_kprobe *op;
 467
 468         if (kprobe_aggrprobe(p)) {
 469                 op = container_of(p, struct optimized_kprobe, kp);
 470                 if (!list_empty(&op->list))
 471                         return 1;
 472         }
 473         return 0;
 474 }
 475
 476 /*
 477  * Return an optimized kprobe whose optimizing code replaces
 478  * instructions including addr (exclude breakpoint).
 479  */
 480 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 481 {
 482         int i;
 483         struct kprobe *p = NULL;
 484         struct optimized_kprobe *op;
 485
 486         /* Don't check i == 0, since that is a breakpoint case. */
 487         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 488                 p = get_kprobe((void *)(addr - i));
 489
 490         if (p && kprobe_optready(p)) {
 491                 op = container_of(p, struct optimized_kprobe, kp);
 492                 if (arch_within_optimized_kprobe(op, addr))
 493                         return p;
 494         }
 495
 496         return NULL;
 497 }
 498
 499 /* Optimization staging list, protected by kprobe_mutex */
 500 static LIST_HEAD(optimizing_list);
 501 static LIST_HEAD(unoptimizing_list);
 502 static LIST_HEAD(freeing_list);
 503
 504 static void kprobe_optimizer(struct work_struct *work);
 505 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 506 #define OPTIMIZE_DELAY 5
 507
 508 /*
 509  * Optimize (replace a breakpoint with a jump) kprobes listed on
 510  * optimizing_list.
 511  */
 512 static void do_optimize_kprobes(void)
 513 {
 514         lockdep_assert_held(&text_mutex);
 515         /*
 516          * The optimization/unoptimization refers online_cpus via
 517          * stop_machine() and cpu-hotplug modifies online_cpus.
 518          * And same time, text_mutex will be held in cpu-hotplug and here.
 519          * This combination can cause a deadlock (cpu-hotplug try to lock
 520          * text_mutex but stop_machine can not be done because online_cpus
 521          * has been changed)
 522          * To avoid this deadlock, caller must have locked cpu hotplug
 523          * for preventing cpu-hotplug outside of text_mutex locking.
 524          */
 525         lockdep_assert_cpus_held();
 526
 527         /* Optimization never be done when disarmed */
 528         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 529             list_empty(&optimizing_list))
 530                 return;
 531
 532         arch_optimize_kprobes(&optimizing_list);
 533 }
 534
 535 /*
 536  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 537  * if need) kprobes listed on unoptimizing_list.
 538  */
 539 static void do_unoptimize_kprobes(void)
 540 {
 541         struct optimized_kprobe *op, *tmp;
 542
 543         lockdep_assert_held(&text_mutex);
 544         /* See comment in do_optimize_kprobes() */
 545         lockdep_assert_cpus_held();
 546
 547         /* Unoptimization must be done anytime */
 548         if (list_empty(&unoptimizing_list))
 549                 return;
 550
 551         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 552         /* Loop free_list for disarming */
 553         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 554                 /* Switching from detour code to origin */
 555                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 556                 /* Disarm probes if marked disabled */
 557                 if (kprobe_disabled(&op->kp))
 558                         arch_disarm_kprobe(&op->kp);
 559                 if (kprobe_unused(&op->kp)) {
 560                         /*
 561                          * Remove unused probes from hash list. After waiting
 562                          * for synchronization, these probes are reclaimed.
 563                          * (reclaiming is done by do_free_cleaned_kprobes.)
 564                          */
 565                         hlist_del_rcu(&op->kp.hlist);
 566                 } else
 567                         list_del_init(&op->list);
 568         }
 569 }
 570
 571 /* Reclaim all kprobes on the free_list */
 572 static void do_free_cleaned_kprobes(void)
 573 {
 574         struct optimized_kprobe *op, *tmp;
 575
 576         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 577                 list_del_init(&op->list);
 578                 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
 579                         /*
 580                          * This must not happen, but if there is a kprobe
 581                          * still in use, keep it on kprobes hash list.
 582                          */
 583                         continue;
 584                 }
 585                 free_aggr_kprobe(&op->kp);
 586         }
 587 }
 588
 589 /* Start optimizer after OPTIMIZE_DELAY passed */
 590 static void kick_kprobe_optimizer(void)
 591 {
 592         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 593 }
 594
 595 /* Kprobe jump optimizer */
 596 static void kprobe_optimizer(struct work_struct *work)
 597 {
 598         mutex_lock(&kprobe_mutex);
 599         cpus_read_lock();
 600         mutex_lock(&text_mutex);
 601
 602         /*
 603          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 604          * kprobes before waiting for quiesence period.
 605          */
 606         do_unoptimize_kprobes();
 607
 608         /*
 609          * Step 2: Wait for quiesence period to ensure all potentially
 610          * preempted tasks to have normally scheduled. Because optprobe
 611          * may modify multiple instructions, there is a chance that Nth
 612          * instruction is preempted. In that case, such tasks can return
 613          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
 614          * Note that on non-preemptive kernel, this is transparently converted
 615          * to synchronoze_sched() to wait for all interrupts to have completed.
 616          */
 617         synchronize_rcu_tasks();
 618
 619         /* Step 3: Optimize kprobes after quiesence period */
 620         do_optimize_kprobes();
 621
 622         /* Step 4: Free cleaned kprobes after quiesence period */
 623         do_free_cleaned_kprobes();
 624
 625         mutex_unlock(&text_mutex);
 626         cpus_read_unlock();
 627
 628         /* Step 5: Kick optimizer again if needed */
 629         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 630                 kick_kprobe_optimizer();
 631
 632         mutex_unlock(&kprobe_mutex);
 633 }
 634
 635 /* Wait for completing optimization and unoptimization */
 636 void wait_for_kprobe_optimizer(void)
 637 {
 638         mutex_lock(&kprobe_mutex);
 639
 640         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 641                 mutex_unlock(&kprobe_mutex);
 642
 643                 /* this will also make optimizing_work execute immmediately */
 644                 flush_delayed_work(&optimizing_work);
 645                 /* @optimizing_work might not have been queued yet, relax */
 646                 cpu_relax();
 647
 648                 mutex_lock(&kprobe_mutex);
 649         }
 650
 651         mutex_unlock(&kprobe_mutex);
 652 }
 653
 654 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
 655 {
 656         struct optimized_kprobe *_op;
 657
 658         list_for_each_entry(_op, &unoptimizing_list, list) {
 659                 if (op == _op)
 660                         return true;
 661         }
 662
 663         return false;
 664 }
 665
 666 /* Optimize kprobe if p is ready to be optimized */
 667 static void optimize_kprobe(struct kprobe *p)
 668 {
 669         struct optimized_kprobe *op;
 670
 671         /* Check if the kprobe is disabled or not ready for optimization. */
 672         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 673             (kprobe_disabled(p) || kprobes_all_disarmed))
 674                 return;
 675
 676         /* kprobes with post_handler can not be optimized */
 677         if (p->post_handler)
 678                 return;
 679
 680         op = container_of(p, struct optimized_kprobe, kp);
 681
 682         /* Check there is no other kprobes at the optimized instructions */
 683         if (arch_check_optimized_kprobe(op) < 0)
 684                 return;
 685
 686         /* Check if it is already optimized. */
 687         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
 688                 if (optprobe_queued_unopt(op)) {
 689                         /* This is under unoptimizing. Just dequeue the probe */
 690                         list_del_init(&op->list);
 691                 }
 692                 return;
 693         }
 694         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 695
 696         /* On unoptimizing/optimizing_list, op must have OPTIMIZED flag */
 697         if (WARN_ON_ONCE(!list_empty(&op->list)))
 698                 return;
 699
 700         list_add(&op->list, &optimizing_list);
 701         kick_kprobe_optimizer();
 702 }
 703
 704 /* Short cut to direct unoptimizing */
 705 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 706 {
 707         lockdep_assert_cpus_held();
 708         arch_unoptimize_kprobe(op);
 709         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 710 }
 711
 712 /* Unoptimize a kprobe if p is optimized */
 713 static void unoptimize_kprobe(struct kprobe *p, bool force)
 714 {
 715         struct optimized_kprobe *op;
 716
 717         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 718                 return; /* This is not an optprobe nor optimized */
 719
 720         op = container_of(p, struct optimized_kprobe, kp);
 721         if (!kprobe_optimized(p))
 722                 return;
 723
 724         if (!list_empty(&op->list)) {
 725                 if (optprobe_queued_unopt(op)) {
 726                         /* Queued in unoptimizing queue */
 727                         if (force) {
 728                                 /*
 729                                  * Forcibly unoptimize the kprobe here, and queue it
 730                                  * in the freeing list for release afterwards.
 731                                  */
 732                                 force_unoptimize_kprobe(op);
 733                                 list_move(&op->list, &freeing_list);
 734                         }
 735                 } else {
 736                         /* Dequeue from the optimizing queue */
 737                         list_del_init(&op->list);
 738                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 739                 }
 740                 return;
 741         }
 742
 743         /* Optimized kprobe case */
 744         if (force) {
 745                 /* Forcibly update the code: this is a special case */
 746                 force_unoptimize_kprobe(op);
 747         } else {
 748                 list_add(&op->list, &unoptimizing_list);
 749                 kick_kprobe_optimizer();
 750         }
 751 }
 752
 753 /* Cancel unoptimizing for reusing */
 754 static int reuse_unused_kprobe(struct kprobe *ap)
 755 {
 756         struct optimized_kprobe *op;
 757
 758         /*
 759          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 760          * there is still a relative jump) and disabled.
