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 void __weak 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 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 int reuse_unused_kprobe(struct kprobe *ap)
 669 {
 670         struct optimized_kprobe *op;
 671         int ret;
 672
 673         BUG_ON(!kprobe_unused(ap));
 674         /*
 675          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 676          * there is still a relative jump) and disabled.
 677          */
 678         op = container_of(ap, struct optimized_kprobe, kp);
 679         if (unlikely(list_empty(&op->list)))
 680                 printk(KERN_WARNING "Warning: found a stray unused "
 681                         "aggrprobe@%p\n", ap->addr);
 682         /* Enable the probe again */
 683         ap->flags &= ~KPROBE_FLAG_DISABLED;
 684         /* Optimize it again (remove from op->list) */
 685         ret = kprobe_optready(ap);
 686         if (ret)
 687                 return ret;
 688
 689         optimize_kprobe(ap);
 690         return 0;
 691 }
 692
 693 /* Remove optimized instructions */
 694 static void kill_optimized_kprobe(struct kprobe *p)
 695 {
 696         struct optimized_kprobe *op;
 697
 698         op = container_of(p, struct optimized_kprobe, kp);
 699         if (!list_empty(&op->list))
 700                 /* Dequeue from the (un)optimization queue */
 701                 list_del_init(&op->list);
 702         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 703
 704         if (kprobe_unused(p)) {
 705                 /* Enqueue if it is unused */
 706                 list_add(&op->list, &freeing_list);
 707                 /*
 708                  * Remove unused probes from the hash list. After waiting
 709                  * for synchronization, this probe is reclaimed.
 710                  * (reclaiming is done by do_free_cleaned_kprobes().)
 711                  */
 712                 hlist_del_rcu(&op->kp.hlist);
 713         }
 714
 715         /* Don't touch the code, because it is already freed. */
 716         arch_remove_optimized_kprobe(op);
 717 }
 718
 719 /* Try to prepare optimized instructions */
 720 static void prepare_optimized_kprobe(struct kprobe *p)
 721 {
 722         struct optimized_kprobe *op;
 723
 724         op = container_of(p, struct optimized_kprobe, kp);
 725         arch_prepare_optimized_kprobe(op, p);
 726 }
 727
 728 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 729 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 730 {
 731         struct optimized_kprobe *op;
 732
 733         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 734         if (!op)
 735                 return NULL;
 736
 737         INIT_LIST_HEAD(&op->list);
 738         op->kp.addr = p->addr;
 739         arch_prepare_optimized_kprobe(op, p);
 740
 741         return &op->kp;
 742 }
 743
 744 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 745
 746 /*
 747  * Prepare an optimized_kprobe and optimize it
 748  * NOTE: p must be a normal registered kprobe
 749  */
 750 static void try_to_optimize_kprobe(struct kprobe *p)
 751 {
 752         struct kprobe *ap;
 753         struct optimized_kprobe *op;
 754
 755         /* Impossible to optimize ftrace-based kprobe */
 756         if (kprobe_ftrace(p))
 757                 return;
 758
 759         /* For preparing optimization, jump_label_text_reserved() is called */
 760         jump_label_lock();
 761         mutex_lock(&text_mutex);
 762
 763         ap = alloc_aggr_kprobe(p);
 764         if (!ap)
 765                 goto out;
 766
 767         op = container_of(ap, struct optimized_kprobe, kp);
 768         if (!arch_prepared_optinsn(&op->optinsn)) {
 769                 /* If failed to setup optimizing, fallback to kprobe */
 770                 arch_remove_optimized_kprobe(op);
 771                 kfree(op);
 772                 goto out;
 773         }
 774
 775         init_aggr_kprobe(ap, p);
 776         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 777
 778 out:
 779         mutex_unlock(&text_mutex);
 780         jump_label_unlock();
 781 }
 782
 783 #ifdef CONFIG_SYSCTL
 784 static void optimize_all_kprobes(void)
 785 {
 786         struct hlist_head *head;
 787         struct kprobe *p;
 788         unsigned int i;
 789
 790         mutex_lock(&kprobe_mutex);
 791         /* If optimization is already allowed, just return */
 792         if (kprobes_allow_optimization)
 793                 goto out;
 794
 795         kprobes_allow_optimization = true;
 796         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 797                 head = &kprobe_table[i];
 798                 hlist_for_each_entry_rcu(p, head, hlist)
 799                         if (!kprobe_disabled(p))
 800                                 optimize_kprobe(p);
 801         }
 802         printk(KERN_INFO "Kprobes globally optimized\n");
 803 out:
 804         mutex_unlock(&kprobe_mutex);
 805 }
 806
 807 static void unoptimize_all_kprobes(void)
 808 {
 809         struct hlist_head *head;
 810         struct kprobe *p;
 811         unsigned int i;
 812
 813         mutex_lock(&kprobe_mutex);
 814         /* If optimization is already prohibited, just return */
 815         if (!kprobes_allow_optimization) {
 816                 mutex_unlock(&kprobe_mutex);
 817                 return;
 818         }
 819
 820         kprobes_allow_optimization = false;
 821         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 822                 head = &kprobe_table[i];
 823                 hlist_for_each_entry_rcu(p, head, hlist) {
 824                         if (!kprobe_disabled(p))
 825                                 unoptimize_kprobe(p, false);
 826                 }
 827         }
 828         mutex_unlock(&kprobe_mutex);
 829
 830         /* Wait for unoptimizing completion */
 831         wait_for_kprobe_optimizer();
 832         printk(KERN_INFO "Kprobes globally unoptimized\n");
 833 }
 834
 835 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 836 int sysctl_kprobes_optimization;
 837 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 838                                       void __user *buffer, size_t *length,
 839                                       loff_t *ppos)
 840 {
 841         int ret;
 842
 843         mutex_lock(&kprobe_sysctl_mutex);
 844         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 845         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 846
 847         if (sysctl_kprobes_optimization)
 848                 optimize_all_kprobes();
 849         else
 850                 unoptimize_all_kprobes();
 851         mutex_unlock(&kprobe_sysctl_mutex);
 852
 853         return ret;
 854 }
 855 #endif /* CONFIG_SYSCTL */
 856
 857 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 858 static void __arm_kprobe(struct kprobe *p)
 859 {
 860         struct kprobe *_p;
 861
 862         /* Check collision with other optimized kprobes */
 863         _p = get_optimized_kprobe((unsigned long)p->addr);
 864         if (unlikely(_p))
 865                 /* Fallback to unoptimized kprobe */
 866                 unoptimize_kprobe(_p, true);
 867
 868         arch_arm_kprobe(p);
 869         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 870 }
 871
 872 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 873 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 874 {
 875         struct kprobe *_p;
 876
 877         /* Try to unoptimize */
 878         unoptimize_kprobe(p, kprobes_all_disarmed);
 879
 880         if (!kprobe_queued(p)) {
 881                 arch_disarm_kprobe(p);
 882                 /* If another kprobe was blocked, optimize it. */
 883                 _p = get_optimized_kprobe((unsigned long)p->addr);
 884                 if (unlikely(_p) && reopt)
 885                         optimize_kprobe(_p);
 886         }
 887         /* TODO: reoptimize others after unoptimized this probe */
 888 }
 889
 890 #else /* !CONFIG_OPTPROBES */
 891
 892 #define optimize_kprobe(p)                      do {} while (0)
 893 #define unoptimize_kprobe(p, f)                 do {} while (0)
 894 #define kill_optimized_kprobe(p)                do {} while (0)
 895 #define prepare_optimized_kprobe(p)             do {} while (0)
 896 #define try_to_optimize_kprobe(p)               do {} while (0)
 897 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 898 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 899 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 900 #define wait_for_kprobe_optimizer()             do {} while (0)
 901
 902 static int reuse_unused_kprobe(struct kprobe *ap)
 903 {
 904         /*
 905          * If the optimized kprobe is NOT supported, the aggr kprobe is
 906          * released at the same time that the last aggregated kprobe is
 907          * unregistered.
