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, p);
 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, p);
 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         /* Try to unoptimize */
 873         unoptimize_kprobe(p, kprobes_all_disarmed);
 874
 875         if (!kprobe_queued(p)) {
 876                 arch_disarm_kprobe(p);
 877                 /* If another kprobe was blocked, optimize it. */
 878                 _p = get_optimized_kprobe((unsigned long)p->addr);
 879                 if (unlikely(_p) && reopt)
 880                         optimize_kprobe(_p);
 881         }
 882         /* TODO: reoptimize others after unoptimized this probe */
 883 }
 884
 885 #else /* !CONFIG_OPTPROBES */
 886
 887 #define optimize_kprobe(p)                      do {} while (0)
 888 #define unoptimize_kprobe(p, f)                 do {} while (0)
 889 #define kill_optimized_kprobe(p)                do {} while (0)
 890 #define prepare_optimized_kprobe(p)             do {} while (0)
 891 #define try_to_optimize_kprobe(p)               do {} while (0)
 892 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 893 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 894 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 895 #define wait_for_kprobe_optimizer()             do {} while (0)
 896
 897 /* There should be no unused kprobes can be reused without optimization */
 898 static void reuse_unused_kprobe(struct kprobe *ap)
 899 {
 900         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 901         BUG_ON(kprobe_unused(ap));
 902 }
 903
 904 static void free_aggr_kprobe(struct kprobe *p)
 905 {
 906         arch_remove_kprobe(p);
 907         kfree(p);
 908 }
 909
 910 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 911 {
 912         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 913 }
 914 #endif /* CONFIG_OPTPROBES */
 915
 916 #ifdef CONFIG_KPROBES_ON_FTRACE
 917 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 918         .func = kprobe_ftrace_handler,
 919         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 920 };
 921 static int kprobe_ftrace_enabled;
 922
 923 /* Must ensure p->addr is really on ftrace */
 924 static int prepare_kprobe(struct kprobe *p)
 925 {
 926         if (!kprobe_ftrace(p))
 927                 return arch_prepare_kprobe(p);
 928
 929         return arch_prepare_kprobe_ftrace(p);
 930 }
 931
 932 /* Caller must lock kprobe_mutex */
 933 static void arm_kprobe_ftrace(struct kprobe *p)
 934 {
 935         int ret;
 936
 937         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 938                                    (unsigned long)p->addr, 0, 0);
 939         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 940         kprobe_ftrace_enabled++;
 941         if (kprobe_ftrace_enabled == 1) {
 942                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 943                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 944         }
 945 }
 946
 947 /* Caller must lock kprobe_mutex */
 948 static void disarm_kprobe_ftrace(struct kprobe *p)
 949 {
 950         int ret;
 951
 952         kprobe_ftrace_enabled--;
 953         if (kprobe_ftrace_enabled == 0) {
 954                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 955                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 956         }
 957         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 958                            (unsigned long)p->addr, 1, 0);
 959         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 960 }
 961 #else   /* !CONFIG_KPROBES_ON_FTRACE */
 962 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 963 #define arm_kprobe_ftrace(p)    do {} while (0)
 964 #define disarm_kprobe_ftrace(p) do {} while (0)
 965 #endif
 966
 967 /* Arm a kprobe with text_mutex */
 968 static void arm_kprobe(struct kprobe *kp)
 969 {
 970         if (unlikely(kprobe_ftrace(kp))) {
 971                 arm_kprobe_ftrace(kp);
 972                 return;
 973         }
 974         /*
 975          * Here, since __arm_kprobe() doesn't use stop_machine(),
 976          * this doesn't cause deadlock on text_mutex. So, we don't
 977          * need get_online_cpus().
 978          */
 979         mutex_lock(&text_mutex);
 980         __arm_kprobe(kp);
 981         mutex_unlock(&text_mutex);
 982 }
 983
 984 /* Disarm a kprobe with text_mutex */
 985 static void disarm_kprobe(struct kprobe *kp, bool reopt)
 986 {
 987         if (unlikely(kprobe_ftrace(kp))) {
 988                 disarm_kprobe_ftrace(kp);
 989                 return;
 990         }
 991         /* Ditto */
 992         mutex_lock(&text_mutex);
 993         __disarm_kprobe(kp, reopt);
 994         mutex_unlock(&text_mutex);
 995 }
 996
 997 /*
 998  * Aggregate handlers for multiple kprobes support - these handlers
 999  * take care of invoking the individual kprobe handlers on p->list
1000  */
1001 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1002 {
1003         struct kprobe *kp;
1004
1005         list_for_each_entry_rcu(kp, &p->list, list) {
1006                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1007                         set_kprobe_instance(kp);
1008                         if (kp->pre_handler(kp, regs))
1009                                 return 1;
1010                 }
1011                 reset_kprobe_instance();
1012         }
1013         return 0;
1014 }
1015 NOKPROBE_SYMBOL(aggr_pre_handler);
1016
1017 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1018                               unsigned long flags)
1019 {
1020         struct kprobe *kp;
1021
1022         list_for_each_entry_rcu(kp, &p->list, list) {
1023                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1024                         set_kprobe_instance(kp);
1025                         kp->post_handler(kp, regs, flags);
1026                         reset_kprobe_instance();
1027                 }
1028         }
1029 }
1030 NOKPROBE_SYMBOL(aggr_post_handler);
1031
1032 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1033                               int trapnr)
1034 {
1035         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1036
1037         /*
1038          * if we faulted "during" the execution of a user specified
1039          * probe handler, invoke just that probe's fault handler
1040          */
1041         if (cur && cur->fault_handler) {
1042                 if (cur->fault_handler(cur, regs, trapnr))
1043                         return 1;
1044         }
1045         return 0;
1046 }
1047 NOKPROBE_SYMBOL(aggr_fault_handler);
1048
1049 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1050 {
1051         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1052         int ret = 0;
1053
1054         if (cur && cur->break_handler) {
1055                 if (cur->break_handler(cur, regs))
1056                         ret = 1;
1057         }
1058         reset_kprobe_instance();
1059         return ret;
1060 }
1061 NOKPROBE_SYMBOL(aggr_break_handler);
1062
1063 /* Walks the list and increments nmissed count for multiprobe case */
1064 void kprobes_inc_nmissed_count(struct kprobe *p)
1065 {
1066         struct kprobe *kp;
1067         if (!kprobe_aggrprobe(p)) {
1068                 p->nmissed++;
1069         } else {
1070                 list_for_each_entry_rcu(kp, &p->list, list)
1071                         kp->nmissed++;
1072         }
1073         return;
1074 }
1075 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1076
1077 void recycle_rp_inst(struct kretprobe_instance *ri,
1078                      struct hlist_head *head)
1079 {
1080         struct kretprobe *rp = ri->rp;
1081
1082         /* remove rp inst off the rprobe_inst_table */
1083         hlist_del(&ri->hlist);
1084         INIT_HLIST_NODE(&ri->hlist);
