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/sections.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/errno.h>
  55 #include <linux/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         /* Since the slot array is not protected by rcu, we need a mutex */
 153         mutex_lock(&c->mutex);
 154  retry:
 155         rcu_read_lock();
 156         list_for_each_entry_rcu(kip, &c->pages, list) {
 157                 if (kip->nused < slots_per_page(c)) {
 158                         int i;
 159                         for (i = 0; i < slots_per_page(c); i++) {
 160                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 161                                         kip->slot_used[i] = SLOT_USED;
 162                                         kip->nused++;
 163                                         slot = kip->insns + (i * c->insn_size);
 164                                         rcu_read_unlock();
 165                                         goto out;
 166                                 }
 167                         }
 168                         /* kip->nused is broken. Fix it. */
 169                         kip->nused = slots_per_page(c);
 170                         WARN_ON(1);
 171                 }
 172         }
 173         rcu_read_unlock();
 174
 175         /* If there are any garbage slots, collect it and try again. */
 176         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 177                 goto retry;
 178
 179         /* All out of space.  Need to allocate a new page. */
 180         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 181         if (!kip)
 182                 goto out;
 183
 184         /*
 185          * Use module_alloc so this page is within +/- 2GB of where the
 186          * kernel image and loaded module images reside. This is required
 187          * so x86_64 can correctly handle the %rip-relative fixups.
 188          */
 189         kip->insns = c->alloc();
 190         if (!kip->insns) {
 191                 kfree(kip);
 192                 goto out;
 193         }
 194         INIT_LIST_HEAD(&kip->list);
 195         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 196         kip->slot_used[0] = SLOT_USED;
 197         kip->nused = 1;
 198         kip->ngarbage = 0;
 199         kip->cache = c;
 200         list_add_rcu(&kip->list, &c->pages);
 201         slot = kip->insns;
 202 out:
 203         mutex_unlock(&c->mutex);
 204         return slot;
 205 }
 206
 207 /* Return 1 if all garbages are collected, otherwise 0. */
 208 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 209 {
 210         kip->slot_used[idx] = SLOT_CLEAN;
 211         kip->nused--;
 212         if (kip->nused == 0) {
 213                 /*
 214                  * Page is no longer in use.  Free it unless
 215                  * it's the last one.  We keep the last one
 216                  * so as not to have to set it up again the
 217                  * next time somebody inserts a probe.
 218                  */
 219                 if (!list_is_singular(&kip->list)) {
 220                         list_del_rcu(&kip->list);
 221                         synchronize_rcu();
 222                         kip->cache->free(kip->insns);
 223                         kfree(kip);
 224                 }
 225                 return 1;
 226         }
 227         return 0;
 228 }
 229
 230 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 231 {
 232         struct kprobe_insn_page *kip, *next;
 233
 234         /* Ensure no-one is interrupted on the garbages */
 235         synchronize_sched();
 236
 237         list_for_each_entry_safe(kip, next, &c->pages, list) {
 238                 int i;
 239                 if (kip->ngarbage == 0)
 240                         continue;
 241                 kip->ngarbage = 0;      /* we will collect all garbages */
 242                 for (i = 0; i < slots_per_page(c); i++) {
 243                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 244                                 break;
 245                 }
 246         }
 247         c->nr_garbage = 0;
 248         return 0;
 249 }
 250
 251 void __free_insn_slot(struct kprobe_insn_cache *c,
 252                       kprobe_opcode_t *slot, int dirty)
 253 {
 254         struct kprobe_insn_page *kip;
 255         long idx;
 256
 257         mutex_lock(&c->mutex);
 258         rcu_read_lock();
 259         list_for_each_entry_rcu(kip, &c->pages, list) {
 260                 idx = ((long)slot - (long)kip->insns) /
 261                         (c->insn_size * sizeof(kprobe_opcode_t));
 262                 if (idx >= 0 && idx < slots_per_page(c))
 263                         goto out;
 264         }
 265         /* Could not find this slot. */
 266         WARN_ON(1);
 267         kip = NULL;
 268 out:
 269         rcu_read_unlock();
 270         /* Mark and sweep: this may sleep */
 271         if (kip) {
 272                 /* Check double free */
 273                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 274                 if (dirty) {
 275                         kip->slot_used[idx] = SLOT_DIRTY;
 276                         kip->ngarbage++;
 277                         if (++c->nr_garbage > slots_per_page(c))
 278                                 collect_garbage_slots(c);
 279                 } else {
 280                         collect_one_slot(kip, idx);
 281                 }
 282         }
 283         mutex_unlock(&c->mutex);
 284 }
 285
 286 /*
 287  * Check given address is on the page of kprobe instruction slots.
