kernel/kprobes.c

   1 /*
   2  *  Kernel Probes (KProbes)
   3  *  kernel/kprobes.c
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; either version 2 of the License, or
   8  * (at your option) any later version.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program; if not, write to the Free Software
  17  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  18  *
  19  * Copyright (C) IBM Corporation, 2002, 2004
  20  *
  21  * 2002-Oct     Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
  22  *              Probes initial implementation (includes suggestions from
  23  *              Rusty Russell).
  24  * 2004-Aug     Updated by Prasanna S Panchamukhi <prasanna@in.ibm.com> with
  25  *              hlists and exceptions notifier as suggested by Andi Kleen.
  26  * 2004-July    Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
  27  *              interface to access function arguments.
  28  * 2004-Sep     Prasanna S Panchamukhi <prasanna@in.ibm.com> Changed Kprobes
  29  *              exceptions notifier to be first on the priority list.
  30  * 2005-May     Hien Nguyen <hien@us.ibm.com>, Jim Keniston
  31  *              <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
  32  *              <prasanna@in.ibm.com> added function-return probes.
  33  */
  34 #include <linux/kprobes.h>
  35 #include <linux/hash.h>
  36 #include <linux/init.h>
  37 #include <linux/slab.h>
  38 #include <linux/stddef.h>
  39 #include <linux/export.h>
  40 #include <linux/moduleloader.h>
  41 #include <linux/kallsyms.h>
  42 #include <linux/freezer.h>
  43 #include <linux/seq_file.h>
  44 #include <linux/debugfs.h>
  45 #include <linux/sysctl.h>
  46 #include <linux/kdebug.h>
  47 #include <linux/memory.h>
  48 #include <linux/ftrace.h>
  49 #include <linux/cpu.h>
  50 #include <linux/jump_label.h>
  51
  52 #include <asm-generic/sections.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/errno.h>
  55 #include <asm/uaccess.h>
  56
  57 #define KPROBE_HASH_BITS 6
  58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  59
  60
  61 /*
  62  * Some oddball architectures like 64bit powerpc have function descriptors
  63  * so this must be overridable.
  64  */
  65 #ifndef kprobe_lookup_name
  66 #define kprobe_lookup_name(name, addr) \
  67         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  68 #endif
  69
  70 static int kprobes_initialized;
  71 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  72 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  73
  74 /* NOTE: change this value only with kprobe_mutex held */
  75 static bool kprobes_all_disarmed;
  76
  77 /* This protects kprobe_table and optimizing_list */
  78 static DEFINE_MUTEX(kprobe_mutex);
  79 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  80 static struct {
  81         raw_spinlock_t lock ____cacheline_aligned_in_smp;
  82 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  83
  84 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  85 {
  86         return &(kretprobe_table_locks[hash].lock);
  87 }
  88
  89 /* Blacklist -- list of struct kprobe_blacklist_entry */
  90 static LIST_HEAD(kprobe_blacklist);
  91
  92 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  93 /*
  94  * kprobe->ainsn.insn points to the copy of the instruction to be
  95  * single-stepped. x86_64, POWER4 and above have no-exec support and
  96  * stepping on the instruction on a vmalloced/kmalloced/data page
  97  * is a recipe for disaster
  98  */
  99 struct kprobe_insn_page {
 100         struct list_head list;
 101         kprobe_opcode_t *insns;         /* Page of instruction slots */
 102         struct kprobe_insn_cache *cache;
 103         int nused;
 104         int ngarbage;
 105         char slot_used[];
 106 };
 107
 108 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
 109         (offsetof(struct kprobe_insn_page, slot_used) + \
 110          (sizeof(char) * (slots)))
 111
 112 static int slots_per_page(struct kprobe_insn_cache *c)
 113 {
 114         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 115 }
 116
 117 enum kprobe_slot_state {
 118         SLOT_CLEAN = 0,
 119         SLOT_DIRTY = 1,
 120         SLOT_USED = 2,
 121 };
 122
 123 static void *alloc_insn_page(void)
 124 {
 125         return module_alloc(PAGE_SIZE);
 126 }
 127
 128 static void free_insn_page(void *page)
 129 {
 130         module_memfree(page);
 131 }
 132
 133 struct kprobe_insn_cache kprobe_insn_slots = {
 134         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 135         .alloc = alloc_insn_page,
 136         .free = free_insn_page,
 137         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 138         .insn_size = MAX_INSN_SIZE,
 139         .nr_garbage = 0,
 140 };
 141 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 142
 143 /**
 144  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 145  * We allocate an executable page if there's no room on existing ones.
 146  */
 147 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 148 {
 149         struct kprobe_insn_page *kip;
 150         kprobe_opcode_t *slot = NULL;
 151
 152         mutex_lock(&c->mutex);
 153  retry:
 154         list_for_each_entry(kip, &c->pages, list) {
 155                 if (kip->nused < slots_per_page(c)) {
 156                         int i;
 157                         for (i = 0; i < slots_per_page(c); i++) {
 158                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 159                                         kip->slot_used[i] = SLOT_USED;
 160                                         kip->nused++;
 161                                         slot = kip->insns + (i * c->insn_size);
 162                                         goto out;
 163                                 }
 164                         }
 165                         /* kip->nused is broken. Fix it. */
 166                         kip->nused = slots_per_page(c);
 167                         WARN_ON(1);
 168                 }
 169         }
 170
 171         /* If there are any garbage slots, collect it and try again. */
 172         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 173                 goto retry;
 174
 175         /* All out of space.  Need to allocate a new page. */
 176         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 177         if (!kip)
 178                 goto out;
 179
 180         /*
 181          * Use module_alloc so this page is within +/- 2GB of where the
 182          * kernel image and loaded module images reside. This is required
 183          * so x86_64 can correctly handle the %rip-relative fixups.
 184          */
 185         kip->insns = c->alloc();
 186         if (!kip->insns) {
 187                 kfree(kip);
 188                 goto out;
 189         }
 190         INIT_LIST_HEAD(&kip->list);
 191         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 192         kip->slot_used[0] = SLOT_USED;
 193         kip->nused = 1;
 194         kip->ngarbage = 0;
 195         kip->cache = c;
 196         list_add(&kip->list, &c->pages);
 197         slot = kip->insns;
 198 out:
 199         mutex_unlock(&c->mutex);
 200         return slot;
 201 }
 202
 203 /* Return 1 if all garbages are collected, otherwise 0. */
 204 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 205 {
 206         kip->slot_used[idx] = SLOT_CLEAN;
 207         kip->nused--;
 208         if (kip->nused == 0) {
 209                 /*
 210                  * Page is no longer in use.  Free it unless
 211                  * it's the last one.  We keep the last one
 212                  * so as not to have to set it up again the
 213                  * next time somebody inserts a probe.
 214                  */
 215                 if (!list_is_singular(&kip->list)) {
 216                         list_del(&kip->list);
 217                         kip->cache->free(kip->insns);
 218                         kfree(kip);
 219                 }
 220                 return 1;
 221         }
 222         return 0;
 223 }
 224
 225 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 226 {
 227         struct kprobe_insn_page *kip, *next;
 228
 229         /* Ensure no-one is interrupted on the garbages */
 230         synchronize_sched();
 231
 232         list_for_each_entry_safe(kip, next, &c->pages, list) {
 233                 int i;
 234                 if (kip->ngarbage == 0)
 235                         continue;
 236                 kip->ngarbage = 0;      /* we will collect all garbages */
 237                 for (i = 0; i < slots_per_page(c); i++) {
 238                         if (kip->slot_used[i] == SLOT_DIRTY &&
 239                             collect_one_slot(kip, i))
 240                                 break;
 241                 }
 242         }
 243         c->nr_garbage = 0;
 244         return 0;
 245 }
 246
 247 void __free_insn_slot(struct kprobe_insn_cache *c,
 248                       kprobe_opcode_t *slot, int dirty)
 249 {
 250         struct kprobe_insn_page *kip;
 251
 252         mutex_lock(&c->mutex);
 253         list_for_each_entry(kip, &c->pages, list) {
 254                 long idx = ((long)slot - (long)kip->insns) /
 255                                 (c->insn_size * sizeof(kprobe_opcode_t));
 256                 if (idx >= 0 && idx < slots_per_page(c)) {
 257                         WARN_ON(kip->slot_used[idx] != SLOT_USED);
 258                         if (dirty) {
 259                                 kip->slot_used[idx] = SLOT_DIRTY;
 260                                 kip->ngarbage++;
 261                                 if (++c->nr_garbage > slots_per_page(c))
 262                                         collect_garbage_slots(c);
 263                         } else
 264                                 collect_one_slot(kip, idx);
 265                         goto out;
 266                 }
 267         }
 268         /* Could not free this slot. */
 269         WARN_ON(1);
 270 out:
 271         mutex_unlock(&c->mutex);
 272 }
 273
 274 #ifdef CONFIG_OPTPROBES
 275 /* For optimized_kprobe buffer */
 276 struct kprobe_insn_cache kprobe_optinsn_slots = {
 277         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 278         .alloc = alloc_insn_page,
 279         .free = free_insn_page,
 280         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 281         /* .insn_size is initialized later */
 282         .nr_garbage = 0,
 283 };
 284 #endif
 285 #endif
 286
 287 /* We have preemption disabled.. so it is safe to use __ versions */
 288 static inline void set_kprobe_instance(struct kprobe *kp)
 289 {
 290         __this_cpu_write(kprobe_instance, kp);
 291 }
 292
 293 static inline void reset_kprobe_instance(void)
 294 {
 295         __this_cpu_write(kprobe_instance, NULL);
 296 }
 297
 298 /*
 299  * This routine is called either:
 300  *      - under the kprobe_mutex - during kprobe_[un]register()
 301  *                              OR
 302  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 303  */
 304 struct kprobe *get_kprobe(void *addr)
 305 {
 306         struct hlist_head *head;
 307         struct kprobe *p;
 308
 309         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 310         hlist_for_each_entry_rcu(p, head, hlist) {
 311                 if (p->addr == addr)
 312                         return p;
 313         }
 314
 315         return NULL;
 316 }
 317 NOKPROBE_SYMBOL(get_kprobe);
 318
 319 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 320
 321 /* Return true if the kprobe is an aggregator */
 322 static inline int kprobe_aggrprobe(struct kprobe *p)
 323 {
 324         return p->pre_handler == aggr_pre_handler;
 325 }
 326
 327 /* Return true(!0) if the kprobe is unused */
 328 static inline int kprobe_unused(struct kprobe *p)
 329 {
 330         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 331                list_empty(&p->list);
 332 }
 333
 334 /*
 335  * Keep all fields in the kprobe consistent
 336  */
 337 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 338 {
 339         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 340         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 341 }
 342
 343 #ifdef CONFIG_OPTPROBES
 344 /* NOTE: change this value only with kprobe_mutex held */
 345 static bool kprobes_allow_optimization;
 346
 347 /*
 348  * Call all pre_handler on the list, but ignores its return value.