 908          * Thus there should be no chance to reuse unused kprobe.
 909          */
 910         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 911         return -EINVAL;
 912 }
 913
 914 static void free_aggr_kprobe(struct kprobe *p)
 915 {
 916         arch_remove_kprobe(p);
 917         kfree(p);
 918 }
 919
 920 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 921 {
 922         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 923 }
 924 #endif /* CONFIG_OPTPROBES */
 925
 926 #ifdef CONFIG_KPROBES_ON_FTRACE
 927 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 928         .func = kprobe_ftrace_handler,
 929         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 930 };
 931 static int kprobe_ftrace_enabled;
 932
 933 /* Must ensure p->addr is really on ftrace */
 934 static int prepare_kprobe(struct kprobe *p)
 935 {
 936         if (!kprobe_ftrace(p))
 937                 return arch_prepare_kprobe(p);
 938
 939         return arch_prepare_kprobe_ftrace(p);
 940 }
 941
 942 /* Caller must lock kprobe_mutex */
 943 static void arm_kprobe_ftrace(struct kprobe *p)
 944 {
 945         int ret;
 946
 947         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 948                                    (unsigned long)p->addr, 0, 0);
 949         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 950         kprobe_ftrace_enabled++;
 951         if (kprobe_ftrace_enabled == 1) {
 952                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 953                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 954         }
 955 }
 956
 957 /* Caller must lock kprobe_mutex */
 958 static void disarm_kprobe_ftrace(struct kprobe *p)
 959 {
 960         int ret;
 961
 962         kprobe_ftrace_enabled--;
 963         if (kprobe_ftrace_enabled == 0) {
 964                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 965                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 966         }
 967         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 968                            (unsigned long)p->addr, 1, 0);
 969         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 970 }
 971 #else   /* !CONFIG_KPROBES_ON_FTRACE */
 972 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 973 #define arm_kprobe_ftrace(p)    do {} while (0)
 974 #define disarm_kprobe_ftrace(p) do {} while (0)
 975 #endif
 976
 977 /* Arm a kprobe with text_mutex */
 978 static void arm_kprobe(struct kprobe *kp)
 979 {
 980         if (unlikely(kprobe_ftrace(kp))) {
 981                 arm_kprobe_ftrace(kp);
 982                 return;
 983         }
 984         /*
 985          * Here, since __arm_kprobe() doesn't use stop_machine(),
 986          * this doesn't cause deadlock on text_mutex. So, we don't
 987          * need get_online_cpus().
 988          */
 989         mutex_lock(&text_mutex);
 990         __arm_kprobe(kp);
 991         mutex_unlock(&text_mutex);
 992 }
 993
 994 /* Disarm a kprobe with text_mutex */
 995 static void disarm_kprobe(struct kprobe *kp, bool reopt)
 996 {
 997         if (unlikely(kprobe_ftrace(kp))) {
 998                 disarm_kprobe_ftrace(kp);
 999                 return;
1000         }
1001         /* Ditto */
1002         mutex_lock(&text_mutex);
1003         __disarm_kprobe(kp, reopt);
1004         mutex_unlock(&text_mutex);
1005 }
1006
1007 /*
1008  * Aggregate handlers for multiple kprobes support - these handlers
1009  * take care of invoking the individual kprobe handlers on p->list
1010  */
1011 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1012 {
1013         struct kprobe *kp;
1014
1015         list_for_each_entry_rcu(kp, &p->list, list) {
1016                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1017                         set_kprobe_instance(kp);
1018                         if (kp->pre_handler(kp, regs))
1019                                 return 1;
1020                 }
1021                 reset_kprobe_instance();
1022         }
1023         return 0;
1024 }
1025 NOKPROBE_SYMBOL(aggr_pre_handler);
1026
1027 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1028                               unsigned long flags)
1029 {
1030         struct kprobe *kp;
1031
1032         list_for_each_entry_rcu(kp, &p->list, list) {
1033                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1034                         set_kprobe_instance(kp);
1035                         kp->post_handler(kp, regs, flags);
1036                         reset_kprobe_instance();
1037                 }
1038         }
1039 }
1040 NOKPROBE_SYMBOL(aggr_post_handler);
1041
1042 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1043                               int trapnr)
1044 {
1045         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1046
1047         /*
1048          * if we faulted "during" the execution of a user specified
1049          * probe handler, invoke just that probe's fault handler
1050          */
1051         if (cur && cur->fault_handler) {
1052                 if (cur->fault_handler(cur, regs, trapnr))
1053                         return 1;
1054         }
1055         return 0;
1056 }
1057 NOKPROBE_SYMBOL(aggr_fault_handler);
1058
1059 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1060 {
1061         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1062         int ret = 0;
1063
1064         if (cur && cur->break_handler) {
1065                 if (cur->break_handler(cur, regs))
1066                         ret = 1;
1067         }
1068         reset_kprobe_instance();
1069         return ret;
1070 }
1071 NOKPROBE_SYMBOL(aggr_break_handler);
1072
1073 /* Walks the list and increments nmissed count for multiprobe case */
1074 void kprobes_inc_nmissed_count(struct kprobe *p)
1075 {
1076         struct kprobe *kp;
1077         if (!kprobe_aggrprobe(p)) {
1078                 p->nmissed++;
1079         } else {
1080                 list_for_each_entry_rcu(kp, &p->list, list)
1081                         kp->nmissed++;
1082         }
1083         return;
1084 }
1085 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1086
1087 void recycle_rp_inst(struct kretprobe_instance *ri,
1088                      struct hlist_head *head)
1089 {
1090         struct kretprobe *rp = ri->rp;
1091
1092         /* remove rp inst off the rprobe_inst_table */
1093         hlist_del(&ri->hlist);
1094         INIT_HLIST_NODE(&ri->hlist);
1095         if (likely(rp)) {
1096                 raw_spin_lock(&rp->lock);
1097                 hlist_add_head(&ri->hlist, &rp->free_instances);
1098                 raw_spin_unlock(&rp->lock);
1099         } else
1100                 /* Unregistering */
1101                 hlist_add_head(&ri->hlist, head);
1102 }
1103 NOKPROBE_SYMBOL(recycle_rp_inst);
1104
1105 void kretprobe_hash_lock(struct task_struct *tsk,
1106                          struct hlist_head **head, unsigned long *flags)
1107 __acquires(hlist_lock)
1108 {
1109         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1110         raw_spinlock_t *hlist_lock;
1111
1112         *head = &kretprobe_inst_table[hash];
1113         hlist_lock = kretprobe_table_lock_ptr(hash);
1114         raw_spin_lock_irqsave(hlist_lock, *flags);
1115 }
1116 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1117
1118 static void kretprobe_table_lock(unsigned long hash,
1119                                  unsigned long *flags)
1120 __acquires(hlist_lock)
1121 {
1122         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1123         raw_spin_lock_irqsave(hlist_lock, *flags);
1124 }
1125 NOKPROBE_SYMBOL(kretprobe_table_lock);
1126
1127 void kretprobe_hash_unlock(struct task_struct *tsk,
1128                            unsigned long *flags)
1129 __releases(hlist_lock)
1130 {
1131         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1132         raw_spinlock_t *hlist_lock;
1133
1134         hlist_lock = kretprobe_table_lock_ptr(hash);
1135         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1136 }
1137 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1138
1139 static void kretprobe_table_unlock(unsigned long hash,
1140                                    unsigned long *flags)
1141 __releases(hlist_lock)
1142 {
1143         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1144         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1145 }
1146 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1147
1148 /*
1149  * This function is called from finish_task_switch when task tk becomes dead,
1150  * so that we can recycle any function-return probe instances associated
1151  * with this task. These left over instances represent probed functions
1152  * that have been called but will never return.