 288  * This will be used for checking whether the address on a stack
 289  * is on a text area or not.
 290  */
 291 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 292 {
 293         struct kprobe_insn_page *kip;
 294         bool ret = false;
 295
 296         rcu_read_lock();
 297         list_for_each_entry_rcu(kip, &c->pages, list) {
 298                 if (addr >= (unsigned long)kip->insns &&
 299                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 300                         ret = true;
 301                         break;
 302                 }
 303         }
 304         rcu_read_unlock();
 305
 306         return ret;
 307 }
 308
 309 #ifdef CONFIG_OPTPROBES
 310 /* For optimized_kprobe buffer */
 311 struct kprobe_insn_cache kprobe_optinsn_slots = {
 312         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 313         .alloc = alloc_insn_page,
 314         .free = free_insn_page,
 315         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 316         /* .insn_size is initialized later */
 317         .nr_garbage = 0,
 318 };
 319 #endif
 320 #endif
 321
 322 /* We have preemption disabled.. so it is safe to use __ versions */
 323 static inline void set_kprobe_instance(struct kprobe *kp)
 324 {
 325         __this_cpu_write(kprobe_instance, kp);
 326 }
 327
 328 static inline void reset_kprobe_instance(void)
 329 {
 330         __this_cpu_write(kprobe_instance, NULL);
 331 }
 332
 333 /*
 334  * This routine is called either:
 335  *      - under the kprobe_mutex - during kprobe_[un]register()
 336  *                              OR
 337  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 338  */
 339 struct kprobe *get_kprobe(void *addr)
 340 {
 341         struct hlist_head *head;
 342         struct kprobe *p;
 343
 344         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 345         hlist_for_each_entry_rcu(p, head, hlist) {
 346                 if (p->addr == addr)
 347                         return p;
 348         }
 349
 350         return NULL;
 351 }
 352 NOKPROBE_SYMBOL(get_kprobe);
 353
 354 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 355
 356 /* Return true if the kprobe is an aggregator */
 357 static inline int kprobe_aggrprobe(struct kprobe *p)
 358 {
 359         return p->pre_handler == aggr_pre_handler;
 360 }
 361
 362 /* Return true(!0) if the kprobe is unused */
 363 static inline int kprobe_unused(struct kprobe *p)
 364 {
 365         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 366                list_empty(&p->list);
 367 }
 368
 369 /*
 370  * Keep all fields in the kprobe consistent
 371  */
 372 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 373 {
 374         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 375         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 376 }
 377
 378 #ifdef CONFIG_OPTPROBES
 379 /* NOTE: change this value only with kprobe_mutex held */
 380 static bool kprobes_allow_optimization;
 381
 382 /*
 383  * Call all pre_handler on the list, but ignores its return value.
 384  * This must be called from arch-dep optimized caller.