 761          */
 762         op = container_of(ap, struct optimized_kprobe, kp);
 763         WARN_ON_ONCE(list_empty(&op->list));
 764         /* Enable the probe again */
 765         ap->flags &= ~KPROBE_FLAG_DISABLED;
 766         /* Optimize it again (remove from op->list) */
 767         if (!kprobe_optready(ap))
 768                 return -EINVAL;
 769
 770         optimize_kprobe(ap);
 771         return 0;
 772 }
 773
 774 /* Remove optimized instructions */
 775 static void kill_optimized_kprobe(struct kprobe *p)
 776 {
 777         struct optimized_kprobe *op;
 778
 779         op = container_of(p, struct optimized_kprobe, kp);
 780         if (!list_empty(&op->list))
 781                 /* Dequeue from the (un)optimization queue */
 782                 list_del_init(&op->list);
 783         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 784
 785         if (kprobe_unused(p)) {
 786                 /* Enqueue if it is unused */
 787                 list_add(&op->list, &freeing_list);
 788                 /*
 789                  * Remove unused probes from the hash list. After waiting
 790                  * for synchronization, this probe is reclaimed.
 791                  * (reclaiming is done by do_free_cleaned_kprobes().)
 792                  */
 793                 hlist_del_rcu(&op->kp.hlist);
 794         }
 795
 796         /* Don't touch the code, because it is already freed. */
 797         arch_remove_optimized_kprobe(op);
 798 }
 799
 800 static inline
 801 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 802 {
 803         if (!kprobe_ftrace(p))
 804                 arch_prepare_optimized_kprobe(op, p);
 805 }
 806
 807 /* Try to prepare optimized instructions */
 808 static void prepare_optimized_kprobe(struct kprobe *p)
 809 {
 810         struct optimized_kprobe *op;
 811
 812         op = container_of(p, struct optimized_kprobe, kp);
 813         __prepare_optimized_kprobe(op, p);
 814 }
 815
 816 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 817 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 818 {
 819         struct optimized_kprobe *op;
 820
 821         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 822         if (!op)
 823                 return NULL;
 824
 825         INIT_LIST_HEAD(&op->list);
 826         op->kp.addr = p->addr;
 827         __prepare_optimized_kprobe(op, p);
 828
 829         return &op->kp;
 830 }
 831
 832 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 833
 834 /*
 835  * Prepare an optimized_kprobe and optimize it
 836  * NOTE: p must be a normal registered kprobe
 837  */
 838 static void try_to_optimize_kprobe(struct kprobe *p)
 839 {
 840         struct kprobe *ap;
 841         struct optimized_kprobe *op;
 842
 843         /* Impossible to optimize ftrace-based kprobe */
 844         if (kprobe_ftrace(p))
 845                 return;
 846
 847         /* For preparing optimization, jump_label_text_reserved() is called */
 848         cpus_read_lock();
 849         jump_label_lock();
 850         mutex_lock(&text_mutex);
 851
 852         ap = alloc_aggr_kprobe(p);
 853         if (!ap)
 854                 goto out;
 855
 856         op = container_of(ap, struct optimized_kprobe, kp);
 857         if (!arch_prepared_optinsn(&op->optinsn)) {
 858                 /* If failed to setup optimizing, fallback to kprobe */
 859                 arch_remove_optimized_kprobe(op);
 860                 kfree(op);
 861                 goto out;
 862         }
 863
 864         init_aggr_kprobe(ap, p);
 865         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 866
 867 out:
 868         mutex_unlock(&text_mutex);
 869         jump_label_unlock();
 870         cpus_read_unlock();
 871 }
 872
 873 #ifdef CONFIG_SYSCTL
 874 static void optimize_all_kprobes(void)
 875 {
 876         struct hlist_head *head;
 877         struct kprobe *p;
 878         unsigned int i;
 879
 880         mutex_lock(&kprobe_mutex);
 881         /* If optimization is already allowed, just return */
 882         if (kprobes_allow_optimization)
 883                 goto out;
 884
 885         cpus_read_lock();
 886         kprobes_allow_optimization = true;
 887         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 888                 head = &kprobe_table[i];
 889                 hlist_for_each_entry(p, head, hlist)
 890                         if (!kprobe_disabled(p))
 891                                 optimize_kprobe(p);
 892         }
 893         cpus_read_unlock();
 894         printk(KERN_INFO "Kprobes globally optimized\n");
 895 out:
 896         mutex_unlock(&kprobe_mutex);
 897 }
 898
 899 static void unoptimize_all_kprobes(void)
 900 {
 901         struct hlist_head *head;
 902         struct kprobe *p;
 903         unsigned int i;
 904
 905         mutex_lock(&kprobe_mutex);
 906         /* If optimization is already prohibited, just return */
 907         if (!kprobes_allow_optimization) {
 908                 mutex_unlock(&kprobe_mutex);
 909                 return;
 910         }
 911
 912         cpus_read_lock();
 913         kprobes_allow_optimization = false;
 914         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 915                 head = &kprobe_table[i];
 916                 hlist_for_each_entry(p, head, hlist) {
 917                         if (!kprobe_disabled(p))
 918                                 unoptimize_kprobe(p, false);
 919                 }
 920         }
 921         cpus_read_unlock();
 922         mutex_unlock(&kprobe_mutex);
 923
 924         /* Wait for unoptimizing completion */
 925         wait_for_kprobe_optimizer();
 926         printk(KERN_INFO "Kprobes globally unoptimized\n");
 927 }
 928
 929 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 930 int sysctl_kprobes_optimization;
 931 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 932                                       void *buffer, size_t *length,
 933                                       loff_t *ppos)
 934 {
 935         int ret;
 936
 937         mutex_lock(&kprobe_sysctl_mutex);
 938         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 939         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 940
 941         if (sysctl_kprobes_optimization)
 942                 optimize_all_kprobes();
 943         else
 944                 unoptimize_all_kprobes();
 945         mutex_unlock(&kprobe_sysctl_mutex);
 946
 947         return ret;
 948 }
 949 #endif /* CONFIG_SYSCTL */
 950
 951 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 952 static void __arm_kprobe(struct kprobe *p)
 953 {
 954         struct kprobe *_p;
 955
 956         /* Check collision with other optimized kprobes */
 957         _p = get_optimized_kprobe((unsigned long)p->addr);
 958         if (unlikely(_p))
 959                 /* Fallback to unoptimized kprobe */
 960                 unoptimize_kprobe(_p, true);
 961
 962         arch_arm_kprobe(p);
 963         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 964 }
 965
 966 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 967 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 968 {
 969         struct kprobe *_p;
 970
 971         /* Try to unoptimize */
 972         unoptimize_kprobe(p, kprobes_all_disarmed);
 973
 974         if (!kprobe_queued(p)) {
 975                 arch_disarm_kprobe(p);
 976                 /* If another kprobe was blocked, optimize it. */
 977                 _p = get_optimized_kprobe((unsigned long)p->addr);
 978                 if (unlikely(_p) && reopt)
 979                         optimize_kprobe(_p);
 980         }
 981         /* TODO: reoptimize others after unoptimized this probe */
 982 }
 983
 984 #else /* !CONFIG_OPTPROBES */
 985
 986 #define optimize_kprobe(p)                      do {} while (0)
 987 #define unoptimize_kprobe(p, f)                 do {} while (0)
 988 #define kill_optimized_kprobe(p)                do {} while (0)
 989 #define prepare_optimized_kprobe(p)             do {} while (0)
 990 #define try_to_optimize_kprobe(p)               do {} while (0)
 991 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 992 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 993 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 994 #define wait_for_kprobe_optimizer()             do {} while (0)
 995
 996 static int reuse_unused_kprobe(struct kprobe *ap)
 997 {
 998         /*
 999          * If the optimized kprobe is NOT supported, the aggr kprobe is
1000          * released at the same time that the last aggregated kprobe is
1001          * unregistered.
1002          * Thus there should be no chance to reuse unused kprobe.