1153  */
1154 void kprobe_flush_task(struct task_struct *tk)
1155 {
1156         struct kretprobe_instance *ri;
1157         struct hlist_head *head, empty_rp;
1158         struct hlist_node *tmp;
1159         unsigned long hash, flags = 0;
1160
1161         if (unlikely(!kprobes_initialized))
1162                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1163                 return;
1164
1165         INIT_HLIST_HEAD(&empty_rp);
1166         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1167         head = &kretprobe_inst_table[hash];
1168         kretprobe_table_lock(hash, &flags);
1169         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1170                 if (ri->task == tk)
1171                         recycle_rp_inst(ri, &empty_rp);
1172         }
1173         kretprobe_table_unlock(hash, &flags);
1174         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1175                 hlist_del(&ri->hlist);
1176                 kfree(ri);
1177         }
1178 }
1179 NOKPROBE_SYMBOL(kprobe_flush_task);
1180
1181 static inline void free_rp_inst(struct kretprobe *rp)
1182 {
1183         struct kretprobe_instance *ri;
1184         struct hlist_node *next;
1185
1186         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1187                 hlist_del(&ri->hlist);
1188                 kfree(ri);
1189         }
1190 }
1191
1192 static void cleanup_rp_inst(struct kretprobe *rp)
1193 {
1194         unsigned long flags, hash;
1195         struct kretprobe_instance *ri;
1196         struct hlist_node *next;
1197         struct hlist_head *head;
1198
1199         /* No race here */
1200         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1201                 kretprobe_table_lock(hash, &flags);
1202                 head = &kretprobe_inst_table[hash];
1203                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1204                         if (ri->rp == rp)
1205                                 ri->rp = NULL;
1206                 }
1207                 kretprobe_table_unlock(hash, &flags);
1208         }
1209         free_rp_inst(rp);
1210 }
1211 NOKPROBE_SYMBOL(cleanup_rp_inst);
1212
1213 /*
1214 * Add the new probe to ap->list. Fail if this is the
1215 * second jprobe at the address - two jprobes can't coexist
1216 */
1217 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1218 {
1219         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1220
1221         if (p->break_handler || p->post_handler)
1222                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1223
1224         if (p->break_handler) {
1225                 if (ap->break_handler)
1226                         return -EEXIST;
1227                 list_add_tail_rcu(&p->list, &ap->list);
1228                 ap->break_handler = aggr_break_handler;
1229         } else
1230                 list_add_rcu(&p->list, &ap->list);
1231         if (p->post_handler && !ap->post_handler)
1232                 ap->post_handler = aggr_post_handler;
1233
1234         return 0;
1235 }
1236
1237 /*
1238  * Fill in the required fields of the "manager kprobe". Replace the
1239  * earlier kprobe in the hlist with the manager kprobe
1240  */
1241 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1242 {
1243         /* Copy p's insn slot to ap */
1244         copy_kprobe(p, ap);
1245         flush_insn_slot(ap);
1246         ap->addr = p->addr;
1247         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1248         ap->pre_handler = aggr_pre_handler;
1249         ap->fault_handler = aggr_fault_handler;
1250         /* We don't care the kprobe which has gone. */
1251         if (p->post_handler && !kprobe_gone(p))
1252                 ap->post_handler = aggr_post_handler;
1253         if (p->break_handler && !kprobe_gone(p))
1254                 ap->break_handler = aggr_break_handler;
1255
1256         INIT_LIST_HEAD(&ap->list);
1257         INIT_HLIST_NODE(&ap->hlist);
1258
1259         list_add_rcu(&p->list, &ap->list);
1260         hlist_replace_rcu(&p->hlist, &ap->hlist);
1261 }
1262
1263 /*
1264  * This is the second or subsequent kprobe at the address - handle
1265  * the intricacies
1266  */
1267 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1268 {
1269         int ret = 0;
1270         struct kprobe *ap = orig_p;
1271
1272         /* For preparing optimization, jump_label_text_reserved() is called */
1273         jump_label_lock();
1274         /*
1275          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1276          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1277          */
1278         get_online_cpus();
1279         mutex_lock(&text_mutex);
1280
1281         if (!kprobe_aggrprobe(orig_p)) {
1282                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1283                 ap = alloc_aggr_kprobe(orig_p);
1284                 if (!ap) {
1285                         ret = -ENOMEM;
1286                         goto out;
1287                 }
1288                 init_aggr_kprobe(ap, orig_p);
1289         } else if (kprobe_unused(ap)) {
1290                 /* This probe is going to die. Rescue it */
1291                 ret = reuse_unused_kprobe(ap);
1292                 if (ret)
1293                         goto out;
1294         }
1295
1296         if (kprobe_gone(ap)) {
1297                 /*
1298                  * Attempting to insert new probe at the same location that
1299                  * had a probe in the module vaddr area which already
1300                  * freed. So, the instruction slot has already been
1301                  * released. We need a new slot for the new probe.
1302                  */
1303                 ret = arch_prepare_kprobe(ap);
1304                 if (ret)
1305                         /*
1306                          * Even if fail to allocate new slot, don't need to
1307                          * free aggr_probe. It will be used next time, or
1308                          * freed by unregister_kprobe.
1309                          */
1310                         goto out;
1311
1312                 /* Prepare optimized instructions if possible. */
1313                 prepare_optimized_kprobe(ap);
1314
1315                 /*
1316                  * Clear gone flag to prevent allocating new slot again, and
1317                  * set disabled flag because it is not armed yet.
1318                  */
1319                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1320                             | KPROBE_FLAG_DISABLED;
1321         }
1322
1323         /* Copy ap's insn slot to p */
1324         copy_kprobe(ap, p);
1325         ret = add_new_kprobe(ap, p);
1326
1327 out:
1328         mutex_unlock(&text_mutex);
1329         put_online_cpus();
1330         jump_label_unlock();
1331
1332         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1333                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1334                 if (!kprobes_all_disarmed)
1335                         /* Arm the breakpoint again. */
1336                         arm_kprobe(ap);
1337         }
1338         return ret;
1339 }
1340
1341 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1342 {
1343         /* The __kprobes marked functions and entry code must not be probed */
1344         return addr >= (unsigned long)__kprobes_text_start &&
1345                addr < (unsigned long)__kprobes_text_end;
1346 }
1347
1348 bool within_kprobe_blacklist(unsigned long addr)
1349 {
1350         struct kprobe_blacklist_entry *ent;
1351
1352         if (arch_within_kprobe_blacklist(addr))
1353                 return true;
1354         /*
1355          * If there exists a kprobe_blacklist, verify and
1356          * fail any probe registration in the prohibited area
1357          */
1358         list_for_each_entry(ent, &kprobe_blacklist, list) {
1359                 if (addr >= ent->start_addr && addr < ent->end_addr)
1360                         return true;
1361         }
1362
1363         return false;
1364 }
1365
1366 /*
1367  * If we have a symbol_name argument, look it up and add the offset field
1368  * to it. This way, we can specify a relative address to a symbol.