 385  */
 386 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 387 {
 388         struct kprobe *kp;
 389
 390         list_for_each_entry_rcu(kp, &p->list, list) {
 391                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 392                         set_kprobe_instance(kp);
 393                         kp->pre_handler(kp, regs);
 394                 }
 395                 reset_kprobe_instance();
 396         }
 397 }
 398 NOKPROBE_SYMBOL(opt_pre_handler);
 399
 400 /* Free optimized instructions and optimized_kprobe */
 401 static void free_aggr_kprobe(struct kprobe *p)
 402 {
 403         struct optimized_kprobe *op;
 404
 405         op = container_of(p, struct optimized_kprobe, kp);
 406         arch_remove_optimized_kprobe(op);
 407         arch_remove_kprobe(p);
 408         kfree(op);
 409 }
 410
 411 /* Return true(!0) if the kprobe is ready for optimization. */
 412 static inline int kprobe_optready(struct kprobe *p)
 413 {
 414         struct optimized_kprobe *op;
 415
 416         if (kprobe_aggrprobe(p)) {
 417                 op = container_of(p, struct optimized_kprobe, kp);
 418                 return arch_prepared_optinsn(&op->optinsn);
 419         }
 420
 421         return 0;
 422 }
 423
 424 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 425 static inline int kprobe_disarmed(struct kprobe *p)
 426 {
 427         struct optimized_kprobe *op;
 428
 429         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 430         if (!kprobe_aggrprobe(p))
 431                 return kprobe_disabled(p);
 432
 433         op = container_of(p, struct optimized_kprobe, kp);
 434
 435         return kprobe_disabled(p) && list_empty(&op->list);
 436 }
 437
 438 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 439 static int kprobe_queued(struct kprobe *p)
 440 {
 441         struct optimized_kprobe *op;
 442
 443         if (kprobe_aggrprobe(p)) {
 444                 op = container_of(p, struct optimized_kprobe, kp);
 445                 if (!list_empty(&op->list))
 446                         return 1;
 447         }
 448         return 0;
 449 }
 450
 451 /*
 452  * Return an optimized kprobe whose optimizing code replaces
 453  * instructions including addr (exclude breakpoint).
 454  */
 455 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 456 {
 457         int i;
 458         struct kprobe *p = NULL;
 459         struct optimized_kprobe *op;
 460
 461         /* Don't check i == 0, since that is a breakpoint case. */
 462         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 463                 p = get_kprobe((void *)(addr - i));
 464
 465         if (p && kprobe_optready(p)) {
 466                 op = container_of(p, struct optimized_kprobe, kp);
 467                 if (arch_within_optimized_kprobe(op, addr))
 468                         return p;
 469         }
 470
 471         return NULL;
 472 }
 473
 474 /* Optimization staging list, protected by kprobe_mutex */
 475 static LIST_HEAD(optimizing_list);
 476 static LIST_HEAD(unoptimizing_list);
 477 static LIST_HEAD(freeing_list);
 478
 479 static void kprobe_optimizer(struct work_struct *work);
 480 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 481 #define OPTIMIZE_DELAY 5
 482
 483 /*
 484  * Optimize (replace a breakpoint with a jump) kprobes listed on
 485  * optimizing_list.
 486  */
 487 static void do_optimize_kprobes(void)
 488 {
 489         /* Optimization never be done when disarmed */
 490         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 491             list_empty(&optimizing_list))
 492                 return;
 493
 494         /*
 495          * The optimization/unoptimization refers online_cpus via
 496          * stop_machine() and cpu-hotplug modifies online_cpus.
 497          * And same time, text_mutex will be held in cpu-hotplug and here.
 498          * This combination can cause a deadlock (cpu-hotplug try to lock
 499          * text_mutex but stop_machine can not be done because online_cpus
 500          * has been changed)
 501          * To avoid this deadlock, we need to call get_online_cpus()
 502          * for preventing cpu-hotplug outside of text_mutex locking.
 503          */
 504         get_online_cpus();
 505         mutex_lock(&text_mutex);
 506         arch_optimize_kprobes(&optimizing_list);
 507         mutex_unlock(&text_mutex);
 508         put_online_cpus();
 509 }
 510
 511 /*
 512  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 513  * if need) kprobes listed on unoptimizing_list.
 514  */
 515 static void do_unoptimize_kprobes(void)
 516 {
 517         struct optimized_kprobe *op, *tmp;
 518
 519         /* Unoptimization must be done anytime */
 520         if (list_empty(&unoptimizing_list))
 521                 return;
 522
 523         /* Ditto to do_optimize_kprobes */
 524         get_online_cpus();
 525         mutex_lock(&text_mutex);
 526         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 527         /* Loop free_list for disarming */
 528         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 529                 /* Disarm probes if marked disabled */
 530                 if (kprobe_disabled(&op->kp))
 531                         arch_disarm_kprobe(&op->kp);
 532                 if (kprobe_unused(&op->kp)) {
 533                         /*
 534                          * Remove unused probes from hash list. After waiting
 535                          * for synchronization, these probes are reclaimed.