1003          */
1004         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
1005         return -EINVAL;
1006 }
1007
1008 static void free_aggr_kprobe(struct kprobe *p)
1009 {
1010         arch_remove_kprobe(p);
1011         kfree(p);
1012 }
1013
1014 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1015 {
1016         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1017 }
1018 #endif /* CONFIG_OPTPROBES */
1019
1020 #ifdef CONFIG_KPROBES_ON_FTRACE
1021 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1022         .func = kprobe_ftrace_handler,
1023         .flags = FTRACE_OPS_FL_SAVE_REGS,
1024 };
1025
1026 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1027         .func = kprobe_ftrace_handler,
1028         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1029 };
1030
1031 static int kprobe_ipmodify_enabled;
1032 static int kprobe_ftrace_enabled;
1033
1034 /* Must ensure p->addr is really on ftrace */
1035 static int prepare_kprobe(struct kprobe *p)
1036 {
1037         if (!kprobe_ftrace(p))
1038                 return arch_prepare_kprobe(p);
1039
1040         return arch_prepare_kprobe_ftrace(p);
1041 }
1042
1043 /* Caller must lock kprobe_mutex */
1044 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1045                                int *cnt)
1046 {
1047         int ret = 0;
1048
1049         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1050         if (ret) {
1051                 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
1052                          p->addr, ret);
1053                 return ret;
1054         }
1055
1056         if (*cnt == 0) {
1057                 ret = register_ftrace_function(ops);
1058                 if (ret) {
1059                         pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
1060                         goto err_ftrace;
1061                 }
1062         }
1063
1064         (*cnt)++;
1065         return ret;
1066
1067 err_ftrace:
1068         /*
1069          * At this point, sinec ops is not registered, we should be sefe from
1070          * registering empty filter.
1071          */
1072         ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1073         return ret;
1074 }
1075
1076 static int arm_kprobe_ftrace(struct kprobe *p)
1077 {
1078         bool ipmodify = (p->post_handler != NULL);
1079
1080         return __arm_kprobe_ftrace(p,
1081                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1082                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1083 }
1084
1085 /* Caller must lock kprobe_mutex */
1086 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1087                                   int *cnt)
1088 {
1089         int ret = 0;
1090
1091         if (*cnt == 1) {
1092                 ret = unregister_ftrace_function(ops);
1093                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1094                         return ret;
1095         }
1096
1097         (*cnt)--;
1098
1099         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1100         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1101                   p->addr, ret);
1102         return ret;
1103 }
1104
1105 static int disarm_kprobe_ftrace(struct kprobe *p)
1106 {
1107         bool ipmodify = (p->post_handler != NULL);
1108
1109         return __disarm_kprobe_ftrace(p,
1110                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1111                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1112 }
1113 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1114 static inline int prepare_kprobe(struct kprobe *p)
1115 {
1116         return arch_prepare_kprobe(p);
1117 }
1118
1119 static inline int arm_kprobe_ftrace(struct kprobe *p)
1120 {
1121         return -ENODEV;
1122 }
1123
1124 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1125 {
1126         return -ENODEV;
1127 }
1128 #endif
1129
1130 /* Arm a kprobe with text_mutex */
1131 static int arm_kprobe(struct kprobe *kp)
1132 {
1133         if (unlikely(kprobe_ftrace(kp)))
1134                 return arm_kprobe_ftrace(kp);
1135
1136         cpus_read_lock();
1137         mutex_lock(&text_mutex);
1138         __arm_kprobe(kp);
1139         mutex_unlock(&text_mutex);
1140         cpus_read_unlock();
1141
1142         return 0;
1143 }
1144
1145 /* Disarm a kprobe with text_mutex */
1146 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1147 {
1148         if (unlikely(kprobe_ftrace(kp)))
1149                 return disarm_kprobe_ftrace(kp);
1150
1151         cpus_read_lock();
1152         mutex_lock(&text_mutex);
1153         __disarm_kprobe(kp, reopt);
1154         mutex_unlock(&text_mutex);
1155         cpus_read_unlock();
1156
1157         return 0;
1158 }
1159
1160 /*
1161  * Aggregate handlers for multiple kprobes support - these handlers
1162  * take care of invoking the individual kprobe handlers on p->list
1163  */
1164 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1165 {
1166         struct kprobe *kp;
1167
1168         list_for_each_entry_rcu(kp, &p->list, list) {
1169                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1170                         set_kprobe_instance(kp);
1171                         if (kp->pre_handler(kp, regs))
1172                                 return 1;
1173                 }
1174                 reset_kprobe_instance();
1175         }
1176         return 0;
1177 }
1178 NOKPROBE_SYMBOL(aggr_pre_handler);
1179
1180 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1181                               unsigned long flags)
1182 {
1183         struct kprobe *kp;
1184
1185         list_for_each_entry_rcu(kp, &p->list, list) {
1186                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1187                         set_kprobe_instance(kp);
1188                         kp->post_handler(kp, regs, flags);
1189                         reset_kprobe_instance();
1190                 }
1191         }
1192 }
1193 NOKPROBE_SYMBOL(aggr_post_handler);
1194
1195 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1196                               int trapnr)
1197 {
1198         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1199
1200         /*
1201          * if we faulted "during" the execution of a user specified
1202          * probe handler, invoke just that probe's fault handler
1203          */
1204         if (cur && cur->fault_handler) {
1205                 if (cur->fault_handler(cur, regs, trapnr))
1206                         return 1;
1207         }
1208         return 0;
1209 }
1210 NOKPROBE_SYMBOL(aggr_fault_handler);
1211
1212 /* Walks the list and increments nmissed count for multiprobe case */
1213 void kprobes_inc_nmissed_count(struct kprobe *p)
1214 {
1215         struct kprobe *kp;
1216         if (!kprobe_aggrprobe(p)) {
1217                 p->nmissed++;
1218         } else {
1219                 list_for_each_entry_rcu(kp, &p->list, list)
1220                         kp->nmissed++;
1221         }
1222         return;
1223 }
1224 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1225
1226 void recycle_rp_inst(struct kretprobe_instance *ri,
1227                      struct hlist_head *head)
1228 {
1229         struct kretprobe *rp = ri->rp;
1230
1231         /* remove rp inst off the rprobe_inst_table */
1232         hlist_del(&ri->hlist);
1233         INIT_HLIST_NODE(&ri->hlist);
1234         if (likely(rp)) {
1235                 raw_spin_lock(&rp->lock);
1236                 hlist_add_head(&ri->hlist, &rp->free_instances);
1237                 raw_spin_unlock(&rp->lock);
1238         } else
1239                 /* Unregistering */
1240                 hlist_add_head(&ri->hlist, head);
1241 }
1242 NOKPROBE_SYMBOL(recycle_rp_inst);
1243
1244 void kretprobe_hash_lock(struct task_struct *tsk,
1245                          struct hlist_head **head, unsigned long *flags)
1246 __acquires(hlist_lock)
1247 {
1248         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1249         raw_spinlock_t *hlist_lock;
1250
1251         *head = &kretprobe_inst_table[hash];
1252         hlist_lock = kretprobe_table_lock_ptr(hash);
1253         raw_spin_lock_irqsave(hlist_lock, *flags);
1254 }
1255 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1256
1257 static void kretprobe_table_lock(unsigned long hash,
1258                                  unsigned long *flags)
1259 __acquires(hlist_lock)
1260 {
1261         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1262         raw_spin_lock_irqsave(hlist_lock, *flags);
1263 }
1264 NOKPROBE_SYMBOL(kretprobe_table_lock);
1265
1266 void kretprobe_hash_unlock(struct task_struct *tsk,
1267                            unsigned long *flags)
1268 __releases(hlist_lock)
1269 {
1270         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1271         raw_spinlock_t *hlist_lock;
1272
1273         hlist_lock = kretprobe_table_lock_ptr(hash);
1274         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1275 }
1276 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1277
1278 static void kretprobe_table_unlock(unsigned long hash,
1279                                    unsigned long *flags)
1280 __releases(hlist_lock)
1281 {
1282         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1283         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1284 }
1285 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1286
1287 struct kprobe kprobe_busy = {
1288         .addr = (void *) get_kprobe,
1289 };
1290
1291 void kprobe_busy_begin(void)
1292 {
1293         struct kprobe_ctlblk *kcb;
1294
1295         preempt_disable();
1296         __this_cpu_write(current_kprobe, &kprobe_busy);
1297         kcb = get_kprobe_ctlblk();
1298         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1299 }
1300
1301 void kprobe_busy_end(void)
1302 {
1303         __this_cpu_write(current_kprobe, NULL);
1304         preempt_enable();
1305 }
1306
1307 /*
1308  * This function is called from finish_task_switch when task tk becomes dead,
1309  * so that we can recycle any function-return probe instances associated
1310  * with this task. These left over instances represent probed functions
1311  * that have been called but will never return.