1369  * This returns encoded errors if it fails to look up symbol or invalid
1370  * combination of parameters.
1371  */
1372 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1373 {
1374         kprobe_opcode_t *addr = p->addr;
1375
1376         if ((p->symbol_name && p->addr) ||
1377             (!p->symbol_name && !p->addr))
1378                 goto invalid;
1379
1380         if (p->symbol_name) {
1381                 kprobe_lookup_name(p->symbol_name, addr);
1382                 if (!addr)
1383                         return ERR_PTR(-ENOENT);
1384         }
1385
1386         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1387         if (addr)
1388                 return addr;
1389
1390 invalid:
1391         return ERR_PTR(-EINVAL);
1392 }
1393
1394 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1395 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1396 {
1397         struct kprobe *ap, *list_p;
1398
1399         ap = get_kprobe(p->addr);
1400         if (unlikely(!ap))
1401                 return NULL;
1402
1403         if (p != ap) {
1404                 list_for_each_entry_rcu(list_p, &ap->list, list)
1405                         if (list_p == p)
1406                         /* kprobe p is a valid probe */
1407                                 goto valid;
1408                 return NULL;
1409         }
1410 valid:
1411         return ap;
1412 }
1413
1414 /* Return error if the kprobe is being re-registered */
1415 static inline int check_kprobe_rereg(struct kprobe *p)
1416 {
1417         int ret = 0;
1418
1419         mutex_lock(&kprobe_mutex);
1420         if (__get_valid_kprobe(p))
1421                 ret = -EINVAL;
1422         mutex_unlock(&kprobe_mutex);
1423
1424         return ret;
1425 }
1426
1427 int __weak arch_check_ftrace_location(struct kprobe *p)
1428 {
1429         unsigned long ftrace_addr;
1430
1431         ftrace_addr = ftrace_location((unsigned long)p->addr);
1432         if (ftrace_addr) {
1433 #ifdef CONFIG_KPROBES_ON_FTRACE
1434                 /* Given address is not on the instruction boundary */
1435                 if ((unsigned long)p->addr != ftrace_addr)
1436                         return -EILSEQ;
1437                 p->flags |= KPROBE_FLAG_FTRACE;
1438 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1439                 return -EINVAL;
1440 #endif
1441         }
1442         return 0;
1443 }
1444
1445 static int check_kprobe_address_safe(struct kprobe *p,
1446                                      struct module **probed_mod)
1447 {
1448         int ret;
1449
1450         ret = arch_check_ftrace_location(p);
1451         if (ret)
1452                 return ret;
1453         jump_label_lock();
1454         preempt_disable();
1455
1456         /* Ensure it is not in reserved area nor out of text */
1457         if (!kernel_text_address((unsigned long) p->addr) ||
1458             within_kprobe_blacklist((unsigned long) p->addr) ||
1459             jump_label_text_reserved(p->addr, p->addr)) {
1460                 ret = -EINVAL;
1461                 goto out;
1462         }
1463
1464         /* Check if are we probing a module */
1465         *probed_mod = __module_text_address((unsigned long) p->addr);
1466         if (*probed_mod) {
1467                 /*
1468                  * We must hold a refcount of the probed module while updating
1469                  * its code to prohibit unexpected unloading.
1470                  */
1471                 if (unlikely(!try_module_get(*probed_mod))) {
1472                         ret = -ENOENT;
1473                         goto out;
1474                 }
1475
1476                 /*
1477                  * If the module freed .init.text, we couldn't insert
1478                  * kprobes in there.
1479                  */
1480                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1481                     (*probed_mod)->state != MODULE_STATE_COMING) {
1482                         module_put(*probed_mod);
1483                         *probed_mod = NULL;
1484                         ret = -ENOENT;
1485                 }
1486         }
1487 out:
1488         preempt_enable();
1489         jump_label_unlock();
1490
1491         return ret;
1492 }
1493
1494 int register_kprobe(struct kprobe *p)
1495 {
1496         int ret;
1497         struct kprobe *old_p;
1498         struct module *probed_mod;
1499         kprobe_opcode_t *addr;
1500
1501         /* Adjust probe address from symbol */
1502         addr = kprobe_addr(p);
1503         if (IS_ERR(addr))
1504                 return PTR_ERR(addr);
1505         p->addr = addr;
1506
1507         ret = check_kprobe_rereg(p);
1508         if (ret)
1509                 return ret;
1510
1511         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1512         p->flags &= KPROBE_FLAG_DISABLED;
1513         p->nmissed = 0;
1514         INIT_LIST_HEAD(&p->list);
1515
1516         ret = check_kprobe_address_safe(p, &probed_mod);
1517         if (ret)
1518                 return ret;
1519
1520         mutex_lock(&kprobe_mutex);
1521
1522         old_p = get_kprobe(p->addr);
1523         if (old_p) {
1524                 /* Since this may unoptimize old_p, locking text_mutex. */
1525                 ret = register_aggr_kprobe(old_p, p);
1526                 goto out;
1527         }
1528
1529         mutex_lock(&text_mutex);        /* Avoiding text modification */
1530         ret = prepare_kprobe(p);
1531         mutex_unlock(&text_mutex);
1532         if (ret)
1533                 goto out;
1534
1535         INIT_HLIST_NODE(&p->hlist);
1536         hlist_add_head_rcu(&p->hlist,
1537                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1538
1539         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1540                 arm_kprobe(p);
1541
1542         /* Try to optimize kprobe */
1543         try_to_optimize_kprobe(p);
1544
1545 out:
1546         mutex_unlock(&kprobe_mutex);
1547
1548         if (probed_mod)
1549                 module_put(probed_mod);
1550
1551         return ret;
1552 }
1553 EXPORT_SYMBOL_GPL(register_kprobe);
1554
1555 /* Check if all probes on the aggrprobe are disabled */
1556 static int aggr_kprobe_disabled(struct kprobe *ap)
1557 {
1558         struct kprobe *kp;
1559
1560         list_for_each_entry_rcu(kp, &ap->list, list)
1561                 if (!kprobe_disabled(kp))
1562                         /*
1563                          * There is an active probe on the list.
1564                          * We can't disable this ap.
1565                          */
1566                         return 0;
1567
1568         return 1;
1569 }
1570
1571 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1572 static struct kprobe *__disable_kprobe(struct kprobe *p)
1573 {
1574         struct kprobe *orig_p;
1575
1576         /* Get an original kprobe for return */
1577         orig_p = __get_valid_kprobe(p);
1578         if (unlikely(orig_p == NULL))
1579                 return NULL;
1580
1581         if (!kprobe_disabled(p)) {
1582                 /* Disable probe if it is a child probe */
1583                 if (p != orig_p)
1584                         p->flags |= KPROBE_FLAG_DISABLED;
1585
1586                 /* Try to disarm and disable this/parent probe */
1587                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1588                         /*
1589                          * If kprobes_all_disarmed is set, orig_p
1590                          * should have already been disarmed, so
1591                          * skip unneed disarming process.
