kernel/kprobes.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  *  Kernel Probes (KProbes)
   4  *
   5  * Copyright (C) IBM Corporation, 2002, 2004
   6  *
   7  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
   8  *              Probes initial implementation (includes suggestions from
   9  *              Rusty Russell).
  10  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  11  *              hlists and exceptions notifier as suggested by Andi Kleen.
  12  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  13  *              interface to access function arguments.
  14  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  15  *              exceptions notifier to be first on the priority list.
  16  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  17  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  18  *              <prasanna@in.ibm.com> added function-return probes.
  19  */
  20
  21 #define pr_fmt(fmt) "kprobes: " fmt
  22
  23 #include <linux/kprobes.h>
  24 #include <linux/hash.h>
  25 #include <linux/init.h>
  26 #include <linux/slab.h>
  27 #include <linux/stddef.h>
  28 #include <linux/export.h>
  29 #include <linux/moduleloader.h>
  30 #include <linux/kallsyms.h>
  31 #include <linux/freezer.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/sysctl.h>
  35 #include <linux/kdebug.h>
  36 #include <linux/memory.h>
  37 #include <linux/ftrace.h>
  38 #include <linux/cpu.h>
  39 #include <linux/jump_label.h>
  40 #include <linux/static_call.h>
  41 #include <linux/perf_event.h>
  42
  43 #include <asm/sections.h>
  44 #include <asm/cacheflush.h>
  45 #include <asm/errno.h>
  46 #include <linux/uaccess.h>
  47
  48 #define KPROBE_HASH_BITS 6
  49 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  50
  51
  52 static int kprobes_initialized;
  53 /* kprobe_table can be accessed by
  54  * - Normal hlist traversal and RCU add/del under 'kprobe_mutex' is held.
  55  * Or
  56  * - RCU hlist traversal under disabling preempt (breakpoint handlers)
  57  */
  58 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  59
  60 /* NOTE: change this value only with 'kprobe_mutex' held */
  61 static bool kprobes_all_disarmed;
  62
  63 /* This protects 'kprobe_table' and 'optimizing_list' */
  64 static DEFINE_MUTEX(kprobe_mutex);
  65 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance);
  66
  67 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  68                                         unsigned int __unused)
  69 {
  70         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  71 }
  72
  73 /*
  74  * Blacklist -- list of 'struct kprobe_blacklist_entry' to store info where
  75  * kprobes can not probe.
  76  */
  77 static LIST_HEAD(kprobe_blacklist);
  78
  79 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  80 /*
  81  * 'kprobe::ainsn.insn' points to the copy of the instruction to be
  82  * single-stepped. x86_64, POWER4 and above have no-exec support and
  83  * stepping on the instruction on a vmalloced/kmalloced/data page
  84  * is a recipe for disaster
  85  */
  86 struct kprobe_insn_page {
  87         struct list_head list;
  88         kprobe_opcode_t *insns;         /* Page of instruction slots */
  89         struct kprobe_insn_cache *cache;
  90         int nused;
  91         int ngarbage;
  92         char slot_used[];
  93 };
  94
  95 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
  96         (offsetof(struct kprobe_insn_page, slot_used) + \
  97          (sizeof(char) * (slots)))
  98
  99 static int slots_per_page(struct kprobe_insn_cache *c)
 100 {
 101         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 102 }
 103
 104 enum kprobe_slot_state {
 105         SLOT_CLEAN = 0,
 106         SLOT_DIRTY = 1,
 107         SLOT_USED = 2,
 108 };
 109
 110 void __weak *alloc_insn_page(void)
 111 {
 112         /*
 113          * Use module_alloc() so this page is within +/- 2GB of where the
 114          * kernel image and loaded module images reside. This is required
 115          * for most of the architectures.
 116          * (e.g. x86-64 needs this to handle the %rip-relative fixups.)
 117          */
 118         return module_alloc(PAGE_SIZE);
 119 }
 120
 121 static void free_insn_page(void *page)
 122 {
 123         module_memfree(page);
 124 }
 125
 126 struct kprobe_insn_cache kprobe_insn_slots = {
 127         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 128         .alloc = alloc_insn_page,
 129         .free = free_insn_page,
 130         .sym = KPROBE_INSN_PAGE_SYM,
 131         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 132         .insn_size = MAX_INSN_SIZE,
 133         .nr_garbage = 0,
 134 };
 135 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 136
 137 /**
 138  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 139  * We allocate an executable page if there's no room on existing ones.
 140  */
 141 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 142 {
 143         struct kprobe_insn_page *kip;
 144         kprobe_opcode_t *slot = NULL;
 145
 146         /* Since the slot array is not protected by rcu, we need a mutex */
 147         mutex_lock(&c->mutex);
 148  retry:
 149         rcu_read_lock();
 150         list_for_each_entry_rcu(kip, &c->pages, list) {
 151                 if (kip->nused < slots_per_page(c)) {
 152                         int i;
 153
 154                         for (i = 0; i < slots_per_page(c); i++) {
 155                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 156                                         kip->slot_used[i] = SLOT_USED;
 157                                         kip->nused++;
 158                                         slot = kip->insns + (i * c->insn_size);
 159                                         rcu_read_unlock();
 160                                         goto out;
 161                                 }
 162                         }
 163                         /* kip->nused is broken. Fix it. */
 164                         kip->nused = slots_per_page(c);
 165                         WARN_ON(1);
 166                 }
 167         }
 168         rcu_read_unlock();
 169
 170         /* If there are any garbage slots, collect it and try again. */
 171         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 172                 goto retry;
 173
 174         /* All out of space.  Need to allocate a new page. */
 175         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 176         if (!kip)
 177                 goto out;
 178
 179         kip->insns = c->alloc();
 180         if (!kip->insns) {
 181                 kfree(kip);
 182                 goto out;
 183         }
 184         INIT_LIST_HEAD(&kip->list);
 185         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 186         kip->slot_used[0] = SLOT_USED;
 187         kip->nused = 1;
 188         kip->ngarbage = 0;
 189         kip->cache = c;
 190         list_add_rcu(&kip->list, &c->pages);
 191         slot = kip->insns;
 192
 193         /* Record the perf ksymbol register event after adding the page */
 194         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL, (unsigned long)kip->insns,
 195                            PAGE_SIZE, false, c->sym);
 196 out:
 197         mutex_unlock(&c->mutex);
 198         return slot;
 199 }
 200
 201 /* Return true if all garbages are collected, otherwise false. */
 202 static bool collect_one_slot(struct kprobe_insn_page *kip, int idx)
 203 {
 204         kip->slot_used[idx] = SLOT_CLEAN;
 205         kip->nused--;
 206         if (kip->nused == 0) {
 207                 /*
 208                  * Page is no longer in use.  Free it unless
 209                  * it's the last one.  We keep the last one
 210                  * so as not to have to set it up again the
 211                  * next time somebody inserts a probe.
 212                  */
 213                 if (!list_is_singular(&kip->list)) {
 214                         /*
 215                          * Record perf ksymbol unregister event before removing
 216                          * the page.
 217                          */
 218                         perf_event_ksymbol(PERF_RECORD_KSYMBOL_TYPE_OOL,
 219                                            (unsigned long)kip->insns, PAGE_SIZE, true,
 220                                            kip->cache->sym);
 221                         list_del_rcu(&kip->list);
 222                         synchronize_rcu();
 223                         kip->cache->free(kip->insns);
 224                         kfree(kip);
 225                 }
 226                 return true;
 227         }
 228         return false;
 229 }
 230
 231 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 232 {
 233         struct kprobe_insn_page *kip, *next;
 234
 235         /* Ensure no-one is interrupted on the garbages */
 236         synchronize_rcu();
 237
 238         list_for_each_entry_safe(kip, next, &c->pages, list) {
 239                 int i;
 240
 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                 strscpy(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 void __weak *alloc_optinsn_page(void)
 334 {
 335         return alloc_insn_page();
 336 }
 337
 338 void __weak free_optinsn_page(void *page)
 339 {
 340         free_insn_page(page);
 341 }
 342
 343 /* For optimized_kprobe buffer */
 344 struct kprobe_insn_cache kprobe_optinsn_slots = {
 345         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 346         .alloc = alloc_optinsn_page,
 347         .free = free_optinsn_page,
 348         .sym = KPROBE_OPTINSN_PAGE_SYM,
 349         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 350         /* .insn_size is initialized later */
 351         .nr_garbage = 0,
 352 };
 353 #endif
 354 #endif
 355
 356 /* We have preemption disabled.. so it is safe to use __ versions */
 357 static inline void set_kprobe_instance(struct kprobe *kp)
 358 {
 359         __this_cpu_write(kprobe_instance, kp);
 360 }
 361
 362 static inline void reset_kprobe_instance(void)
 363 {
 364         __this_cpu_write(kprobe_instance, NULL);
 365 }
 366
 367 /*
 368  * This routine is called either:
 369  *      - under the 'kprobe_mutex' - during kprobe_[un]register().
 370  *                              OR
 371  *      - with preemption disabled - from architecture specific code.
 372  */
 373 struct kprobe *get_kprobe(void *addr)
 374 {
 375         struct hlist_head *head;
 376         struct kprobe *p;
 377
 378         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 379         hlist_for_each_entry_rcu(p, head, hlist,
 380                                  lockdep_is_held(&kprobe_mutex)) {
 381                 if (p->addr == addr)
 382                         return p;
 383         }
 384
 385         return NULL;
 386 }
 387 NOKPROBE_SYMBOL(get_kprobe);
 388
 389 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 390
 391 /* Return true if 'p' is an aggregator */
 392 static inline bool kprobe_aggrprobe(struct kprobe *p)
 393 {
 394         return p->pre_handler == aggr_pre_handler;
 395 }
 396
 397 /* Return true if 'p' is unused */
 398 static inline bool kprobe_unused(struct kprobe *p)
 399 {
 400         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 401                list_empty(&p->list);
 402 }
 403
 404 /* Keep all fields in the kprobe consistent. */
 405 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 406 {
 407         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 408         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 409 }
 410
 411 #ifdef CONFIG_OPTPROBES
 412 /* NOTE: This is protected by 'kprobe_mutex'. */
 413 static bool kprobes_allow_optimization;
 414
 415 /*
 416  * Call all 'kprobe::pre_handler' on the list, but ignores its return value.
 417  * This must be called from arch-dep optimized caller.
 418  */
 419 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 420 {
 421         struct kprobe *kp;
 422
 423         list_for_each_entry_rcu(kp, &p->list, list) {
 424                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 425                         set_kprobe_instance(kp);
 426                         kp->pre_handler(kp, regs);
 427                 }
 428                 reset_kprobe_instance();
 429         }
 430 }
 431 NOKPROBE_SYMBOL(opt_pre_handler);
 432
 433 /* Free optimized instructions and optimized_kprobe */
 434 static void free_aggr_kprobe(struct kprobe *p)
 435 {
 436         struct optimized_kprobe *op;
 437
 438         op = container_of(p, struct optimized_kprobe, kp);
 439         arch_remove_optimized_kprobe(op);
 440         arch_remove_kprobe(p);
 441         kfree(op);
 442 }
 443
 444 /* Return true if the kprobe is ready for optimization. */
 445 static inline int kprobe_optready(struct kprobe *p)
 446 {
 447         struct optimized_kprobe *op;
 448
 449         if (kprobe_aggrprobe(p)) {
 450                 op = container_of(p, struct optimized_kprobe, kp);
 451                 return arch_prepared_optinsn(&op->optinsn);
 452         }
 453
 454         return 0;
 455 }
 456
 457 /* Return true if the kprobe is disarmed. Note: p must be on hash list */
 458 static inline bool kprobe_disarmed(struct kprobe *p)
 459 {
 460         struct optimized_kprobe *op;
 461
 462         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 463         if (!kprobe_aggrprobe(p))
 464                 return kprobe_disabled(p);
 465
 466         op = container_of(p, struct optimized_kprobe, kp);
 467
 468         return kprobe_disabled(p) && list_empty(&op->list);
 469 }
 470
 471 /* Return true if the probe is queued on (un)optimizing lists */
 472 static bool kprobe_queued(struct kprobe *p)
 473 {
 474         struct optimized_kprobe *op;
 475
 476         if (kprobe_aggrprobe(p)) {
 477                 op = container_of(p, struct optimized_kprobe, kp);
 478                 if (!list_empty(&op->list))
 479                         return true;
 480         }
 481         return false;
 482 }
 483
 484 /*
 485  * Return an optimized kprobe whose optimizing code replaces
 486  * instructions including 'addr' (exclude breakpoint).