1085         if (likely(rp)) {
1086                 raw_spin_lock(&rp->lock);
1087                 hlist_add_head(&ri->hlist, &rp->free_instances);
1088                 raw_spin_unlock(&rp->lock);
1089         } else
1090                 /* Unregistering */
1091                 hlist_add_head(&ri->hlist, head);
1092 }
1093 NOKPROBE_SYMBOL(recycle_rp_inst);
1094
1095 void kretprobe_hash_lock(struct task_struct *tsk,
1096                          struct hlist_head **head, unsigned long *flags)
1097 __acquires(hlist_lock)
1098 {
1099         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1100         raw_spinlock_t *hlist_lock;
1101
1102         *head = &kretprobe_inst_table[hash];
1103         hlist_lock = kretprobe_table_lock_ptr(hash);
1104         raw_spin_lock_irqsave(hlist_lock, *flags);
1105 }
1106 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1107
1108 static void kretprobe_table_lock(unsigned long hash,
1109                                  unsigned long *flags)
1110 __acquires(hlist_lock)
1111 {
1112         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1113         raw_spin_lock_irqsave(hlist_lock, *flags);
1114 }
1115 NOKPROBE_SYMBOL(kretprobe_table_lock);
1116
1117 void kretprobe_hash_unlock(struct task_struct *tsk,
1118                            unsigned long *flags)
1119 __releases(hlist_lock)
1120 {
1121         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1122         raw_spinlock_t *hlist_lock;
1123
1124         hlist_lock = kretprobe_table_lock_ptr(hash);
1125         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1126 }
1127 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1128
1129 static void kretprobe_table_unlock(unsigned long hash,
1130                                    unsigned long *flags)
1131 __releases(hlist_lock)
1132 {
1133         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1134         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1135 }
1136 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1137
1138 /*
1139  * This function is called from finish_task_switch when task tk becomes dead,
1140  * so that we can recycle any function-return probe instances associated
1141  * with this task. These left over instances represent probed functions
1142  * that have been called but will never return.
1143  */
1144 void kprobe_flush_task(struct task_struct *tk)
1145 {
1146         struct kretprobe_instance *ri;
1147         struct hlist_head *head, empty_rp;
1148         struct hlist_node *tmp;
1149         unsigned long hash, flags = 0;
1150
1151         if (unlikely(!kprobes_initialized))
1152                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1153                 return;
1154
1155         INIT_HLIST_HEAD(&empty_rp);
1156         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1157         head = &kretprobe_inst_table[hash];
1158         kretprobe_table_lock(hash, &flags);
1159         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1160                 if (ri->task == tk)
1161                         recycle_rp_inst(ri, &empty_rp);
1162         }
1163         kretprobe_table_unlock(hash, &flags);
1164         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1165                 hlist_del(&ri->hlist);
1166                 kfree(ri);
1167         }
1168 }
1169 NOKPROBE_SYMBOL(kprobe_flush_task);
1170
1171 static inline void free_rp_inst(struct kretprobe *rp)
1172 {
1173         struct kretprobe_instance *ri;
1174         struct hlist_node *next;
1175
1176         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1177                 hlist_del(&ri->hlist);
1178                 kfree(ri);
1179         }
1180 }
1181
1182 static void cleanup_rp_inst(struct kretprobe *rp)
1183 {
1184         unsigned long flags, hash;
1185         struct kretprobe_instance *ri;
1186         struct hlist_node *next;
1187         struct hlist_head *head;
1188
1189         /* No race here */
1190         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1191                 kretprobe_table_lock(hash, &flags);
1192                 head = &kretprobe_inst_table[hash];
1193                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1194                         if (ri->rp == rp)
1195                                 ri->rp = NULL;
1196                 }
1197                 kretprobe_table_unlock(hash, &flags);
1198         }
1199         free_rp_inst(rp);
1200 }
1201 NOKPROBE_SYMBOL(cleanup_rp_inst);
1202
1203 /*
1204 * Add the new probe to ap->list. Fail if this is the
1205 * second jprobe at the address - two jprobes can't coexist
1206 */
1207 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1208 {
1209         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1210
1211         if (p->break_handler || p->post_handler)
1212                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1213
1214         if (p->break_handler) {
1215                 if (ap->break_handler)
1216                         return -EEXIST;
1217                 list_add_tail_rcu(&p->list, &ap->list);
1218                 ap->break_handler = aggr_break_handler;
1219         } else
1220                 list_add_rcu(&p->list, &ap->list);
1221         if (p->post_handler && !ap->post_handler)
1222                 ap->post_handler = aggr_post_handler;
1223
1224         return 0;
1225 }
1226
1227 /*
1228  * Fill in the required fields of the "manager kprobe". Replace the
1229  * earlier kprobe in the hlist with the manager kprobe
1230  */
1231 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1232 {
1233         /* Copy p's insn slot to ap */
1234         copy_kprobe(p, ap);
1235         flush_insn_slot(ap);
1236         ap->addr = p->addr;
1237         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1238         ap->pre_handler = aggr_pre_handler;
1239         ap->fault_handler = aggr_fault_handler;
1240         /* We don't care the kprobe which has gone. */
1241         if (p->post_handler && !kprobe_gone(p))
1242                 ap->post_handler = aggr_post_handler;
1243         if (p->break_handler && !kprobe_gone(p))
1244                 ap->break_handler = aggr_break_handler;
1245
1246         INIT_LIST_HEAD(&ap->list);
1247         INIT_HLIST_NODE(&ap->hlist);
1248
1249         list_add_rcu(&p->list, &ap->list);
1250         hlist_replace_rcu(&p->hlist, &ap->hlist);
1251 }
1252
1253 /*
1254  * This is the second or subsequent kprobe at the address - handle
1255  * the intricacies
1256  */
1257 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1258 {
1259         int ret = 0;
1260         struct kprobe *ap = orig_p;
1261
1262         /* For preparing optimization, jump_label_text_reserved() is called */
1263         jump_label_lock();
1264         /*
1265          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1266          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1267          */
1268         get_online_cpus();
1269         mutex_lock(&text_mutex);
1270
1271         if (!kprobe_aggrprobe(orig_p)) {
1272                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1273                 ap = alloc_aggr_kprobe(orig_p);
1274                 if (!ap) {
1275                         ret = -ENOMEM;
1276                         goto out;
1277                 }
1278                 init_aggr_kprobe(ap, orig_p);
1279         } else if (kprobe_unused(ap))
1280                 /* This probe is going to die. Rescue it */
1281                 reuse_unused_kprobe(ap);
1282
1283         if (kprobe_gone(ap)) {
1284                 /*
1285                  * Attempting to insert new probe at the same location that
1286                  * had a probe in the module vaddr area which already
1287                  * freed. So, the instruction slot has already been
1288                  * released. We need a new slot for the new probe.