1312  */
1313 void kprobe_flush_task(struct task_struct *tk)
1314 {
1315         struct kretprobe_instance *ri;
1316         struct hlist_head *head, empty_rp;
1317         struct hlist_node *tmp;
1318         unsigned long hash, flags = 0;
1319
1320         if (unlikely(!kprobes_initialized))
1321                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1322                 return;
1323
1324         kprobe_busy_begin();
1325
1326         INIT_HLIST_HEAD(&empty_rp);
1327         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1328         head = &kretprobe_inst_table[hash];
1329         kretprobe_table_lock(hash, &flags);
1330         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1331                 if (ri->task == tk)
1332                         recycle_rp_inst(ri, &empty_rp);
1333         }
1334         kretprobe_table_unlock(hash, &flags);
1335         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1336                 hlist_del(&ri->hlist);
1337                 kfree(ri);
1338         }
1339
1340         kprobe_busy_end();
1341 }
1342 NOKPROBE_SYMBOL(kprobe_flush_task);
1343
1344 static inline void free_rp_inst(struct kretprobe *rp)
1345 {
1346         struct kretprobe_instance *ri;
1347         struct hlist_node *next;
1348
1349         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1350                 hlist_del(&ri->hlist);
1351                 kfree(ri);
1352         }
1353 }
1354
1355 static void cleanup_rp_inst(struct kretprobe *rp)
1356 {
1357         unsigned long flags, hash;
1358         struct kretprobe_instance *ri;
1359         struct hlist_node *next;
1360         struct hlist_head *head;
1361
1362         /* No race here */
1363         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1364                 kretprobe_table_lock(hash, &flags);
1365                 head = &kretprobe_inst_table[hash];
1366                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1367                         if (ri->rp == rp)
1368                                 ri->rp = NULL;
1369                 }
1370                 kretprobe_table_unlock(hash, &flags);
1371         }
1372         free_rp_inst(rp);
1373 }
1374 NOKPROBE_SYMBOL(cleanup_rp_inst);
1375
1376 /* Add the new probe to ap->list */
1377 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1378 {
1379         if (p->post_handler)
1380                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1381
1382         list_add_rcu(&p->list, &ap->list);
1383         if (p->post_handler && !ap->post_handler)
1384                 ap->post_handler = aggr_post_handler;
1385
1386         return 0;
1387 }
1388
1389 /*
1390  * Fill in the required fields of the "manager kprobe". Replace the
1391  * earlier kprobe in the hlist with the manager kprobe
1392  */
1393 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1394 {
1395         /* Copy p's insn slot to ap */
1396         copy_kprobe(p, ap);
1397         flush_insn_slot(ap);
1398         ap->addr = p->addr;
1399         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1400         ap->pre_handler = aggr_pre_handler;
1401         ap->fault_handler = aggr_fault_handler;
1402         /* We don't care the kprobe which has gone. */
1403         if (p->post_handler && !kprobe_gone(p))
1404                 ap->post_handler = aggr_post_handler;
1405
1406         INIT_LIST_HEAD(&ap->list);
1407         INIT_HLIST_NODE(&ap->hlist);
1408
1409         list_add_rcu(&p->list, &ap->list);
1410         hlist_replace_rcu(&p->hlist, &ap->hlist);
1411 }
1412
1413 /*
1414  * This is the second or subsequent kprobe at the address - handle
1415  * the intricacies
1416  */
1417 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1418 {
1419         int ret = 0;
1420         struct kprobe *ap = orig_p;
1421
1422         cpus_read_lock();
1423
1424         /* For preparing optimization, jump_label_text_reserved() is called */
1425         jump_label_lock();
1426         mutex_lock(&text_mutex);
1427
1428         if (!kprobe_aggrprobe(orig_p)) {
1429                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1430                 ap = alloc_aggr_kprobe(orig_p);
1431                 if (!ap) {
1432                         ret = -ENOMEM;
1433                         goto out;
1434                 }
1435                 init_aggr_kprobe(ap, orig_p);
1436         } else if (kprobe_unused(ap)) {
1437                 /* This probe is going to die. Rescue it */
1438                 ret = reuse_unused_kprobe(ap);
1439                 if (ret)
1440                         goto out;
1441         }
1442
1443         if (kprobe_gone(ap)) {
1444                 /*
1445                  * Attempting to insert new probe at the same location that
1446                  * had a probe in the module vaddr area which already
1447                  * freed. So, the instruction slot has already been
1448                  * released. We need a new slot for the new probe.
1449                  */
1450                 ret = arch_prepare_kprobe(ap);
1451                 if (ret)
1452                         /*
1453                          * Even if fail to allocate new slot, don't need to
1454                          * free aggr_probe. It will be used next time, or
1455                          * freed by unregister_kprobe.
1456                          */
1457                         goto out;
1458
1459                 /* Prepare optimized instructions if possible. */
1460                 prepare_optimized_kprobe(ap);
1461
1462                 /*
1463                  * Clear gone flag to prevent allocating new slot again, and
1464                  * set disabled flag because it is not armed yet.
1465                  */
1466                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1467                             | KPROBE_FLAG_DISABLED;
1468         }
1469
1470         /* Copy ap's insn slot to p */
1471         copy_kprobe(ap, p);
1472         ret = add_new_kprobe(ap, p);
1473
1474 out:
1475         mutex_unlock(&text_mutex);
1476         jump_label_unlock();
1477         cpus_read_unlock();
1478
1479         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1480                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1481                 if (!kprobes_all_disarmed) {
1482                         /* Arm the breakpoint again. */
1483                         ret = arm_kprobe(ap);
1484                         if (ret) {
1485                                 ap->flags |= KPROBE_FLAG_DISABLED;
1486                                 list_del_rcu(&p->list);
1487                                 synchronize_rcu();
1488                         }
1489                 }
1490         }
1491         return ret;
1492 }
1493
1494 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1495 {
1496         /* The __kprobes marked functions and entry code must not be probed */
1497         return addr >= (unsigned long)__kprobes_text_start &&
1498                addr < (unsigned long)__kprobes_text_end;
1499 }
1500
1501 static bool __within_kprobe_blacklist(unsigned long addr)
1502 {
1503         struct kprobe_blacklist_entry *ent;
1504
1505         if (arch_within_kprobe_blacklist(addr))
1506                 return true;
1507         /*
1508          * If there exists a kprobe_blacklist, verify and
1509          * fail any probe registration in the prohibited area
1510          */
1511         list_for_each_entry(ent, &kprobe_blacklist, list) {
1512                 if (addr >= ent->start_addr && addr < ent->end_addr)
1513                         return true;
1514         }
1515         return false;
1516 }
1517
1518 bool within_kprobe_blacklist(unsigned long addr)
1519 {
1520         char symname[KSYM_NAME_LEN], *p;
1521
1522         if (__within_kprobe_blacklist(addr))
1523                 return true;
1524
1525         /* Check if the address is on a suffixed-symbol */
1526         if (!lookup_symbol_name(addr, symname)) {
1527                 p = strchr(symname, '.');
1528                 if (!p)
1529                         return false;
1530                 *p = '\0';
1531                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1532                 if (addr)
1533                         return __within_kprobe_blacklist(addr);
1534         }
1535         return false;
1536 }
1537
1538 /*
1539  * If we have a symbol_name argument, look it up and add the offset field
1540  * to it. This way, we can specify a relative address to a symbol.
1541  * This returns encoded errors if it fails to look up symbol or invalid
1542  * combination of parameters.
1543  */
1544 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1545                         const char *symbol_name, unsigned int offset)
1546 {
1547         if ((symbol_name && addr) || (!symbol_name && !addr))
1548                 goto invalid;
1549
1550         if (symbol_name) {
1551                 addr = kprobe_lookup_name(symbol_name, offset);
1552                 if (!addr)
1553                         return ERR_PTR(-ENOENT);
1554         }
1555
1556         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1557         if (addr)
1558                 return addr;
1559
1560 invalid:
1561         return ERR_PTR(-EINVAL);
1562 }
1563
1564 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1565 {
1566         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1567 }
1568
1569 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1570 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1571 {
1572         struct kprobe *ap, *list_p;
1573
1574         lockdep_assert_held(&kprobe_mutex);
1575
1576         ap = get_kprobe(p->addr);
1577         if (unlikely(!ap))
1578                 return NULL;
1579
1580         if (p != ap) {
1581                 list_for_each_entry(list_p, &ap->list, list)
1582                         if (list_p == p)
1583                         /* kprobe p is a valid probe */
1584                                 goto valid;
1585                 return NULL;
1586         }
1587 valid:
1588         return ap;
1589 }
1590
1591 /* Return error if the kprobe is being re-registered */
1592 static inline int check_kprobe_rereg(struct kprobe *p)
1593 {
1594         int ret = 0;
1595
1596         mutex_lock(&kprobe_mutex);
1597         if (__get_valid_kprobe(p))
1598                 ret = -EINVAL;
1599         mutex_unlock(&kprobe_mutex);
1600
1601         return ret;
1602 }
1603
1604 int __weak arch_check_ftrace_location(struct kprobe *p)
1605 {
1606         unsigned long ftrace_addr;
1607
1608         ftrace_addr = ftrace_location((unsigned long)p->addr);
1609         if (ftrace_addr) {
1610 #ifdef CONFIG_KPROBES_ON_FTRACE
1611                 /* Given address is not on the instruction boundary */
1612                 if ((unsigned long)p->addr != ftrace_addr)
1613                         return -EILSEQ;
1614                 p->flags |= KPROBE_FLAG_FTRACE;
1615 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1616                 return -EINVAL;
1617 #endif
1618         }
1619         return 0;
1620 }
1621
1622 static int check_kprobe_address_safe(struct kprobe *p,
1623                                      struct module **probed_mod)
1624 {
1625         int ret;
1626
1627         ret = arch_check_ftrace_location(p);
1628         if (ret)
1629                 return ret;
1630         jump_label_lock();
1631         preempt_disable();
1632
1633         /* Ensure it is not in reserved area nor out of text */
1634         if (!kernel_text_address((unsigned long) p->addr) ||
1635             within_kprobe_blacklist((unsigned long) p->addr) ||
1636             jump_label_text_reserved(p->addr, p->addr) ||
1637             find_bug((unsigned long)p->addr)) {
1638                 ret = -EINVAL;
1639                 goto out;
1640         }
1641
1642         /* Check if are we probing a module */
1643         *probed_mod = __module_text_address((unsigned long) p->addr);
1644         if (*probed_mod) {
1645                 /*
1646                  * We must hold a refcount of the probed module while updating
1647                  * its code to prohibit unexpected unloading.