 487  */
 488 static struct kprobe *get_optimized_kprobe(kprobe_opcode_t *addr)
 489 {
 490         int i;
 491         struct kprobe *p = NULL;
 492         struct optimized_kprobe *op;
 493
 494         /* Don't check i == 0, since that is a breakpoint case. */
 495         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH / sizeof(kprobe_opcode_t); i++)
 496                 p = get_kprobe(addr - i);
 497
 498         if (p && kprobe_optready(p)) {
 499                 op = container_of(p, struct optimized_kprobe, kp);
 500                 if (arch_within_optimized_kprobe(op, addr))
 501                         return p;
 502         }
 503
 504         return NULL;
 505 }
 506
 507 /* Optimization staging list, protected by 'kprobe_mutex' */
 508 static LIST_HEAD(optimizing_list);
 509 static LIST_HEAD(unoptimizing_list);
 510 static LIST_HEAD(freeing_list);
 511
 512 static void kprobe_optimizer(struct work_struct *work);
 513 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 514 #define OPTIMIZE_DELAY 5
 515
 516 /*
 517  * Optimize (replace a breakpoint with a jump) kprobes listed on
 518  * 'optimizing_list'.
 519  */
 520 static void do_optimize_kprobes(void)
 521 {
 522         lockdep_assert_held(&text_mutex);
 523         /*
 524          * The optimization/unoptimization refers 'online_cpus' via
 525          * stop_machine() and cpu-hotplug modifies the 'online_cpus'.
 526          * And same time, 'text_mutex' will be held in cpu-hotplug and here.
 527          * This combination can cause a deadlock (cpu-hotplug tries to lock
 528          * 'text_mutex' but stop_machine() can not be done because
 529          * the 'online_cpus' has been changed)
 530          * To avoid this deadlock, caller must have locked cpu-hotplug
 531          * for preventing cpu-hotplug outside of 'text_mutex' locking.
 532          */
 533         lockdep_assert_cpus_held();
 534
 535         /* Optimization never be done when disarmed */
 536         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 537             list_empty(&optimizing_list))
 538                 return;
 539
 540         arch_optimize_kprobes(&optimizing_list);
 541 }
 542
 543 /*
 544  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 545  * if need) kprobes listed on 'unoptimizing_list'.
 546  */
 547 static void do_unoptimize_kprobes(void)
 548 {
 549         struct optimized_kprobe *op, *tmp;
 550
 551         lockdep_assert_held(&text_mutex);
 552         /* See comment in do_optimize_kprobes() */
 553         lockdep_assert_cpus_held();
 554
 555         /* Unoptimization must be done anytime */
 556         if (list_empty(&unoptimizing_list))
 557                 return;
 558
 559         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 560         /* Loop on 'freeing_list' for disarming */
 561         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 562                 /* Switching from detour code to origin */
 563                 op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 564                 /* Disarm probes if marked disabled */
 565                 if (kprobe_disabled(&op->kp))
 566                         arch_disarm_kprobe(&op->kp);
 567                 if (kprobe_unused(&op->kp)) {
 568                         /*
 569                          * Remove unused probes from hash list. After waiting
 570                          * for synchronization, these probes are reclaimed.
 571                          * (reclaiming is done by do_free_cleaned_kprobes().)
 572                          */
 573                         hlist_del_rcu(&op->kp.hlist);
 574                 } else
 575                         list_del_init(&op->list);
 576         }
 577 }
 578
 579 /* Reclaim all kprobes on the 'freeing_list' */
 580 static void do_free_cleaned_kprobes(void)
 581 {
 582         struct optimized_kprobe *op, *tmp;
 583
 584         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 585                 list_del_init(&op->list);
 586                 if (WARN_ON_ONCE(!kprobe_unused(&op->kp))) {
 587                         /*
 588                          * This must not happen, but if there is a kprobe
 589                          * still in use, keep it on kprobes hash list.
 590                          */
 591                         continue;
 592                 }
 593                 free_aggr_kprobe(&op->kp);
 594         }
 595 }
 596
 597 /* Start optimizer after OPTIMIZE_DELAY passed */
 598 static void kick_kprobe_optimizer(void)
 599 {
 600         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 601 }
 602
 603 /* Kprobe jump optimizer */
 604 static void kprobe_optimizer(struct work_struct *work)
 605 {
 606         mutex_lock(&kprobe_mutex);
 607         cpus_read_lock();
 608         mutex_lock(&text_mutex);
 609
 610         /*
 611          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 612          * kprobes before waiting for quiesence period.
 613          */
 614         do_unoptimize_kprobes();
 615
 616         /*
 617          * Step 2: Wait for quiesence period to ensure all potentially
 618          * preempted tasks to have normally scheduled. Because optprobe
 619          * may modify multiple instructions, there is a chance that Nth
 620          * instruction is preempted. In that case, such tasks can return
 621          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
 622          * Note that on non-preemptive kernel, this is transparently converted
 623          * to synchronoze_sched() to wait for all interrupts to have completed.
 624          */
 625         synchronize_rcu_tasks();
 626
 627         /* Step 3: Optimize kprobes after quiesence period */
 628         do_optimize_kprobes();
 629
 630         /* Step 4: Free cleaned kprobes after quiesence period */
 631         do_free_cleaned_kprobes();
 632
 633         mutex_unlock(&text_mutex);
 634         cpus_read_unlock();
 635
 636         /* Step 5: Kick optimizer again if needed */
 637         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 638                 kick_kprobe_optimizer();
 639
 640         mutex_unlock(&kprobe_mutex);
 641 }
 642
 643 /* Wait for completing optimization and unoptimization */
 644 void wait_for_kprobe_optimizer(void)
 645 {
 646         mutex_lock(&kprobe_mutex);
 647
 648         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 649                 mutex_unlock(&kprobe_mutex);
 650
 651                 /* This will also make 'optimizing_work' execute immmediately */
 652                 flush_delayed_work(&optimizing_work);
 653                 /* 'optimizing_work' might not have been queued yet, relax */
 654                 cpu_relax();
 655
 656                 mutex_lock(&kprobe_mutex);
 657         }
 658
 659         mutex_unlock(&kprobe_mutex);
 660 }
 661
 662 static bool optprobe_queued_unopt(struct optimized_kprobe *op)
 663 {
 664         struct optimized_kprobe *_op;
 665
 666         list_for_each_entry(_op, &unoptimizing_list, list) {
 667                 if (op == _op)
 668                         return true;
 669         }
 670
 671         return false;
 672 }
 673
 674 /* Optimize kprobe if p is ready to be optimized */
 675 static void optimize_kprobe(struct kprobe *p)
 676 {
 677         struct optimized_kprobe *op;
 678
 679         /* Check if the kprobe is disabled or not ready for optimization. */
 680         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 681             (kprobe_disabled(p) || kprobes_all_disarmed))
 682                 return;
 683
 684         /* kprobes with 'post_handler' can not be optimized */
 685         if (p->post_handler)
 686                 return;
 687
 688         op = container_of(p, struct optimized_kprobe, kp);
 689
 690         /* Check there is no other kprobes at the optimized instructions */
 691         if (arch_check_optimized_kprobe(op) < 0)
 692                 return;
 693
 694         /* Check if it is already optimized. */
 695         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED) {
 696                 if (optprobe_queued_unopt(op)) {
 697                         /* This is under unoptimizing. Just dequeue the probe */
 698                         list_del_init(&op->list);
 699                 }
 700                 return;
 701         }
 702         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 703
 704         /*
 705          * On the 'unoptimizing_list' and 'optimizing_list',
 706          * 'op' must have OPTIMIZED flag
 707          */
 708         if (WARN_ON_ONCE(!list_empty(&op->list)))
 709                 return;
 710
 711         list_add(&op->list, &optimizing_list);
 712         kick_kprobe_optimizer();
 713 }
 714
 715 /* Short cut to direct unoptimizing */
 716 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 717 {
 718         lockdep_assert_cpus_held();
 719         arch_unoptimize_kprobe(op);
 720         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 721 }
 722
 723 /* Unoptimize a kprobe if p is optimized */
 724 static void unoptimize_kprobe(struct kprobe *p, bool force)
 725 {
 726         struct optimized_kprobe *op;
 727
 728         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 729                 return; /* This is not an optprobe nor optimized */
 730
 731         op = container_of(p, struct optimized_kprobe, kp);
 732         if (!kprobe_optimized(p))
 733                 return;
 734
 735         if (!list_empty(&op->list)) {
 736                 if (optprobe_queued_unopt(op)) {
 737                         /* Queued in unoptimizing queue */
 738                         if (force) {
 739                                 /*
 740                                  * Forcibly unoptimize the kprobe here, and queue it
 741                                  * in the freeing list for release afterwards.
 742                                  */
 743                                 force_unoptimize_kprobe(op);
 744                                 list_move(&op->list, &freeing_list);
 745                         }
 746                 } else {
 747                         /* Dequeue from the optimizing queue */
 748                         list_del_init(&op->list);
 749                         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 750                 }
 751                 return;
 752         }
 753
 754         /* Optimized kprobe case */
 755         if (force) {
 756                 /* Forcibly update the code: this is a special case */
 757                 force_unoptimize_kprobe(op);
 758         } else {
 759                 list_add(&op->list, &unoptimizing_list);
 760                 kick_kprobe_optimizer();
 761         }
 762 }
 763
 764 /* Cancel unoptimizing for reusing */
 765 static int reuse_unused_kprobe(struct kprobe *ap)
 766 {
 767         struct optimized_kprobe *op;
 768
 769         /*
 770          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 771          * there is still a relative jump) and disabled.
 772          */
 773         op = container_of(ap, struct optimized_kprobe, kp);
 774         WARN_ON_ONCE(list_empty(&op->list));
 775         /* Enable the probe again */
 776         ap->flags &= ~KPROBE_FLAG_DISABLED;
 777         /* Optimize it again. (remove from 'op->list') */
 778         if (!kprobe_optready(ap))
 779                 return -EINVAL;
 780
 781         optimize_kprobe(ap);
 782         return 0;
 783 }
 784
 785 /* Remove optimized instructions */
 786 static void kill_optimized_kprobe(struct kprobe *p)
 787 {
 788         struct optimized_kprobe *op;
 789
 790         op = container_of(p, struct optimized_kprobe, kp);
 791         if (!list_empty(&op->list))
 792                 /* Dequeue from the (un)optimization queue */
 793                 list_del_init(&op->list);
 794         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 795
 796         if (kprobe_unused(p)) {
 797                 /* Enqueue if it is unused */
 798                 list_add(&op->list, &freeing_list);
 799                 /*
 800                  * Remove unused probes from the hash list. After waiting
 801                  * for synchronization, this probe is reclaimed.