 536                          * (reclaiming is done by do_free_cleaned_kprobes.)
 537                          */
 538                         hlist_del_rcu(&op->kp.hlist);
 539                 } else
 540                         list_del_init(&op->list);
 541         }
 542         mutex_unlock(&text_mutex);
 543         put_online_cpus();
 544 }
 545
 546 /* Reclaim all kprobes on the free_list */
 547 static void do_free_cleaned_kprobes(void)
 548 {
 549         struct optimized_kprobe *op, *tmp;
 550
 551         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 552                 BUG_ON(!kprobe_unused(&op->kp));
 553                 list_del_init(&op->list);
 554                 free_aggr_kprobe(&op->kp);
 555         }
 556 }
 557
 558 /* Start optimizer after OPTIMIZE_DELAY passed */
 559 static void kick_kprobe_optimizer(void)
 560 {
 561         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 562 }
 563
 564 /* Kprobe jump optimizer */
 565 static void kprobe_optimizer(struct work_struct *work)
 566 {
 567         mutex_lock(&kprobe_mutex);
 568         /* Lock modules while optimizing kprobes */
 569         mutex_lock(&module_mutex);
 570
 571         /*
 572          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 573          * kprobes before waiting for quiesence period.
 574          */
 575         do_unoptimize_kprobes();
 576
 577         /*
 578          * Step 2: Wait for quiesence period to ensure all running interrupts
 579          * are done. Because optprobe may modify multiple instructions
 580          * there is a chance that Nth instruction is interrupted. In that
 581          * case, running interrupt can return to 2nd-Nth byte of jump
 582          * instruction. This wait is for avoiding it.
 583          */
 584         synchronize_sched();
 585
 586         /* Step 3: Optimize kprobes after quiesence period */
 587         do_optimize_kprobes();
 588
 589         /* Step 4: Free cleaned kprobes after quiesence period */
 590         do_free_cleaned_kprobes();
 591
 592         mutex_unlock(&module_mutex);
 593         mutex_unlock(&kprobe_mutex);
 594
 595         /* Step 5: Kick optimizer again if needed */
 596         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 597                 kick_kprobe_optimizer();
 598 }
 599
 600 /* Wait for completing optimization and unoptimization */
 601 static void wait_for_kprobe_optimizer(void)
 602 {
 603         mutex_lock(&kprobe_mutex);
 604
 605         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 606                 mutex_unlock(&kprobe_mutex);
 607
 608                 /* this will also make optimizing_work execute immmediately */
 609                 flush_delayed_work(&optimizing_work);
 610                 /* @optimizing_work might not have been queued yet, relax */
 611                 cpu_relax();
 612
 613                 mutex_lock(&kprobe_mutex);
 614         }
 615
 616         mutex_unlock(&kprobe_mutex);
 617 }
 618
 619 /* Optimize kprobe if p is ready to be optimized */
 620 static void optimize_kprobe(struct kprobe *p)
 621 {
 622         struct optimized_kprobe *op;
 623
 624         /* Check if the kprobe is disabled or not ready for optimization. */
 625         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 626             (kprobe_disabled(p) || kprobes_all_disarmed))
 627                 return;
 628
 629         /* Both of break_handler and post_handler are not supported. */
 630         if (p->break_handler || p->post_handler)
 631                 return;
 632
 633         op = container_of(p, struct optimized_kprobe, kp);
 634
 635         /* Check there is no other kprobes at the optimized instructions */
 636         if (arch_check_optimized_kprobe(op) < 0)
 637                 return;
 638
 639         /* Check if it is already optimized. */
 640         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 641                 return;