1648                  */
1649                 if (unlikely(!try_module_get(*probed_mod))) {
1650                         ret = -ENOENT;
1651                         goto out;
1652                 }
1653
1654                 /*
1655                  * If the module freed .init.text, we couldn't insert
1656                  * kprobes in there.
1657                  */
1658                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1659                     (*probed_mod)->state != MODULE_STATE_COMING) {
1660                         module_put(*probed_mod);
1661                         *probed_mod = NULL;
1662                         ret = -ENOENT;
1663                 }
1664         }
1665 out:
1666         preempt_enable();
1667         jump_label_unlock();
1668
1669         return ret;
1670 }
1671
1672 int register_kprobe(struct kprobe *p)
1673 {
1674         int ret;
1675         struct kprobe *old_p;
1676         struct module *probed_mod;
1677         kprobe_opcode_t *addr;
1678
1679         /* Adjust probe address from symbol */
1680         addr = kprobe_addr(p);
1681         if (IS_ERR(addr))
1682                 return PTR_ERR(addr);
1683         p->addr = addr;
1684
1685         ret = check_kprobe_rereg(p);
1686         if (ret)
1687                 return ret;
1688
1689         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1690         p->flags &= KPROBE_FLAG_DISABLED;
1691         p->nmissed = 0;
1692         INIT_LIST_HEAD(&p->list);
1693
1694         ret = check_kprobe_address_safe(p, &probed_mod);
1695         if (ret)
1696                 return ret;
1697
1698         mutex_lock(&kprobe_mutex);
1699
1700         old_p = get_kprobe(p->addr);
1701         if (old_p) {
1702                 /* Since this may unoptimize old_p, locking text_mutex. */
1703                 ret = register_aggr_kprobe(old_p, p);
1704                 goto out;
1705         }
1706
1707         cpus_read_lock();
1708         /* Prevent text modification */
1709         mutex_lock(&text_mutex);
1710         ret = prepare_kprobe(p);
1711         mutex_unlock(&text_mutex);
1712         cpus_read_unlock();
1713         if (ret)
1714                 goto out;
1715
1716         INIT_HLIST_NODE(&p->hlist);
1717         hlist_add_head_rcu(&p->hlist,
1718                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1719
1720         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1721                 ret = arm_kprobe(p);
1722                 if (ret) {
1723                         hlist_del_rcu(&p->hlist);
1724                         synchronize_rcu();
1725                         goto out;
1726                 }
1727         }
1728
1729         /* Try to optimize kprobe */
1730         try_to_optimize_kprobe(p);
1731 out:
1732         mutex_unlock(&kprobe_mutex);
1733
1734         if (probed_mod)
1735                 module_put(probed_mod);
1736
1737         return ret;
1738 }
1739 EXPORT_SYMBOL_GPL(register_kprobe);
1740
1741 /* Check if all probes on the aggrprobe are disabled */
1742 static int aggr_kprobe_disabled(struct kprobe *ap)
1743 {
1744         struct kprobe *kp;
1745
1746         lockdep_assert_held(&kprobe_mutex);
1747
1748         list_for_each_entry(kp, &ap->list, list)
1749                 if (!kprobe_disabled(kp))
1750                         /*
1751                          * There is an active probe on the list.
1752                          * We can't disable this ap.
1753                          */
1754                         return 0;
1755
1756         return 1;
1757 }
1758
1759 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1760 static struct kprobe *__disable_kprobe(struct kprobe *p)
1761 {
1762         struct kprobe *orig_p;
1763         int ret;
1764
1765         /* Get an original kprobe for return */
1766         orig_p = __get_valid_kprobe(p);
1767         if (unlikely(orig_p == NULL))
1768                 return ERR_PTR(-EINVAL);
1769
1770         if (!kprobe_disabled(p)) {
1771                 /* Disable probe if it is a child probe */
1772                 if (p != orig_p)
1773                         p->flags |= KPROBE_FLAG_DISABLED;
1774
1775                 /* Try to disarm and disable this/parent probe */
1776                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1777                         /*
1778                          * If kprobes_all_disarmed is set, orig_p
1779                          * should have already been disarmed, so
1780                          * skip unneed disarming process.
1781                          */
1782                         if (!kprobes_all_disarmed) {
1783                                 ret = disarm_kprobe(orig_p, true);
1784                                 if (ret) {
1785                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1786                                         return ERR_PTR(ret);
1787                                 }
1788                         }
1789                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1790                 }
1791         }
1792
1793         return orig_p;
1794 }
1795
1796 /*
1797  * Unregister a kprobe without a scheduler synchronization.
1798  */
1799 static int __unregister_kprobe_top(struct kprobe *p)
1800 {
1801         struct kprobe *ap, *list_p;
1802
1803         /* Disable kprobe. This will disarm it if needed. */
1804         ap = __disable_kprobe(p);
1805         if (IS_ERR(ap))
1806                 return PTR_ERR(ap);
1807
1808         if (ap == p)
1809                 /*
1810                  * This probe is an independent(and non-optimized) kprobe
1811                  * (not an aggrprobe). Remove from the hash list.
1812                  */
1813                 goto disarmed;
1814
1815         /* Following process expects this probe is an aggrprobe */
1816         WARN_ON(!kprobe_aggrprobe(ap));
1817
1818         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1819                 /*
1820                  * !disarmed could be happen if the probe is under delayed
1821                  * unoptimizing.
1822                  */
1823                 goto disarmed;
1824         else {
1825                 /* If disabling probe has special handlers, update aggrprobe */
1826                 if (p->post_handler && !kprobe_gone(p)) {
1827                         list_for_each_entry(list_p, &ap->list, list) {
1828                                 if ((list_p != p) && (list_p->post_handler))
1829                                         goto noclean;
1830                         }
1831                         ap->post_handler = NULL;
1832                 }
1833 noclean:
1834                 /*
1835                  * Remove from the aggrprobe: this path will do nothing in
1836                  * __unregister_kprobe_bottom().
1837                  */
1838                 list_del_rcu(&p->list);
1839                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1840                         /*
1841                          * Try to optimize this probe again, because post
1842                          * handler may have been changed.
1843                          */
1844                         optimize_kprobe(ap);
1845         }
1846         return 0;
1847
1848 disarmed:
1849         hlist_del_rcu(&ap->hlist);
1850         return 0;
1851 }
1852
1853 static void __unregister_kprobe_bottom(struct kprobe *p)
1854 {
1855         struct kprobe *ap;
1856
1857         if (list_empty(&p->list))
1858                 /* This is an independent kprobe */
1859                 arch_remove_kprobe(p);
1860         else if (list_is_singular(&p->list)) {
1861                 /* This is the last child of an aggrprobe */
1862                 ap = list_entry(p->list.next, struct kprobe, list);
1863                 list_del(&p->list);
1864                 free_aggr_kprobe(ap);
1865         }
1866         /* Otherwise, do nothing. */
1867 }
1868
1869 int register_kprobes(struct kprobe **kps, int num)
1870 {
1871         int i, ret = 0;
1872
1873         if (num <= 0)
1874                 return -EINVAL;
1875         for (i = 0; i < num; i++) {
1876                 ret = register_kprobe(kps[i]);
1877                 if (ret < 0) {
1878                         if (i > 0)
1879                                 unregister_kprobes(kps, i);
1880                         break;
1881                 }
1882         }
1883         return ret;
1884 }
1885 EXPORT_SYMBOL_GPL(register_kprobes);
1886
1887 void unregister_kprobe(struct kprobe *p)
1888 {
1889         unregister_kprobes(&p, 1);
1890 }
1891 EXPORT_SYMBOL_GPL(unregister_kprobe);
1892
1893 void unregister_kprobes(struct kprobe **kps, int num)
1894 {
1895         int i;
1896
1897         if (num <= 0)
1898                 return;
1899         mutex_lock(&kprobe_mutex);
1900         for (i = 0; i < num; i++)
1901                 if (__unregister_kprobe_top(kps[i]) < 0)
1902                         kps[i]->addr = NULL;
1903         mutex_unlock(&kprobe_mutex);
1904
1905         synchronize_rcu();
1906         for (i = 0; i < num; i++)
1907                 if (kps[i]->addr)
1908                         __unregister_kprobe_bottom(kps[i]);
1909 }
1910 EXPORT_SYMBOL_GPL(unregister_kprobes);
1911
1912 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1913                                         unsigned long val, void *data)
1914 {
1915         return NOTIFY_DONE;
1916 }
1917 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1918
1919 static struct notifier_block kprobe_exceptions_nb = {
1920         .notifier_call = kprobe_exceptions_notify,
1921         .priority = 0x7fffffff /* we need to be notified first */
1922 };
1923
1924 unsigned long __weak arch_deref_entry_point(void *entry)
1925 {
1926         return (unsigned long)entry;
1927 }
1928
1929 #ifdef CONFIG_KRETPROBES
1930 /*
1931  * This kprobe pre_handler is registered with every kretprobe. When probe
1932  * hits it will set up the return probe.