 802                  * (reclaiming is done by do_free_cleaned_kprobes().)
 803                  */
 804                 hlist_del_rcu(&op->kp.hlist);
 805         }
 806
 807         /* Don't touch the code, because it is already freed. */
 808         arch_remove_optimized_kprobe(op);
 809 }
 810
 811 static inline
 812 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 813 {
 814         if (!kprobe_ftrace(p))
 815                 arch_prepare_optimized_kprobe(op, p);
 816 }
 817
 818 /* Try to prepare optimized instructions */
 819 static void prepare_optimized_kprobe(struct kprobe *p)
 820 {
 821         struct optimized_kprobe *op;
 822
 823         op = container_of(p, struct optimized_kprobe, kp);
 824         __prepare_optimized_kprobe(op, p);
 825 }
 826
 827 /* Allocate new optimized_kprobe and try to prepare optimized instructions. */
 828 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 829 {
 830         struct optimized_kprobe *op;
 831
 832         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 833         if (!op)
 834                 return NULL;
 835
 836         INIT_LIST_HEAD(&op->list);
 837         op->kp.addr = p->addr;
 838         __prepare_optimized_kprobe(op, p);
 839
 840         return &op->kp;
 841 }
 842
 843 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 844
 845 /*
 846  * Prepare an optimized_kprobe and optimize it.
 847  * NOTE: 'p' must be a normal registered kprobe.
 848  */
 849 static void try_to_optimize_kprobe(struct kprobe *p)
 850 {
 851         struct kprobe *ap;
 852         struct optimized_kprobe *op;
 853
 854         /* Impossible to optimize ftrace-based kprobe. */
 855         if (kprobe_ftrace(p))
 856                 return;
 857
 858         /* For preparing optimization, jump_label_text_reserved() is called. */
 859         cpus_read_lock();
 860         jump_label_lock();
 861         mutex_lock(&text_mutex);
 862
 863         ap = alloc_aggr_kprobe(p);
 864         if (!ap)
 865                 goto out;
 866
 867         op = container_of(ap, struct optimized_kprobe, kp);
 868         if (!arch_prepared_optinsn(&op->optinsn)) {
 869                 /* If failed to setup optimizing, fallback to kprobe. */
 870                 arch_remove_optimized_kprobe(op);
 871                 kfree(op);
 872                 goto out;
 873         }
 874
 875         init_aggr_kprobe(ap, p);
 876         optimize_kprobe(ap);    /* This just kicks optimizer thread. */
 877
 878 out:
 879         mutex_unlock(&text_mutex);
 880         jump_label_unlock();
 881         cpus_read_unlock();
 882 }
 883
 884 static void optimize_all_kprobes(void)
 885 {
 886         struct hlist_head *head;
 887         struct kprobe *p;
 888         unsigned int i;
 889
 890         mutex_lock(&kprobe_mutex);
 891         /* If optimization is already allowed, just return. */
 892         if (kprobes_allow_optimization)
 893                 goto out;
 894
 895         cpus_read_lock();
 896         kprobes_allow_optimization = true;
 897         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 898                 head = &kprobe_table[i];
 899                 hlist_for_each_entry(p, head, hlist)
 900                         if (!kprobe_disabled(p))
 901                                 optimize_kprobe(p);
 902         }
 903         cpus_read_unlock();
 904         pr_info("kprobe jump-optimization is enabled. All kprobes are optimized if possible.\n");
 905 out:
 906         mutex_unlock(&kprobe_mutex);
 907 }
 908
 909 #ifdef CONFIG_SYSCTL
 910 static void unoptimize_all_kprobes(void)
 911 {
 912         struct hlist_head *head;
 913         struct kprobe *p;
 914         unsigned int i;
 915
 916         mutex_lock(&kprobe_mutex);
 917         /* If optimization is already prohibited, just return. */
 918         if (!kprobes_allow_optimization) {
 919                 mutex_unlock(&kprobe_mutex);
 920                 return;
 921         }
 922
 923         cpus_read_lock();
 924         kprobes_allow_optimization = false;
 925         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 926                 head = &kprobe_table[i];
 927                 hlist_for_each_entry(p, head, hlist) {
 928                         if (!kprobe_disabled(p))
 929                                 unoptimize_kprobe(p, false);
 930                 }
 931         }
 932         cpus_read_unlock();
 933         mutex_unlock(&kprobe_mutex);
 934
 935         /* Wait for unoptimizing completion. */
 936         wait_for_kprobe_optimizer();
 937         pr_info("kprobe jump-optimization is disabled. All kprobes are based on software breakpoint.\n");
 938 }
 939
 940 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 941 int sysctl_kprobes_optimization;
 942 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 943                                       void *buffer, size_t *length,
 944                                       loff_t *ppos)
 945 {
 946         int ret;
 947
 948         mutex_lock(&kprobe_sysctl_mutex);
 949         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 950         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 951
 952         if (sysctl_kprobes_optimization)
 953                 optimize_all_kprobes();
 954         else
 955                 unoptimize_all_kprobes();
 956         mutex_unlock(&kprobe_sysctl_mutex);
 957
 958         return ret;
 959 }
 960 #endif /* CONFIG_SYSCTL */
 961
 962 /* Put a breakpoint for a probe. */
 963 static void __arm_kprobe(struct kprobe *p)
 964 {
 965         struct kprobe *_p;
 966
 967         lockdep_assert_held(&text_mutex);
 968
 969         /* Find the overlapping optimized kprobes. */
 970         _p = get_optimized_kprobe(p->addr);
 971         if (unlikely(_p))
 972                 /* Fallback to unoptimized kprobe */
 973                 unoptimize_kprobe(_p, true);
 974
 975         arch_arm_kprobe(p);
 976         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 977 }
 978
 979 /* Remove the breakpoint of a probe. */
 980 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 981 {
 982         struct kprobe *_p;
 983
 984         lockdep_assert_held(&text_mutex);
 985
 986         /* Try to unoptimize */
 987         unoptimize_kprobe(p, kprobes_all_disarmed);
 988
 989         if (!kprobe_queued(p)) {
 990                 arch_disarm_kprobe(p);
 991                 /* If another kprobe was blocked, re-optimize it. */
 992                 _p = get_optimized_kprobe(p->addr);
 993                 if (unlikely(_p) && reopt)
 994                         optimize_kprobe(_p);
 995         }
 996         /*
 997          * TODO: Since unoptimization and real disarming will be done by
 998          * the worker thread, we can not check whether another probe are
 999          * unoptimized because of this probe here. It should be re-optimized
1000          * by the worker thread.
1001          */
1002 }
1003
1004 #else /* !CONFIG_OPTPROBES */
1005
1006 #define optimize_kprobe(p)                      do {} while (0)
1007 #define unoptimize_kprobe(p, f)                 do {} while (0)
1008 #define kill_optimized_kprobe(p)                do {} while (0)
1009 #define prepare_optimized_kprobe(p)             do {} while (0)
1010 #define try_to_optimize_kprobe(p)               do {} while (0)
1011 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
1012 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
1013 #define kprobe_disarmed(p)                      kprobe_disabled(p)
1014 #define wait_for_kprobe_optimizer()             do {} while (0)
1015
1016 static int reuse_unused_kprobe(struct kprobe *ap)
1017 {
1018         /*
1019          * If the optimized kprobe is NOT supported, the aggr kprobe is
1020          * released at the same time that the last aggregated kprobe is
1021          * unregistered.
1022          * Thus there should be no chance to reuse unused kprobe.
1023          */
1024         WARN_ON_ONCE(1);
1025         return -EINVAL;
1026 }
1027
1028 static void free_aggr_kprobe(struct kprobe *p)
1029 {
1030         arch_remove_kprobe(p);
1031         kfree(p);
1032 }
1033
1034 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
1035 {
1036         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
1037 }
1038 #endif /* CONFIG_OPTPROBES */
1039
1040 #ifdef CONFIG_KPROBES_ON_FTRACE
1041 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
1042         .func = kprobe_ftrace_handler,
1043         .flags = FTRACE_OPS_FL_SAVE_REGS,
1044 };
1045
1046 static struct ftrace_ops kprobe_ipmodify_ops __read_mostly = {
1047         .func = kprobe_ftrace_handler,
1048         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
1049 };
1050
1051 static int kprobe_ipmodify_enabled;
1052 static int kprobe_ftrace_enabled;
1053
1054 static int __arm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1055                                int *cnt)
1056 {
1057         int ret = 0;
1058
1059         lockdep_assert_held(&kprobe_mutex);
1060
1061         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 0, 0);
1062         if (WARN_ONCE(ret < 0, "Failed to arm kprobe-ftrace at %pS (error %d)\n", p->addr, ret))
1063                 return ret;
1064
1065         if (*cnt == 0) {
1066                 ret = register_ftrace_function(ops);
1067                 if (WARN(ret < 0, "Failed to register kprobe-ftrace (error %d)\n", ret))
1068                         goto err_ftrace;
1069         }
1070
1071         (*cnt)++;
1072         return ret;
1073
1074 err_ftrace:
1075         /*
1076          * At this point, sinec ops is not registered, we should be sefe from
1077          * registering empty filter.