 642         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 643
 644         if (!list_empty(&op->list))
 645                 /* This is under unoptimizing. Just dequeue the probe */
 646                 list_del_init(&op->list);
 647         else {
 648                 list_add(&op->list, &optimizing_list);
 649                 kick_kprobe_optimizer();
 650         }
 651 }
 652
 653 /* Short cut to direct unoptimizing */
 654 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 655 {
 656         get_online_cpus();
 657         arch_unoptimize_kprobe(op);
 658         put_online_cpus();
 659         if (kprobe_disabled(&op->kp))
 660                 arch_disarm_kprobe(&op->kp);
 661 }
 662
 663 /* Unoptimize a kprobe if p is optimized */
 664 static void unoptimize_kprobe(struct kprobe *p, bool force)
 665 {
 666         struct optimized_kprobe *op;
 667
 668         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 669                 return; /* This is not an optprobe nor optimized */
 670
 671         op = container_of(p, struct optimized_kprobe, kp);
 672         if (!kprobe_optimized(p)) {
 673                 /* Unoptimized or unoptimizing case */
 674                 if (force && !list_empty(&op->list)) {
 675                         /*
 676                          * Only if this is unoptimizing kprobe and forced,
 677                          * forcibly unoptimize it. (No need to unoptimize
 678                          * unoptimized kprobe again :)
 679                          */
 680                         list_del_init(&op->list);
 681                         force_unoptimize_kprobe(op);
 682                 }
 683                 return;
 684         }
 685
 686         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 687         if (!list_empty(&op->list)) {
 688                 /* Dequeue from the optimization queue */
 689                 list_del_init(&op->list);
 690                 return;
 691         }
 692         /* Optimized kprobe case */
 693         if (force)
 694                 /* Forcibly update the code: this is a special case */
 695                 force_unoptimize_kprobe(op);
 696         else {
 697                 list_add(&op->list, &unoptimizing_list);
 698                 kick_kprobe_optimizer();
 699         }
 700 }
 701
 702 /* Cancel unoptimizing for reusing */
 703 static void reuse_unused_kprobe(struct kprobe *ap)
 704 {
 705         struct optimized_kprobe *op;
 706
 707         BUG_ON(!kprobe_unused(ap));
 708         /*
 709          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 710          * there is still a relative jump) and disabled.
 711          */
 712         op = container_of(ap, struct optimized_kprobe, kp);
 713         if (unlikely(list_empty(&op->list)))
 714                 printk(KERN_WARNING "Warning: found a stray unused "
 715                         "aggrprobe@%p\n", ap->addr);
 716         /* Enable the probe again */
 717         ap->flags &= ~KPROBE_FLAG_DISABLED;
 718         /* Optimize it again (remove from op->list) */
 719         BUG_ON(!kprobe_optready(ap));
 720         optimize_kprobe(ap);
 721 }
 722
 723 /* Remove optimized instructions */
 724 static void kill_optimized_kprobe(struct kprobe *p)
 725 {
 726         struct optimized_kprobe *op;
 727
 728         op = container_of(p, struct optimized_kprobe, kp);
 729         if (!list_empty(&op->list))
 730                 /* Dequeue from the (un)optimization queue */
 731                 list_del_init(&op->list);
 732         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 733
 734         if (kprobe_unused(p)) {
 735                 /* Enqueue if it is unused */
 736                 list_add(&op->list, &freeing_list);
 737                 /*
 738                  * Remove unused probes from the hash list. After waiting
 739                  * for synchronization, this probe is reclaimed.
 740                  * (reclaiming is done by do_free_cleaned_kprobes().)