1933  */
1934 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1935 {
1936         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1937         unsigned long hash, flags = 0;
1938         struct kretprobe_instance *ri;
1939
1940         /*
1941          * To avoid deadlocks, prohibit return probing in NMI contexts,
1942          * just skip the probe and increase the (inexact) 'nmissed'
1943          * statistical counter, so that the user is informed that
1944          * something happened:
1945          */
1946         if (unlikely(in_nmi())) {
1947                 rp->nmissed++;
1948                 return 0;
1949         }
1950
1951         /* TODO: consider to only swap the RA after the last pre_handler fired */
1952         hash = hash_ptr(current, KPROBE_HASH_BITS);
1953         raw_spin_lock_irqsave(&rp->lock, flags);
1954         if (!hlist_empty(&rp->free_instances)) {
1955                 ri = hlist_entry(rp->free_instances.first,
1956                                 struct kretprobe_instance, hlist);
1957                 hlist_del(&ri->hlist);
1958                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1959
1960                 ri->rp = rp;
1961                 ri->task = current;
1962
1963                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1964                         raw_spin_lock_irqsave(&rp->lock, flags);
1965                         hlist_add_head(&ri->hlist, &rp->free_instances);
1966                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1967                         return 0;
1968                 }
1969
1970                 arch_prepare_kretprobe(ri, regs);
1971
1972                 /* XXX(hch): why is there no hlist_move_head? */
1973                 INIT_HLIST_NODE(&ri->hlist);
1974                 kretprobe_table_lock(hash, &flags);
1975                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1976                 kretprobe_table_unlock(hash, &flags);
1977         } else {
1978                 rp->nmissed++;
1979                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1980         }
1981         return 0;
1982 }
1983 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1984
1985 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1986 {
1987         return !offset;
1988 }
1989
1990 bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1991 {
1992         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1993
1994         if (IS_ERR(kp_addr))
1995                 return false;
1996
1997         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) ||
1998                                                 !arch_kprobe_on_func_entry(offset))
1999                 return false;
2000
2001         return true;
2002 }
2003
2004 int register_kretprobe(struct kretprobe *rp)
2005 {
2006         int ret = 0;
2007         struct kretprobe_instance *inst;
2008         int i;
2009         void *addr;
2010
2011         if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset))
2012                 return -EINVAL;
2013
2014         if (kretprobe_blacklist_size) {
2015                 addr = kprobe_addr(&rp->kp);
2016                 if (IS_ERR(addr))
2017                         return PTR_ERR(addr);
2018
2019                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2020                         if (kretprobe_blacklist[i].addr == addr)
2021                                 return -EINVAL;
2022                 }
2023         }
2024
2025         rp->kp.pre_handler = pre_handler_kretprobe;
2026         rp->kp.post_handler = NULL;
2027         rp->kp.fault_handler = NULL;
2028
2029         /* Pre-allocate memory for max kretprobe instances */
2030         if (rp->maxactive <= 0) {
2031 #ifdef CONFIG_PREEMPTION
2032                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2033 #else
2034                 rp->maxactive = num_possible_cpus();
2035 #endif
2036         }
2037         raw_spin_lock_init(&rp->lock);
2038         INIT_HLIST_HEAD(&rp->free_instances);
2039         for (i = 0; i < rp->maxactive; i++) {
2040                 inst = kmalloc(sizeof(struct kretprobe_instance) +
2041                                rp->data_size, GFP_KERNEL);
2042                 if (inst == NULL) {
2043                         free_rp_inst(rp);
2044                         return -ENOMEM;
2045                 }
2046                 INIT_HLIST_NODE(&inst->hlist);
2047                 hlist_add_head(&inst->hlist, &rp->free_instances);
2048         }
2049
2050         rp->nmissed = 0;
2051         /* Establish function entry probe point */
2052         ret = register_kprobe(&rp->kp);
2053         if (ret != 0)
2054                 free_rp_inst(rp);
2055         return ret;
2056 }
2057 EXPORT_SYMBOL_GPL(register_kretprobe);
2058
2059 int register_kretprobes(struct kretprobe **rps, int num)
2060 {
2061         int ret = 0, i;
2062
2063         if (num <= 0)
2064                 return -EINVAL;
2065         for (i = 0; i < num; i++) {
2066                 ret = register_kretprobe(rps[i]);
2067                 if (ret < 0) {
2068                         if (i > 0)
2069                                 unregister_kretprobes(rps, i);
2070                         break;
2071                 }
2072         }
2073         return ret;
2074 }
2075 EXPORT_SYMBOL_GPL(register_kretprobes);
2076
2077 void unregister_kretprobe(struct kretprobe *rp)
2078 {
2079         unregister_kretprobes(&rp, 1);
2080 }
2081 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2082
2083 void unregister_kretprobes(struct kretprobe **rps, int num)
2084 {
2085         int i;
2086
2087         if (num <= 0)
2088                 return;
2089         mutex_lock(&kprobe_mutex);
2090         for (i = 0; i < num; i++)
2091                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2092                         rps[i]->kp.addr = NULL;
2093         mutex_unlock(&kprobe_mutex);
2094
2095         synchronize_rcu();
2096         for (i = 0; i < num; i++) {
2097                 if (rps[i]->kp.addr) {
2098                         __unregister_kprobe_bottom(&rps[i]->kp);
2099                         cleanup_rp_inst(rps[i]);
2100                 }
2101         }
2102 }
2103 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2104
2105 #else /* CONFIG_KRETPROBES */
2106 int register_kretprobe(struct kretprobe *rp)
2107 {
2108         return -ENOSYS;
2109 }
2110 EXPORT_SYMBOL_GPL(register_kretprobe);
2111
2112 int register_kretprobes(struct kretprobe **rps, int num)
2113 {
2114         return -ENOSYS;
2115 }
2116 EXPORT_SYMBOL_GPL(register_kretprobes);
2117
2118 void unregister_kretprobe(struct kretprobe *rp)
2119 {
2120 }
2121 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2122
2123 void unregister_kretprobes(struct kretprobe **rps, int num)
2124 {
2125 }
2126 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2127
2128 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2129 {
2130         return 0;
2131 }
2132 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2133
2134 #endif /* CONFIG_KRETPROBES */
2135
2136 /* Set the kprobe gone and remove its instruction buffer. */
2137 static void kill_kprobe(struct kprobe *p)
2138 {
2139         struct kprobe *kp;
2140
2141         lockdep_assert_held(&kprobe_mutex);
2142
2143         p->flags |= KPROBE_FLAG_GONE;
2144         if (kprobe_aggrprobe(p)) {
2145                 /*
2146                  * If this is an aggr_kprobe, we have to list all the
2147                  * chained probes and mark them GONE.
2148                  */
2149                 list_for_each_entry(kp, &p->list, list)
2150                         kp->flags |= KPROBE_FLAG_GONE;
2151                 p->post_handler = NULL;
2152                 kill_optimized_kprobe(p);
2153         }
2154         /*
2155          * Here, we can remove insn_slot safely, because no thread calls
2156          * the original probed function (which will be freed soon) any more.
2157          */
2158         arch_remove_kprobe(p);
2159
2160         /*
2161          * The module is going away. We should disarm the kprobe which
2162          * is using ftrace.