1078          */
1079         ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1080         return ret;
1081 }
1082
1083 static int arm_kprobe_ftrace(struct kprobe *p)
1084 {
1085         bool ipmodify = (p->post_handler != NULL);
1086
1087         return __arm_kprobe_ftrace(p,
1088                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1089                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1090 }
1091
1092 static int __disarm_kprobe_ftrace(struct kprobe *p, struct ftrace_ops *ops,
1093                                   int *cnt)
1094 {
1095         int ret = 0;
1096
1097         lockdep_assert_held(&kprobe_mutex);
1098
1099         if (*cnt == 1) {
1100                 ret = unregister_ftrace_function(ops);
1101                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (error %d)\n", ret))
1102                         return ret;
1103         }
1104
1105         (*cnt)--;
1106
1107         ret = ftrace_set_filter_ip(ops, (unsigned long)p->addr, 1, 0);
1108         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (error %d)\n",
1109                   p->addr, ret);
1110         return ret;
1111 }
1112
1113 static int disarm_kprobe_ftrace(struct kprobe *p)
1114 {
1115         bool ipmodify = (p->post_handler != NULL);
1116
1117         return __disarm_kprobe_ftrace(p,
1118                 ipmodify ? &kprobe_ipmodify_ops : &kprobe_ftrace_ops,
1119                 ipmodify ? &kprobe_ipmodify_enabled : &kprobe_ftrace_enabled);
1120 }
1121 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1122 static inline int arm_kprobe_ftrace(struct kprobe *p)
1123 {
1124         return -ENODEV;
1125 }
1126
1127 static inline int disarm_kprobe_ftrace(struct kprobe *p)
1128 {
1129         return -ENODEV;
1130 }
1131 #endif
1132
1133 static int prepare_kprobe(struct kprobe *p)
1134 {
1135         /* Must ensure p->addr is really on ftrace */
1136         if (kprobe_ftrace(p))
1137                 return arch_prepare_kprobe_ftrace(p);
1138
1139         return arch_prepare_kprobe(p);
1140 }
1141
1142 static int arm_kprobe(struct kprobe *kp)
1143 {
1144         if (unlikely(kprobe_ftrace(kp)))
1145                 return arm_kprobe_ftrace(kp);
1146
1147         cpus_read_lock();
1148         mutex_lock(&text_mutex);
1149         __arm_kprobe(kp);
1150         mutex_unlock(&text_mutex);
1151         cpus_read_unlock();
1152
1153         return 0;
1154 }
1155
1156 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1157 {
1158         if (unlikely(kprobe_ftrace(kp)))
1159                 return disarm_kprobe_ftrace(kp);
1160
1161         cpus_read_lock();
1162         mutex_lock(&text_mutex);
1163         __disarm_kprobe(kp, reopt);
1164         mutex_unlock(&text_mutex);
1165         cpus_read_unlock();
1166
1167         return 0;
1168 }
1169
1170 /*
1171  * Aggregate handlers for multiple kprobes support - these handlers
1172  * take care of invoking the individual kprobe handlers on p->list
1173  */
1174 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1175 {
1176         struct kprobe *kp;
1177
1178         list_for_each_entry_rcu(kp, &p->list, list) {
1179                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1180                         set_kprobe_instance(kp);
1181                         if (kp->pre_handler(kp, regs))
1182                                 return 1;
1183                 }
1184                 reset_kprobe_instance();
1185         }
1186         return 0;
1187 }
1188 NOKPROBE_SYMBOL(aggr_pre_handler);
1189
1190 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1191                               unsigned long flags)
1192 {
1193         struct kprobe *kp;
1194
1195         list_for_each_entry_rcu(kp, &p->list, list) {
1196                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1197                         set_kprobe_instance(kp);
1198                         kp->post_handler(kp, regs, flags);
1199                         reset_kprobe_instance();
1200                 }
1201         }
1202 }
1203 NOKPROBE_SYMBOL(aggr_post_handler);
1204
1205 /* Walks the list and increments 'nmissed' if 'p' has child probes. */
1206 void kprobes_inc_nmissed_count(struct kprobe *p)
1207 {
1208         struct kprobe *kp;
1209
1210         if (!kprobe_aggrprobe(p)) {
1211                 p->nmissed++;
1212         } else {
1213                 list_for_each_entry_rcu(kp, &p->list, list)
1214                         kp->nmissed++;
1215         }
1216 }
1217 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1218
1219 static void free_rp_inst_rcu(struct rcu_head *head)
1220 {
1221         struct kretprobe_instance *ri = container_of(head, struct kretprobe_instance, rcu);
1222
1223         if (refcount_dec_and_test(&ri->rph->ref))
1224                 kfree(ri->rph);
1225         kfree(ri);
1226 }
1227 NOKPROBE_SYMBOL(free_rp_inst_rcu);
1228
1229 static void recycle_rp_inst(struct kretprobe_instance *ri)
1230 {
1231         struct kretprobe *rp = get_kretprobe(ri);
1232
1233         if (likely(rp))
1234                 freelist_add(&ri->freelist, &rp->freelist);
1235         else
1236                 call_rcu(&ri->rcu, free_rp_inst_rcu);
1237 }
1238 NOKPROBE_SYMBOL(recycle_rp_inst);
1239
1240 static struct kprobe kprobe_busy = {
1241         .addr = (void *) get_kprobe,
1242 };
1243
1244 void kprobe_busy_begin(void)
1245 {
1246         struct kprobe_ctlblk *kcb;
1247
1248         preempt_disable();
1249         __this_cpu_write(current_kprobe, &kprobe_busy);
1250         kcb = get_kprobe_ctlblk();
1251         kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1252 }
1253
1254 void kprobe_busy_end(void)
1255 {
1256         __this_cpu_write(current_kprobe, NULL);
1257         preempt_enable();
1258 }
1259
1260 /*
1261  * This function is called from delayed_put_task_struct() when a task is
1262  * dead and cleaned up to recycle any kretprobe instances associated with
1263  * this task. These left over instances represent probed functions that
1264  * have been called but will never return.
1265  */
1266 void kprobe_flush_task(struct task_struct *tk)
1267 {
1268         struct kretprobe_instance *ri;
1269         struct llist_node *node;
1270
1271         /* Early boot, not yet initialized. */
1272         if (unlikely(!kprobes_initialized))
1273                 return;
1274
1275         kprobe_busy_begin();
1276
1277         node = __llist_del_all(&tk->kretprobe_instances);
1278         while (node) {
1279                 ri = container_of(node, struct kretprobe_instance, llist);
1280                 node = node->next;
1281
1282                 recycle_rp_inst(ri);
1283         }
1284
1285         kprobe_busy_end();
1286 }
1287 NOKPROBE_SYMBOL(kprobe_flush_task);
1288
1289 static inline void free_rp_inst(struct kretprobe *rp)
1290 {
1291         struct kretprobe_instance *ri;
1292         struct freelist_node *node;
1293         int count = 0;
1294
1295         node = rp->freelist.head;
1296         while (node) {
1297                 ri = container_of(node, struct kretprobe_instance, freelist);
1298                 node = node->next;
1299
1300                 kfree(ri);
1301                 count++;
1302         }
1303
1304         if (refcount_sub_and_test(count, &rp->rph->ref)) {
1305                 kfree(rp->rph);
1306                 rp->rph = NULL;
1307         }
1308 }
1309
1310 /* Add the new probe to 'ap->list'. */
1311 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1312 {
1313         if (p->post_handler)
1314                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1315
1316         list_add_rcu(&p->list, &ap->list);
1317         if (p->post_handler && !ap->post_handler)
1318                 ap->post_handler = aggr_post_handler;
1319
1320         return 0;
1321 }
1322
1323 /*
1324  * Fill in the required fields of the aggregator kprobe. Replace the
1325  * earlier kprobe in the hlist with the aggregator kprobe.
1326  */
1327 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1328 {
1329         /* Copy the insn slot of 'p' to 'ap'. */
1330         copy_kprobe(p, ap);
1331         flush_insn_slot(ap);
1332         ap->addr = p->addr;
1333         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1334         ap->pre_handler = aggr_pre_handler;
1335         /* We don't care the kprobe which has gone. */
1336         if (p->post_handler && !kprobe_gone(p))
1337                 ap->post_handler = aggr_post_handler;
1338
1339         INIT_LIST_HEAD(&ap->list);
1340         INIT_HLIST_NODE(&ap->hlist);
1341
1342         list_add_rcu(&p->list, &ap->list);
1343         hlist_replace_rcu(&p->hlist, &ap->hlist);
1344 }
1345
1346 /*
1347  * This registers the second or subsequent kprobe at the same address.
1348  */
1349 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1350 {
1351         int ret = 0;
1352         struct kprobe *ap = orig_p;
1353
1354         cpus_read_lock();
1355
1356         /* For preparing optimization, jump_label_text_reserved() is called */
1357         jump_label_lock();
1358         mutex_lock(&text_mutex);
1359
1360         if (!kprobe_aggrprobe(orig_p)) {
1361                 /* If 'orig_p' is not an 'aggr_kprobe', create new one. */
1362                 ap = alloc_aggr_kprobe(orig_p);
1363                 if (!ap) {
1364                         ret = -ENOMEM;
1365                         goto out;
1366                 }
1367                 init_aggr_kprobe(ap, orig_p);
1368         } else if (kprobe_unused(ap)) {
1369                 /* This probe is going to die. Rescue it */
1370                 ret = reuse_unused_kprobe(ap);
1371                 if (ret)
1372                         goto out;
1373         }
1374
1375         if (kprobe_gone(ap)) {
1376                 /*
1377                  * Attempting to insert new probe at the same location that
1378                  * had a probe in the module vaddr area which already
1379                  * freed. So, the instruction slot has already been
1380                  * released. We need a new slot for the new probe.
1381                  */
1382                 ret = arch_prepare_kprobe(ap);
1383                 if (ret)
1384                         /*
1385                          * Even if fail to allocate new slot, don't need to
1386                          * free the 'ap'. It will be used next time, or
1387                          * freed by unregister_kprobe().
1388                          */
1389                         goto out;
1390
1391                 /* Prepare optimized instructions if possible. */
1392                 prepare_optimized_kprobe(ap);
1393
1394                 /*
1395                  * Clear gone flag to prevent allocating new slot again, and
1396                  * set disabled flag because it is not armed yet.
1397                  */
1398                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1399                             | KPROBE_FLAG_DISABLED;
1400         }
1401
1402         /* Copy the insn slot of 'p' to 'ap'. */
1403         copy_kprobe(ap, p);
1404         ret = add_new_kprobe(ap, p);
1405
1406 out:
1407         mutex_unlock(&text_mutex);
1408         jump_label_unlock();
1409         cpus_read_unlock();
1410
1411         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1412                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1413                 if (!kprobes_all_disarmed) {
1414                         /* Arm the breakpoint again. */
1415                         ret = arm_kprobe(ap);
1416                         if (ret) {
1417                                 ap->flags |= KPROBE_FLAG_DISABLED;
1418                                 list_del_rcu(&p->list);
1419                                 synchronize_rcu();
1420                         }
1421                 }
1422         }
1423         return ret;
1424 }
1425
1426 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1427 {
1428         /* The '__kprobes' functions and entry code must not be probed. */
1429         return addr >= (unsigned long)__kprobes_text_start &&
1430                addr < (unsigned long)__kprobes_text_end;
1431 }
1432
1433 static bool __within_kprobe_blacklist(unsigned long addr)
1434 {
1435         struct kprobe_blacklist_entry *ent;
1436
1437         if (arch_within_kprobe_blacklist(addr))
1438                 return true;
1439         /*
1440          * If 'kprobe_blacklist' is defined, check the address and
1441          * reject any probe registration in the prohibited area.
1442          */
1443         list_for_each_entry(ent, &kprobe_blacklist, list) {
1444                 if (addr >= ent->start_addr && addr < ent->end_addr)
1445                         return true;
1446         }
1447         return false;
1448 }
1449
1450 bool within_kprobe_blacklist(unsigned long addr)
1451 {
1452         char symname[KSYM_NAME_LEN], *p;
1453
1454         if (__within_kprobe_blacklist(addr))
1455                 return true;
1456
1457         /* Check if the address is on a suffixed-symbol */
1458         if (!lookup_symbol_name(addr, symname)) {
1459                 p = strchr(symname, '.');
1460                 if (!p)
1461                         return false;
1462                 *p = '\0';
1463                 addr = (unsigned long)kprobe_lookup_name(symname, 0);
1464                 if (addr)
1465                         return __within_kprobe_blacklist(addr);
1466         }
1467         return false;
1468 }
1469
1470 /*
1471  * If 'symbol_name' is specified, look it up and add the 'offset'
1472  * to it. This way, we can specify a relative address to a symbol.
1473  * This returns encoded errors if it fails to look up symbol or invalid
1474  * combination of parameters.
1475  */
1476 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1477                         const char *symbol_name, unsigned int offset)
1478 {
1479         if ((symbol_name && addr) || (!symbol_name && !addr))
1480                 goto invalid;
1481
1482         if (symbol_name) {
1483                 addr = kprobe_lookup_name(symbol_name, offset);
1484                 if (!addr)
1485                         return ERR_PTR(-ENOENT);
1486         }
1487
1488         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1489         if (addr)
1490                 return addr;
1491
1492 invalid:
1493         return ERR_PTR(-EINVAL);
1494 }
1495
1496 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1497 {
1498         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1499 }
1500
1501 /*
1502  * Check the 'p' is valid and return the aggregator kprobe
1503  * at the same address.