 741                  */
 742                 hlist_del_rcu(&op->kp.hlist);
 743         }
 744
 745         /* Don't touch the code, because it is already freed. */
 746         arch_remove_optimized_kprobe(op);
 747 }
 748
 749 /* Try to prepare optimized instructions */
 750 static void prepare_optimized_kprobe(struct kprobe *p)
 751 {
 752         struct optimized_kprobe *op;
 753
 754         op = container_of(p, struct optimized_kprobe, kp);
 755         arch_prepare_optimized_kprobe(op, p);
 756 }
 757
 758 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 759 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 760 {
 761         struct optimized_kprobe *op;
 762
 763         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 764         if (!op)
 765                 return NULL;
 766
 767         INIT_LIST_HEAD(&op->list);
 768         op->kp.addr = p->addr;
 769         arch_prepare_optimized_kprobe(op, p);
 770
 771         return &op->kp;
 772 }
 773
 774 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 775
 776 /*
 777  * Prepare an optimized_kprobe and optimize it
 778  * NOTE: p must be a normal registered kprobe
 779  */
 780 static void try_to_optimize_kprobe(struct kprobe *p)
 781 {
 782         struct kprobe *ap;
 783         struct optimized_kprobe *op;
 784
 785         /* Impossible to optimize ftrace-based kprobe */
 786         if (kprobe_ftrace(p))
 787                 return;
 788
 789         /* For preparing optimization, jump_label_text_reserved() is called */
 790         jump_label_lock();
 791         mutex_lock(&text_mutex);
 792
 793         ap = alloc_aggr_kprobe(p);
 794         if (!ap)
 795                 goto out;
 796
 797         op = container_of(ap, struct optimized_kprobe, kp);
 798         if (!arch_prepared_optinsn(&op->optinsn)) {
 799                 /* If failed to setup optimizing, fallback to kprobe */
 800                 arch_remove_optimized_kprobe(op);
 801                 kfree(op);
 802                 goto out;
 803         }
 804
 805         init_aggr_kprobe(ap, p);
 806         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 807
 808 out:
 809         mutex_unlock(&text_mutex);
 810         jump_label_unlock();
 811 }
 812
 813 #ifdef CONFIG_SYSCTL
 814 static void optimize_all_kprobes(void)
 815 {
 816         struct hlist_head *head;
 817         struct kprobe *p;
 818         unsigned int i;
 819
 820         mutex_lock(&kprobe_mutex);
 821         /* If optimization is already allowed, just return */
 822         if (kprobes_allow_optimization)
 823                 goto out;
 824
 825         kprobes_allow_optimization = true;
 826         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 827                 head = &kprobe_table[i];
 828                 hlist_for_each_entry_rcu(p, head, hlist)
 829                         if (!kprobe_disabled(p))
 830                                 optimize_kprobe(p);
 831         }
 832         printk(KERN_INFO "Kprobes globally optimized\n");
 833 out:
 834         mutex_unlock(&kprobe_mutex);
 835 }
 836
 837 static void unoptimize_all_kprobes(void)
 838 {
 839         struct hlist_head *head;
 840         struct kprobe *p;
 841         unsigned int i;
 842
 843         mutex_lock(&kprobe_mutex);
 844         /* If optimization is already prohibited, just return */
 845         if (!kprobes_allow_optimization) {
 846                 mutex_unlock(&kprobe_mutex);
 847                 return;
 848         }
 849
 850         kprobes_allow_optimization = false;
 851         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 852                 head = &kprobe_table[i];
 853                 hlist_for_each_entry_rcu(p, head, hlist) {
 854                         if (!kprobe_disabled(p))
 855                                 unoptimize_kprobe(p, false);
 856                 }
 857         }
 858         mutex_unlock(&kprobe_mutex);
 859
 860         /* Wait for unoptimizing completion */
 861         wait_for_kprobe_optimizer();
 862         printk(KERN_INFO "Kprobes globally unoptimized\n");
 863 }
 864
 865 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 866 int sysctl_kprobes_optimization;
 867 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 868                                       void __user *buffer, size_t *length,
 869                                       loff_t *ppos)
 870 {
 871         int ret;
 872
 873         mutex_lock(&kprobe_sysctl_mutex);
 874         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 875         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 876
 877         if (sysctl_kprobes_optimization)
 878                 optimize_all_kprobes();
 879         else
 880                 unoptimize_all_kprobes();
 881         mutex_unlock(&kprobe_sysctl_mutex);
 882