2163          */
2164         if (kprobe_ftrace(p))
2165                 disarm_kprobe_ftrace(p);
2166 }
2167
2168 /* Disable one kprobe */
2169 int disable_kprobe(struct kprobe *kp)
2170 {
2171         int ret = 0;
2172         struct kprobe *p;
2173
2174         mutex_lock(&kprobe_mutex);
2175
2176         /* Disable this kprobe */
2177         p = __disable_kprobe(kp);
2178         if (IS_ERR(p))
2179                 ret = PTR_ERR(p);
2180
2181         mutex_unlock(&kprobe_mutex);
2182         return ret;
2183 }
2184 EXPORT_SYMBOL_GPL(disable_kprobe);
2185
2186 /* Enable one kprobe */
2187 int enable_kprobe(struct kprobe *kp)
2188 {
2189         int ret = 0;
2190         struct kprobe *p;
2191
2192         mutex_lock(&kprobe_mutex);
2193
2194         /* Check whether specified probe is valid. */
2195         p = __get_valid_kprobe(kp);
2196         if (unlikely(p == NULL)) {
2197                 ret = -EINVAL;
2198                 goto out;
2199         }
2200
2201         if (kprobe_gone(kp)) {
2202                 /* This kprobe has gone, we couldn't enable it. */
2203                 ret = -EINVAL;
2204                 goto out;
2205         }
2206
2207         if (p != kp)
2208                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2209
2210         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2211                 p->flags &= ~KPROBE_FLAG_DISABLED;
2212                 ret = arm_kprobe(p);
2213                 if (ret)
2214                         p->flags |= KPROBE_FLAG_DISABLED;
2215         }
2216 out:
2217         mutex_unlock(&kprobe_mutex);
2218         return ret;
2219 }
2220 EXPORT_SYMBOL_GPL(enable_kprobe);
2221
2222 /* Caller must NOT call this in usual path. This is only for critical case */
2223 void dump_kprobe(struct kprobe *kp)
2224 {
2225         pr_err("Dumping kprobe:\n");
2226         pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2227                kp->symbol_name, kp->offset, kp->addr);
2228 }
2229 NOKPROBE_SYMBOL(dump_kprobe);
2230
2231 int kprobe_add_ksym_blacklist(unsigned long entry)
2232 {
2233         struct kprobe_blacklist_entry *ent;
2234         unsigned long offset = 0, size = 0;
2235
2236         if (!kernel_text_address(entry) ||
2237             !kallsyms_lookup_size_offset(entry, &size, &offset))
2238                 return -EINVAL;
2239
2240         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2241         if (!ent)
2242                 return -ENOMEM;
2243         ent->start_addr = entry;
2244         ent->end_addr = entry + size;
2245         INIT_LIST_HEAD(&ent->list);
2246         list_add_tail(&ent->list, &kprobe_blacklist);
2247
2248         return (int)size;
2249 }
2250
2251 /* Add all symbols in given area into kprobe blacklist */
2252 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2253 {
2254         unsigned long entry;
2255         int ret = 0;
2256
2257         for (entry = start; entry < end; entry += ret) {
2258                 ret = kprobe_add_ksym_blacklist(entry);
2259                 if (ret < 0)
2260                         return ret;
2261                 if (ret == 0)   /* In case of alias symbol */
2262                         ret = 1;
2263         }
2264         return 0;
2265 }
2266
2267 /* Remove all symbols in given area from kprobe blacklist */
2268 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2269 {
2270         struct kprobe_blacklist_entry *ent, *n;
2271
2272         list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2273                 if (ent->start_addr < start || ent->start_addr >= end)
2274                         continue;
2275                 list_del(&ent->list);
2276                 kfree(ent);
2277         }
2278 }
2279
2280 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2281 {
2282         kprobe_remove_area_blacklist(entry, entry + 1);
2283 }
2284
2285 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2286                                    char *type, char *sym)
2287 {
2288         return -ERANGE;
2289 }
2290
2291 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2292                        char *sym)
2293 {
2294 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2295         if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2296                 return 0;
2297 #ifdef CONFIG_OPTPROBES
2298         if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2299                 return 0;
2300 #endif
2301 #endif
2302         if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2303                 return 0;
2304         return -ERANGE;
2305 }
2306
2307 int __init __weak arch_populate_kprobe_blacklist(void)
2308 {
2309         return 0;
2310 }
2311
2312 /*
2313  * Lookup and populate the kprobe_blacklist.
2314  *
2315  * Unlike the kretprobe blacklist, we'll need to determine
2316  * the range of addresses that belong to the said functions,
2317  * since a kprobe need not necessarily be at the beginning
2318  * of a function.
2319  */
2320 static int __init populate_kprobe_blacklist(unsigned long *start,
2321                                              unsigned long *end)
2322 {
2323         unsigned long entry;
2324         unsigned long *iter;
2325         int ret;
2326
2327         for (iter = start; iter < end; iter++) {
2328                 entry = arch_deref_entry_point((void *)*iter);
2329                 ret = kprobe_add_ksym_blacklist(entry);
2330                 if (ret == -EINVAL)
2331                         continue;
2332                 if (ret < 0)
2333                         return ret;
2334         }
2335
2336         /* Symbols in __kprobes_text are blacklisted */
2337         ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2338                                         (unsigned long)__kprobes_text_end);
2339         if (ret)
2340                 return ret;
2341
2342         /* Symbols in noinstr section are blacklisted */
2343         ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2344                                         (unsigned long)__noinstr_text_end);
2345
2346         return ret ? : arch_populate_kprobe_blacklist();
2347 }
2348
2349 static void add_module_kprobe_blacklist(struct module *mod)
2350 {
2351         unsigned long start, end;
2352         int i;
2353
2354         if (mod->kprobe_blacklist) {
2355                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2356                         kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2357         }
2358
2359         start = (unsigned long)mod->kprobes_text_start;
2360         if (start) {
2361                 end = start + mod->kprobes_text_size;
2362                 kprobe_add_area_blacklist(start, end);
2363         }
2364
2365         start = (unsigned long)mod->noinstr_text_start;
2366         if (start) {
2367                 end = start + mod->noinstr_text_size;
2368                 kprobe_add_area_blacklist(start, end);
2369         }
2370 }
2371
2372 static void remove_module_kprobe_blacklist(struct module *mod)
2373 {
2374         unsigned long start, end;
2375         int i;
2376
2377         if (mod->kprobe_blacklist) {
2378                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2379                         kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2380         }
2381
2382         start = (unsigned long)mod->kprobes_text_start;
2383         if (start) {
2384                 end = start + mod->kprobes_text_size;
2385                 kprobe_remove_area_blacklist(start, end);
2386         }
2387
2388         start = (unsigned long)mod->noinstr_text_start;
2389         if (start) {
2390                 end = start + mod->noinstr_text_size;
2391                 kprobe_remove_area_blacklist(start, end);
2392         }
2393 }
2394
2395 /* Module notifier call back, checking kprobes on the module */
2396 static int kprobes_module_callback(struct notifier_block *nb,
2397                                    unsigned long val, void *data)
2398 {
2399         struct module *mod = data;
2400         struct hlist_head *head;
2401         struct kprobe *p;
2402         unsigned int i;
2403         int checkcore = (val == MODULE_STATE_GOING);
2404
2405         if (val == MODULE_STATE_COMING) {
2406                 mutex_lock(&kprobe_mutex);
2407                 add_module_kprobe_blacklist(mod);
2408                 mutex_unlock(&kprobe_mutex);
2409         }
2410         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2411                 return NOTIFY_DONE;
2412
2413         /*
2414          * When MODULE_STATE_GOING was notified, both of module .text and
2415          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2416          * notified, only .init.text section would be freed. We need to
2417          * disable kprobes which have been inserted in the sections.
2418          */
2419         mutex_lock(&kprobe_mutex);
2420         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2421                 head = &kprobe_table[i];
2422                 hlist_for_each_entry(p, head, hlist)
2423                         if (within_module_init((unsigned long)p->addr, mod) ||
2424                             (checkcore &&
2425                              within_module_core((unsigned long)p->addr, mod))) {
2426                                 /*
2427                                  * The vaddr this probe is installed will soon
2428                                  * be vfreed buy not synced to disk. Hence,
2429                                  * disarming the breakpoint isn't needed.
2430                                  *
2431                                  * Note, this will also move any optimized probes
2432                                  * that are pending to be removed from their
2433                                  * corresponding lists to the freeing_list and
2434                                  * will not be touched by the delayed
2435                                  * kprobe_optimizer work handler.