1504  */
1505 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1506 {
1507         struct kprobe *ap, *list_p;
1508
1509         lockdep_assert_held(&kprobe_mutex);
1510
1511         ap = get_kprobe(p->addr);
1512         if (unlikely(!ap))
1513                 return NULL;
1514
1515         if (p != ap) {
1516                 list_for_each_entry(list_p, &ap->list, list)
1517                         if (list_p == p)
1518                         /* kprobe p is a valid probe */
1519                                 goto valid;
1520                 return NULL;
1521         }
1522 valid:
1523         return ap;
1524 }
1525
1526 /*
1527  * Warn and return error if the kprobe is being re-registered since
1528  * there must be a software bug.
1529  */
1530 static inline int warn_kprobe_rereg(struct kprobe *p)
1531 {
1532         int ret = 0;
1533
1534         mutex_lock(&kprobe_mutex);
1535         if (WARN_ON_ONCE(__get_valid_kprobe(p)))
1536                 ret = -EINVAL;
1537         mutex_unlock(&kprobe_mutex);
1538
1539         return ret;
1540 }
1541
1542 static int check_ftrace_location(struct kprobe *p)
1543 {
1544         unsigned long ftrace_addr;
1545
1546         ftrace_addr = ftrace_location((unsigned long)p->addr);
1547         if (ftrace_addr) {
1548 #ifdef CONFIG_KPROBES_ON_FTRACE
1549                 /* Given address is not on the instruction boundary */
1550                 if ((unsigned long)p->addr != ftrace_addr)
1551                         return -EILSEQ;
1552                 p->flags |= KPROBE_FLAG_FTRACE;
1553 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1554                 return -EINVAL;
1555 #endif
1556         }
1557         return 0;
1558 }
1559
1560 static int check_kprobe_address_safe(struct kprobe *p,
1561                                      struct module **probed_mod)
1562 {
1563         int ret;
1564
1565         ret = check_ftrace_location(p);
1566         if (ret)
1567                 return ret;
1568         jump_label_lock();
1569         preempt_disable();
1570
1571         /* Ensure it is not in reserved area nor out of text */
1572         if (!kernel_text_address((unsigned long) p->addr) ||
1573             within_kprobe_blacklist((unsigned long) p->addr) ||
1574             jump_label_text_reserved(p->addr, p->addr) ||
1575             static_call_text_reserved(p->addr, p->addr) ||
1576             find_bug((unsigned long)p->addr)) {
1577                 ret = -EINVAL;
1578                 goto out;
1579         }
1580
1581         /* Check if 'p' is probing a module. */
1582         *probed_mod = __module_text_address((unsigned long) p->addr);
1583         if (*probed_mod) {
1584                 /*
1585                  * We must hold a refcount of the probed module while updating
1586                  * its code to prohibit unexpected unloading.
1587                  */
1588                 if (unlikely(!try_module_get(*probed_mod))) {
1589                         ret = -ENOENT;
1590                         goto out;
1591                 }
1592
1593                 /*
1594                  * If the module freed '.init.text', we couldn't insert
1595                  * kprobes in there.
1596                  */
1597                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1598                     (*probed_mod)->state != MODULE_STATE_COMING) {
1599                         module_put(*probed_mod);
1600                         *probed_mod = NULL;
1601                         ret = -ENOENT;
1602                 }
1603         }
1604 out:
1605         preempt_enable();
1606         jump_label_unlock();
1607
1608         return ret;
1609 }
1610
1611 int register_kprobe(struct kprobe *p)
1612 {
1613         int ret;
1614         struct kprobe *old_p;
1615         struct module *probed_mod;
1616         kprobe_opcode_t *addr;
1617
1618         /* Adjust probe address from symbol */
1619         addr = kprobe_addr(p);
1620         if (IS_ERR(addr))
1621                 return PTR_ERR(addr);
1622         p->addr = addr;
1623
1624         ret = warn_kprobe_rereg(p);
1625         if (ret)
1626                 return ret;
1627
1628         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1629         p->flags &= KPROBE_FLAG_DISABLED;
1630         p->nmissed = 0;
1631         INIT_LIST_HEAD(&p->list);
1632
1633         ret = check_kprobe_address_safe(p, &probed_mod);
1634         if (ret)
1635                 return ret;
1636
1637         mutex_lock(&kprobe_mutex);
1638
1639         old_p = get_kprobe(p->addr);
1640         if (old_p) {
1641                 /* Since this may unoptimize 'old_p', locking 'text_mutex'. */
1642                 ret = register_aggr_kprobe(old_p, p);
1643                 goto out;
1644         }
1645
1646         cpus_read_lock();
1647         /* Prevent text modification */
1648         mutex_lock(&text_mutex);
1649         ret = prepare_kprobe(p);
1650         mutex_unlock(&text_mutex);
1651         cpus_read_unlock();
1652         if (ret)
1653                 goto out;
1654
1655         INIT_HLIST_NODE(&p->hlist);
1656         hlist_add_head_rcu(&p->hlist,
1657                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1658
1659         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1660                 ret = arm_kprobe(p);
1661                 if (ret) {
1662                         hlist_del_rcu(&p->hlist);
1663                         synchronize_rcu();
1664                         goto out;
1665                 }
1666         }
1667
1668         /* Try to optimize kprobe */
1669         try_to_optimize_kprobe(p);
1670 out:
1671         mutex_unlock(&kprobe_mutex);
1672
1673         if (probed_mod)
1674                 module_put(probed_mod);
1675
1676         return ret;
1677 }
1678 EXPORT_SYMBOL_GPL(register_kprobe);
1679
1680 /* Check if all probes on the 'ap' are disabled. */
1681 static bool aggr_kprobe_disabled(struct kprobe *ap)
1682 {
1683         struct kprobe *kp;
1684
1685         lockdep_assert_held(&kprobe_mutex);
1686
1687         list_for_each_entry(kp, &ap->list, list)
1688                 if (!kprobe_disabled(kp))
1689                         /*
1690                          * Since there is an active probe on the list,
1691                          * we can't disable this 'ap'.
1692                          */
1693                         return false;
1694
1695         return true;
1696 }
1697
1698 static struct kprobe *__disable_kprobe(struct kprobe *p)
1699 {
1700         struct kprobe *orig_p;
1701         int ret;
1702
1703         lockdep_assert_held(&kprobe_mutex);
1704
1705         /* Get an original kprobe for return */
1706         orig_p = __get_valid_kprobe(p);
1707         if (unlikely(orig_p == NULL))
1708                 return ERR_PTR(-EINVAL);
1709
1710         if (!kprobe_disabled(p)) {
1711                 /* Disable probe if it is a child probe */
1712                 if (p != orig_p)
1713                         p->flags |= KPROBE_FLAG_DISABLED;
1714
1715                 /* Try to disarm and disable this/parent probe */
1716                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1717                         /*
1718                          * If 'kprobes_all_disarmed' is set, 'orig_p'
1719                          * should have already been disarmed, so
1720                          * skip unneed disarming process.
1721                          */
1722                         if (!kprobes_all_disarmed) {
1723                                 ret = disarm_kprobe(orig_p, true);
1724                                 if (ret) {
1725                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1726                                         return ERR_PTR(ret);
1727                                 }
1728                         }
1729                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1730                 }
1731         }
1732
1733         return orig_p;
1734 }
1735
1736 /*
1737  * Unregister a kprobe without a scheduler synchronization.
1738  */
1739 static int __unregister_kprobe_top(struct kprobe *p)
1740 {
1741         struct kprobe *ap, *list_p;
1742
1743         /* Disable kprobe. This will disarm it if needed. */
1744         ap = __disable_kprobe(p);
1745         if (IS_ERR(ap))
1746                 return PTR_ERR(ap);
1747
1748         if (ap == p)
1749                 /*
1750                  * This probe is an independent(and non-optimized) kprobe
1751                  * (not an aggrprobe). Remove from the hash list.
1752                  */
1753                 goto disarmed;
1754
1755         /* Following process expects this probe is an aggrprobe */
1756         WARN_ON(!kprobe_aggrprobe(ap));
1757
1758         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1759                 /*
1760                  * !disarmed could be happen if the probe is under delayed
1761                  * unoptimizing.
1762                  */
1763                 goto disarmed;
1764         else {
1765                 /* If disabling probe has special handlers, update aggrprobe */
1766                 if (p->post_handler && !kprobe_gone(p)) {
1767                         list_for_each_entry(list_p, &ap->list, list) {
1768                                 if ((list_p != p) && (list_p->post_handler))
1769                                         goto noclean;
1770                         }
1771                         ap->post_handler = NULL;
1772                 }
1773 noclean:
1774                 /*
1775                  * Remove from the aggrprobe: this path will do nothing in
1776                  * __unregister_kprobe_bottom().
1777                  */
1778                 list_del_rcu(&p->list);
1779                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1780                         /*
1781                          * Try to optimize this probe again, because post
1782                          * handler may have been changed.
1783                          */
1784                         optimize_kprobe(ap);
1785         }
1786         return 0;
1787
1788 disarmed:
1789         hlist_del_rcu(&ap->hlist);
1790         return 0;
1791 }
1792
1793 static void __unregister_kprobe_bottom(struct kprobe *p)
1794 {
1795         struct kprobe *ap;
1796
1797         if (list_empty(&p->list))
1798                 /* This is an independent kprobe */
1799                 arch_remove_kprobe(p);
1800         else if (list_is_singular(&p->list)) {
1801                 /* This is the last child of an aggrprobe */
1802                 ap = list_entry(p->list.next, struct kprobe, list);
1803                 list_del(&p->list);
1804                 free_aggr_kprobe(ap);
1805         }
1806         /* Otherwise, do nothing. */
1807 }
1808
1809 int register_kprobes(struct kprobe **kps, int num)
1810 {
1811         int i, ret = 0;
1812
1813         if (num <= 0)
1814                 return -EINVAL;
1815         for (i = 0; i < num; i++) {
1816                 ret = register_kprobe(kps[i]);
1817                 if (ret < 0) {
1818                         if (i > 0)
1819                                 unregister_kprobes(kps, i);
1820                         break;
1821                 }
1822         }
1823         return ret;
1824 }
1825 EXPORT_SYMBOL_GPL(register_kprobes);
1826
1827 void unregister_kprobe(struct kprobe *p)
1828 {
1829         unregister_kprobes(&p, 1);
1830 }
1831 EXPORT_SYMBOL_GPL(unregister_kprobe);
1832
1833 void unregister_kprobes(struct kprobe **kps, int num)
1834 {
1835         int i;
1836
1837         if (num <= 0)
1838                 return;
1839         mutex_lock(&kprobe_mutex);
1840         for (i = 0; i < num; i++)
1841                 if (__unregister_kprobe_top(kps[i]) < 0)
1842                         kps[i]->addr = NULL;
1843         mutex_unlock(&kprobe_mutex);
1844
1845         synchronize_rcu();
1846         for (i = 0; i < num; i++)
1847                 if (kps[i]->addr)
1848                         __unregister_kprobe_bottom(kps[i]);
1849 }
1850 EXPORT_SYMBOL_GPL(unregister_kprobes);
1851
1852 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1853                                         unsigned long val, void *data)
1854 {
1855         return NOTIFY_DONE;
1856 }
1857 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1858
1859 static struct notifier_block kprobe_exceptions_nb = {
1860         .notifier_call = kprobe_exceptions_notify,
1861         .priority = 0x7fffffff /* we need to be notified first */
1862 };
1863
1864 #ifdef CONFIG_KRETPROBES
1865
1866 /* This assumes the 'tsk' is the current task or the is not running. */
1867 static kprobe_opcode_t *__kretprobe_find_ret_addr(struct task_struct *tsk,
1868                                                   struct llist_node **cur)
1869 {
1870         struct kretprobe_instance *ri = NULL;
1871         struct llist_node *node = *cur;
1872
1873         if (!node)
1874                 node = tsk->kretprobe_instances.first;
1875         else
1876                 node = node->next;
1877
1878         while (node) {
1879                 ri = container_of(node, struct kretprobe_instance, llist);
1880                 if (ri->ret_addr != kretprobe_trampoline_addr()) {
1881                         *cur = node;
1882                         return ri->ret_addr;
1883                 }
1884                 node = node->next;
1885         }
1886         return NULL;
1887 }
1888 NOKPROBE_SYMBOL(__kretprobe_find_ret_addr);
1889
1890 /**
1891  * kretprobe_find_ret_addr -- Find correct return address modified by kretprobe
1892  * @tsk: Target task
1893  * @fp: A frame pointer
1894  * @cur: a storage of the loop cursor llist_node pointer for next call
1895  *
1896  * Find the correct return address modified by a kretprobe on @tsk in unsigned
1897  * long type. If it finds the return address, this returns that address value,
1898  * or this returns 0.