 349  * This must be called from arch-dep optimized caller.
 350  */
 351 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 352 {
 353         struct kprobe *kp;
 354
 355         list_for_each_entry_rcu(kp, &p->list, list) {
 356                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 357                         set_kprobe_instance(kp);
 358                         kp->pre_handler(kp, regs);
 359                 }
 360                 reset_kprobe_instance();
 361         }
 362 }
 363 NOKPROBE_SYMBOL(opt_pre_handler);
 364
 365 /* Free optimized instructions and optimized_kprobe */
 366 static void free_aggr_kprobe(struct kprobe *p)
 367 {
 368         struct optimized_kprobe *op;
 369
 370         op = container_of(p, struct optimized_kprobe, kp);
 371         arch_remove_optimized_kprobe(op);
 372         arch_remove_kprobe(p);
 373         kfree(op);
 374 }
 375
 376 /* Return true(!0) if the kprobe is ready for optimization. */
 377 static inline int kprobe_optready(struct kprobe *p)
 378 {
 379         struct optimized_kprobe *op;
 380
 381         if (kprobe_aggrprobe(p)) {
 382                 op = container_of(p, struct optimized_kprobe, kp);
 383                 return arch_prepared_optinsn(&op->optinsn);
 384         }
 385
 386         return 0;
 387 }
 388
 389 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 390 static inline int kprobe_disarmed(struct kprobe *p)
 391 {
 392         struct optimized_kprobe *op;
 393
 394         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 395         if (!kprobe_aggrprobe(p))
 396                 return kprobe_disabled(p);
 397
 398         op = container_of(p, struct optimized_kprobe, kp);
 399
 400         return kprobe_disabled(p) && list_empty(&op->list);
 401 }
 402
 403 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 404 static int kprobe_queued(struct kprobe *p)
 405 {
 406         struct optimized_kprobe *op;
 407
 408         if (kprobe_aggrprobe(p)) {
 409                 op = container_of(p, struct optimized_kprobe, kp);
 410                 if (!list_empty(&op->list))
 411                         return 1;
 412         }
 413         return 0;
 414 }
 415
 416 /*
 417  * Return an optimized kprobe whose optimizing code replaces
 418  * instructions including addr (exclude breakpoint).
 419  */
 420 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 421 {
 422         int i;
 423         struct kprobe *p = NULL;
 424         struct optimized_kprobe *op;
 425
 426         /* Don't check i == 0, since that is a breakpoint case. */
 427         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 428                 p = get_kprobe((void *)(addr - i));
 429
 430         if (p && kprobe_optready(p)) {
 431                 op = container_of(p, struct optimized_kprobe, kp);
 432                 if (arch_within_optimized_kprobe(op, addr))
 433                         return p;
 434         }
 435
 436         return NULL;
 437 }
 438
 439 /* Optimization staging list, protected by kprobe_mutex */
 440 static LIST_HEAD(optimizing_list);
 441 static LIST_HEAD(unoptimizing_list);
 442 static LIST_HEAD(freeing_list);
 443
 444 static void kprobe_optimizer(struct work_struct *work);
 445 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 446 #define OPTIMIZE_DELAY 5
 447
 448 /*
 449  * Optimize (replace a breakpoint with a jump) kprobes listed on
 450  * optimizing_list.
 451  */
 452 static void do_optimize_kprobes(void)
 453 {
 454         /* Optimization never be done when disarmed */
 455         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 456             list_empty(&optimizing_list))
 457                 return;
 458
 459         /*
 460          * The optimization/unoptimization refers online_cpus via
 461          * stop_machine() and cpu-hotplug modifies online_cpus.
 462          * And same time, text_mutex will be held in cpu-hotplug and here.
 463          * This combination can cause a deadlock (cpu-hotplug try to lock
 464          * text_mutex but stop_machine can not be done because online_cpus
 465          * has been changed)
 466          * To avoid this deadlock, we need to call get_online_cpus()
 467          * for preventing cpu-hotplug outside of text_mutex locking.
 468          */
 469         get_online_cpus();
 470         mutex_lock(&text_mutex);
 471         arch_optimize_kprobes(&optimizing_list);
 472         mutex_unlock(&text_mutex);
 473         put_online_cpus();
 474 }
 475
 476 /*
 477  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 478  * if need) kprobes listed on unoptimizing_list.
 479  */
 480 static void do_unoptimize_kprobes(void)
 481 {
 482         struct optimized_kprobe *op, *tmp;
 483
 484         /* Unoptimization must be done anytime */
 485         if (list_empty(&unoptimizing_list))
 486                 return;
 487
 488         /* Ditto to do_optimize_kprobes */
 489         get_online_cpus();
 490         mutex_lock(&text_mutex);
 491         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 492         /* Loop free_list for disarming */
 493         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 494                 /* Disarm probes if marked disabled */
 495                 if (kprobe_disabled(&op->kp))
 496                         arch_disarm_kprobe(&op->kp);
 497                 if (kprobe_unused(&op->kp)) {
 498                         /*
 499                          * Remove unused probes from hash list. After waiting
 500                          * for synchronization, these probes are reclaimed.
 501                          * (reclaiming is done by do_free_cleaned_kprobes.)
 502                          */
 503                         hlist_del_rcu(&op->kp.hlist);
 504                 } else
 505                         list_del_init(&op->list);
 506         }
 507         mutex_unlock(&text_mutex);
 508         put_online_cpus();
 509 }
 510
 511 /* Reclaim all kprobes on the free_list */
 512 static void do_free_cleaned_kprobes(void)
 513 {
 514         struct optimized_kprobe *op, *tmp;
 515
 516         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 517                 BUG_ON(!kprobe_unused(&op->kp));
 518                 list_del_init(&op->list);
 519                 free_aggr_kprobe(&op->kp);
 520         }
 521 }
 522
 523 /* Start optimizer after OPTIMIZE_DELAY passed */
 524 static void kick_kprobe_optimizer(void)
 525 {
 526         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 527 }
 528
 529 /* Kprobe jump optimizer */
 530 static void kprobe_optimizer(struct work_struct *work)
 531 {
 532         mutex_lock(&kprobe_mutex);
 533         /* Lock modules while optimizing kprobes */
 534         mutex_lock(&module_mutex);
 535
 536         /*
 537          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 538          * kprobes before waiting for quiesence period.
 539          */
 540         do_unoptimize_kprobes();
 541
 542         /*
 543          * Step 2: Wait for quiesence period to ensure all running interrupts
 544          * are done. Because optprobe may modify multiple instructions
 545          * there is a chance that Nth instruction is interrupted. In that
 546          * case, running interrupt can return to 2nd-Nth byte of jump
 547          * instruction. This wait is for avoiding it.
 548          */
 549         synchronize_sched();
 550
 551         /* Step 3: Optimize kprobes after quiesence period */
 552         do_optimize_kprobes();
 553
 554         /* Step 4: Free cleaned kprobes after quiesence period */
 555         do_free_cleaned_kprobes();
 556
 557         mutex_unlock(&module_mutex);
 558         mutex_unlock(&kprobe_mutex);
 559
 560         /* Step 5: Kick optimizer again if needed */
 561         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 562                 kick_kprobe_optimizer();
 563 }
 564
 565 /* Wait for completing optimization and unoptimization */
 566 static void wait_for_kprobe_optimizer(void)
 567 {
 568         mutex_lock(&kprobe_mutex);
 569
 570         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 571                 mutex_unlock(&kprobe_mutex);
 572
 573                 /* this will also make optimizing_work execute immmediately */
 574                 flush_delayed_work(&optimizing_work);
 575                 /* @optimizing_work might not have been queued yet, relax */
 576                 cpu_relax();
 577
 578                 mutex_lock(&kprobe_mutex);
 579         }
 580
 581         mutex_unlock(&kprobe_mutex);
 582 }
 583
 584 /* Optimize kprobe if p is ready to be optimized */
 585 static void optimize_kprobe(struct kprobe *p)
 586 {
 587         struct optimized_kprobe *op;
 588
 589         /* Check if the kprobe is disabled or not ready for optimization. */
 590         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 591             (kprobe_disabled(p) || kprobes_all_disarmed))
 592                 return;
 593
 594         /* Both of break_handler and post_handler are not supported. */
 595         if (p->break_handler || p->post_handler)
 596                 return;
 597
 598         op = container_of(p, struct optimized_kprobe, kp);
 599
 600         /* Check there is no other kprobes at the optimized instructions */
 601         if (arch_check_optimized_kprobe(op) < 0)
 602                 return;
 603
 604         /* Check if it is already optimized. */
 605         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 606                 return;
 607         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 608
 609         if (!list_empty(&op->list))
 610                 /* This is under unoptimizing. Just dequeue the probe */
 611                 list_del_init(&op->list);
 612         else {
 613                 list_add(&op->list, &optimizing_list);
 614                 kick_kprobe_optimizer();
 615         }
 616 }
 617
 618 /* Short cut to direct unoptimizing */
 619 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 620 {
 621         get_online_cpus();
 622         arch_unoptimize_kprobe(op);
 623         put_online_cpus();
 624         if (kprobe_disabled(&op->kp))
 625                 arch_disarm_kprobe(&op->kp);
 626 }
 627
 628 /* Unoptimize a kprobe if p is optimized */
 629 static void unoptimize_kprobe(struct kprobe *p, bool force)
 630 {
 631         struct optimized_kprobe *op;
 632
 633         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 634                 return; /* This is not an optprobe nor optimized */
 635
 636         op = container_of(p, struct optimized_kprobe, kp);
 637         if (!kprobe_optimized(p)) {
 638                 /* Unoptimized or unoptimizing case */
 639                 if (force && !list_empty(&op->list)) {
 640                         /*
 641                          * Only if this is unoptimizing kprobe and forced,
 642                          * forcibly unoptimize it. (No need to unoptimize
 643                          * unoptimized kprobe again :)
 644                          */
 645                         list_del_init(&op->list);
 646                         force_unoptimize_kprobe(op);
 647                 }
 648                 return;
 649         }
 650
 651         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 652         if (!list_empty(&op->list)) {
 653                 /* Dequeue from the optimization queue */
 654                 list_del_init(&op->list);
 655                 return;
 656         }
 657         /* Optimized kprobe case */
 658         if (force)
 659                 /* Forcibly update the code: this is a special case */
 660                 force_unoptimize_kprobe(op);
 661         else {
 662                 list_add(&op->list, &unoptimizing_list);
 663                 kick_kprobe_optimizer();
 664         }
 665 }
 666
 667 /* Cancel unoptimizing for reusing */
 668 static void reuse_unused_kprobe(struct kprobe *ap)
 669 {
 670         struct optimized_kprobe *op;
 671
 672         BUG_ON(!kprobe_unused(ap));
 673         /*
 674          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 675          * there is still a relative jump) and disabled.