 883         return ret;
 884 }
 885 #endif /* CONFIG_SYSCTL */
 886
 887 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 888 static void __arm_kprobe(struct kprobe *p)
 889 {
 890         struct kprobe *_p;
 891
 892         /* Check collision with other optimized kprobes */
 893         _p = get_optimized_kprobe((unsigned long)p->addr);
 894         if (unlikely(_p))
 895                 /* Fallback to unoptimized kprobe */
 896                 unoptimize_kprobe(_p, true);
 897
 898         arch_arm_kprobe(p);
 899         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 900 }
 901
 902 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 903 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 904 {
 905         struct kprobe *_p;
 906
 907         /* Try to unoptimize */
 908         unoptimize_kprobe(p, kprobes_all_disarmed);
 909
 910         if (!kprobe_queued(p)) {
 911                 arch_disarm_kprobe(p);
 912                 /* If another kprobe was blocked, optimize it. */
 913                 _p = get_optimized_kprobe((unsigned long)p->addr);
 914                 if (unlikely(_p) && reopt)
 915                         optimize_kprobe(_p);
 916         }
 917         /* TODO: reoptimize others after unoptimized this probe */
 918 }
 919
 920 #else /* !CONFIG_OPTPROBES */
 921
 922 #define optimize_kprobe(p)                      do {} while (0)
 923 #define unoptimize_kprobe(p, f)                 do {} while (0)
 924 #define kill_optimized_kprobe(p)                do {} while (0)
 925 #define prepare_optimized_kprobe(p)             do {} while (0)
 926 #define try_to_optimize_kprobe(p)               do {} while (0)
 927 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 928 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 929 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 930 #define wait_for_kprobe_optimizer()             do {} while (0)
 931
 932 /* There should be no unused kprobes can be reused without optimization */
 933 static void reuse_unused_kprobe(struct kprobe *ap)
 934 {
 935         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 936         BUG_ON(kprobe_unused(ap));
 937 }
 938
 939 static void free_aggr_kprobe(struct kprobe *p)
 940 {
 941         arch_remove_kprobe(p);
 942         kfree(p);
 943 }
 944
 945 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 946 {
 947         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 948 }
 949 #endif /* CONFIG_OPTPROBES */
 950
 951 #ifdef CONFIG_KPROBES_ON_FTRACE
 952 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 953         .func = kprobe_ftrace_handler,
 954         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 955 };
 956 static int kprobe_ftrace_enabled;
 957
 958 /* Must ensure p->addr is really on ftrace */
 959 static int prepare_kprobe(struct kprobe *p)
 960 {
 961         if (!kprobe_ftrace(p))
 962                 return arch_prepare_kprobe(p);
 963
 964         return arch_prepare_kprobe_ftrace(p);
 965 }
 966
 967 /* Caller must lock kprobe_mutex */
 968 static void arm_kprobe_ftrace(struct kprobe *p)
 969 {
 970         int ret;
 971
 972         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 973                                    (unsigned long)p->addr, 0, 0);
 974         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 975         kprobe_ftrace_enabled++;
 976         if (kprobe_ftrace_enabled == 1) {
 977                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 978                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 979         }
 980 }
 981
 982 /* Caller must lock kprobe_mutex */
 983 static void disarm_kprobe_ftrace(struct kprobe *p)
 984 {
 985         int ret;
 986
 987         kprobe_ftrace_enabled--;
 988         if (kprobe_ftrace_enabled == 0) {
 989                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 990                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 991         }
 992         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 993                            (unsigned long)p->addr, 1, 0);
 994         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 995 }
 996 #else   /* !CONFIG_KPROBES_ON_FTRACE */
 997 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 998 #define arm_kprobe_ftrace(p)    do {} while (0)
 999 #define disarm_kprobe_ftrace(p) do {} while (0)
1000 #endif
1001
1002 /* Arm a kprobe with text_mutex */
1003 static void arm_kprobe(struct kprobe *kp)
1004 {
1005         if (unlikely(kprobe_ftrace(kp))) {
1006                 arm_kprobe_ftrace(kp);
1007                 return;
1008         }
1009         /*
1010          * Here, since __arm_kprobe() doesn't use stop_machine(),
1011          * this doesn't cause deadlock on text_mutex. So, we don't
1012          * need get_online_cpus().