1899  * The @tsk must be 'current' or a task which is not running. @fp is a hint
1900  * to get the currect return address - which is compared with the
1901  * kretprobe_instance::fp field. The @cur is a loop cursor for searching the
1902  * kretprobe return addresses on the @tsk. The '*@cur' should be NULL at the
1903  * first call, but '@cur' itself must NOT NULL.
1904  */
1905 unsigned long kretprobe_find_ret_addr(struct task_struct *tsk, void *fp,
1906                                       struct llist_node **cur)
1907 {
1908         struct kretprobe_instance *ri = NULL;
1909         kprobe_opcode_t *ret;
1910
1911         if (WARN_ON_ONCE(!cur))
1912                 return 0;
1913
1914         do {
1915                 ret = __kretprobe_find_ret_addr(tsk, cur);
1916                 if (!ret)
1917                         break;
1918                 ri = container_of(*cur, struct kretprobe_instance, llist);
1919         } while (ri->fp != fp);
1920
1921         return (unsigned long)ret;
1922 }
1923 NOKPROBE_SYMBOL(kretprobe_find_ret_addr);
1924
1925 void __weak arch_kretprobe_fixup_return(struct pt_regs *regs,
1926                                         kprobe_opcode_t *correct_ret_addr)
1927 {
1928         /*
1929          * Do nothing by default. Please fill this to update the fake return
1930          * address on the stack with the correct one on each arch if possible.
1931          */
1932 }
1933
1934 unsigned long __kretprobe_trampoline_handler(struct pt_regs *regs,
1935                                              void *frame_pointer)
1936 {
1937         kprobe_opcode_t *correct_ret_addr = NULL;
1938         struct kretprobe_instance *ri = NULL;
1939         struct llist_node *first, *node = NULL;
1940         struct kretprobe *rp;
1941
1942         /* Find correct address and all nodes for this frame. */
1943         correct_ret_addr = __kretprobe_find_ret_addr(current, &node);
1944         if (!correct_ret_addr) {
1945                 pr_err("kretprobe: Return address not found, not execute handler. Maybe there is a bug in the kernel.\n");
1946                 BUG_ON(1);
1947         }
1948
1949         /*
1950          * Set the return address as the instruction pointer, because if the
1951          * user handler calls stack_trace_save_regs() with this 'regs',
1952          * the stack trace will start from the instruction pointer.
1953          */
1954         instruction_pointer_set(regs, (unsigned long)correct_ret_addr);
1955
1956         /* Run the user handler of the nodes. */
1957         first = current->kretprobe_instances.first;
1958         while (first) {
1959                 ri = container_of(first, struct kretprobe_instance, llist);
1960
1961                 if (WARN_ON_ONCE(ri->fp != frame_pointer))
1962                         break;
1963
1964                 rp = get_kretprobe(ri);
1965                 if (rp && rp->handler) {
1966                         struct kprobe *prev = kprobe_running();
1967
1968                         __this_cpu_write(current_kprobe, &rp->kp);
1969                         ri->ret_addr = correct_ret_addr;
1970                         rp->handler(ri, regs);
1971                         __this_cpu_write(current_kprobe, prev);
1972                 }
1973                 if (first == node)
1974                         break;
1975
1976                 first = first->next;
1977         }
1978
1979         arch_kretprobe_fixup_return(regs, correct_ret_addr);
1980
1981         /* Unlink all nodes for this frame. */
1982         first = current->kretprobe_instances.first;
1983         current->kretprobe_instances.first = node->next;
1984         node->next = NULL;
1985
1986         /* Recycle free instances. */
1987         while (first) {
1988                 ri = container_of(first, struct kretprobe_instance, llist);
1989                 first = first->next;
1990
1991                 recycle_rp_inst(ri);
1992         }
1993
1994         return (unsigned long)correct_ret_addr;
1995 }
1996 NOKPROBE_SYMBOL(__kretprobe_trampoline_handler)
1997
1998 /*
1999  * This kprobe pre_handler is registered with every kretprobe. When probe
2000  * hits it will set up the return probe.
2001  */
2002 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2003 {
2004         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
2005         struct kretprobe_instance *ri;
2006         struct freelist_node *fn;
2007
2008         fn = freelist_try_get(&rp->freelist);
2009         if (!fn) {
2010                 rp->nmissed++;
2011                 return 0;
2012         }
2013
2014         ri = container_of(fn, struct kretprobe_instance, freelist);
2015
2016         if (rp->entry_handler && rp->entry_handler(ri, regs)) {
2017                 freelist_add(&ri->freelist, &rp->freelist);
2018                 return 0;
2019         }
2020
2021         arch_prepare_kretprobe(ri, regs);
2022
2023         __llist_add(&ri->llist, &current->kretprobe_instances);
2024
2025         return 0;
2026 }
2027 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2028
2029 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
2030 {
2031         return !offset;
2032 }
2033
2034 /**
2035  * kprobe_on_func_entry() -- check whether given address is function entry
2036  * @addr: Target address
2037  * @sym:  Target symbol name
2038  * @offset: The offset from the symbol or the address
2039  *
2040  * This checks whether the given @addr+@offset or @sym+@offset is on the
2041  * function entry address or not.
2042  * This returns 0 if it is the function entry, or -EINVAL if it is not.
2043  * And also it returns -ENOENT if it fails the symbol or address lookup.
2044  * Caller must pass @addr or @sym (either one must be NULL), or this
2045  * returns -EINVAL.
2046  */
2047 int kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
2048 {
2049         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
2050
2051         if (IS_ERR(kp_addr))
2052                 return PTR_ERR(kp_addr);
2053
2054         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset))
2055                 return -ENOENT;
2056
2057         if (!arch_kprobe_on_func_entry(offset))
2058                 return -EINVAL;
2059
2060         return 0;
2061 }
2062
2063 int register_kretprobe(struct kretprobe *rp)
2064 {
2065         int ret;
2066         struct kretprobe_instance *inst;
2067         int i;
2068         void *addr;
2069
2070         ret = kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset);
2071         if (ret)
2072                 return ret;
2073
2074         /* If only 'rp->kp.addr' is specified, check reregistering kprobes */
2075         if (rp->kp.addr && warn_kprobe_rereg(&rp->kp))
2076                 return -EINVAL;
2077
2078         if (kretprobe_blacklist_size) {
2079                 addr = kprobe_addr(&rp->kp);
2080                 if (IS_ERR(addr))
2081                         return PTR_ERR(addr);
2082
2083                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2084                         if (kretprobe_blacklist[i].addr == addr)
2085                                 return -EINVAL;
2086                 }
2087         }
2088
2089         if (rp->data_size > KRETPROBE_MAX_DATA_SIZE)
2090                 return -E2BIG;
2091
2092         rp->kp.pre_handler = pre_handler_kretprobe;
2093         rp->kp.post_handler = NULL;
2094
2095         /* Pre-allocate memory for max kretprobe instances */
2096         if (rp->maxactive <= 0) {
2097 #ifdef CONFIG_PREEMPTION
2098                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
2099 #else
2100                 rp->maxactive = num_possible_cpus();
2101 #endif
2102         }
2103         rp->freelist.head = NULL;
2104         rp->rph = kzalloc(sizeof(struct kretprobe_holder), GFP_KERNEL);
2105         if (!rp->rph)
2106                 return -ENOMEM;
2107
2108         rp->rph->rp = rp;
2109         for (i = 0; i < rp->maxactive; i++) {
2110                 inst = kzalloc(sizeof(struct kretprobe_instance) +
2111                                rp->data_size, GFP_KERNEL);
2112                 if (inst == NULL) {
2113                         refcount_set(&rp->rph->ref, i);
2114                         free_rp_inst(rp);
2115                         return -ENOMEM;
2116                 }
2117                 inst->rph = rp->rph;
2118                 freelist_add(&inst->freelist, &rp->freelist);
2119         }
2120         refcount_set(&rp->rph->ref, i);
2121
2122         rp->nmissed = 0;
2123         /* Establish function entry probe point */
2124         ret = register_kprobe(&rp->kp);
2125         if (ret != 0)
2126                 free_rp_inst(rp);
2127         return ret;
2128 }
2129 EXPORT_SYMBOL_GPL(register_kretprobe);
2130
2131 int register_kretprobes(struct kretprobe **rps, int num)
2132 {
2133         int ret = 0, i;
2134
2135         if (num <= 0)
2136                 return -EINVAL;
2137         for (i = 0; i < num; i++) {
2138                 ret = register_kretprobe(rps[i]);
2139                 if (ret < 0) {
2140                         if (i > 0)
2141                                 unregister_kretprobes(rps, i);
2142                         break;
2143                 }
2144         }
2145         return ret;
2146 }
2147 EXPORT_SYMBOL_GPL(register_kretprobes);
2148
2149 void unregister_kretprobe(struct kretprobe *rp)
2150 {
2151         unregister_kretprobes(&rp, 1);
2152 }
2153 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2154
2155 void unregister_kretprobes(struct kretprobe **rps, int num)
2156 {
2157         int i;
2158
2159         if (num <= 0)
2160                 return;
2161         mutex_lock(&kprobe_mutex);
2162         for (i = 0; i < num; i++) {
2163                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
2164                         rps[i]->kp.addr = NULL;
2165                 rps[i]->rph->rp = NULL;
2166         }
2167         mutex_unlock(&kprobe_mutex);
2168
2169         synchronize_rcu();
2170         for (i = 0; i < num; i++) {
2171                 if (rps[i]->kp.addr) {
2172                         __unregister_kprobe_bottom(&rps[i]->kp);
2173                         free_rp_inst(rps[i]);
2174                 }
2175         }
2176 }
2177 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2178
2179 #else /* CONFIG_KRETPROBES */
2180 int register_kretprobe(struct kretprobe *rp)
2181 {
2182         return -EOPNOTSUPP;
2183 }
2184 EXPORT_SYMBOL_GPL(register_kretprobe);
2185
2186 int register_kretprobes(struct kretprobe **rps, int num)
2187 {
2188         return -EOPNOTSUPP;
2189 }
2190 EXPORT_SYMBOL_GPL(register_kretprobes);
2191
2192 void unregister_kretprobe(struct kretprobe *rp)
2193 {
2194 }
2195 EXPORT_SYMBOL_GPL(unregister_kretprobe);
2196
2197 void unregister_kretprobes(struct kretprobe **rps, int num)
2198 {
2199 }
2200 EXPORT_SYMBOL_GPL(unregister_kretprobes);
2201
2202 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2203 {
2204         return 0;
2205 }
2206 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2207
2208 #endif /* CONFIG_KRETPROBES */
2209
2210 /* Set the kprobe gone and remove its instruction buffer. */
2211 static void kill_kprobe(struct kprobe *p)
2212 {
2213         struct kprobe *kp;
2214
2215         lockdep_assert_held(&kprobe_mutex);
2216
2217         p->flags |= KPROBE_FLAG_GONE;
2218         if (kprobe_aggrprobe(p)) {
2219                 /*
2220                  * If this is an aggr_kprobe, we have to list all the
2221                  * chained probes and mark them GONE.