1289                  */
1290                 ret = arch_prepare_kprobe(ap);
1291                 if (ret)
1292                         /*
1293                          * Even if fail to allocate new slot, don't need to
1294                          * free aggr_probe. It will be used next time, or
1295                          * freed by unregister_kprobe.
1296                          */
1297                         goto out;
1298
1299                 /* Prepare optimized instructions if possible. */
1300                 prepare_optimized_kprobe(ap);
1301
1302                 /*
1303                  * Clear gone flag to prevent allocating new slot again, and
1304                  * set disabled flag because it is not armed yet.
1305                  */
1306                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1307                             | KPROBE_FLAG_DISABLED;
1308         }
1309
1310         /* Copy ap's insn slot to p */
1311         copy_kprobe(ap, p);
1312         ret = add_new_kprobe(ap, p);
1313
1314 out:
1315         mutex_unlock(&text_mutex);
1316         put_online_cpus();
1317         jump_label_unlock();
1318
1319         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1320                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1321                 if (!kprobes_all_disarmed)
1322                         /* Arm the breakpoint again. */
1323                         arm_kprobe(ap);
1324         }
1325         return ret;
1326 }
1327
1328 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1329 {
1330         /* The __kprobes marked functions and entry code must not be probed */
1331         return addr >= (unsigned long)__kprobes_text_start &&
1332                addr < (unsigned long)__kprobes_text_end;
1333 }
1334
1335 bool within_kprobe_blacklist(unsigned long addr)
1336 {
1337         struct kprobe_blacklist_entry *ent;
1338
1339         if (arch_within_kprobe_blacklist(addr))
1340                 return true;
1341         /*
1342          * If there exists a kprobe_blacklist, verify and
1343          * fail any probe registration in the prohibited area
1344          */
1345         list_for_each_entry(ent, &kprobe_blacklist, list) {
1346                 if (addr >= ent->start_addr && addr < ent->end_addr)
1347                         return true;
1348         }
1349
1350         return false;
1351 }
1352
1353 /*
1354  * If we have a symbol_name argument, look it up and add the offset field
1355  * to it. This way, we can specify a relative address to a symbol.
1356  * This returns encoded errors if it fails to look up symbol or invalid
1357  * combination of parameters.
1358  */
1359 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1360 {
1361         kprobe_opcode_t *addr = p->addr;
1362
1363         if ((p->symbol_name && p->addr) ||
1364             (!p->symbol_name && !p->addr))
1365                 goto invalid;
1366
1367         if (p->symbol_name) {
1368                 kprobe_lookup_name(p->symbol_name, addr);
1369                 if (!addr)
1370                         return ERR_PTR(-ENOENT);
1371         }
1372
1373         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1374         if (addr)
1375                 return addr;
1376
1377 invalid:
1378         return ERR_PTR(-EINVAL);
1379 }
1380
1381 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1382 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1383 {
1384         struct kprobe *ap, *list_p;
1385
1386         ap = get_kprobe(p->addr);
1387         if (unlikely(!ap))
1388                 return NULL;
1389
1390         if (p != ap) {
1391                 list_for_each_entry_rcu(list_p, &ap->list, list)
1392                         if (list_p == p)
1393                         /* kprobe p is a valid probe */
1394                                 goto valid;
1395                 return NULL;
1396         }
1397 valid:
1398         return ap;
1399 }
1400
1401 /* Return error if the kprobe is being re-registered */
1402 static inline int check_kprobe_rereg(struct kprobe *p)
1403 {
1404         int ret = 0;
1405
1406         mutex_lock(&kprobe_mutex);
1407         if (__get_valid_kprobe(p))
1408                 ret = -EINVAL;
1409         mutex_unlock(&kprobe_mutex);
1410
1411         return ret;
1412 }
1413
1414 int __weak arch_check_ftrace_location(struct kprobe *p)
1415 {
1416         unsigned long ftrace_addr;
1417
1418         ftrace_addr = ftrace_location((unsigned long)p->addr);
1419         if (ftrace_addr) {
1420 #ifdef CONFIG_KPROBES_ON_FTRACE
1421                 /* Given address is not on the instruction boundary */
1422                 if ((unsigned long)p->addr != ftrace_addr)
1423                         return -EILSEQ;
1424                 p->flags |= KPROBE_FLAG_FTRACE;
1425 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1426                 return -EINVAL;
1427 #endif
1428         }
1429         return 0;
1430 }
1431
1432 static int check_kprobe_address_safe(struct kprobe *p,
1433                                      struct module **probed_mod)
1434 {
1435         int ret;
1436
1437         ret = arch_check_ftrace_location(p);
1438         if (ret)
1439                 return ret;
1440         jump_label_lock();
1441         preempt_disable();
1442
1443         /* Ensure it is not in reserved area nor out of text */
1444         if (!kernel_text_address((unsigned long) p->addr) ||
1445             within_kprobe_blacklist((unsigned long) p->addr) ||
1446             jump_label_text_reserved(p->addr, p->addr)) {
1447                 ret = -EINVAL;
1448                 goto out;
1449         }
1450
1451         /* Check if are we probing a module */
1452         *probed_mod = __module_text_address((unsigned long) p->addr);
1453         if (*probed_mod) {
1454                 /*
1455                  * We must hold a refcount of the probed module while updating
1456                  * its code to prohibit unexpected unloading.
1457                  */
1458                 if (unlikely(!try_module_get(*probed_mod))) {
1459                         ret = -ENOENT;
1460                         goto out;
1461                 }
1462
1463                 /*
1464                  * If the module freed .init.text, we couldn't insert
1465                  * kprobes in there.