 676          */
 677         op = container_of(ap, struct optimized_kprobe, kp);
 678         if (unlikely(list_empty(&op->list)))
 679                 printk(KERN_WARNING "Warning: found a stray unused "
 680                         "aggrprobe@%p\n", ap->addr);
 681         /* Enable the probe again */
 682         ap->flags &= ~KPROBE_FLAG_DISABLED;
 683         /* Optimize it again (remove from op->list) */
 684         BUG_ON(!kprobe_optready(ap));
 685         optimize_kprobe(ap);
 686 }
 687
 688 /* Remove optimized instructions */
 689 static void kill_optimized_kprobe(struct kprobe *p)
 690 {
 691         struct optimized_kprobe *op;
 692
 693         op = container_of(p, struct optimized_kprobe, kp);
 694         if (!list_empty(&op->list))
 695                 /* Dequeue from the (un)optimization queue */
 696                 list_del_init(&op->list);
 697         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 698
 699         if (kprobe_unused(p)) {
 700                 /* Enqueue if it is unused */
 701                 list_add(&op->list, &freeing_list);
 702                 /*
 703                  * Remove unused probes from the hash list. After waiting
 704                  * for synchronization, this probe is reclaimed.
 705                  * (reclaiming is done by do_free_cleaned_kprobes().)
 706                  */
 707                 hlist_del_rcu(&op->kp.hlist);
 708         }
 709
 710         /* Don't touch the code, because it is already freed. */
 711         arch_remove_optimized_kprobe(op);
 712 }
 713
 714 /* Try to prepare optimized instructions */
 715 static void prepare_optimized_kprobe(struct kprobe *p)
 716 {
 717         struct optimized_kprobe *op;
 718
 719         op = container_of(p, struct optimized_kprobe, kp);
 720         arch_prepare_optimized_kprobe(op);
 721 }
 722
 723 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 724 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 725 {
 726         struct optimized_kprobe *op;
 727
 728         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 729         if (!op)
 730                 return NULL;
 731
 732         INIT_LIST_HEAD(&op->list);
 733         op->kp.addr = p->addr;
 734         arch_prepare_optimized_kprobe(op);
 735
 736         return &op->kp;
 737 }
 738
 739 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 740
 741 /*
 742  * Prepare an optimized_kprobe and optimize it
 743  * NOTE: p must be a normal registered kprobe
 744  */
 745 static void try_to_optimize_kprobe(struct kprobe *p)
 746 {
 747         struct kprobe *ap;
 748         struct optimized_kprobe *op;
 749
 750         /* Impossible to optimize ftrace-based kprobe */
 751         if (kprobe_ftrace(p))
 752                 return;
 753
 754         /* For preparing optimization, jump_label_text_reserved() is called */
 755         jump_label_lock();
 756         mutex_lock(&text_mutex);
 757
 758         ap = alloc_aggr_kprobe(p);
 759         if (!ap)
 760                 goto out;
 761
 762         op = container_of(ap, struct optimized_kprobe, kp);
 763         if (!arch_prepared_optinsn(&op->optinsn)) {
 764                 /* If failed to setup optimizing, fallback to kprobe */
 765                 arch_remove_optimized_kprobe(op);
 766                 kfree(op);
 767                 goto out;
 768         }
 769
 770         init_aggr_kprobe(ap, p);
 771         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 772
 773 out:
 774         mutex_unlock(&text_mutex);
 775         jump_label_unlock();
 776 }
 777
 778 #ifdef CONFIG_SYSCTL
 779 static void optimize_all_kprobes(void)
 780 {
 781         struct hlist_head *head;
 782         struct kprobe *p;
 783         unsigned int i;
 784
 785         mutex_lock(&kprobe_mutex);
 786         /* If optimization is already allowed, just return */
 787         if (kprobes_allow_optimization)
 788                 goto out;
 789
 790         kprobes_allow_optimization = true;
 791         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 792                 head = &kprobe_table[i];
 793                 hlist_for_each_entry_rcu(p, head, hlist)
 794                         if (!kprobe_disabled(p))
 795                                 optimize_kprobe(p);
 796         }
 797         printk(KERN_INFO "Kprobes globally optimized\n");
 798 out:
 799         mutex_unlock(&kprobe_mutex);
 800 }
 801
 802 static void unoptimize_all_kprobes(void)
 803 {
 804         struct hlist_head *head;
 805         struct kprobe *p;
 806         unsigned int i;
 807
 808         mutex_lock(&kprobe_mutex);
 809         /* If optimization is already prohibited, just return */
 810         if (!kprobes_allow_optimization) {
 811                 mutex_unlock(&kprobe_mutex);
 812                 return;
 813         }
 814
 815         kprobes_allow_optimization = false;
 816         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 817                 head = &kprobe_table[i];
 818                 hlist_for_each_entry_rcu(p, head, hlist) {
 819                         if (!kprobe_disabled(p))
 820                                 unoptimize_kprobe(p, false);
 821                 }
 822         }
 823         mutex_unlock(&kprobe_mutex);
 824
 825         /* Wait for unoptimizing completion */
 826         wait_for_kprobe_optimizer();
 827         printk(KERN_INFO "Kprobes globally unoptimized\n");
 828 }
 829
 830 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 831 int sysctl_kprobes_optimization;
 832 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 833                                       void __user *buffer, size_t *length,
 834                                       loff_t *ppos)
 835 {
 836         int ret;
 837
 838         mutex_lock(&kprobe_sysctl_mutex);
 839         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 840         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 841
 842         if (sysctl_kprobes_optimization)
 843                 optimize_all_kprobes();
 844         else
 845                 unoptimize_all_kprobes();
 846         mutex_unlock(&kprobe_sysctl_mutex);
 847
 848         return ret;
 849 }
 850 #endif /* CONFIG_SYSCTL */
 851
 852 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 853 static void __arm_kprobe(struct kprobe *p)
 854 {
 855         struct kprobe *_p;
 856
 857         /* Check collision with other optimized kprobes */
 858         _p = get_optimized_kprobe((unsigned long)p->addr);
 859         if (unlikely(_p))
 860                 /* Fallback to unoptimized kprobe */
 861                 unoptimize_kprobe(_p, true);
 862
 863         arch_arm_kprobe(p);
 864         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 865 }
 866
 867 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 868 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 869 {
 870         struct kprobe *_p;
 871
 872         unoptimize_kprobe(p, false);    /* Try to unoptimize */
 873
 874         if (!kprobe_queued(p)) {
 875                 arch_disarm_kprobe(p);
 876                 /* If another kprobe was blocked, optimize it. */
 877                 _p = get_optimized_kprobe((unsigned long)p->addr);
 878                 if (unlikely(_p) && reopt)
 879                         optimize_kprobe(_p);
 880         }
 881         /* TODO: reoptimize others after unoptimized this probe */
 882 }
 883
 884 #else /* !CONFIG_OPTPROBES */
 885
 886 #define optimize_kprobe(p)                      do {} while (0)
 887 #define unoptimize_kprobe(p, f)                 do {} while (0)
 888 #define kill_optimized_kprobe(p)                do {} while (0)
 889 #define prepare_optimized_kprobe(p)             do {} while (0)
 890 #define try_to_optimize_kprobe(p)               do {} while (0)
 891 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 892 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 893 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 894 #define wait_for_kprobe_optimizer()             do {} while (0)
 895
 896 /* There should be no unused kprobes can be reused without optimization */
 897 static void reuse_unused_kprobe(struct kprobe *ap)
 898 {
 899         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 900         BUG_ON(kprobe_unused(ap));
 901 }
 902
 903 static void free_aggr_kprobe(struct kprobe *p)
 904 {
 905         arch_remove_kprobe(p);
 906         kfree(p);
 907 }
 908
 909 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 910 {
 911         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 912 }
 913 #endif /* CONFIG_OPTPROBES */
 914
 915 #ifdef CONFIG_KPROBES_ON_FTRACE
 916 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 917         .func = kprobe_ftrace_handler,
 918         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 919 };
 920 static int kprobe_ftrace_enabled;
 921
 922 /* Must ensure p->addr is really on ftrace */
 923 static int prepare_kprobe(struct kprobe *p)
 924 {
 925         if (!kprobe_ftrace(p))
 926                 return arch_prepare_kprobe(p);
 927
 928         return arch_prepare_kprobe_ftrace(p);
 929 }
 930
 931 /* Caller must lock kprobe_mutex */
 932 static void arm_kprobe_ftrace(struct kprobe *p)
 933 {
 934         int ret;
 935
 936         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 937                                    (unsigned long)p->addr, 0, 0);
 938         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 939         kprobe_ftrace_enabled++;
 940         if (kprobe_ftrace_enabled == 1) {
 941                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 942                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 943         }
 944 }
 945
 946 /* Caller must lock kprobe_mutex */
 947 static void disarm_kprobe_ftrace(struct kprobe *p)
 948 {
 949         int ret;
 950
 951         kprobe_ftrace_enabled--;
 952         if (kprobe_ftrace_enabled == 0) {
 953                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 954                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 955         }
 956         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 957                            (unsigned long)p->addr, 1, 0);
 958         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 959 }
 960 #else   /* !CONFIG_KPROBES_ON_FTRACE */
 961 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 962 #define arm_kprobe_ftrace(p)    do {} while (0)
 963 #define disarm_kprobe_ftrace(p) do {} while (0)
 964 #endif
 965
 966 /* Arm a kprobe with text_mutex */
 967 static void arm_kprobe(struct kprobe *kp)
 968 {
 969         if (unlikely(kprobe_ftrace(kp))) {
 970                 arm_kprobe_ftrace(kp);
 971                 return;
 972         }
 973         /*
 974          * Here, since __arm_kprobe() doesn't use stop_machine(),
 975          * this doesn't cause deadlock on text_mutex. So, we don't
 976          * need get_online_cpus().