1592                          */
1593                         if (!kprobes_all_disarmed)
1594                                 disarm_kprobe(orig_p, true);
1595                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1596                 }
1597         }
1598
1599         return orig_p;
1600 }
1601
1602 /*
1603  * Unregister a kprobe without a scheduler synchronization.
1604  */
1605 static int __unregister_kprobe_top(struct kprobe *p)
1606 {
1607         struct kprobe *ap, *list_p;
1608
1609         /* Disable kprobe. This will disarm it if needed. */
1610         ap = __disable_kprobe(p);
1611         if (ap == NULL)
1612                 return -EINVAL;
1613
1614         if (ap == p)
1615                 /*
1616                  * This probe is an independent(and non-optimized) kprobe
1617                  * (not an aggrprobe). Remove from the hash list.
1618                  */
1619                 goto disarmed;
1620
1621         /* Following process expects this probe is an aggrprobe */
1622         WARN_ON(!kprobe_aggrprobe(ap));
1623
1624         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1625                 /*
1626                  * !disarmed could be happen if the probe is under delayed
1627                  * unoptimizing.
1628                  */
1629                 goto disarmed;
1630         else {
1631                 /* If disabling probe has special handlers, update aggrprobe */
1632                 if (p->break_handler && !kprobe_gone(p))
1633                         ap->break_handler = NULL;
1634                 if (p->post_handler && !kprobe_gone(p)) {
1635                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1636                                 if ((list_p != p) && (list_p->post_handler))
1637                                         goto noclean;
1638                         }
1639                         ap->post_handler = NULL;
1640                 }
1641 noclean:
1642                 /*
1643                  * Remove from the aggrprobe: this path will do nothing in
1644                  * __unregister_kprobe_bottom().
1645                  */
1646                 list_del_rcu(&p->list);
1647                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1648                         /*
1649                          * Try to optimize this probe again, because post
1650                          * handler may have been changed.
1651                          */
1652                         optimize_kprobe(ap);
1653         }
1654         return 0;
1655
1656 disarmed:
1657         BUG_ON(!kprobe_disarmed(ap));
1658         hlist_del_rcu(&ap->hlist);
1659         return 0;
1660 }
1661
1662 static void __unregister_kprobe_bottom(struct kprobe *p)
1663 {
1664         struct kprobe *ap;
1665
1666         if (list_empty(&p->list))
1667                 /* This is an independent kprobe */
1668                 arch_remove_kprobe(p);
1669         else if (list_is_singular(&p->list)) {
1670                 /* This is the last child of an aggrprobe */
1671                 ap = list_entry(p->list.next, struct kprobe, list);
1672                 list_del(&p->list);
1673                 free_aggr_kprobe(ap);
1674         }
1675         /* Otherwise, do nothing. */
1676 }
1677
1678 int register_kprobes(struct kprobe **kps, int num)
1679 {
1680         int i, ret = 0;
1681
1682         if (num <= 0)
1683                 return -EINVAL;
1684         for (i = 0; i < num; i++) {
1685                 ret = register_kprobe(kps[i]);
1686                 if (ret < 0) {
1687                         if (i > 0)
1688                                 unregister_kprobes(kps, i);
1689                         break;
1690                 }
1691         }
1692         return ret;
1693 }
1694 EXPORT_SYMBOL_GPL(register_kprobes);
1695
1696 void unregister_kprobe(struct kprobe *p)
1697 {
1698         unregister_kprobes(&p, 1);
1699 }
1700 EXPORT_SYMBOL_GPL(unregister_kprobe);
1701
1702 void unregister_kprobes(struct kprobe **kps, int num)
1703 {
1704         int i;
1705
1706         if (num <= 0)
1707                 return;
1708         mutex_lock(&kprobe_mutex);
1709         for (i = 0; i < num; i++)
1710                 if (__unregister_kprobe_top(kps[i]) < 0)
1711                         kps[i]->addr = NULL;
1712         mutex_unlock(&kprobe_mutex);
1713
1714         synchronize_sched();
1715         for (i = 0; i < num; i++)
1716                 if (kps[i]->addr)
1717                         __unregister_kprobe_bottom(kps[i]);
1718 }
1719 EXPORT_SYMBOL_GPL(unregister_kprobes);
1720
1721 static struct notifier_block kprobe_exceptions_nb = {
1722         .notifier_call = kprobe_exceptions_notify,
1723         .priority = 0x7fffffff /* we need to be notified first */
1724 };
1725
1726 unsigned long __weak arch_deref_entry_point(void *entry)
1727 {
1728         return (unsigned long)entry;
1729 }
1730
1731 int register_jprobes(struct jprobe **jps, int num)
1732 {
1733         struct jprobe *jp;
1734         int ret = 0, i;
1735
1736         if (num <= 0)
1737                 return -EINVAL;
1738         for (i = 0; i < num; i++) {
1739                 unsigned long addr, offset;
1740                 jp = jps[i];
1741                 addr = arch_deref_entry_point(jp->entry);
1742
1743                 /* Verify probepoint is a function entry point */
1744                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1745                     offset == 0) {
1746                         jp->kp.pre_handler = setjmp_pre_handler;
1747                         jp->kp.break_handler = longjmp_break_handler;
1748                         ret = register_kprobe(&jp->kp);
1749                 } else
1750                         ret = -EINVAL;
1751
1752                 if (ret < 0) {
1753                         if (i > 0)
1754                                 unregister_jprobes(jps, i);
1755                         break;
1756                 }
1757         }
1758         return ret;
1759 }
1760 EXPORT_SYMBOL_GPL(register_jprobes);
1761
1762 int register_jprobe(struct jprobe *jp)
1763 {
1764         return register_jprobes(&jp, 1);
1765 }
1766 EXPORT_SYMBOL_GPL(register_jprobe);
1767
1768 void unregister_jprobe(struct jprobe *jp)
1769 {
1770         unregister_jprobes(&jp, 1);
1771 }
1772 EXPORT_SYMBOL_GPL(unregister_jprobe);
1773
1774 void unregister_jprobes(struct jprobe **jps, int num)
1775 {
1776         int i;
1777
1778         if (num <= 0)
1779                 return;
1780         mutex_lock(&kprobe_mutex);
1781         for (i = 0; i < num; i++)
1782                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1783                         jps[i]->kp.addr = NULL;
1784         mutex_unlock(&kprobe_mutex);
1785
1786         synchronize_sched();
1787         for (i = 0; i < num; i++) {
1788                 if (jps[i]->kp.addr)
1789                         __unregister_kprobe_bottom(&jps[i]->kp);
1790         }
1791 }
1792 EXPORT_SYMBOL_GPL(unregister_jprobes);
1793
1794 #ifdef CONFIG_KRETPROBES
1795 /*
1796  * This kprobe pre_handler is registered with every kretprobe. When probe
1797  * hits it will set up the return probe.