1466                  */
1467                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1468                     (*probed_mod)->state != MODULE_STATE_COMING) {
1469                         module_put(*probed_mod);
1470                         *probed_mod = NULL;
1471                         ret = -ENOENT;
1472                 }
1473         }
1474 out:
1475         preempt_enable();
1476         jump_label_unlock();
1477
1478         return ret;
1479 }
1480
1481 int register_kprobe(struct kprobe *p)
1482 {
1483         int ret;
1484         struct kprobe *old_p;
1485         struct module *probed_mod;
1486         kprobe_opcode_t *addr;
1487
1488         /* Adjust probe address from symbol */
1489         addr = kprobe_addr(p);
1490         if (IS_ERR(addr))
1491                 return PTR_ERR(addr);
1492         p->addr = addr;
1493
1494         ret = check_kprobe_rereg(p);
1495         if (ret)
1496                 return ret;
1497
1498         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1499         p->flags &= KPROBE_FLAG_DISABLED;
1500         p->nmissed = 0;
1501         INIT_LIST_HEAD(&p->list);
1502
1503         ret = check_kprobe_address_safe(p, &probed_mod);
1504         if (ret)
1505                 return ret;
1506
1507         mutex_lock(&kprobe_mutex);
1508
1509         old_p = get_kprobe(p->addr);
1510         if (old_p) {
1511                 /* Since this may unoptimize old_p, locking text_mutex. */
1512                 ret = register_aggr_kprobe(old_p, p);
1513                 goto out;
1514         }
1515
1516         mutex_lock(&text_mutex);        /* Avoiding text modification */
1517         ret = prepare_kprobe(p);
1518         mutex_unlock(&text_mutex);
1519         if (ret)
1520                 goto out;
1521
1522         INIT_HLIST_NODE(&p->hlist);
1523         hlist_add_head_rcu(&p->hlist,
1524                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1525
1526         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1527                 arm_kprobe(p);
1528
1529         /* Try to optimize kprobe */
1530         try_to_optimize_kprobe(p);
1531
1532 out:
1533         mutex_unlock(&kprobe_mutex);
1534
1535         if (probed_mod)
1536                 module_put(probed_mod);
1537
1538         return ret;
1539 }
1540 EXPORT_SYMBOL_GPL(register_kprobe);
1541
1542 /* Check if all probes on the aggrprobe are disabled */
1543 static int aggr_kprobe_disabled(struct kprobe *ap)
1544 {
1545         struct kprobe *kp;
1546
1547         list_for_each_entry_rcu(kp, &ap->list, list)
1548                 if (!kprobe_disabled(kp))
1549                         /*
1550                          * There is an active probe on the list.
1551                          * We can't disable this ap.
1552                          */
1553                         return 0;
1554
1555         return 1;
1556 }
1557
1558 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1559 static struct kprobe *__disable_kprobe(struct kprobe *p)
1560 {
1561         struct kprobe *orig_p;
1562
1563         /* Get an original kprobe for return */
1564         orig_p = __get_valid_kprobe(p);
1565         if (unlikely(orig_p == NULL))
1566                 return NULL;
1567
1568         if (!kprobe_disabled(p)) {
1569                 /* Disable probe if it is a child probe */
1570                 if (p != orig_p)
1571                         p->flags |= KPROBE_FLAG_DISABLED;
1572
1573                 /* Try to disarm and disable this/parent probe */
1574                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1575                         /*
1576                          * If kprobes_all_disarmed is set, orig_p
1577                          * should have already been disarmed, so
1578                          * skip unneed disarming process.
1579                          */
1580                         if (!kprobes_all_disarmed)
1581                                 disarm_kprobe(orig_p, true);
1582                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1583                 }
1584         }
1585
1586         return orig_p;
1587 }
1588
1589 /*
1590  * Unregister a kprobe without a scheduler synchronization.
1591  */
1592 static int __unregister_kprobe_top(struct kprobe *p)
1593 {
1594         struct kprobe *ap, *list_p;
1595
1596         /* Disable kprobe. This will disarm it if needed. */
1597         ap = __disable_kprobe(p);
1598         if (ap == NULL)
1599                 return -EINVAL;
1600
1601         if (ap == p)
1602                 /*
1603                  * This probe is an independent(and non-optimized) kprobe
1604                  * (not an aggrprobe). Remove from the hash list.
1605                  */
1606                 goto disarmed;
1607
1608         /* Following process expects this probe is an aggrprobe */
1609         WARN_ON(!kprobe_aggrprobe(ap));
1610
1611         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1612                 /*
1613                  * !disarmed could be happen if the probe is under delayed
1614                  * unoptimizing.
1615                  */
1616                 goto disarmed;
1617         else {
1618                 /* If disabling probe has special handlers, update aggrprobe */
1619                 if (p->break_handler && !kprobe_gone(p))
1620                         ap->break_handler = NULL;
1621                 if (p->post_handler && !kprobe_gone(p)) {
1622                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1623                                 if ((list_p != p) && (list_p->post_handler))
1624                                         goto noclean;
1625                         }
1626                         ap->post_handler = NULL;
1627                 }
1628 noclean:
1629                 /*
1630                  * Remove from the aggrprobe: this path will do nothing in
1631                  * __unregister_kprobe_bottom().
1632                  */
1633                 list_del_rcu(&p->list);
1634                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1635                         /*
1636                          * Try to optimize this probe again, because post
1637                          * handler may have been changed.