 977          */
 978         mutex_lock(&text_mutex);
 979         __arm_kprobe(kp);
 980         mutex_unlock(&text_mutex);
 981 }
 982
 983 /* Disarm a kprobe with text_mutex */
 984 static void disarm_kprobe(struct kprobe *kp, bool reopt)
 985 {
 986         if (unlikely(kprobe_ftrace(kp))) {
 987                 disarm_kprobe_ftrace(kp);
 988                 return;
 989         }
 990         /* Ditto */
 991         mutex_lock(&text_mutex);
 992         __disarm_kprobe(kp, reopt);
 993         mutex_unlock(&text_mutex);
 994 }
 995
 996 /*
 997  * Aggregate handlers for multiple kprobes support - these handlers
 998  * take care of invoking the individual kprobe handlers on p->list
 999  */
1000 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1001 {
1002         struct kprobe *kp;
1003
1004         list_for_each_entry_rcu(kp, &p->list, list) {
1005                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1006                         set_kprobe_instance(kp);
1007                         if (kp->pre_handler(kp, regs))
1008                                 return 1;
1009                 }
1010                 reset_kprobe_instance();
1011         }
1012         return 0;
1013 }
1014 NOKPROBE_SYMBOL(aggr_pre_handler);
1015
1016 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1017                               unsigned long flags)
1018 {
1019         struct kprobe *kp;
1020
1021         list_for_each_entry_rcu(kp, &p->list, list) {
1022                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1023                         set_kprobe_instance(kp);
1024                         kp->post_handler(kp, regs, flags);
1025                         reset_kprobe_instance();
1026                 }
1027         }
1028 }
1029 NOKPROBE_SYMBOL(aggr_post_handler);
1030
1031 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1032                               int trapnr)
1033 {
1034         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1035
1036         /*
1037          * if we faulted "during" the execution of a user specified
1038          * probe handler, invoke just that probe's fault handler
1039          */
1040         if (cur && cur->fault_handler) {
1041                 if (cur->fault_handler(cur, regs, trapnr))
1042                         return 1;
1043         }
1044         return 0;
1045 }
1046 NOKPROBE_SYMBOL(aggr_fault_handler);
1047
1048 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1049 {
1050         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1051         int ret = 0;
1052
1053         if (cur && cur->break_handler) {
1054                 if (cur->break_handler(cur, regs))
1055                         ret = 1;
1056         }
1057         reset_kprobe_instance();
1058         return ret;
1059 }
1060 NOKPROBE_SYMBOL(aggr_break_handler);
1061
1062 /* Walks the list and increments nmissed count for multiprobe case */
1063 void kprobes_inc_nmissed_count(struct kprobe *p)
1064 {
1065         struct kprobe *kp;
1066         if (!kprobe_aggrprobe(p)) {
1067                 p->nmissed++;
1068         } else {
1069                 list_for_each_entry_rcu(kp, &p->list, list)
1070                         kp->nmissed++;
1071         }
1072         return;
1073 }
1074 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1075
1076 void recycle_rp_inst(struct kretprobe_instance *ri,
1077                      struct hlist_head *head)
1078 {
1079         struct kretprobe *rp = ri->rp;
1080
1081         /* remove rp inst off the rprobe_inst_table */
1082         hlist_del(&ri->hlist);
1083         INIT_HLIST_NODE(&ri->hlist);
1084         if (likely(rp)) {
1085                 raw_spin_lock(&rp->lock);
1086                 hlist_add_head(&ri->hlist, &rp->free_instances);
1087                 raw_spin_unlock(&rp->lock);
1088         } else
1089                 /* Unregistering */
1090                 hlist_add_head(&ri->hlist, head);
1091 }
1092 NOKPROBE_SYMBOL(recycle_rp_inst);
1093
1094 void kretprobe_hash_lock(struct task_struct *tsk,
1095                          struct hlist_head **head, unsigned long *flags)
1096 __acquires(hlist_lock)
1097 {
1098         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1099         raw_spinlock_t *hlist_lock;
1100
1101         *head = &kretprobe_inst_table[hash];
1102         hlist_lock = kretprobe_table_lock_ptr(hash);
1103         raw_spin_lock_irqsave(hlist_lock, *flags);
1104 }
1105 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1106
1107 static void kretprobe_table_lock(unsigned long hash,
1108                                  unsigned long *flags)
1109 __acquires(hlist_lock)
1110 {
1111         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1112         raw_spin_lock_irqsave(hlist_lock, *flags);
1113 }
1114 NOKPROBE_SYMBOL(kretprobe_table_lock);
1115
1116 void kretprobe_hash_unlock(struct task_struct *tsk,
1117                            unsigned long *flags)
1118 __releases(hlist_lock)
1119 {
1120         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1121         raw_spinlock_t *hlist_lock;
1122
1123         hlist_lock = kretprobe_table_lock_ptr(hash);
1124         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1125 }
1126 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1127
1128 static void kretprobe_table_unlock(unsigned long hash,
1129                                    unsigned long *flags)
1130 __releases(hlist_lock)
1131 {
1132         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1133         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1134 }
1135 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1136
1137 /*
1138  * This function is called from finish_task_switch when task tk becomes dead,
1139  * so that we can recycle any function-return probe instances associated
1140  * with this task. These left over instances represent probed functions
1141  * that have been called but will never return.
1142  */
1143 void kprobe_flush_task(struct task_struct *tk)
1144 {
1145         struct kretprobe_instance *ri;
1146         struct hlist_head *head, empty_rp;
1147         struct hlist_node *tmp;
1148         unsigned long hash, flags = 0;
1149
1150         if (unlikely(!kprobes_initialized))
1151                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1152                 return;
1153
1154         INIT_HLIST_HEAD(&empty_rp);
1155         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1156         head = &kretprobe_inst_table[hash];
1157         kretprobe_table_lock(hash, &flags);
1158         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1159                 if (ri->task == tk)
1160                         recycle_rp_inst(ri, &empty_rp);
1161         }
1162         kretprobe_table_unlock(hash, &flags);
1163         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1164                 hlist_del(&ri->hlist);
1165                 kfree(ri);
1166         }
1167 }
1168 NOKPROBE_SYMBOL(kprobe_flush_task);
1169
1170 static inline void free_rp_inst(struct kretprobe *rp)
1171 {
1172         struct kretprobe_instance *ri;
1173         struct hlist_node *next;
1174
1175         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1176                 hlist_del(&ri->hlist);
1177                 kfree(ri);
1178         }
1179 }
1180
1181 static void cleanup_rp_inst(struct kretprobe *rp)
1182 {
1183         unsigned long flags, hash;
1184         struct kretprobe_instance *ri;
1185         struct hlist_node *next;
1186         struct hlist_head *head;
1187
1188         /* No race here */
1189         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1190                 kretprobe_table_lock(hash, &flags);
1191                 head = &kretprobe_inst_table[hash];
1192                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1193                         if (ri->rp == rp)
1194                                 ri->rp = NULL;
1195                 }
1196                 kretprobe_table_unlock(hash, &flags);
1197         }
1198         free_rp_inst(rp);
1199 }
1200 NOKPROBE_SYMBOL(cleanup_rp_inst);
1201
1202 /*
1203 * Add the new probe to ap->list. Fail if this is the
1204 * second jprobe at the address - two jprobes can't coexist
1205 */
1206 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1207 {
1208         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1209
1210         if (p->break_handler || p->post_handler)
1211                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1212
1213         if (p->break_handler) {
1214                 if (ap->break_handler)
1215                         return -EEXIST;
1216                 list_add_tail_rcu(&p->list, &ap->list);
1217                 ap->break_handler = aggr_break_handler;
1218         } else
1219                 list_add_rcu(&p->list, &ap->list);
1220         if (p->post_handler && !ap->post_handler)
1221                 ap->post_handler = aggr_post_handler;
1222
1223         return 0;
1224 }
1225
1226 /*
1227  * Fill in the required fields of the "manager kprobe". Replace the
1228  * earlier kprobe in the hlist with the manager kprobe
1229  */
1230 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1231 {
1232         /* Copy p's insn slot to ap */
1233         copy_kprobe(p, ap);
1234         flush_insn_slot(ap);
1235         ap->addr = p->addr;
1236         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1237         ap->pre_handler = aggr_pre_handler;
1238         ap->fault_handler = aggr_fault_handler;
1239         /* We don't care the kprobe which has gone. */
1240         if (p->post_handler && !kprobe_gone(p))
1241                 ap->post_handler = aggr_post_handler;
1242         if (p->break_handler && !kprobe_gone(p))
1243                 ap->break_handler = aggr_break_handler;
1244
1245         INIT_LIST_HEAD(&ap->list);
1246         INIT_HLIST_NODE(&ap->hlist);
1247
1248         list_add_rcu(&p->list, &ap->list);
1249         hlist_replace_rcu(&p->hlist, &ap->hlist);
1250 }
1251
1252 /*
1253  * This is the second or subsequent kprobe at the address - handle
1254  * the intricacies
1255  */
1256 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1257 {
1258         int ret = 0;
1259         struct kprobe *ap = orig_p;
1260
1261         /* For preparing optimization, jump_label_text_reserved() is called */
1262         jump_label_lock();
1263         /*
1264          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1265          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1266          */
1267         get_online_cpus();
1268         mutex_lock(&text_mutex);
1269
1270         if (!kprobe_aggrprobe(orig_p)) {
1271                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1272                 ap = alloc_aggr_kprobe(orig_p);
1273                 if (!ap) {
1274                         ret = -ENOMEM;
1275                         goto out;
1276                 }
1277                 init_aggr_kprobe(ap, orig_p);
1278         } else if (kprobe_unused(ap))
1279                 /* This probe is going to die. Rescue it */
1280                 reuse_unused_kprobe(ap);
1281
1282         if (kprobe_gone(ap)) {
1283                 /*
1284                  * Attempting to insert new probe at the same location that
1285                  * had a probe in the module vaddr area which already
1286                  * freed. So, the instruction slot has already been
1287                  * released. We need a new slot for the new probe.