1798  */
1799 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1800 {
1801         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1802         unsigned long hash, flags = 0;
1803         struct kretprobe_instance *ri;
1804
1805         /*
1806          * To avoid deadlocks, prohibit return probing in NMI contexts,
1807          * just skip the probe and increase the (inexact) 'nmissed'
1808          * statistical counter, so that the user is informed that
1809          * something happened:
1810          */
1811         if (unlikely(in_nmi())) {
1812                 rp->nmissed++;
1813                 return 0;
1814         }
1815
1816         /* TODO: consider to only swap the RA after the last pre_handler fired */
1817         hash = hash_ptr(current, KPROBE_HASH_BITS);
1818         raw_spin_lock_irqsave(&rp->lock, flags);
1819         if (!hlist_empty(&rp->free_instances)) {
1820                 ri = hlist_entry(rp->free_instances.first,
1821                                 struct kretprobe_instance, hlist);
1822                 hlist_del(&ri->hlist);
1823                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1824
1825                 ri->rp = rp;
1826                 ri->task = current;
1827
1828                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1829                         raw_spin_lock_irqsave(&rp->lock, flags);
1830                         hlist_add_head(&ri->hlist, &rp->free_instances);
1831                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1832                         return 0;
1833                 }
1834
1835                 arch_prepare_kretprobe(ri, regs);
1836
1837                 /* XXX(hch): why is there no hlist_move_head? */
1838                 INIT_HLIST_NODE(&ri->hlist);
1839                 kretprobe_table_lock(hash, &flags);
1840                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1841                 kretprobe_table_unlock(hash, &flags);
1842         } else {
1843                 rp->nmissed++;
1844                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1845         }
1846         return 0;
1847 }
1848 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1849
1850 int register_kretprobe(struct kretprobe *rp)
1851 {
1852         int ret = 0;
1853         struct kretprobe_instance *inst;
1854         int i;
1855         void *addr;
1856
1857         if (kretprobe_blacklist_size) {
1858                 addr = kprobe_addr(&rp->kp);
1859                 if (IS_ERR(addr))
1860                         return PTR_ERR(addr);
1861
1862                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1863                         if (kretprobe_blacklist[i].addr == addr)
1864                                 return -EINVAL;
1865                 }
1866         }
1867
1868         rp->kp.pre_handler = pre_handler_kretprobe;
1869         rp->kp.post_handler = NULL;
1870         rp->kp.fault_handler = NULL;
1871         rp->kp.break_handler = NULL;
1872
1873         /* Pre-allocate memory for max kretprobe instances */
1874         if (rp->maxactive <= 0) {
1875 #ifdef CONFIG_PREEMPT
1876                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1877 #else
1878                 rp->maxactive = num_possible_cpus();
1879 #endif
1880         }
1881         raw_spin_lock_init(&rp->lock);
1882         INIT_HLIST_HEAD(&rp->free_instances);
1883         for (i = 0; i < rp->maxactive; i++) {
1884                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1885                                rp->data_size, GFP_KERNEL);
1886                 if (inst == NULL) {
1887                         free_rp_inst(rp);
1888                         return -ENOMEM;
1889                 }
1890                 INIT_HLIST_NODE(&inst->hlist);
1891                 hlist_add_head(&inst->hlist, &rp->free_instances);
1892         }
1893
1894         rp->nmissed = 0;
1895         /* Establish function entry probe point */
1896         ret = register_kprobe(&rp->kp);
1897         if (ret != 0)
1898                 free_rp_inst(rp);
1899         return ret;
1900 }
1901 EXPORT_SYMBOL_GPL(register_kretprobe);
1902
1903 int register_kretprobes(struct kretprobe **rps, int num)
1904 {
1905         int ret = 0, i;
1906
1907         if (num <= 0)
1908                 return -EINVAL;
1909         for (i = 0; i < num; i++) {
1910                 ret = register_kretprobe(rps[i]);
1911                 if (ret < 0) {
1912                         if (i > 0)
1913                                 unregister_kretprobes(rps, i);
1914                         break;
1915                 }
1916         }
1917         return ret;
1918 }
1919 EXPORT_SYMBOL_GPL(register_kretprobes);
1920
1921 void unregister_kretprobe(struct kretprobe *rp)
1922 {
1923         unregister_kretprobes(&rp, 1);
1924 }
1925 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1926
1927 void unregister_kretprobes(struct kretprobe **rps, int num)
1928 {
1929         int i;
1930
1931         if (num <= 0)
1932                 return;
1933         mutex_lock(&kprobe_mutex);
1934         for (i = 0; i < num; i++)
1935                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1936                         rps[i]->kp.addr = NULL;
1937         mutex_unlock(&kprobe_mutex);
1938
1939         synchronize_sched();
1940         for (i = 0; i < num; i++) {
1941                 if (rps[i]->kp.addr) {
1942                         __unregister_kprobe_bottom(&rps[i]->kp);
1943                         cleanup_rp_inst(rps[i]);
1944                 }
1945         }
1946 }
1947 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1948
1949 #else /* CONFIG_KRETPROBES */
1950 int register_kretprobe(struct kretprobe *rp)
1951 {
1952         return -ENOSYS;
1953 }
1954 EXPORT_SYMBOL_GPL(register_kretprobe);
1955
1956 int register_kretprobes(struct kretprobe **rps, int num)
1957 {
1958         return -ENOSYS;
1959 }
1960 EXPORT_SYMBOL_GPL(register_kretprobes);
1961
1962 void unregister_kretprobe(struct kretprobe *rp)
1963 {
1964 }
1965 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1966
1967 void unregister_kretprobes(struct kretprobe **rps, int num)
1968 {
1969 }
1970 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1971
1972 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1973 {
1974         return 0;
1975 }
1976 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1977
1978 #endif /* CONFIG_KRETPROBES */
1979
1980 /* Set the kprobe gone and remove its instruction buffer. */
1981 static void kill_kprobe(struct kprobe *p)
1982 {
1983         struct kprobe *kp;
1984
1985         p->flags |= KPROBE_FLAG_GONE;
1986         if (kprobe_aggrprobe(p)) {
1987                 /*
1988                  * If this is an aggr_kprobe, we have to list all the
1989                  * chained probes and mark them GONE.
1990                  */
1991                 list_for_each_entry_rcu(kp, &p->list, list)
1992                         kp->flags |= KPROBE_FLAG_GONE;
1993                 p->post_handler = NULL;
1994                 p->break_handler = NULL;
1995                 kill_optimized_kprobe(p);
1996         }
1997         /*
1998          * Here, we can remove insn_slot safely, because no thread calls
1999          * the original probed function (which will be freed soon) any more.