2222                  */
2223                 list_for_each_entry(kp, &p->list, list)
2224                         kp->flags |= KPROBE_FLAG_GONE;
2225                 p->post_handler = NULL;
2226                 kill_optimized_kprobe(p);
2227         }
2228         /*
2229          * Here, we can remove insn_slot safely, because no thread calls
2230          * the original probed function (which will be freed soon) any more.
2231          */
2232         arch_remove_kprobe(p);
2233
2234         /*
2235          * The module is going away. We should disarm the kprobe which
2236          * is using ftrace, because ftrace framework is still available at
2237          * 'MODULE_STATE_GOING' notification.
2238          */
2239         if (kprobe_ftrace(p) && !kprobe_disabled(p) && !kprobes_all_disarmed)
2240                 disarm_kprobe_ftrace(p);
2241 }
2242
2243 /* Disable one kprobe */
2244 int disable_kprobe(struct kprobe *kp)
2245 {
2246         int ret = 0;
2247         struct kprobe *p;
2248
2249         mutex_lock(&kprobe_mutex);
2250
2251         /* Disable this kprobe */
2252         p = __disable_kprobe(kp);
2253         if (IS_ERR(p))
2254                 ret = PTR_ERR(p);
2255
2256         mutex_unlock(&kprobe_mutex);
2257         return ret;
2258 }
2259 EXPORT_SYMBOL_GPL(disable_kprobe);
2260
2261 /* Enable one kprobe */
2262 int enable_kprobe(struct kprobe *kp)
2263 {
2264         int ret = 0;
2265         struct kprobe *p;
2266
2267         mutex_lock(&kprobe_mutex);
2268
2269         /* Check whether specified probe is valid. */
2270         p = __get_valid_kprobe(kp);
2271         if (unlikely(p == NULL)) {
2272                 ret = -EINVAL;
2273                 goto out;
2274         }
2275
2276         if (kprobe_gone(kp)) {
2277                 /* This kprobe has gone, we couldn't enable it. */
2278                 ret = -EINVAL;
2279                 goto out;
2280         }
2281
2282         if (p != kp)
2283                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2284
2285         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2286                 p->flags &= ~KPROBE_FLAG_DISABLED;
2287                 ret = arm_kprobe(p);
2288                 if (ret)
2289                         p->flags |= KPROBE_FLAG_DISABLED;
2290         }
2291 out:
2292         mutex_unlock(&kprobe_mutex);
2293         return ret;
2294 }
2295 EXPORT_SYMBOL_GPL(enable_kprobe);
2296
2297 /* Caller must NOT call this in usual path. This is only for critical case */
2298 void dump_kprobe(struct kprobe *kp)
2299 {
2300         pr_err("Dump kprobe:\n.symbol_name = %s, .offset = %x, .addr = %pS\n",
2301                kp->symbol_name, kp->offset, kp->addr);
2302 }
2303 NOKPROBE_SYMBOL(dump_kprobe);
2304
2305 int kprobe_add_ksym_blacklist(unsigned long entry)
2306 {
2307         struct kprobe_blacklist_entry *ent;
2308         unsigned long offset = 0, size = 0;
2309
2310         if (!kernel_text_address(entry) ||
2311             !kallsyms_lookup_size_offset(entry, &size, &offset))
2312                 return -EINVAL;
2313
2314         ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2315         if (!ent)
2316                 return -ENOMEM;
2317         ent->start_addr = entry;
2318         ent->end_addr = entry + size;
2319         INIT_LIST_HEAD(&ent->list);
2320         list_add_tail(&ent->list, &kprobe_blacklist);
2321
2322         return (int)size;
2323 }
2324
2325 /* Add all symbols in given area into kprobe blacklist */
2326 int kprobe_add_area_blacklist(unsigned long start, unsigned long end)
2327 {
2328         unsigned long entry;
2329         int ret = 0;
2330
2331         for (entry = start; entry < end; entry += ret) {
2332                 ret = kprobe_add_ksym_blacklist(entry);
2333                 if (ret < 0)
2334                         return ret;
2335                 if (ret == 0)   /* In case of alias symbol */
2336                         ret = 1;
2337         }
2338         return 0;
2339 }
2340
2341 /* Remove all symbols in given area from kprobe blacklist */
2342 static void kprobe_remove_area_blacklist(unsigned long start, unsigned long end)
2343 {
2344         struct kprobe_blacklist_entry *ent, *n;
2345
2346         list_for_each_entry_safe(ent, n, &kprobe_blacklist, list) {
2347                 if (ent->start_addr < start || ent->start_addr >= end)
2348                         continue;
2349                 list_del(&ent->list);
2350                 kfree(ent);
2351         }
2352 }
2353
2354 static void kprobe_remove_ksym_blacklist(unsigned long entry)
2355 {
2356         kprobe_remove_area_blacklist(entry, entry + 1);
2357 }
2358
2359 int __weak arch_kprobe_get_kallsym(unsigned int *symnum, unsigned long *value,
2360                                    char *type, char *sym)
2361 {
2362         return -ERANGE;
2363 }
2364
2365 int kprobe_get_kallsym(unsigned int symnum, unsigned long *value, char *type,
2366                        char *sym)
2367 {
2368 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
2369         if (!kprobe_cache_get_kallsym(&kprobe_insn_slots, &symnum, value, type, sym))
2370                 return 0;
2371 #ifdef CONFIG_OPTPROBES
2372         if (!kprobe_cache_get_kallsym(&kprobe_optinsn_slots, &symnum, value, type, sym))
2373                 return 0;
2374 #endif
2375 #endif
2376         if (!arch_kprobe_get_kallsym(&symnum, value, type, sym))
2377                 return 0;
2378         return -ERANGE;
2379 }
2380
2381 int __init __weak arch_populate_kprobe_blacklist(void)
2382 {
2383         return 0;
2384 }
2385
2386 /*
2387  * Lookup and populate the kprobe_blacklist.
2388  *
2389  * Unlike the kretprobe blacklist, we'll need to determine
2390  * the range of addresses that belong to the said functions,
2391  * since a kprobe need not necessarily be at the beginning
2392  * of a function.
2393  */
2394 static int __init populate_kprobe_blacklist(unsigned long *start,
2395                                              unsigned long *end)
2396 {
2397         unsigned long entry;
2398         unsigned long *iter;
2399         int ret;
2400
2401         for (iter = start; iter < end; iter++) {
2402                 entry = (unsigned long)dereference_symbol_descriptor((void *)*iter);
2403                 ret = kprobe_add_ksym_blacklist(entry);
2404                 if (ret == -EINVAL)
2405                         continue;
2406                 if (ret < 0)
2407                         return ret;
2408         }
2409
2410         /* Symbols in '__kprobes_text' are blacklisted */
2411         ret = kprobe_add_area_blacklist((unsigned long)__kprobes_text_start,
2412                                         (unsigned long)__kprobes_text_end);
2413         if (ret)
2414                 return ret;
2415
2416         /* Symbols in 'noinstr' section are blacklisted */
2417         ret = kprobe_add_area_blacklist((unsigned long)__noinstr_text_start,
2418                                         (unsigned long)__noinstr_text_end);
2419
2420         return ret ? : arch_populate_kprobe_blacklist();
2421 }
2422
2423 static void add_module_kprobe_blacklist(struct module *mod)
2424 {
2425         unsigned long start, end;
2426         int i;
2427
2428         if (mod->kprobe_blacklist) {
2429                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2430                         kprobe_add_ksym_blacklist(mod->kprobe_blacklist[i]);
2431         }
2432
2433         start = (unsigned long)mod->kprobes_text_start;
2434         if (start) {
2435                 end = start + mod->kprobes_text_size;
2436                 kprobe_add_area_blacklist(start, end);
2437         }
2438
2439         start = (unsigned long)mod->noinstr_text_start;
2440         if (start) {
2441                 end = start + mod->noinstr_text_size;
2442                 kprobe_add_area_blacklist(start, end);
2443         }
2444 }
2445
2446 static void remove_module_kprobe_blacklist(struct module *mod)
2447 {
2448         unsigned long start, end;
2449         int i;
2450
2451         if (mod->kprobe_blacklist) {
2452                 for (i = 0; i < mod->num_kprobe_blacklist; i++)
2453                         kprobe_remove_ksym_blacklist(mod->kprobe_blacklist[i]);
2454         }
2455
2456         start = (unsigned long)mod->kprobes_text_start;
2457         if (start) {
2458                 end = start + mod->kprobes_text_size;
2459                 kprobe_remove_area_blacklist(start, end);
2460         }
2461
2462         start = (unsigned long)mod->noinstr_text_start;
2463         if (start) {
2464                 end = start + mod->noinstr_text_size;
2465                 kprobe_remove_area_blacklist(start, end);
2466         }
2467 }
2468
2469 /* Module notifier call back, checking kprobes on the module */
2470 static int kprobes_module_callback(struct notifier_block *nb,
2471                                    unsigned long val, void *data)
2472 {
2473         struct module *mod = data;
2474         struct hlist_head *head;
2475         struct kprobe *p;
2476         unsigned int i;
2477         int checkcore = (val == MODULE_STATE_GOING);
2478
2479         if (val == MODULE_STATE_COMING) {
2480                 mutex_lock(&kprobe_mutex);
2481                 add_module_kprobe_blacklist(mod);
2482                 mutex_unlock(&kprobe_mutex);
2483         }
2484         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2485                 return NOTIFY_DONE;
2486
2487         /*
2488          * When 'MODULE_STATE_GOING' was notified, both of module '.text' and
2489          * '.init.text' sections would be freed. When 'MODULE_STATE_LIVE' was
2490          * notified, only '.init.text' section would be freed. We need to
2491          * disable kprobes which have been inserted in the sections.
2492          */
2493         mutex_lock(&kprobe_mutex);
2494         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2495                 head = &kprobe_table[i];
2496                 hlist_for_each_entry(p, head, hlist)
2497                         if (within_module_init((unsigned long)p->addr, mod) ||
2498                             (checkcore &&
2499                              within_module_core((unsigned long)p->addr, mod))) {
2500                                 /*
2501                                  * The vaddr this probe is installed will soon
2502                                  * be vfreed buy not synced to disk. Hence,
2503                                  * disarming the breakpoint isn't needed.
2504                                  *
2505                                  * Note, this will also move any optimized probes
2506                                  * that are pending to be removed from their
2507                                  * corresponding lists to the 'freeing_list' and
2508                                  * will not be touched by the delayed
2509                                  * kprobe_optimizer() work handler.