1288                  */
1289                 ret = arch_prepare_kprobe(ap);
1290                 if (ret)
1291                         /*
1292                          * Even if fail to allocate new slot, don't need to
1293                          * free aggr_probe. It will be used next time, or
1294                          * freed by unregister_kprobe.
1295                          */
1296                         goto out;
1297
1298                 /* Prepare optimized instructions if possible. */
1299                 prepare_optimized_kprobe(ap);
1300
1301                 /*
1302                  * Clear gone flag to prevent allocating new slot again, and
1303                  * set disabled flag because it is not armed yet.
1304                  */
1305                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1306                             | KPROBE_FLAG_DISABLED;
1307         }
1308
1309         /* Copy ap's insn slot to p */
1310         copy_kprobe(ap, p);
1311         ret = add_new_kprobe(ap, p);
1312
1313 out:
1314         mutex_unlock(&text_mutex);
1315         put_online_cpus();
1316         jump_label_unlock();
1317
1318         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1319                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1320                 if (!kprobes_all_disarmed)
1321                         /* Arm the breakpoint again. */
1322                         arm_kprobe(ap);
1323         }
1324         return ret;
1325 }
1326
1327 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1328 {
1329         /* The __kprobes marked functions and entry code must not be probed */
1330         return addr >= (unsigned long)__kprobes_text_start &&
1331                addr < (unsigned long)__kprobes_text_end;
1332 }
1333
1334 static bool within_kprobe_blacklist(unsigned long addr)
1335 {
1336         struct kprobe_blacklist_entry *ent;
1337
1338         if (arch_within_kprobe_blacklist(addr))
1339                 return true;
1340         /*
1341          * If there exists a kprobe_blacklist, verify and
1342          * fail any probe registration in the prohibited area
1343          */
1344         list_for_each_entry(ent, &kprobe_blacklist, list) {
1345                 if (addr >= ent->start_addr && addr < ent->end_addr)
1346                         return true;
1347         }
1348
1349         return false;
1350 }
1351
1352 /*
1353  * If we have a symbol_name argument, look it up and add the offset field
1354  * to it. This way, we can specify a relative address to a symbol.
1355  * This returns encoded errors if it fails to look up symbol or invalid
1356  * combination of parameters.
1357  */
1358 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1359 {
1360         kprobe_opcode_t *addr = p->addr;
1361
1362         if ((p->symbol_name && p->addr) ||
1363             (!p->symbol_name && !p->addr))
1364                 goto invalid;
1365
1366         if (p->symbol_name) {
1367                 kprobe_lookup_name(p->symbol_name, addr);
1368                 if (!addr)
1369                         return ERR_PTR(-ENOENT);
1370         }
1371
1372         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1373         if (addr)
1374                 return addr;
1375
1376 invalid:
1377         return ERR_PTR(-EINVAL);
1378 }
1379
1380 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1381 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1382 {
1383         struct kprobe *ap, *list_p;
1384
1385         ap = get_kprobe(p->addr);
1386         if (unlikely(!ap))
1387                 return NULL;
1388
1389         if (p != ap) {
1390                 list_for_each_entry_rcu(list_p, &ap->list, list)
1391                         if (list_p == p)
1392                         /* kprobe p is a valid probe */
1393                                 goto valid;
1394                 return NULL;
1395         }
1396 valid:
1397         return ap;
1398 }
1399
1400 /* Return error if the kprobe is being re-registered */
1401 static inline int check_kprobe_rereg(struct kprobe *p)
1402 {
1403         int ret = 0;
1404
1405         mutex_lock(&kprobe_mutex);
1406         if (__get_valid_kprobe(p))
1407                 ret = -EINVAL;
1408         mutex_unlock(&kprobe_mutex);
1409
1410         return ret;
1411 }
1412
1413 int __weak arch_check_ftrace_location(struct kprobe *p)
1414 {
1415         unsigned long ftrace_addr;
1416
1417         ftrace_addr = ftrace_location((unsigned long)p->addr);
1418         if (ftrace_addr) {
1419 #ifdef CONFIG_KPROBES_ON_FTRACE
1420                 /* Given address is not on the instruction boundary */
1421                 if ((unsigned long)p->addr != ftrace_addr)
1422                         return -EILSEQ;
1423                 p->flags |= KPROBE_FLAG_FTRACE;
1424 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1425                 return -EINVAL;
1426 #endif
1427         }
1428         return 0;
1429 }
1430
1431 static int check_kprobe_address_safe(struct kprobe *p,
1432                                      struct module **probed_mod)
1433 {
1434         int ret;
1435
1436         ret = arch_check_ftrace_location(p);
1437         if (ret)
1438                 return ret;
1439         jump_label_lock();
1440         preempt_disable();
1441
1442         /* Ensure it is not in reserved area nor out of text */
1443         if (!kernel_text_address((unsigned long) p->addr) ||
1444             within_kprobe_blacklist((unsigned long) p->addr) ||
1445             jump_label_text_reserved(p->addr, p->addr)) {
1446                 ret = -EINVAL;
1447                 goto out;
1448         }
1449
1450         /* Check if are we probing a module */
1451         *probed_mod = __module_text_address((unsigned long) p->addr);
1452         if (*probed_mod) {
1453                 /*
1454                  * We must hold a refcount of the probed module while updating
1455                  * its code to prohibit unexpected unloading.
1456                  */
1457                 if (unlikely(!try_module_get(*probed_mod))) {
1458                         ret = -ENOENT;
1459                         goto out;
1460                 }
1461
1462                 /*
1463                  * If the module freed .init.text, we couldn't insert
1464                  * kprobes in there.
1465                  */
1466                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1467                     (*probed_mod)->state != MODULE_STATE_COMING) {
1468                         module_put(*probed_mod);
1469                         *probed_mod = NULL;
1470                         ret = -ENOENT;
1471                 }
1472         }
1473 out:
1474         preempt_enable();
1475         jump_label_unlock();
1476
1477         return ret;
1478 }
1479
1480 int register_kprobe(struct kprobe *p)
1481 {
1482         int ret;
1483         struct kprobe *old_p;
1484         struct module *probed_mod;
1485         kprobe_opcode_t *addr;
1486
1487         /* Adjust probe address from symbol */
1488         addr = kprobe_addr(p);
1489         if (IS_ERR(addr))
1490                 return PTR_ERR(addr);
1491         p->addr = addr;
1492
1493         ret = check_kprobe_rereg(p);
1494         if (ret)
1495                 return ret;
1496
1497         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1498         p->flags &= KPROBE_FLAG_DISABLED;
1499         p->nmissed = 0;
1500         INIT_LIST_HEAD(&p->list);
1501
1502         ret = check_kprobe_address_safe(p, &probed_mod);
1503         if (ret)
1504                 return ret;
1505
1506         mutex_lock(&kprobe_mutex);
1507
1508         old_p = get_kprobe(p->addr);
1509         if (old_p) {
1510                 /* Since this may unoptimize old_p, locking text_mutex. */
1511                 ret = register_aggr_kprobe(old_p, p);
1512                 goto out;
1513         }
1514
1515         mutex_lock(&text_mutex);        /* Avoiding text modification */
1516         ret = prepare_kprobe(p);
1517         mutex_unlock(&text_mutex);
1518         if (ret)
1519                 goto out;
1520
1521         INIT_HLIST_NODE(&p->hlist);
1522         hlist_add_head_rcu(&p->hlist,
1523                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1524
1525         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1526                 arm_kprobe(p);
1527
1528         /* Try to optimize kprobe */
1529         try_to_optimize_kprobe(p);
1530
1531 out:
1532         mutex_unlock(&kprobe_mutex);
1533
1534         if (probed_mod)
1535                 module_put(probed_mod);
1536
1537         return ret;
1538 }
1539 EXPORT_SYMBOL_GPL(register_kprobe);
1540
1541 /* Check if all probes on the aggrprobe are disabled */
1542 static int aggr_kprobe_disabled(struct kprobe *ap)
1543 {
1544         struct kprobe *kp;
1545
1546         list_for_each_entry_rcu(kp, &ap->list, list)
1547                 if (!kprobe_disabled(kp))
1548                         /*
1549                          * There is an active probe on the list.