1638                          */
1639                         optimize_kprobe(ap);
1640         }
1641         return 0;
1642
1643 disarmed:
1644         BUG_ON(!kprobe_disarmed(ap));
1645         hlist_del_rcu(&ap->hlist);
1646         return 0;
1647 }
1648
1649 static void __unregister_kprobe_bottom(struct kprobe *p)
1650 {
1651         struct kprobe *ap;
1652
1653         if (list_empty(&p->list))
1654                 /* This is an independent kprobe */
1655                 arch_remove_kprobe(p);
1656         else if (list_is_singular(&p->list)) {
1657                 /* This is the last child of an aggrprobe */
1658                 ap = list_entry(p->list.next, struct kprobe, list);
1659                 list_del(&p->list);
1660                 free_aggr_kprobe(ap);
1661         }
1662         /* Otherwise, do nothing. */
1663 }
1664
1665 int register_kprobes(struct kprobe **kps, int num)
1666 {
1667         int i, ret = 0;
1668
1669         if (num <= 0)
1670                 return -EINVAL;
1671         for (i = 0; i < num; i++) {
1672                 ret = register_kprobe(kps[i]);
1673                 if (ret < 0) {
1674                         if (i > 0)
1675                                 unregister_kprobes(kps, i);
1676                         break;
1677                 }
1678         }
1679         return ret;
1680 }
1681 EXPORT_SYMBOL_GPL(register_kprobes);
1682
1683 void unregister_kprobe(struct kprobe *p)
1684 {
1685         unregister_kprobes(&p, 1);
1686 }
1687 EXPORT_SYMBOL_GPL(unregister_kprobe);
1688
1689 void unregister_kprobes(struct kprobe **kps, int num)
1690 {
1691         int i;
1692
1693         if (num <= 0)
1694                 return;
1695         mutex_lock(&kprobe_mutex);
1696         for (i = 0; i < num; i++)
1697                 if (__unregister_kprobe_top(kps[i]) < 0)
1698                         kps[i]->addr = NULL;
1699         mutex_unlock(&kprobe_mutex);
1700
1701         synchronize_sched();
1702         for (i = 0; i < num; i++)
1703                 if (kps[i]->addr)
1704                         __unregister_kprobe_bottom(kps[i]);
1705 }
1706 EXPORT_SYMBOL_GPL(unregister_kprobes);
1707
1708 static struct notifier_block kprobe_exceptions_nb = {
1709         .notifier_call = kprobe_exceptions_notify,
1710         .priority = 0x7fffffff /* we need to be notified first */
1711 };
1712
1713 unsigned long __weak arch_deref_entry_point(void *entry)
1714 {
1715         return (unsigned long)entry;
1716 }
1717
1718 int register_jprobes(struct jprobe **jps, int num)
1719 {
1720         struct jprobe *jp;
1721         int ret = 0, i;
1722
1723         if (num <= 0)
1724                 return -EINVAL;
1725         for (i = 0; i < num; i++) {
1726                 unsigned long addr, offset;
1727                 jp = jps[i];
1728                 addr = arch_deref_entry_point(jp->entry);
1729
1730                 /* Verify probepoint is a function entry point */
1731                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1732                     offset == 0) {
1733                         jp->kp.pre_handler = setjmp_pre_handler;
1734                         jp->kp.break_handler = longjmp_break_handler;
1735                         ret = register_kprobe(&jp->kp);
1736                 } else
1737                         ret = -EINVAL;
1738
1739                 if (ret < 0) {
1740                         if (i > 0)
1741                                 unregister_jprobes(jps, i);
1742                         break;
1743                 }
1744         }
1745         return ret;
1746 }
1747 EXPORT_SYMBOL_GPL(register_jprobes);
1748
1749 int register_jprobe(struct jprobe *jp)
1750 {
1751         return register_jprobes(&jp, 1);
1752 }
1753 EXPORT_SYMBOL_GPL(register_jprobe);
1754
1755 void unregister_jprobe(struct jprobe *jp)
1756 {
1757         unregister_jprobes(&jp, 1);
1758 }
1759 EXPORT_SYMBOL_GPL(unregister_jprobe);
1760
1761 void unregister_jprobes(struct jprobe **jps, int num)
1762 {
1763         int i;
1764
1765         if (num <= 0)
1766                 return;
1767         mutex_lock(&kprobe_mutex);
1768         for (i = 0; i < num; i++)
1769                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1770                         jps[i]->kp.addr = NULL;
1771         mutex_unlock(&kprobe_mutex);
1772
1773         synchronize_sched();
1774         for (i = 0; i < num; i++) {
1775                 if (jps[i]->kp.addr)
1776                         __unregister_kprobe_bottom(&jps[i]->kp);
1777         }
1778 }
1779 EXPORT_SYMBOL_GPL(unregister_jprobes);
1780
1781 #ifdef CONFIG_KRETPROBES
1782 /*
1783  * This kprobe pre_handler is registered with every kretprobe. When probe
1784  * hits it will set up the return probe.
1785  */
1786 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1787 {
1788         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1789         unsigned long hash, flags = 0;
1790         struct kretprobe_instance *ri;
1791
1792         /*
1793          * To avoid deadlocks, prohibit return probing in NMI contexts,
1794          * just skip the probe and increase the (inexact) 'nmissed'
1795          * statistical counter, so that the user is informed that
1796          * something happened:
1797          */
1798         if (unlikely(in_nmi())) {
1799                 rp->nmissed++;
1800                 return 0;
1801         }
1802
1803         /* TODO: consider to only swap the RA after the last pre_handler fired */
1804         hash = hash_ptr(current, KPROBE_HASH_BITS);
1805         raw_spin_lock_irqsave(&rp->lock, flags);
1806         if (!hlist_empty(&rp->free_instances)) {
1807                 ri = hlist_entry(rp->free_instances.first,
1808                                 struct kretprobe_instance, hlist);
1809                 hlist_del(&ri->hlist);
1810                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1811
1812                 ri->rp = rp;
1813                 ri->task = current;
1814
1815                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1816                         raw_spin_lock_irqsave(&rp->lock, flags);
1817                         hlist_add_head(&ri->hlist, &rp->free_instances);
1818                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1819                         return 0;
1820                 }
1821
1822                 arch_prepare_kretprobe(ri, regs);
1823
1824                 /* XXX(hch): why is there no hlist_move_head? */
1825                 INIT_HLIST_NODE(&ri->hlist);
1826                 kretprobe_table_lock(hash, &flags);
1827                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1828                 kretprobe_table_unlock(hash, &flags);
1829         } else {
1830                 rp->nmissed++;
1831                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1832         }
1833         return 0;
1834 }
1835 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1836
1837 int register_kretprobe(struct kretprobe *rp)
1838 {
1839         int ret = 0;