2510                                  */
2511                                 kill_kprobe(p);
2512                         }
2513         }
2514         if (val == MODULE_STATE_GOING)
2515                 remove_module_kprobe_blacklist(mod);
2516         mutex_unlock(&kprobe_mutex);
2517         return NOTIFY_DONE;
2518 }
2519
2520 static struct notifier_block kprobe_module_nb = {
2521         .notifier_call = kprobes_module_callback,
2522         .priority = 0
2523 };
2524
2525 void kprobe_free_init_mem(void)
2526 {
2527         void *start = (void *)(&__init_begin);
2528         void *end = (void *)(&__init_end);
2529         struct hlist_head *head;
2530         struct kprobe *p;
2531         int i;
2532
2533         mutex_lock(&kprobe_mutex);
2534
2535         /* Kill all kprobes on initmem because the target code has been freed. */
2536         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2537                 head = &kprobe_table[i];
2538                 hlist_for_each_entry(p, head, hlist) {
2539                         if (start <= (void *)p->addr && (void *)p->addr < end)
2540                                 kill_kprobe(p);
2541                 }
2542         }
2543
2544         mutex_unlock(&kprobe_mutex);
2545 }
2546
2547 static int __init init_kprobes(void)
2548 {
2549         int i, err = 0;
2550
2551         /* FIXME allocate the probe table, currently defined statically */
2552         /* initialize all list heads */
2553         for (i = 0; i < KPROBE_TABLE_SIZE; i++)
2554                 INIT_HLIST_HEAD(&kprobe_table[i]);
2555
2556         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2557                                         __stop_kprobe_blacklist);
2558         if (err)
2559                 pr_err("Failed to populate blacklist (error %d), kprobes not restricted, be careful using them!\n", err);
2560
2561         if (kretprobe_blacklist_size) {
2562                 /* lookup the function address from its name */
2563                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2564                         kretprobe_blacklist[i].addr =
2565                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2566                         if (!kretprobe_blacklist[i].addr)
2567                                 pr_err("Failed to lookup symbol '%s' for kretprobe blacklist. Maybe the target function is removed or renamed.\n",
2568                                        kretprobe_blacklist[i].name);
2569                 }
2570         }
2571
2572         /* By default, kprobes are armed */
2573         kprobes_all_disarmed = false;
2574
2575 #if defined(CONFIG_OPTPROBES) && defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2576         /* Init 'kprobe_optinsn_slots' for allocation */
2577         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2578 #endif
2579
2580         err = arch_init_kprobes();
2581         if (!err)
2582                 err = register_die_notifier(&kprobe_exceptions_nb);
2583         if (!err)
2584                 err = register_module_notifier(&kprobe_module_nb);
2585
2586         kprobes_initialized = (err == 0);
2587         return err;
2588 }
2589 early_initcall(init_kprobes);
2590
2591 #if defined(CONFIG_OPTPROBES)
2592 static int __init init_optprobes(void)
2593 {
2594         /*
2595          * Enable kprobe optimization - this kicks the optimizer which
2596          * depends on synchronize_rcu_tasks() and ksoftirqd, that is
2597          * not spawned in early initcall. So delay the optimization.
2598          */
2599         optimize_all_kprobes();
2600
2601         return 0;
2602 }
2603 subsys_initcall(init_optprobes);
2604 #endif
2605
2606 #ifdef CONFIG_DEBUG_FS
2607 static void report_probe(struct seq_file *pi, struct kprobe *p,
2608                 const char *sym, int offset, char *modname, struct kprobe *pp)
2609 {
2610         char *kprobe_type;
2611         void *addr = p->addr;
2612
2613         if (p->pre_handler == pre_handler_kretprobe)
2614                 kprobe_type = "r";
2615         else
2616                 kprobe_type = "k";
2617
2618         if (!kallsyms_show_value(pi->file->f_cred))
2619                 addr = NULL;
2620
2621         if (sym)
2622                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2623                         addr, kprobe_type, sym, offset,
2624                         (modname ? modname : " "));
2625         else    /* try to use %pS */
2626                 seq_printf(pi, "%px  %s  %pS ",
2627                         addr, kprobe_type, p->addr);
2628
2629         if (!pp)
2630                 pp = p;
2631         seq_printf(pi, "%s%s%s%s\n",
2632                 (kprobe_gone(p) ? "[GONE]" : ""),
2633                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2634                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2635                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2636 }
2637
2638 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2639 {
2640         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2641 }
2642
2643 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2644 {
2645         (*pos)++;
2646         if (*pos >= KPROBE_TABLE_SIZE)
2647                 return NULL;
2648         return pos;
2649 }
2650
2651 static void kprobe_seq_stop(struct seq_file *f, void *v)
2652 {
2653         /* Nothing to do */
2654 }
2655
2656 static int show_kprobe_addr(struct seq_file *pi, void *v)
2657 {
2658         struct hlist_head *head;
2659         struct kprobe *p, *kp;
2660         const char *sym = NULL;
2661         unsigned int i = *(loff_t *) v;
2662         unsigned long offset = 0;
2663         char *modname, namebuf[KSYM_NAME_LEN];
2664
2665         head = &kprobe_table[i];
2666         preempt_disable();
2667         hlist_for_each_entry_rcu(p, head, hlist) {
2668                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2669                                         &offset, &modname, namebuf);
2670                 if (kprobe_aggrprobe(p)) {
2671                         list_for_each_entry_rcu(kp, &p->list, list)
2672                                 report_probe(pi, kp, sym, offset, modname, p);
2673                 } else
2674                         report_probe(pi, p, sym, offset, modname, NULL);
2675         }
2676         preempt_enable();
2677         return 0;
2678 }
2679
2680 static const struct seq_operations kprobes_sops = {
2681         .start = kprobe_seq_start,
2682         .next  = kprobe_seq_next,
2683         .stop  = kprobe_seq_stop,
2684         .show  = show_kprobe_addr
2685 };
2686
2687 DEFINE_SEQ_ATTRIBUTE(kprobes);
2688
2689 /* kprobes/blacklist -- shows which functions can not be probed */
2690 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2691 {
2692         mutex_lock(&kprobe_mutex);
2693         return seq_list_start(&kprobe_blacklist, *pos);
2694 }
2695
2696 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2697 {
2698         return seq_list_next(v, &kprobe_blacklist, pos);
2699 }
2700
2701 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2702 {
2703         struct kprobe_blacklist_entry *ent =
2704                 list_entry(v, struct kprobe_blacklist_entry, list);
2705
2706         /*
2707          * If '/proc/kallsyms' is not showing kernel address, we won't
2708          * show them here either.
2709          */
2710         if (!kallsyms_show_value(m->file->f_cred))
2711                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2712                            (void *)ent->start_addr);
2713         else
2714                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2715                            (void *)ent->end_addr, (void *)ent->start_addr);
2716         return 0;
2717 }
2718
2719 static void kprobe_blacklist_seq_stop(struct seq_file *f, void *v)
2720 {
2721         mutex_unlock(&kprobe_mutex);
2722 }
2723
2724 static const struct seq_operations kprobe_blacklist_sops = {
2725         .start = kprobe_blacklist_seq_start,
2726         .next  = kprobe_blacklist_seq_next,
2727         .stop  = kprobe_blacklist_seq_stop,
2728         .show  = kprobe_blacklist_seq_show,
2729 };
2730 DEFINE_SEQ_ATTRIBUTE(kprobe_blacklist);
2731
2732 static int arm_all_kprobes(void)
2733 {
2734         struct hlist_head *head;
2735         struct kprobe *p;
2736         unsigned int i, total = 0, errors = 0;
2737         int err, ret = 0;
2738
2739         mutex_lock(&kprobe_mutex);
2740
2741         /* If kprobes are armed, just return */
2742         if (!kprobes_all_disarmed)
2743                 goto already_enabled;
2744
2745         /*
2746          * optimize_kprobe() called by arm_kprobe() checks
2747          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2748          * arm_kprobe.
2749          */
2750         kprobes_all_disarmed = false;
2751         /* Arming kprobes doesn't optimize kprobe itself */
2752         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2753                 head = &kprobe_table[i];
2754                 /* Arm all kprobes on a best-effort basis */
2755                 hlist_for_each_entry(p, head, hlist) {
2756                         if (!kprobe_disabled(p)) {
2757                                 err = arm_kprobe(p);
2758                                 if (err)  {
2759                                         errors++;
2760                                         ret = err;
2761                                 }
2762                                 total++;
2763                         }
2764                 }
2765         }
2766
2767         if (errors)
2768                 pr_warn("Kprobes globally enabled, but failed to enable %d out of %d probes. Please check which kprobes are kept disabled via debugfs.\n",
2769                         errors, total);
2770         else
2771                 pr_info("Kprobes globally enabled\n");
2772
2773 already_enabled:
2774         mutex_unlock(&kprobe_mutex);
2775         return ret;
2776 }
2777
2778 static int disarm_all_kprobes(void)
2779 {
2780         struct hlist_head *head;
2781         struct kprobe *p;
2782         unsigned int i, total = 0, errors = 0;
2783         int err, ret = 0;
2784
2785         mutex_lock(&kprobe_mutex);
2786
2787         /* If kprobes are already disarmed, just return */
2788         if (kprobes_all_disarmed) {
2789                 mutex_unlock(&kprobe_mutex);
2790                 return 0;
2791         }
2792
2793         kprobes_all_disarmed = true;
2794
2795         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2796                 head = &kprobe_table[i];
2797                 /* Disarm all kprobes on a best-effort basis */
2798                 hlist_for_each_entry(p, head, hlist) {
2799                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2800                                 err = disarm_kprobe(p, false);
2801                                 if (err) {
2802                                         errors++;
2803                                         ret = err;
2804                                 }
2805                                 total++;
2806                         }
2807                 }
2808         }
2809
2810         if (errors)
2811                 pr_warn("Kprobes globally disabled, but failed to disable %d out of %d probes. Please check which kprobes are kept enabled via debugfs.\n",
2812                         errors, total);
2813         else
2814                 pr_info("Kprobes globally disabled\n");
2815
2816         mutex_unlock(&kprobe_mutex);
2817
2818         /* Wait for disarming all kprobes by optimizer */
2819         wait_for_kprobe_optimizer();
2820
2821         return ret;
2822 }
2823
2824 /*
2825  * XXX: The debugfs bool file interface doesn't allow for callbacks
2826  * when the bool state is switched. We can reuse that facility when
2827  * available
2828  */
2829 static ssize_t read_enabled_file_bool(struct file *file,
2830                char __user *user_buf, size_t count, loff_t *ppos)
2831 {
2832         char buf[3];
2833
2834         if (!kprobes_all_disarmed)
2835                 buf[0] = '1';
2836         else
2837                 buf[0] = '0';
2838         buf[1] = '\n';
2839         buf[2] = 0x00;
2840         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2841 }
2842
2843 static ssize_t write_enabled_file_bool(struct file *file,
2844                const char __user *user_buf, size_t count, loff_t *ppos)
2845 {
2846         bool enable;
2847         int ret;
2848
2849         ret = kstrtobool_from_user(user_buf, count, &enable);
2850         if (ret)
2851                 return ret;
2852
2853         ret = enable ? arm_all_kprobes() : disarm_all_kprobes();
2854         if (ret)
2855                 return ret;
2856
2857         return count;
2858 }
2859
2860 static const struct file_operations fops_kp = {
2861         .read =         read_enabled_file_bool,
2862         .write =        write_enabled_file_bool,
2863         .llseek =       default_llseek,
2864 };
2865
2866 static int __init debugfs_kprobe_init(void)
2867 {
2868         struct dentry *dir;
2869
2870         dir = debugfs_create_dir("kprobes", NULL);
2871
2872         debugfs_create_file("list", 0400, dir, NULL, &kprobes_fops);
2873
2874         debugfs_create_file("enabled", 0600, dir, NULL, &fops_kp);
2875
2876         debugfs_create_file("blacklist", 0400, dir, NULL,
2877                             &kprobe_blacklist_fops);
2878
2879         return 0;
2880 }
2881
2882 late_initcall(debugfs_kprobe_init);
2883 #endif /* CONFIG_DEBUG_FS */