1550                          * We can't disable this ap.
1551                          */
1552                         return 0;
1553
1554         return 1;
1555 }
1556
1557 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1558 static struct kprobe *__disable_kprobe(struct kprobe *p)
1559 {
1560         struct kprobe *orig_p;
1561
1562         /* Get an original kprobe for return */
1563         orig_p = __get_valid_kprobe(p);
1564         if (unlikely(orig_p == NULL))
1565                 return NULL;
1566
1567         if (!kprobe_disabled(p)) {
1568                 /* Disable probe if it is a child probe */
1569                 if (p != orig_p)
1570                         p->flags |= KPROBE_FLAG_DISABLED;
1571
1572                 /* Try to disarm and disable this/parent probe */
1573                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1574                         disarm_kprobe(orig_p, true);
1575                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1576                 }
1577         }
1578
1579         return orig_p;
1580 }
1581
1582 /*
1583  * Unregister a kprobe without a scheduler synchronization.
1584  */
1585 static int __unregister_kprobe_top(struct kprobe *p)
1586 {
1587         struct kprobe *ap, *list_p;
1588
1589         /* Disable kprobe. This will disarm it if needed. */
1590         ap = __disable_kprobe(p);
1591         if (ap == NULL)
1592                 return -EINVAL;
1593
1594         if (ap == p)
1595                 /*
1596                  * This probe is an independent(and non-optimized) kprobe
1597                  * (not an aggrprobe). Remove from the hash list.
1598                  */
1599                 goto disarmed;
1600
1601         /* Following process expects this probe is an aggrprobe */
1602         WARN_ON(!kprobe_aggrprobe(ap));
1603
1604         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1605                 /*
1606                  * !disarmed could be happen if the probe is under delayed
1607                  * unoptimizing.
1608                  */
1609                 goto disarmed;
1610         else {
1611                 /* If disabling probe has special handlers, update aggrprobe */
1612                 if (p->break_handler && !kprobe_gone(p))
1613                         ap->break_handler = NULL;
1614                 if (p->post_handler && !kprobe_gone(p)) {
1615                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1616                                 if ((list_p != p) && (list_p->post_handler))
1617                                         goto noclean;
1618                         }
1619                         ap->post_handler = NULL;
1620                 }
1621 noclean:
1622                 /*
1623                  * Remove from the aggrprobe: this path will do nothing in
1624                  * __unregister_kprobe_bottom().
1625                  */
1626                 list_del_rcu(&p->list);
1627                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1628                         /*
1629                          * Try to optimize this probe again, because post
1630                          * handler may have been changed.
1631                          */
1632                         optimize_kprobe(ap);
1633         }
1634         return 0;
1635
1636 disarmed:
1637         BUG_ON(!kprobe_disarmed(ap));
1638         hlist_del_rcu(&ap->hlist);
1639         return 0;
1640 }
1641
1642 static void __unregister_kprobe_bottom(struct kprobe *p)
1643 {
1644         struct kprobe *ap;
1645
1646         if (list_empty(&p->list))
1647                 /* This is an independent kprobe */
1648                 arch_remove_kprobe(p);
1649         else if (list_is_singular(&p->list)) {
1650                 /* This is the last child of an aggrprobe */
1651                 ap = list_entry(p->list.next, struct kprobe, list);
1652                 list_del(&p->list);
1653                 free_aggr_kprobe(ap);
1654         }
1655         /* Otherwise, do nothing. */
1656 }
1657
1658 int register_kprobes(struct kprobe **kps, int num)
1659 {
1660         int i, ret = 0;
1661
1662         if (num <= 0)
1663                 return -EINVAL;
1664         for (i = 0; i < num; i++) {
1665                 ret = register_kprobe(kps[i]);
1666                 if (ret < 0) {
1667                         if (i > 0)
1668                                 unregister_kprobes(kps, i);
1669                         break;
1670                 }
1671         }
1672         return ret;
1673 }
1674 EXPORT_SYMBOL_GPL(register_kprobes);
1675
1676 void unregister_kprobe(struct kprobe *p)
1677 {
1678         unregister_kprobes(&p, 1);
1679 }
1680 EXPORT_SYMBOL_GPL(unregister_kprobe);
1681
1682 void unregister_kprobes(struct kprobe **kps, int num)
1683 {
1684         int i;
1685
1686         if (num <= 0)
1687                 return;
1688         mutex_lock(&kprobe_mutex);
1689         for (i = 0; i < num; i++)
1690                 if (__unregister_kprobe_top(kps[i]) < 0)
1691                         kps[i]->addr = NULL;
1692         mutex_unlock(&kprobe_mutex);
1693
1694         synchronize_sched();
1695         for (i = 0; i < num; i++)
1696                 if (kps[i]->addr)
1697                         __unregister_kprobe_bottom(kps[i]);
1698 }
1699 EXPORT_SYMBOL_GPL(unregister_kprobes);
1700
1701 static struct notifier_block kprobe_exceptions_nb = {
1702         .notifier_call = kprobe_exceptions_notify,
1703         .priority = 0x7fffffff /* we need to be notified first */
1704 };
1705
1706 unsigned long __weak arch_deref_entry_point(void *entry)
1707 {
1708         return (unsigned long)entry;
1709 }
1710
1711 int register_jprobes(struct jprobe **jps, int num)
1712 {
1713         struct jprobe *jp;
1714         int ret = 0, i;
1715
1716         if (num <= 0)
1717                 return -EINVAL;
1718         for (i = 0; i < num; i++) {
1719                 unsigned long addr, offset;
1720                 jp = jps[i];
1721                 addr = arch_deref_entry_point(jp->entry);
1722
1723                 /* Verify probepoint is a function entry point */
1724                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1725                     offset == 0) {
1726                         jp->kp.pre_handler = setjmp_pre_handler;
1727                         jp->kp.break_handler = longjmp_break_handler;
1728                         ret = register_kprobe(&jp->kp);
1729                 } else
1730                         ret = -EINVAL;
1731
1732                 if (ret < 0) {
1733                         if (i > 0)
1734                                 unregister_jprobes(jps, i);
1735                         break;
1736                 }
1737         }
1738         return ret;
1739 }
1740 EXPORT_SYMBOL_GPL(register_jprobes);
1741
1742 int register_jprobe(struct jprobe *jp)
1743 {
1744         return register_jprobes(&jp, 1);
1745 }
1746 EXPORT_SYMBOL_GPL(register_jprobe);
1747
1748 void unregister_jprobe(struct jprobe *jp)
1749 {
1750         unregister_jprobes(&jp, 1);
1751 }
1752 EXPORT_SYMBOL_GPL(unregister_jprobe);
1753
1754 void unregister_jprobes(struct jprobe **jps, int num)
1755 {
1756         int i;
1757
1758         if (num <= 0)
1759                 return;
1760         mutex_lock(&kprobe_mutex);
1761         for (i = 0; i < num; i++)
1762                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1763                         jps[i]->kp.addr = NULL;
1764         mutex_unlock(&kprobe_mutex);
1765
1766         synchronize_sched();
1767         for (i = 0; i < num; i++) {
1768                 if (jps[i]->kp.addr)
1769                         __unregister_kprobe_bottom(&jps[i]->kp);
1770         }
1771 }
1772 EXPORT_SYMBOL_GPL(unregister_jprobes);
1773
1774 #ifdef CONFIG_KRETPROBES
1775 /*
1776  * This kprobe pre_handler is registered with every kretprobe. When probe
1777  * hits it will set up the return probe.