1840         struct kretprobe_instance *inst;
1841         int i;
1842         void *addr;
1843
1844         if (kretprobe_blacklist_size) {
1845                 addr = kprobe_addr(&rp->kp);
1846                 if (IS_ERR(addr))
1847                         return PTR_ERR(addr);
1848
1849                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1850                         if (kretprobe_blacklist[i].addr == addr)
1851                                 return -EINVAL;
1852                 }
1853         }
1854
1855         rp->kp.pre_handler = pre_handler_kretprobe;
1856         rp->kp.post_handler = NULL;
1857         rp->kp.fault_handler = NULL;
1858         rp->kp.break_handler = NULL;
1859
1860         /* Pre-allocate memory for max kretprobe instances */
1861         if (rp->maxactive <= 0) {
1862 #ifdef CONFIG_PREEMPT
1863                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1864 #else
1865                 rp->maxactive = num_possible_cpus();
1866 #endif
1867         }
1868         raw_spin_lock_init(&rp->lock);
1869         INIT_HLIST_HEAD(&rp->free_instances);
1870         for (i = 0; i < rp->maxactive; i++) {
1871                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1872                                rp->data_size, GFP_KERNEL);
1873                 if (inst == NULL) {
1874                         free_rp_inst(rp);
1875                         return -ENOMEM;
1876                 }
1877                 INIT_HLIST_NODE(&inst->hlist);
1878                 hlist_add_head(&inst->hlist, &rp->free_instances);
1879         }
1880
1881         rp->nmissed = 0;
1882         /* Establish function entry probe point */
1883         ret = register_kprobe(&rp->kp);
1884         if (ret != 0)
1885                 free_rp_inst(rp);
1886         return ret;
1887 }
1888 EXPORT_SYMBOL_GPL(register_kretprobe);
1889
1890 int register_kretprobes(struct kretprobe **rps, int num)
1891 {
1892         int ret = 0, i;
1893
1894         if (num <= 0)
1895                 return -EINVAL;
1896         for (i = 0; i < num; i++) {
1897                 ret = register_kretprobe(rps[i]);
1898                 if (ret < 0) {
1899                         if (i > 0)
1900                                 unregister_kretprobes(rps, i);
1901                         break;
1902                 }
1903         }
1904         return ret;
1905 }
1906 EXPORT_SYMBOL_GPL(register_kretprobes);
1907
1908 void unregister_kretprobe(struct kretprobe *rp)
1909 {
1910         unregister_kretprobes(&rp, 1);
1911 }
1912 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1913
1914 void unregister_kretprobes(struct kretprobe **rps, int num)
1915 {
1916         int i;
1917
1918         if (num <= 0)
1919                 return;
1920         mutex_lock(&kprobe_mutex);
1921         for (i = 0; i < num; i++)
1922                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1923                         rps[i]->kp.addr = NULL;
1924         mutex_unlock(&kprobe_mutex);
1925
1926         synchronize_sched();
1927         for (i = 0; i < num; i++) {
1928                 if (rps[i]->kp.addr) {
1929                         __unregister_kprobe_bottom(&rps[i]->kp);
1930                         cleanup_rp_inst(rps[i]);
1931                 }
1932         }
1933 }
1934 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1935
1936 #else /* CONFIG_KRETPROBES */
1937 int register_kretprobe(struct kretprobe *rp)
1938 {
1939         return -ENOSYS;
1940 }
1941 EXPORT_SYMBOL_GPL(register_kretprobe);
1942
1943 int register_kretprobes(struct kretprobe **rps, int num)
1944 {
1945         return -ENOSYS;
1946 }
1947 EXPORT_SYMBOL_GPL(register_kretprobes);
1948
1949 void unregister_kretprobe(struct kretprobe *rp)
1950 {
1951 }
1952 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1953
1954 void unregister_kretprobes(struct kretprobe **rps, int num)
1955 {
1956 }
1957 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1958
1959 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1960 {
1961         return 0;
1962 }
1963 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1964
1965 #endif /* CONFIG_KRETPROBES */
1966
1967 /* Set the kprobe gone and remove its instruction buffer. */
1968 static void kill_kprobe(struct kprobe *p)
1969 {
1970         struct kprobe *kp;
1971
1972         p->flags |= KPROBE_FLAG_GONE;
1973         if (kprobe_aggrprobe(p)) {
1974                 /*
1975                  * If this is an aggr_kprobe, we have to list all the
1976                  * chained probes and mark them GONE.
1977                  */
1978                 list_for_each_entry_rcu(kp, &p->list, list)
1979                         kp->flags |= KPROBE_FLAG_GONE;
1980                 p->post_handler = NULL;
1981                 p->break_handler = NULL;
1982                 kill_optimized_kprobe(p);
1983         }
1984         /*
1985          * Here, we can remove insn_slot safely, because no thread calls
1986          * the original probed function (which will be freed soon) any more.
1987          */
1988         arch_remove_kprobe(p);
1989 }
1990
1991 /* Disable one kprobe */
1992 int disable_kprobe(struct kprobe *kp)
1993 {
1994         int ret = 0;
1995
1996         mutex_lock(&kprobe_mutex);
1997
1998         /* Disable this kprobe */
1999         if (__disable_kprobe(kp) == NULL)
2000                 ret = -EINVAL;
2001
2002         mutex_unlock(&kprobe_mutex);
2003         return ret;
2004 }
2005 EXPORT_SYMBOL_GPL(disable_kprobe);
2006
2007 /* Enable one kprobe */
2008 int enable_kprobe(struct kprobe *kp)
2009 {
2010         int ret = 0;
2011         struct kprobe *p;
2012
2013         mutex_lock(&kprobe_mutex);
2014
2015         /* Check whether specified probe is valid. */
2016         p = __get_valid_kprobe(kp);
2017         if (unlikely(p == NULL)) {
2018                 ret = -EINVAL;
2019                 goto out;
2020         }
2021
2022         if (kprobe_gone(kp)) {
2023                 /* This kprobe has gone, we couldn't enable it. */
2024                 ret = -EINVAL;
2025                 goto out;
2026         }
2027
2028         if (p != kp)
2029                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2030
2031         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2032                 p->flags &= ~KPROBE_FLAG_DISABLED;
2033                 arm_kprobe(p);
2034         }
2035 out:
2036         mutex_unlock(&kprobe_mutex);
2037         return ret;
2038 }
2039 EXPORT_SYMBOL_GPL(enable_kprobe);
2040
2041 void dump_kprobe(struct kprobe *kp)
2042 {
2043         printk(KERN_WARNING "Dumping kprobe:\n");
2044         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2045                kp->symbol_name, kp->addr, kp->offset);
2046 }
2047 NOKPROBE_SYMBOL(dump_kprobe);
2048
2049 /*
2050  * Lookup and populate the kprobe_blacklist.
2051  *
2052  * Unlike the kretprobe blacklist, we'll need to determine
2053  * the range of addresses that belong to the said functions,
2054  * since a kprobe need not necessarily be at the beginning
2055  * of a function.