2436                                  */
2437                                 kill_kprobe(p);
2438                         }
2439         }
2440         if (val == MODULE_STATE_GOING)
2441                 remove_module_kprobe_blacklist(mod);
2442         mutex_unlock(&kprobe_mutex);
2443         return NOTIFY_DONE;
2444 }
2445
2446 static struct notifier_block kprobe_module_nb = {
2447         .notifier_call = kprobes_module_callback,
2448         .priority = 0
2449 };
2450
2451 /* Markers of _kprobe_blacklist section */
2452 extern unsigned long __start_kprobe_blacklist[];
2453 extern unsigned long __stop_kprobe_blacklist[];
2454
2455 static int __init init_kprobes(void)
2456 {
2457         int i, err = 0;
2458
2459         /* FIXME allocate the probe table, currently defined statically */
2460         /* initialize all list heads */
2461         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2462                 INIT_HLIST_HEAD(&kprobe_table[i]);
2463                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2464                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2465         }
2466
2467         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2468                                         __stop_kprobe_blacklist);
2469         if (err) {
2470                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2471                 pr_err("Please take care of using kprobes.\n");
2472         }
2473
2474         if (kretprobe_blacklist_size) {
2475                 /* lookup the function address from its name */
2476                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2477                         kretprobe_blacklist[i].addr =
2478                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2479                         if (!kretprobe_blacklist[i].addr)
2480                                 printk("kretprobe: lookup failed: %s\n",
2481                                        kretprobe_blacklist[i].name);
2482                 }
2483         }
2484
2485 #if defined(CONFIG_OPTPROBES)
2486 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2487         /* Init kprobe_optinsn_slots */
2488         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2489 #endif
2490         /* By default, kprobes can be optimized */
2491         kprobes_allow_optimization = true;
2492 #endif
2493
2494         /* By default, kprobes are armed */
2495         kprobes_all_disarmed = false;
2496
2497         err = arch_init_kprobes();
2498         if (!err)
2499                 err = register_die_notifier(&kprobe_exceptions_nb);
2500         if (!err)
2501                 err = register_module_notifier(&kprobe_module_nb);
2502
2503         kprobes_initialized = (err == 0);
2504
2505         if (!err)
2506                 init_test_probes();
2507         return err;
2508 }
2509 subsys_initcall(init_kprobes);
2510
2511 #ifdef CONFIG_DEBUG_FS
2512 static void report_probe(struct seq_file *pi, struct kprobe *p,
2513                 const char *sym, int offset, char *modname, struct kprobe *pp)
2514 {
2515         char *kprobe_type;
2516         void *addr = p->addr;
2517
2518         if (p->pre_handler == pre_handler_kretprobe)
2519                 kprobe_type = "r";
2520         else
2521                 kprobe_type = "k";
2522
2523         if (!kallsyms_show_value(pi->file->f_cred))
2524                 addr = NULL;
2525
2526         if (sym)
2527                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2528                         addr, kprobe_type, sym, offset,
2529                         (modname ? modname : " "));
2530         else    /* try to use %pS */
2531                 seq_printf(pi, "%px  %s  %pS ",
2532                         addr, kprobe_type, p->addr);
2533
2534         if (!pp)
2535                 pp = p;
2536         seq_printf(pi, "%s%s%s%s\n",
2537                 (kprobe_gone(p) ? "[GONE]" : ""),
2538                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2539                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2540                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2541 }
2542
2543 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2544 {
2545         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2546 }
2547
2548 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2549 {
2550         (*pos)++;
2551         if (*pos >= KPROBE_TABLE_SIZE)
2552                 return NULL;
2553         return pos;
2554 }
2555
2556 static void kprobe_seq_stop(struct seq_file *f, void *v)
2557 {
2558         /* Nothing to do */
2559 }
2560
2561 static int show_kprobe_addr(struct seq_file *pi, void *v)
2562 {
2563         struct hlist_head *head;
2564         struct kprobe *p, *kp;
2565         const char *sym = NULL;
2566         unsigned int i = *(loff_t *) v;
2567         unsigned long offset = 0;
2568         char *modname, namebuf[KSYM_NAME_LEN];
2569
2570         head = &kprobe_table[i];
2571         preempt_disable();
2572         hlist_for_each_entry_rcu(p, head, hlist) {
2573                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2574                                         &offset, &modname, namebuf);
2575                 if (kprobe_aggrprobe(p)) {
2576                         list_for_each_entry_rcu(kp, &p->list, list)
2577                                 report_probe(pi, kp, sym, offset, modname, p);
2578                 } else
2579                         report_probe(pi, p, sym, offset, modname, NULL);
2580         }
2581         preempt_enable();
2582         return 0;
2583 }
2584
2585 static const struct seq_operations kprobes_sops = {
2586         .start = kprobe_seq_start,
2587         .next  = kprobe_seq_next,
2588         .stop  = kprobe_seq_stop,
2589         .show  = show_kprobe_addr
2590 };
2591
2592 DEFINE_SEQ_ATTRIBUTE(kprobes);
2593
2594 /* kprobes/blacklist -- shows which functions can not be probed */
2595 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2596 {
2597         mutex_lock(&kprobe_mutex);
2598         return seq_list_start(&kprobe_blacklist, *pos);
2599 }
2600
2601 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2602 {
2603         return seq_list_next(v, &kprobe_blacklist, pos);
2604 }
2605
2606 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2607 {
2608         struct kprobe_blacklist_entry *ent =
2609                 list_entry(v, struct kprobe_blacklist_entry, list);
2610
2611         /*
2612          * If /proc/kallsyms is not showing kernel address, we won't
2613          * show them here either.
2614          */
2615         if (!kallsyms_show_value(m->file->f_cred))
2616                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2617                            (void *)ent->start_addr);
2618         else
2619                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2620                            (void *)ent->end_addr, (void *)ent->start_addr);
2621         return 0;
2622 }
2623
2624 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2625 {
2626         mutex_unlock(&kprobe_mutex);
2627 }
2628
2629 static const struct seq_operations kprobe_blacklist_sops = {
2630         .start = kprobe_blacklist_seq_start,
2631         .next  = kprobe_blacklist_seq_next,
2632         .stop  = kprobe_blacklist_seq_stop,
2633         .show  = kprobe_blacklist_seq_show,
2634 };
2635 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2636
2637 static int arm_all_kprobes(void)
2638 {
2639         struct hlist_head *head;
2640         struct kprobe *p;
2641         unsigned int i, total = 0, errors = 0;
2642         int err, ret = 0;
2643
2644         mutex_lock(&kprobe_mutex);
2645
2646         /* If kprobes are armed, just return */
2647         if (!kprobes_all_disarmed)
2648                 goto already_enabled;
2649
2650         /*
2651          * optimize_kprobe() called by arm_kprobe() checks
2652          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2653          * arm_kprobe.
2654          */
2655         kprobes_all_disarmed = false;
2656         /* Arming kprobes doesn't optimize kprobe itself */
2657         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2658                 head = &kprobe_table[i];
2659                 /* Arm all kprobes on a best-effort basis */
2660                 hlist_for_each_entry(p, head, hlist) {
2661                         if (!kprobe_disabled(p)) {
2662                                 err = arm_kprobe(p);
2663                                 if (err)  {
2664                                         errors++;
2665                                         ret = err;
2666                                 }
2667                                 total++;
2668                         }
2669                 }
2670         }
2671
2672         if (errors)
2673                 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2674                         errors, total);
2675         else
2676                 pr_info("Kprobes globally enabled\n");
2677
2678 already_enabled:
2679         mutex_unlock(&kprobe_mutex);
2680         return ret;
2681 }
2682
2683 static int disarm_all_kprobes(void)
2684 {
2685         struct hlist_head *head;
2686         struct kprobe *p;
2687         unsigned int i, total = 0, errors = 0;
2688         int err, ret = 0;
2689
2690         mutex_lock(&kprobe_mutex);
2691
2692         /* If kprobes are already disarmed, just return */
2693         if (kprobes_all_disarmed) {
2694                 mutex_unlock(&kprobe_mutex);
2695                 return 0;
2696         }
2697
2698         kprobes_all_disarmed = true;
2699
2700         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2701                 head = &kprobe_table[i];
2702                 /* Disarm all kprobes on a best-effort basis */
2703                 hlist_for_each_entry(p, head, hlist) {
2704                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2705                                 err = disarm_kprobe(p, false);
2706                                 if (err) {
2707                                         errors++;
2708                                         ret = err;
2709                                 }
2710                                 total++;
2711                         }
2712                 }
2713         }
2714
2715         if (errors)
2716                 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2717                         errors, total);
2718         else
2719                 pr_info("Kprobes globally disabled\n");
2720
2721         mutex_unlock(&kprobe_mutex);
2722
2723         /* Wait for disarming all kprobes by optimizer */
2724         wait_for_kprobe_optimizer();
2725
2726         return ret;
2727 }
2728
2729 /*
2730  * XXX: The debugfs bool file interface doesn't allow for callbacks
2731  * when the bool state is switched. We can reuse that facility when
2732  * available
2733  */
2734 static ssize_t read_enabled_file_bool(struct file *file,
2735                char __user *user_buf, size_t count, loff_t *ppos)
2736 {
2737         char buf[3];
2738
2739         if (!kprobes_all_disarmed)
2740                 buf[0] = '1';
2741         else
2742                 buf[0] = '0';
2743         buf[1] = '\n';
2744         buf[2] = 0x00;
2745         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2746 }
2747
2748 static ssize_t write_enabled_file_bool(struct file *file,
2749                const char __user *user_buf, size_t count, loff_t *ppos)
2750 {
2751         char buf[32];
2752         size_t buf_size;
2753         int ret = 0;
2754
2755         buf_size = min(count, (sizeof(buf)-1));
2756         if (copy_from_user(buf, user_buf, buf_size))
2757                 return -EFAULT;
2758
2759         buf[buf_size] = '\0';
2760         switch (buf[0]) {
2761         case 'y':
2762         case 'Y':
2763         case '1':
2764                 ret = arm_all_kprobes();
2765                 break;
2766         case 'n':
2767         case 'N':
2768         case '0':
2769                 ret = disarm_all_kprobes();
2770                 break;
2771         default:
2772                 return -EINVAL;
2773         }
2774
2775         if (ret)
2776                 return ret;
2777
2778         return count;
2779 }
2780
2781 static const struct file_operations fops_kp = {
2782         .read =         read_enabled_file_bool,
2783         .write =        write_enabled_file_bool,
2784         .llseek =       default_llseek,
2785 };
2786
2787 static int __init debugfs_kprobe_init(void)
2788 {
2789         struct dentry *dir;
2790         unsigned int value = 1;
2791
2792         dir = debugfs_create_dir("kprobes", NULL);
2793
2794         debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2795
2796         debugfs_create_file("enabled", 0600, dir, &value, &fops_kp);
2797
2798         debugfs_create_file("blacklist", 0400, dir, NULL,
2799                             &kprobe_blacklist_fops);
2800
2801         return 0;
2802 }
2803
2804 late_initcall(debugfs_kprobe_init);
2805 #endif /* CONFIG_DEBUG_FS */