1778  */
1779 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1780 {
1781         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1782         unsigned long hash, flags = 0;
1783         struct kretprobe_instance *ri;
1784
1785         /*
1786          * To avoid deadlocks, prohibit return probing in NMI contexts,
1787          * just skip the probe and increase the (inexact) 'nmissed'
1788          * statistical counter, so that the user is informed that
1789          * something happened:
1790          */
1791         if (unlikely(in_nmi())) {
1792                 rp->nmissed++;
1793                 return 0;
1794         }
1795
1796         /* TODO: consider to only swap the RA after the last pre_handler fired */
1797         hash = hash_ptr(current, KPROBE_HASH_BITS);
1798         raw_spin_lock_irqsave(&rp->lock, flags);
1799         if (!hlist_empty(&rp->free_instances)) {
1800                 ri = hlist_entry(rp->free_instances.first,
1801                                 struct kretprobe_instance, hlist);
1802                 hlist_del(&ri->hlist);
1803                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1804
1805                 ri->rp = rp;
1806                 ri->task = current;
1807
1808                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1809                         raw_spin_lock_irqsave(&rp->lock, flags);
1810                         hlist_add_head(&ri->hlist, &rp->free_instances);
1811                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1812                         return 0;
1813                 }
1814
1815                 arch_prepare_kretprobe(ri, regs);
1816
1817                 /* XXX(hch): why is there no hlist_move_head? */
1818                 INIT_HLIST_NODE(&ri->hlist);
1819                 kretprobe_table_lock(hash, &flags);
1820                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1821                 kretprobe_table_unlock(hash, &flags);
1822         } else {
1823                 rp->nmissed++;
1824                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1825         }
1826         return 0;
1827 }
1828 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1829
1830 int register_kretprobe(struct kretprobe *rp)
1831 {
1832         int ret = 0;
1833         struct kretprobe_instance *inst;
1834         int i;
1835         void *addr;
1836
1837         if (kretprobe_blacklist_size) {
1838                 addr = kprobe_addr(&rp->kp);
1839                 if (IS_ERR(addr))
1840                         return PTR_ERR(addr);
1841
1842                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1843                         if (kretprobe_blacklist[i].addr == addr)
1844                                 return -EINVAL;
1845                 }
1846         }
1847
1848         rp->kp.pre_handler = pre_handler_kretprobe;
1849         rp->kp.post_handler = NULL;
1850         rp->kp.fault_handler = NULL;
1851         rp->kp.break_handler = NULL;
1852
1853         /* Pre-allocate memory for max kretprobe instances */
1854         if (rp->maxactive <= 0) {
1855 #ifdef CONFIG_PREEMPT
1856                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1857 #else
1858                 rp->maxactive = num_possible_cpus();
1859 #endif
1860         }
1861         raw_spin_lock_init(&rp->lock);
1862         INIT_HLIST_HEAD(&rp->free_instances);
1863         for (i = 0; i < rp->maxactive; i++) {
1864                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1865                                rp->data_size, GFP_KERNEL);
1866                 if (inst == NULL) {
1867                         free_rp_inst(rp);
1868                         return -ENOMEM;
1869                 }
1870                 INIT_HLIST_NODE(&inst->hlist);
1871                 hlist_add_head(&inst->hlist, &rp->free_instances);
1872         }
1873
1874         rp->nmissed = 0;
1875         /* Establish function entry probe point */
1876         ret = register_kprobe(&rp->kp);
1877         if (ret != 0)
1878                 free_rp_inst(rp);
1879         return ret;
1880 }
1881 EXPORT_SYMBOL_GPL(register_kretprobe);
1882
1883 int register_kretprobes(struct kretprobe **rps, int num)
1884 {
1885         int ret = 0, i;
1886
1887         if (num <= 0)
1888                 return -EINVAL;
1889         for (i = 0; i < num; i++) {
1890                 ret = register_kretprobe(rps[i]);
1891                 if (ret < 0) {
1892                         if (i > 0)
1893                                 unregister_kretprobes(rps, i);
1894                         break;
1895                 }
1896         }
1897         return ret;
1898 }
1899 EXPORT_SYMBOL_GPL(register_kretprobes);
1900
1901 void unregister_kretprobe(struct kretprobe *rp)
1902 {
1903         unregister_kretprobes(&rp, 1);
1904 }
1905 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1906
1907 void unregister_kretprobes(struct kretprobe **rps, int num)
1908 {
1909         int i;
1910
1911         if (num <= 0)
1912                 return;
1913         mutex_lock(&kprobe_mutex);
1914         for (i = 0; i < num; i++)
1915                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1916                         rps[i]->kp.addr = NULL;
1917         mutex_unlock(&kprobe_mutex);
1918
1919         synchronize_sched();
1920         for (i = 0; i < num; i++) {
1921                 if (rps[i]->kp.addr) {
1922                         __unregister_kprobe_bottom(&rps[i]->kp);
1923                         cleanup_rp_inst(rps[i]);
1924                 }
1925         }
1926 }
1927 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1928
1929 #else /* CONFIG_KRETPROBES */
1930 int register_kretprobe(struct kretprobe *rp)
1931 {
1932         return -ENOSYS;
1933 }
1934 EXPORT_SYMBOL_GPL(register_kretprobe);
1935
1936 int register_kretprobes(struct kretprobe **rps, int num)
1937 {
1938         return -ENOSYS;
1939 }
1940 EXPORT_SYMBOL_GPL(register_kretprobes);
1941
1942 void unregister_kretprobe(struct kretprobe *rp)
1943 {
1944 }
1945 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1946
1947 void unregister_kretprobes(struct kretprobe **rps, int num)
1948 {
1949 }
1950 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1951
1952 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1953 {
1954         return 0;
1955 }
1956 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1957
1958 #endif /* CONFIG_KRETPROBES */
1959
1960 /* Set the kprobe gone and remove its instruction buffer. */
1961 static void kill_kprobe(struct kprobe *p)
1962 {
1963         struct kprobe *kp;
1964
1965         p->flags |= KPROBE_FLAG_GONE;
1966         if (kprobe_aggrprobe(p)) {
1967                 /*
1968                  * If this is an aggr_kprobe, we have to list all the
1969                  * chained probes and mark them GONE.
1970                  */
1971                 list_for_each_entry_rcu(kp, &p->list, list)
1972                         kp->flags |= KPROBE_FLAG_GONE;
1973                 p->post_handler = NULL;
1974                 p->break_handler = NULL;
1975                 kill_optimized_kprobe(p);
1976         }
1977         /*
1978          * Here, we can remove insn_slot safely, because no thread calls
1979          * the original probed function (which will be freed soon) any more.
1980          */
1981         arch_remove_kprobe(p);
1982 }
1983
1984 /* Disable one kprobe */
1985 int disable_kprobe(struct kprobe *kp)
1986 {
1987         int ret = 0;
1988
1989         mutex_lock(&kprobe_mutex);
1990
1991         /* Disable this kprobe */
1992         if (__disable_kprobe(kp) == NULL)
1993                 ret = -EINVAL;
1994
1995         mutex_unlock(&kprobe_mutex);
1996         return ret;
1997 }
1998 EXPORT_SYMBOL_GPL(disable_kprobe);
1999
2000 /* Enable one kprobe */
2001 int enable_kprobe(struct kprobe *kp)
2002 {
2003         int ret = 0;
2004         struct kprobe *p;
2005
2006         mutex_lock(&kprobe_mutex);
2007
2008         /* Check whether specified probe is valid. */
2009         p = __get_valid_kprobe(kp);
2010         if (unlikely(p == NULL)) {
2011                 ret = -EINVAL;
2012                 goto out;
2013         }
2014
2015         if (kprobe_gone(kp)) {
2016                 /* This kprobe has gone, we couldn't enable it. */
2017                 ret = -EINVAL;
2018                 goto out;
2019         }
2020
2021         if (p != kp)
2022                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2023
2024         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2025                 p->flags &= ~KPROBE_FLAG_DISABLED;
2026                 arm_kprobe(p);
2027         }
2028 out:
2029         mutex_unlock(&kprobe_mutex);
2030         return ret;
2031 }
2032 EXPORT_SYMBOL_GPL(enable_kprobe);
2033
2034 void dump_kprobe(struct kprobe *kp)
2035 {
2036         printk(KERN_WARNING "Dumping kprobe:\n");
2037         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2038                kp->symbol_name, kp->addr, kp->offset);
2039 }
2040 NOKPROBE_SYMBOL(dump_kprobe);
2041
2042 /*
2043  * Lookup and populate the kprobe_blacklist.
2044  *
2045  * Unlike the kretprobe blacklist, we'll need to determine
2046  * the range of addresses that belong to the said functions,
2047  * since a kprobe need not necessarily be at the beginning
2048  * of a function.
2049  */
2050 static int __init populate_kprobe_blacklist(unsigned long *start,
2051                                              unsigned long *end)
2052 {
2053         unsigned long *iter;
2054         struct kprobe_blacklist_entry *ent;
2055         unsigned long entry, offset = 0, size = 0;
2056
2057         for (iter = start; iter < end; iter++) {
2058                 entry = arch_deref_entry_point((void *)*iter);
2059
2060                 if (!kernel_text_address(entry) ||
2061                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2062                         pr_err("Failed to find blacklist at %p\n",
2063                                 (void *)entry);
2064                         continue;
2065                 }
2066
2067                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2068                 if (!ent)
2069                         return -ENOMEM;
2070                 ent->start_addr = entry;
2071                 ent->end_addr = entry + size;
2072                 INIT_LIST_HEAD(&ent->list);
2073                 list_add_tail(&ent->list, &kprobe_blacklist);
2074         }
2075         return 0;
2076 }
2077
2078 /* Module notifier call back, checking kprobes on the module */
2079 static int kprobes_module_callback(struct notifier_block *nb,
2080                                    unsigned long val, void *data)
2081 {
2082         struct module *mod = data;
2083         struct hlist_head *head;
2084         struct kprobe *p;
2085         unsigned int i;
2086         int checkcore = (val == MODULE_STATE_GOING);
2087
2088         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2089                 return NOTIFY_DONE;
2090
2091         /*
2092          * When MODULE_STATE_GOING was notified, both of module .text and
2093          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2094          * notified, only .init.text section would be freed. We need to
2095          * disable kprobes which have been inserted in the sections.
2096          */
2097         mutex_lock(&kprobe_mutex);
2098         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2099                 head = &kprobe_table[i];
2100                 hlist_for_each_entry_rcu(p, head, hlist)
2101                         if (within_module_init((unsigned long)p->addr, mod) ||
2102                             (checkcore &&
2103                              within_module_core((unsigned long)p->addr, mod))) {
2104                                 /*
2105                                  * The vaddr this probe is installed will soon
2106                                  * be vfreed buy not synced to disk. Hence,
2107                                  * disarming the breakpoint isn't needed.