2056  */
2057 static int __init populate_kprobe_blacklist(unsigned long *start,
2058                                              unsigned long *end)
2059 {
2060         unsigned long *iter;
2061         struct kprobe_blacklist_entry *ent;
2062         unsigned long entry, offset = 0, size = 0;
2063
2064         for (iter = start; iter < end; iter++) {
2065                 entry = arch_deref_entry_point((void *)*iter);
2066
2067                 if (!kernel_text_address(entry) ||
2068                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2069                         pr_err("Failed to find blacklist at %p\n",
2070                                 (void *)entry);
2071                         continue;
2072                 }
2073
2074                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2075                 if (!ent)
2076                         return -ENOMEM;
2077                 ent->start_addr = entry;
2078                 ent->end_addr = entry + size;
2079                 INIT_LIST_HEAD(&ent->list);
2080                 list_add_tail(&ent->list, &kprobe_blacklist);
2081         }
2082         return 0;
2083 }
2084
2085 /* Module notifier call back, checking kprobes on the module */
2086 static int kprobes_module_callback(struct notifier_block *nb,
2087                                    unsigned long val, void *data)
2088 {
2089         struct module *mod = data;
2090         struct hlist_head *head;
2091         struct kprobe *p;
2092         unsigned int i;
2093         int checkcore = (val == MODULE_STATE_GOING);
2094
2095         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2096                 return NOTIFY_DONE;
2097
2098         /*
2099          * When MODULE_STATE_GOING was notified, both of module .text and
2100          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2101          * notified, only .init.text section would be freed. We need to
2102          * disable kprobes which have been inserted in the sections.
2103          */
2104         mutex_lock(&kprobe_mutex);
2105         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2106                 head = &kprobe_table[i];
2107                 hlist_for_each_entry_rcu(p, head, hlist)
2108                         if (within_module_init((unsigned long)p->addr, mod) ||
2109                             (checkcore &&
2110                              within_module_core((unsigned long)p->addr, mod))) {
2111                                 /*
2112                                  * The vaddr this probe is installed will soon
2113                                  * be vfreed buy not synced to disk. Hence,
2114                                  * disarming the breakpoint isn't needed.
2115                                  */
2116                                 kill_kprobe(p);
2117                         }
2118         }
2119         mutex_unlock(&kprobe_mutex);
2120         return NOTIFY_DONE;
2121 }
2122
2123 static struct notifier_block kprobe_module_nb = {
2124         .notifier_call = kprobes_module_callback,
2125         .priority = 0
2126 };
2127
2128 /* Markers of _kprobe_blacklist section */
2129 extern unsigned long __start_kprobe_blacklist[];
2130 extern unsigned long __stop_kprobe_blacklist[];
2131
2132 static int __init init_kprobes(void)
2133 {
2134         int i, err = 0;
2135
2136         /* FIXME allocate the probe table, currently defined statically */
2137         /* initialize all list heads */
2138         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2139                 INIT_HLIST_HEAD(&kprobe_table[i]);
2140                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2141                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2142         }
2143
2144         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2145                                         __stop_kprobe_blacklist);
2146         if (err) {
2147                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2148                 pr_err("Please take care of using kprobes.\n");
2149         }
2150
2151         if (kretprobe_blacklist_size) {
2152                 /* lookup the function address from its name */
2153                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2154                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2155                                            kretprobe_blacklist[i].addr);
2156                         if (!kretprobe_blacklist[i].addr)
2157                                 printk("kretprobe: lookup failed: %s\n",
2158                                        kretprobe_blacklist[i].name);
2159                 }
2160         }
2161
2162 #if defined(CONFIG_OPTPROBES)
2163 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2164         /* Init kprobe_optinsn_slots */
2165         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2166 #endif
2167         /* By default, kprobes can be optimized */
2168         kprobes_allow_optimization = true;
2169 #endif
2170
2171         /* By default, kprobes are armed */
2172         kprobes_all_disarmed = false;
2173
2174         err = arch_init_kprobes();
2175         if (!err)
2176                 err = register_die_notifier(&kprobe_exceptions_nb);
2177         if (!err)
2178                 err = register_module_notifier(&kprobe_module_nb);
2179
2180         kprobes_initialized = (err == 0);
2181
2182         if (!err)
2183                 init_test_probes();
2184         return err;
2185 }
2186
2187 #ifdef CONFIG_DEBUG_FS
2188 static void report_probe(struct seq_file *pi, struct kprobe *p,
2189                 const char *sym, int offset, char *modname, struct kprobe *pp)
2190 {
2191         char *kprobe_type;
2192
2193         if (p->pre_handler == pre_handler_kretprobe)
2194                 kprobe_type = "r";
2195         else if (p->pre_handler == setjmp_pre_handler)
2196                 kprobe_type = "j";
2197         else
2198                 kprobe_type = "k";
2199
2200         if (sym)
2201                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2202                         p->addr, kprobe_type, sym, offset,
2203                         (modname ? modname : " "));
2204         else
2205                 seq_printf(pi, "%p  %s  %p ",
2206                         p->addr, kprobe_type, p->addr);
2207
2208         if (!pp)
2209                 pp = p;
2210         seq_printf(pi, "%s%s%s%s\n",
2211                 (kprobe_gone(p) ? "[GONE]" : ""),
2212                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2213                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2214                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2215 }
2216
2217 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2218 {
2219         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2220 }
2221
2222 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2223 {
2224         (*pos)++;
2225         if (*pos >= KPROBE_TABLE_SIZE)
2226                 return NULL;
2227         return pos;
2228 }
2229
2230 static void kprobe_seq_stop(struct seq_file *f, void *v)
2231 {
2232         /* Nothing to do */
2233 }
2234
2235 static int show_kprobe_addr(struct seq_file *pi, void *v)
2236 {
2237         struct hlist_head *head;
2238         struct kprobe *p, *kp;
2239         const char *sym = NULL;
2240         unsigned int i = *(loff_t *) v;
2241         unsigned long offset = 0;
2242         char *modname, namebuf[KSYM_NAME_LEN];
2243
2244         head = &kprobe_table[i];
2245         preempt_disable();