2000          */
2001         arch_remove_kprobe(p);
2002 }
2003
2004 /* Disable one kprobe */
2005 int disable_kprobe(struct kprobe *kp)
2006 {
2007         int ret = 0;
2008
2009         mutex_lock(&kprobe_mutex);
2010
2011         /* Disable this kprobe */
2012         if (__disable_kprobe(kp) == NULL)
2013                 ret = -EINVAL;
2014
2015         mutex_unlock(&kprobe_mutex);
2016         return ret;
2017 }
2018 EXPORT_SYMBOL_GPL(disable_kprobe);
2019
2020 /* Enable one kprobe */
2021 int enable_kprobe(struct kprobe *kp)
2022 {
2023         int ret = 0;
2024         struct kprobe *p;
2025
2026         mutex_lock(&kprobe_mutex);
2027
2028         /* Check whether specified probe is valid. */
2029         p = __get_valid_kprobe(kp);
2030         if (unlikely(p == NULL)) {
2031                 ret = -EINVAL;
2032                 goto out;
2033         }
2034
2035         if (kprobe_gone(kp)) {
2036                 /* This kprobe has gone, we couldn't enable it. */
2037                 ret = -EINVAL;
2038                 goto out;
2039         }
2040
2041         if (p != kp)
2042                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2043
2044         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2045                 p->flags &= ~KPROBE_FLAG_DISABLED;
2046                 arm_kprobe(p);
2047         }
2048 out:
2049         mutex_unlock(&kprobe_mutex);
2050         return ret;
2051 }
2052 EXPORT_SYMBOL_GPL(enable_kprobe);
2053
2054 void dump_kprobe(struct kprobe *kp)
2055 {
2056         printk(KERN_WARNING "Dumping kprobe:\n");
2057         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2058                kp->symbol_name, kp->addr, kp->offset);
2059 }
2060 NOKPROBE_SYMBOL(dump_kprobe);
2061
2062 /*
2063  * Lookup and populate the kprobe_blacklist.
2064  *
2065  * Unlike the kretprobe blacklist, we'll need to determine
2066  * the range of addresses that belong to the said functions,
2067  * since a kprobe need not necessarily be at the beginning
2068  * of a function.
2069  */
2070 static int __init populate_kprobe_blacklist(unsigned long *start,
2071                                              unsigned long *end)
2072 {
2073         unsigned long *iter;
2074         struct kprobe_blacklist_entry *ent;
2075         unsigned long entry, offset = 0, size = 0;
2076
2077         for (iter = start; iter < end; iter++) {
2078                 entry = arch_deref_entry_point((void *)*iter);
2079
2080                 if (!kernel_text_address(entry) ||
2081                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2082                         pr_err("Failed to find blacklist at %p\n",
2083                                 (void *)entry);
2084                         continue;
2085                 }
2086
2087                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2088                 if (!ent)
2089                         return -ENOMEM;
2090                 ent->start_addr = entry;
2091                 ent->end_addr = entry + size;
2092                 INIT_LIST_HEAD(&ent->list);
2093                 list_add_tail(&ent->list, &kprobe_blacklist);
2094         }
2095         return 0;
2096 }
2097
2098 /* Module notifier call back, checking kprobes on the module */
2099 static int kprobes_module_callback(struct notifier_block *nb,
2100                                    unsigned long val, void *data)
2101 {
2102         struct module *mod = data;
2103         struct hlist_head *head;
2104         struct kprobe *p;
2105         unsigned int i;
2106         int checkcore = (val == MODULE_STATE_GOING);
2107
2108         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2109                 return NOTIFY_DONE;
2110
2111         /*
2112          * When MODULE_STATE_GOING was notified, both of module .text and
2113          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2114          * notified, only .init.text section would be freed. We need to
2115          * disable kprobes which have been inserted in the sections.
2116          */
2117         mutex_lock(&kprobe_mutex);
2118         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2119                 head = &kprobe_table[i];
2120                 hlist_for_each_entry_rcu(p, head, hlist)
2121                         if (within_module_init((unsigned long)p->addr, mod) ||
2122                             (checkcore &&
2123                              within_module_core((unsigned long)p->addr, mod))) {
2124                                 /*
2125                                  * The vaddr this probe is installed will soon
2126                                  * be vfreed buy not synced to disk. Hence,
2127                                  * disarming the breakpoint isn't needed.
2128                                  */
2129                                 kill_kprobe(p);
2130                         }
2131         }
2132         mutex_unlock(&kprobe_mutex);
2133         return NOTIFY_DONE;
2134 }
2135
2136 static struct notifier_block kprobe_module_nb = {
2137         .notifier_call = kprobes_module_callback,
2138         .priority = 0
2139 };
2140
2141 /* Markers of _kprobe_blacklist section */
2142 extern unsigned long __start_kprobe_blacklist[];
2143 extern unsigned long __stop_kprobe_blacklist[];
2144
2145 static int __init init_kprobes(void)
2146 {
2147         int i, err = 0;
2148
2149         /* FIXME allocate the probe table, currently defined statically */
2150         /* initialize all list heads */
2151         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2152                 INIT_HLIST_HEAD(&kprobe_table[i]);
2153                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2154                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2155         }
2156
2157         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2158                                         __stop_kprobe_blacklist);
2159         if (err) {
2160                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2161                 pr_err("Please take care of using kprobes.\n");
2162         }
2163
2164         if (kretprobe_blacklist_size) {
2165                 /* lookup the function address from its name */
2166                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2167                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2168                                            kretprobe_blacklist[i].addr);
2169                         if (!kretprobe_blacklist[i].addr)
2170                                 printk("kretprobe: lookup failed: %s\n",
2171                                        kretprobe_blacklist[i].name);
2172                 }
2173         }
2174
2175 #if defined(CONFIG_OPTPROBES)
2176 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2177         /* Init kprobe_optinsn_slots */
2178         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2179 #endif
2180         /* By default, kprobes can be optimized */
2181         kprobes_allow_optimization = true;
2182 #endif
2183
2184         /* By default, kprobes are armed */
2185         kprobes_all_disarmed = false;
2186
2187         err = arch_init_kprobes();
2188         if (!err)
2189                 err = register_die_notifier(&kprobe_exceptions_nb);
2190         if (!err)
2191                 err = register_module_notifier(&kprobe_module_nb);
2192
2193         kprobes_initialized = (err == 0);
2194
2195         if (!err)
2196                 init_test_probes();
2197         return err;
2198 }
2199
2200 #ifdef CONFIG_DEBUG_FS
2201 static void report_probe(struct seq_file *pi, struct kprobe *p,
2202                 const char *sym, int offset, char *modname, struct kprobe *pp)
2203 {
2204         char *kprobe_type;
2205
2206         if (p->pre_handler == pre_handler_kretprobe)
2207                 kprobe_type = "r";
2208         else if (p->pre_handler == setjmp_pre_handler)
2209                 kprobe_type = "j";
2210         else
2211                 kprobe_type = "k";
2212
2213         if (sym)
2214                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2215                         p->addr, kprobe_type, sym, offset,
2216                         (modname ? modname : " "));
2217         else
2218                 seq_printf(pi, "%p  %s  %p ",
2219                         p->addr, kprobe_type, p->addr);
2220
2221         if (!pp)
2222                 pp = p;
2223         seq_printf(pi, "%s%s%s%s\n",
2224                 (kprobe_gone(p) ? "[GONE]" : ""),