2108                                  */
2109                                 kill_kprobe(p);
2110                         }
2111         }
2112         mutex_unlock(&kprobe_mutex);
2113         return NOTIFY_DONE;
2114 }
2115
2116 static struct notifier_block kprobe_module_nb = {
2117         .notifier_call = kprobes_module_callback,
2118         .priority = 0
2119 };
2120
2121 /* Markers of _kprobe_blacklist section */
2122 extern unsigned long __start_kprobe_blacklist[];
2123 extern unsigned long __stop_kprobe_blacklist[];
2124
2125 static int __init init_kprobes(void)
2126 {
2127         int i, err = 0;
2128
2129         /* FIXME allocate the probe table, currently defined statically */
2130         /* initialize all list heads */
2131         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2132                 INIT_HLIST_HEAD(&kprobe_table[i]);
2133                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2134                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2135         }
2136
2137         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2138                                         __stop_kprobe_blacklist);
2139         if (err) {
2140                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2141                 pr_err("Please take care of using kprobes.\n");
2142         }
2143
2144         if (kretprobe_blacklist_size) {
2145                 /* lookup the function address from its name */
2146                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2147                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2148                                            kretprobe_blacklist[i].addr);
2149                         if (!kretprobe_blacklist[i].addr)
2150                                 printk("kretprobe: lookup failed: %s\n",
2151                                        kretprobe_blacklist[i].name);
2152                 }
2153         }
2154
2155 #if defined(CONFIG_OPTPROBES)
2156 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2157         /* Init kprobe_optinsn_slots */
2158         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2159 #endif
2160         /* By default, kprobes can be optimized */
2161         kprobes_allow_optimization = true;
2162 #endif
2163
2164         /* By default, kprobes are armed */
2165         kprobes_all_disarmed = false;
2166
2167         err = arch_init_kprobes();
2168         if (!err)
2169                 err = register_die_notifier(&kprobe_exceptions_nb);
2170         if (!err)
2171                 err = register_module_notifier(&kprobe_module_nb);
2172
2173         kprobes_initialized = (err == 0);
2174
2175         if (!err)
2176                 init_test_probes();
2177         return err;
2178 }
2179
2180 #ifdef CONFIG_DEBUG_FS
2181 static void report_probe(struct seq_file *pi, struct kprobe *p,
2182                 const char *sym, int offset, char *modname, struct kprobe *pp)
2183 {
2184         char *kprobe_type;
2185
2186         if (p->pre_handler == pre_handler_kretprobe)
2187                 kprobe_type = "r";
2188         else if (p->pre_handler == setjmp_pre_handler)
2189                 kprobe_type = "j";
2190         else
2191                 kprobe_type = "k";
2192
2193         if (sym)
2194                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2195                         p->addr, kprobe_type, sym, offset,
2196                         (modname ? modname : " "));
2197         else
2198                 seq_printf(pi, "%p  %s  %p ",
2199                         p->addr, kprobe_type, p->addr);
2200
2201         if (!pp)
2202                 pp = p;
2203         seq_printf(pi, "%s%s%s%s\n",
2204                 (kprobe_gone(p) ? "[GONE]" : ""),
2205                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2206                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2207                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2208 }
2209
2210 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2211 {
2212         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2213 }
2214
2215 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2216 {
2217         (*pos)++;
2218         if (*pos >= KPROBE_TABLE_SIZE)
2219                 return NULL;
2220         return pos;
2221 }
2222
2223 static void kprobe_seq_stop(struct seq_file *f, void *v)
2224 {
2225         /* Nothing to do */
2226 }
2227
2228 static int show_kprobe_addr(struct seq_file *pi, void *v)
2229 {
2230         struct hlist_head *head;
2231         struct kprobe *p, *kp;
2232         const char *sym = NULL;
2233         unsigned int i = *(loff_t *) v;
2234         unsigned long offset = 0;
2235         char *modname, namebuf[KSYM_NAME_LEN];
2236
2237         head = &kprobe_table[i];
2238         preempt_disable();
2239         hlist_for_each_entry_rcu(p, head, hlist) {
2240                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2241                                         &offset, &modname, namebuf);
2242                 if (kprobe_aggrprobe(p)) {
2243                         list_for_each_entry_rcu(kp, &p->list, list)
2244                                 report_probe(pi, kp, sym, offset, modname, p);
2245                 } else
2246                         report_probe(pi, p, sym, offset, modname, NULL);
2247         }
2248         preempt_enable();
2249         return 0;
2250 }
2251
2252 static const struct seq_operations kprobes_seq_ops = {
2253         .start = kprobe_seq_start,
2254         .next  = kprobe_seq_next,
2255         .stop  = kprobe_seq_stop,
2256         .show  = show_kprobe_addr
2257 };
2258
2259 static int kprobes_open(struct inode *inode, struct file *filp)
2260 {
2261         return seq_open(filp, &kprobes_seq_ops);
2262 }
2263
2264 static const struct file_operations debugfs_kprobes_operations = {
2265         .open           = kprobes_open,
2266         .read           = seq_read,
2267         .llseek         = seq_lseek,
2268         .release        = seq_release,
2269 };
2270
2271 /* kprobes/blacklist -- shows which functions can not be probed */
2272 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2273 {
2274         return seq_list_start(&kprobe_blacklist, *pos);
2275 }
2276
2277 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2278 {
2279         return seq_list_next(v, &kprobe_blacklist, pos);
2280 }
2281
2282 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2283 {
2284         struct kprobe_blacklist_entry *ent =
2285                 list_entry(v, struct kprobe_blacklist_entry, list);
2286
2287         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2288                    (void *)ent->end_addr, (void *)ent->start_addr);
2289         return 0;
2290 }
2291
2292 static const struct seq_operations kprobe_blacklist_seq_ops = {
2293         .start = kprobe_blacklist_seq_start,
2294         .next  = kprobe_blacklist_seq_next,
2295         .stop  = kprobe_seq_stop,       /* Reuse void function */
2296         .show  = kprobe_blacklist_seq_show,
2297 };
2298
2299 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2300 {
2301         return seq_open(filp, &kprobe_blacklist_seq_ops);
2302 }
2303
2304 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2305         .open           = kprobe_blacklist_open,
2306         .read           = seq_read,
2307         .llseek         = seq_lseek,
2308         .release        = seq_release,
2309 };
2310
2311 static void arm_all_kprobes(void)
2312 {
2313         struct hlist_head *head;
2314         struct kprobe *p;
2315         unsigned int i;
2316
2317         mutex_lock(&kprobe_mutex);
2318
2319         /* If kprobes are armed, just return */
2320         if (!kprobes_all_disarmed)
2321                 goto already_enabled;
2322
2323         /* Arming kprobes doesn't optimize kprobe itself */
2324         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2325                 head = &kprobe_table[i];
2326                 hlist_for_each_entry_rcu(p, head, hlist)
2327                         if (!kprobe_disabled(p))
2328                                 arm_kprobe(p);
2329         }
2330
2331         kprobes_all_disarmed = false;
2332         printk(KERN_INFO "Kprobes globally enabled\n");
2333
2334 already_enabled:
2335         mutex_unlock(&kprobe_mutex);
2336         return;
2337 }
2338
2339 static void disarm_all_kprobes(void)
2340 {
2341         struct hlist_head *head;
2342         struct kprobe *p;
2343         unsigned int i;
2344
2345         mutex_lock(&kprobe_mutex);
2346
2347         /* If kprobes are already disarmed, just return */
2348         if (kprobes_all_disarmed) {
2349                 mutex_unlock(&kprobe_mutex);
2350                 return;
2351         }
2352
2353         kprobes_all_disarmed = true;
2354         printk(KERN_INFO "Kprobes globally disabled\n");
2355
2356         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2357                 head = &kprobe_table[i];
2358                 hlist_for_each_entry_rcu(p, head, hlist) {
2359                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2360                                 disarm_kprobe(p, false);
2361                 }
2362         }
2363         mutex_unlock(&kprobe_mutex);
2364
2365         /* Wait for disarming all kprobes by optimizer */
2366         wait_for_kprobe_optimizer();
2367 }
2368
2369 /*
2370  * XXX: The debugfs bool file interface doesn't allow for callbacks
2371  * when the bool state is switched. We can reuse that facility when
2372  * available
2373  */
2374 static ssize_t read_enabled_file_bool(struct file *file,
2375                char __user *user_buf, size_t count, loff_t *ppos)
2376 {
2377         char buf[3];
2378
2379         if (!kprobes_all_disarmed)
2380                 buf[0] = '1';
2381         else
2382                 buf[0] = '0';
2383         buf[1] = '\n';
2384         buf[2] = 0x00;
2385         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2386 }
2387
2388 static ssize_t write_enabled_file_bool(struct file *file,
2389                const char __user *user_buf, size_t count, loff_t *ppos)
2390 {
2391         char buf[32];
2392         size_t buf_size;
2393
2394         buf_size = min(count, (sizeof(buf)-1));
2395         if (copy_from_user(buf, user_buf, buf_size))
2396                 return -EFAULT;
2397
2398         buf[buf_size] = '\0';
2399         switch (buf[0]) {
2400         case 'y':
2401         case 'Y':
2402         case '1':
2403                 arm_all_kprobes();
2404                 break;
2405         case 'n':
2406         case 'N':
2407         case '0':
2408                 disarm_all_kprobes();
2409                 break;
2410         default:
2411                 return -EINVAL;
2412         }
2413
2414         return count;
2415 }
2416
2417 static const struct file_operations fops_kp = {
2418         .read =         read_enabled_file_bool,
2419         .write =        write_enabled_file_bool,
2420         .llseek =       default_llseek,
2421 };
2422
2423 static int __init debugfs_kprobe_init(void)
2424 {
2425         struct dentry *dir, *file;
2426         unsigned int value = 1;
2427
2428         dir = debugfs_create_dir("kprobes", NULL);
2429         if (!dir)
2430                 return -ENOMEM;
2431
2432         file = debugfs_create_file("list", 0444, dir, NULL,
2433                                 &debugfs_kprobes_operations);
2434         if (!file)
2435                 goto error;
2436
2437         file = debugfs_create_file("enabled", 0600, dir,
2438                                         &value, &fops_kp);
2439         if (!file)
2440                 goto error;
2441
2442         file = debugfs_create_file("blacklist", 0444, dir, NULL,
2443                                 &debugfs_kprobe_blacklist_ops);
2444         if (!file)
2445                 goto error;
2446
2447         return 0;
2448
2449 error:
2450         debugfs_remove(dir);
2451         return -ENOMEM;
2452 }
2453
2454 late_initcall(debugfs_kprobe_init);
2455 #endif /* CONFIG_DEBUG_FS */
2456
2457 module_init(init_kprobes);
2458
2459 /* defined in arch/.../kernel/kprobes.c */
2460 EXPORT_SYMBOL_GPL(jprobe_return);