2246         hlist_for_each_entry_rcu(p, head, hlist) {
2247                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2248                                         &offset, &modname, namebuf);
2249                 if (kprobe_aggrprobe(p)) {
2250                         list_for_each_entry_rcu(kp, &p->list, list)
2251                                 report_probe(pi, kp, sym, offset, modname, p);
2252                 } else
2253                         report_probe(pi, p, sym, offset, modname, NULL);
2254         }
2255         preempt_enable();
2256         return 0;
2257 }
2258
2259 static const struct seq_operations kprobes_seq_ops = {
2260         .start = kprobe_seq_start,
2261         .next  = kprobe_seq_next,
2262         .stop  = kprobe_seq_stop,
2263         .show  = show_kprobe_addr
2264 };
2265
2266 static int kprobes_open(struct inode *inode, struct file *filp)
2267 {
2268         return seq_open(filp, &kprobes_seq_ops);
2269 }
2270
2271 static const struct file_operations debugfs_kprobes_operations = {
2272         .open           = kprobes_open,
2273         .read           = seq_read,
2274         .llseek         = seq_lseek,
2275         .release        = seq_release,
2276 };
2277
2278 /* kprobes/blacklist -- shows which functions can not be probed */
2279 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2280 {
2281         return seq_list_start(&kprobe_blacklist, *pos);
2282 }
2283
2284 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2285 {
2286         return seq_list_next(v, &kprobe_blacklist, pos);
2287 }
2288
2289 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2290 {
2291         struct kprobe_blacklist_entry *ent =
2292                 list_entry(v, struct kprobe_blacklist_entry, list);
2293
2294         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2295                    (void *)ent->end_addr, (void *)ent->start_addr);
2296         return 0;
2297 }
2298
2299 static const struct seq_operations kprobe_blacklist_seq_ops = {
2300         .start = kprobe_blacklist_seq_start,
2301         .next  = kprobe_blacklist_seq_next,
2302         .stop  = kprobe_seq_stop,       /* Reuse void function */
2303         .show  = kprobe_blacklist_seq_show,
2304 };
2305
2306 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2307 {
2308         return seq_open(filp, &kprobe_blacklist_seq_ops);
2309 }
2310
2311 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2312         .open           = kprobe_blacklist_open,
2313         .read           = seq_read,
2314         .llseek         = seq_lseek,
2315         .release        = seq_release,
2316 };
2317
2318 static void arm_all_kprobes(void)
2319 {
2320         struct hlist_head *head;
2321         struct kprobe *p;
2322         unsigned int i;
2323
2324         mutex_lock(&kprobe_mutex);
2325
2326         /* If kprobes are armed, just return */
2327         if (!kprobes_all_disarmed)
2328                 goto already_enabled;
2329
2330         /*
2331          * optimize_kprobe() called by arm_kprobe() checks
2332          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2333          * arm_kprobe.
2334          */
2335         kprobes_all_disarmed = false;
2336         /* Arming kprobes doesn't optimize kprobe itself */
2337         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2338                 head = &kprobe_table[i];
2339                 hlist_for_each_entry_rcu(p, head, hlist)
2340                         if (!kprobe_disabled(p))
2341                                 arm_kprobe(p);
2342         }
2343
2344         printk(KERN_INFO "Kprobes globally enabled\n");
2345
2346 already_enabled:
2347         mutex_unlock(&kprobe_mutex);
2348         return;
2349 }
2350
2351 static void disarm_all_kprobes(void)
2352 {
2353         struct hlist_head *head;
2354         struct kprobe *p;
2355         unsigned int i;
2356
2357         mutex_lock(&kprobe_mutex);
2358
2359         /* If kprobes are already disarmed, just return */
2360         if (kprobes_all_disarmed) {
2361                 mutex_unlock(&kprobe_mutex);
2362                 return;
2363         }
2364
2365         kprobes_all_disarmed = true;
2366         printk(KERN_INFO "Kprobes globally disabled\n");
2367
2368         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2369                 head = &kprobe_table[i];
2370                 hlist_for_each_entry_rcu(p, head, hlist) {
2371                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2372                                 disarm_kprobe(p, false);
2373                 }
2374         }
2375         mutex_unlock(&kprobe_mutex);
2376
2377         /* Wait for disarming all kprobes by optimizer */
2378         wait_for_kprobe_optimizer();
2379 }
2380
2381 /*
2382  * XXX: The debugfs bool file interface doesn't allow for callbacks
2383  * when the bool state is switched. We can reuse that facility when
2384  * available
2385  */
2386 static ssize_t read_enabled_file_bool(struct file *file,
2387                char __user *user_buf, size_t count, loff_t *ppos)
2388 {
2389         char buf[3];
2390
2391         if (!kprobes_all_disarmed)
2392                 buf[0] = '1';
2393         else
2394                 buf[0] = '0';
2395         buf[1] = '\n';
2396         buf[2] = 0x00;
2397         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2398 }
2399
2400 static ssize_t write_enabled_file_bool(struct file *file,
2401                const char __user *user_buf, size_t count, loff_t *ppos)
2402 {
2403         char buf[32];
2404         size_t buf_size;
2405
2406         buf_size = min(count, (sizeof(buf)-1));
2407         if (copy_from_user(buf, user_buf, buf_size))
2408                 return -EFAULT;
2409
2410         buf[buf_size] = '\0';
2411         switch (buf[0]) {
2412         case 'y':
2413         case 'Y':
2414         case '1':
2415                 arm_all_kprobes();
2416                 break;
2417         case 'n':
2418         case 'N':
2419         case '0':
2420                 disarm_all_kprobes();
2421                 break;
2422         default:
2423                 return -EINVAL;
2424         }
2425
2426         return count;
2427 }
2428
2429 static const struct file_operations fops_kp = {
2430         .read =         read_enabled_file_bool,
2431         .write =        write_enabled_file_bool,
2432         .llseek =       default_llseek,
2433 };
2434
2435 static int __init debugfs_kprobe_init(void)
2436 {
2437         struct dentry *dir, *file;
2438         unsigned int value = 1;
2439
2440         dir = debugfs_create_dir("kprobes", NULL);
2441         if (!dir)
2442                 return -ENOMEM;
2443
2444         file = debugfs_create_file("list", 0444, dir, NULL,
2445                                 &debugfs_kprobes_operations);
2446         if (!file)
2447                 goto error;
2448
2449         file = debugfs_create_file("enabled", 0600, dir,
2450                                         &value, &fops_kp);
2451         if (!file)
2452                 goto error;
2453
2454         file = debugfs_create_file("blacklist", 0444, dir, NULL,
2455                                 &debugfs_kprobe_blacklist_ops);
2456         if (!file)
2457                 goto error;
2458
2459         return 0;
2460
2461 error:
2462         debugfs_remove(dir);
2463         return -ENOMEM;
2464 }
2465
2466 late_initcall(debugfs_kprobe_init);
2467 #endif /* CONFIG_DEBUG_FS */
2468
2469 module_init(init_kprobes);
2470
2471 /* defined in arch/.../kernel/kprobes.c */
2472 EXPORT_SYMBOL_GPL(jprobe_return);