2225                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2226                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2227                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2228 }
2229
2230 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2231 {
2232         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2233 }
2234
2235 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2236 {
2237         (*pos)++;
2238         if (*pos >= KPROBE_TABLE_SIZE)
2239                 return NULL;
2240         return pos;
2241 }
2242
2243 static void kprobe_seq_stop(struct seq_file *f, void *v)
2244 {
2245         /* Nothing to do */
2246 }
2247
2248 static int show_kprobe_addr(struct seq_file *pi, void *v)
2249 {
2250         struct hlist_head *head;
2251         struct kprobe *p, *kp;
2252         const char *sym = NULL;
2253         unsigned int i = *(loff_t *) v;
2254         unsigned long offset = 0;
2255         char *modname, namebuf[KSYM_NAME_LEN];
2256
2257         head = &kprobe_table[i];
2258         preempt_disable();
2259         hlist_for_each_entry_rcu(p, head, hlist) {
2260                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2261                                         &offset, &modname, namebuf);
2262                 if (kprobe_aggrprobe(p)) {
2263                         list_for_each_entry_rcu(kp, &p->list, list)
2264                                 report_probe(pi, kp, sym, offset, modname, p);
2265                 } else
2266                         report_probe(pi, p, sym, offset, modname, NULL);
2267         }
2268         preempt_enable();
2269         return 0;
2270 }
2271
2272 static const struct seq_operations kprobes_seq_ops = {
2273         .start = kprobe_seq_start,
2274         .next  = kprobe_seq_next,
2275         .stop  = kprobe_seq_stop,
2276         .show  = show_kprobe_addr
2277 };
2278
2279 static int kprobes_open(struct inode *inode, struct file *filp)
2280 {
2281         return seq_open(filp, &kprobes_seq_ops);
2282 }
2283
2284 static const struct file_operations debugfs_kprobes_operations = {
2285         .open           = kprobes_open,
2286         .read           = seq_read,
2287         .llseek         = seq_lseek,
2288         .release        = seq_release,
2289 };
2290
2291 /* kprobes/blacklist -- shows which functions can not be probed */
2292 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2293 {
2294         return seq_list_start(&kprobe_blacklist, *pos);
2295 }
2296
2297 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2298 {
2299         return seq_list_next(v, &kprobe_blacklist, pos);
2300 }
2301
2302 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2303 {
2304         struct kprobe_blacklist_entry *ent =
2305                 list_entry(v, struct kprobe_blacklist_entry, list);
2306
2307         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2308                    (void *)ent->end_addr, (void *)ent->start_addr);
2309         return 0;
2310 }
2311
2312 static const struct seq_operations kprobe_blacklist_seq_ops = {
2313         .start = kprobe_blacklist_seq_start,
2314         .next  = kprobe_blacklist_seq_next,
2315         .stop  = kprobe_seq_stop,       /* Reuse void function */
2316         .show  = kprobe_blacklist_seq_show,
2317 };
2318
2319 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2320 {
2321         return seq_open(filp, &kprobe_blacklist_seq_ops);
2322 }
2323
2324 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2325         .open           = kprobe_blacklist_open,
2326         .read           = seq_read,
2327         .llseek         = seq_lseek,
2328         .release        = seq_release,
2329 };
2330
2331 static void arm_all_kprobes(void)
2332 {
2333         struct hlist_head *head;
2334         struct kprobe *p;
2335         unsigned int i;
2336
2337         mutex_lock(&kprobe_mutex);
2338
2339         /* If kprobes are armed, just return */
2340         if (!kprobes_all_disarmed)
2341                 goto already_enabled;
2342
2343         /*
2344          * optimize_kprobe() called by arm_kprobe() checks
2345          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2346          * arm_kprobe.
2347          */
2348         kprobes_all_disarmed = false;
2349         /* Arming kprobes doesn't optimize kprobe itself */
2350         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2351                 head = &kprobe_table[i];
2352                 hlist_for_each_entry_rcu(p, head, hlist)
2353                         if (!kprobe_disabled(p))
2354                                 arm_kprobe(p);
2355         }
2356
2357         printk(KERN_INFO "Kprobes globally enabled\n");
2358
2359 already_enabled:
2360         mutex_unlock(&kprobe_mutex);
2361         return;
2362 }
2363
2364 static void disarm_all_kprobes(void)
2365 {
2366         struct hlist_head *head;
2367         struct kprobe *p;
2368         unsigned int i;
2369
2370         mutex_lock(&kprobe_mutex);
2371
2372         /* If kprobes are already disarmed, just return */
2373         if (kprobes_all_disarmed) {
2374                 mutex_unlock(&kprobe_mutex);
2375                 return;
2376         }
2377
2378         kprobes_all_disarmed = true;
2379         printk(KERN_INFO "Kprobes globally disabled\n");
2380
2381         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2382                 head = &kprobe_table[i];
2383                 hlist_for_each_entry_rcu(p, head, hlist) {
2384                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2385                                 disarm_kprobe(p, false);
2386                 }
2387         }
2388         mutex_unlock(&kprobe_mutex);
2389
2390         /* Wait for disarming all kprobes by optimizer */
2391         wait_for_kprobe_optimizer();
2392 }
2393
2394 /*
2395  * XXX: The debugfs bool file interface doesn't allow for callbacks
2396  * when the bool state is switched. We can reuse that facility when
2397  * available
2398  */
2399 static ssize_t read_enabled_file_bool(struct file *file,
2400                char __user *user_buf, size_t count, loff_t *ppos)
2401 {
2402         char buf[3];
2403
2404         if (!kprobes_all_disarmed)
2405                 buf[0] = '1';
2406         else
2407                 buf[0] = '0';
2408         buf[1] = '\n';
2409         buf[2] = 0x00;
2410         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2411 }
2412
2413 static ssize_t write_enabled_file_bool(struct file *file,
2414                const char __user *user_buf, size_t count, loff_t *ppos)
2415 {
2416         char buf[32];
2417         size_t buf_size;
2418
2419         buf_size = min(count, (sizeof(buf)-1));
2420         if (copy_from_user(buf, user_buf, buf_size))
2421                 return -EFAULT;
2422
2423         buf[buf_size] = '\0';
2424         switch (buf[0]) {
2425         case 'y':
2426         case 'Y':
2427         case '1':
2428                 arm_all_kprobes();
2429                 break;
2430         case 'n':
2431         case 'N':
2432         case '0':
2433                 disarm_all_kprobes();
2434                 break;
2435         default:
2436                 return -EINVAL;
2437         }
2438
2439         return count;
2440 }
2441
2442 static const struct file_operations fops_kp = {
2443         .read =         read_enabled_file_bool,
2444         .write =        write_enabled_file_bool,
2445         .llseek =       default_llseek,
2446 };
2447
2448 static int __init debugfs_kprobe_init(void)
2449 {
2450         struct dentry *dir, *file;
2451         unsigned int value = 1;
2452
2453         dir = debugfs_create_dir("kprobes", NULL);
2454         if (!dir)
2455                 return -ENOMEM;
2456
2457         file = debugfs_create_file("list", 0400, dir, NULL,
2458                                 &debugfs_kprobes_operations);
2459         if (!file)
2460                 goto error;
2461
2462         file = debugfs_create_file("enabled", 0600, dir,
2463                                         &value, &fops_kp);
2464         if (!file)
2465                 goto error;
2466
2467         file = debugfs_create_file("blacklist", 0400, dir, NULL,
2468                                 &debugfs_kprobe_blacklist_ops);
2469         if (!file)
2470                 goto error;
2471
2472         return 0;
2473
2474 error:
2475         debugfs_remove(dir);
2476         return -ENOMEM;
2477 }
2478
2479 late_initcall(debugfs_kprobe_init);
2480 #endif /* CONFIG_DEBUG_FS */
2481
2482 module_init(init_kprobes);
2483
2484 /* defined in arch/.../kernel/kprobes.c */
2485 EXPORT_SYMBOL_GPL(jprobe_return);