1013          */
1014         mutex_lock(&text_mutex);
1015         __arm_kprobe(kp);
1016         mutex_unlock(&text_mutex);
1017 }
1018
1019 /* Disarm a kprobe with text_mutex */
1020 static void disarm_kprobe(struct kprobe *kp, bool reopt)
1021 {
1022         if (unlikely(kprobe_ftrace(kp))) {
1023                 disarm_kprobe_ftrace(kp);
1024                 return;
1025         }
1026         /* Ditto */
1027         mutex_lock(&text_mutex);
1028         __disarm_kprobe(kp, reopt);
1029         mutex_unlock(&text_mutex);
1030 }
1031
1032 /*
1033  * Aggregate handlers for multiple kprobes support - these handlers
1034  * take care of invoking the individual kprobe handlers on p->list
1035  */
1036 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1037 {
1038         struct kprobe *kp;
1039
1040         list_for_each_entry_rcu(kp, &p->list, list) {
1041                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1042                         set_kprobe_instance(kp);
1043                         if (kp->pre_handler(kp, regs))
1044                                 return 1;
1045                 }
1046                 reset_kprobe_instance();
1047         }
1048         return 0;
1049 }
1050 NOKPROBE_SYMBOL(aggr_pre_handler);
1051
1052 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1053                               unsigned long flags)
1054 {
1055         struct kprobe *kp;
1056
1057         list_for_each_entry_rcu(kp, &p->list, list) {
1058                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1059                         set_kprobe_instance(kp);
1060                         kp->post_handler(kp, regs, flags);
1061                         reset_kprobe_instance();
1062                 }
1063         }
1064 }
1065 NOKPROBE_SYMBOL(aggr_post_handler);
1066
1067 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1068                               int trapnr)
1069 {
1070         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1071
1072         /*
1073          * if we faulted "during" the execution of a user specified
1074          * probe handler, invoke just that probe's fault handler
1075          */
1076         if (cur && cur->fault_handler) {
1077                 if (cur->fault_handler(cur, regs, trapnr))
1078                         return 1;
1079         }
1080         return 0;
1081 }
1082 NOKPROBE_SYMBOL(aggr_fault_handler);
1083
1084 static int aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1085 {
1086         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1087         int ret = 0;
1088
1089         if (cur && cur->break_handler) {
1090                 if (cur->break_handler(cur, regs))
1091                         ret = 1;
1092         }
1093         reset_kprobe_instance();
1094         return ret;
1095 }
1096 NOKPROBE_SYMBOL(aggr_break_handler);
1097
1098 /* Walks the list and increments nmissed count for multiprobe case */
1099 void kprobes_inc_nmissed_count(struct kprobe *p)
1100 {
1101         struct kprobe *kp;
1102         if (!kprobe_aggrprobe(p)) {
1103                 p->nmissed++;
1104         } else {
1105                 list_for_each_entry_rcu(kp, &p->list, list)
1106                         kp->nmissed++;
1107         }
1108         return;
1109 }
1110 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1111
1112 void recycle_rp_inst(struct kretprobe_instance *ri,
1113                      struct hlist_head *head)
1114 {
1115         struct kretprobe *rp = ri->rp;
1116
1117         /* remove rp inst off the rprobe_inst_table */
1118         hlist_del(&ri->hlist);
1119         INIT_HLIST_NODE(&ri->hlist);
1120         if (likely(rp)) {
1121                 raw_spin_lock(&rp->lock);
1122                 hlist_add_head(&ri->hlist, &rp->free_instances);
1123                 raw_spin_unlock(&rp->lock);
1124         } else
1125                 /* Unregistering */
1126                 hlist_add_head(&ri->hlist, head);
1127 }
1128 NOKPROBE_SYMBOL(recycle_rp_inst);
1129
1130 void kretprobe_hash_lock(struct task_struct *tsk,
1131                          struct hlist_head **head, unsigned long *flags)
1132 __acquires(hlist_lock)
1133 {
1134         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1135         raw_spinlock_t *hlist_lock;
1136
1137         *head = &kretprobe_inst_table[hash];
1138         hlist_lock = kretprobe_table_lock_ptr(hash);
1139         raw_spin_lock_irqsave(hlist_lock, *flags);
1140 }
1141 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1142
1143 static void kretprobe_table_lock(unsigned long hash,
1144                                  unsigned long *flags)
1145 __acquires(hlist_lock)
1146 {
1147         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1148         raw_spin_lock_irqsave(hlist_lock, *flags);
1149 }
1150 NOKPROBE_SYMBOL(kretprobe_table_lock);
1151
1152 void kretprobe_hash_unlock(struct task_struct *tsk,
1153                            unsigned long *flags)
1154 __releases(hlist_lock)
1155 {
1156         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1157         raw_spinlock_t *hlist_lock;
1158
1159         hlist_lock = kretprobe_table_lock_ptr(hash);
1160         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1161 }
1162 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1163
1164 static void kretprobe_table_unlock(unsigned long hash,
1165                                    unsigned long *flags)
1166 __releases(hlist_lock)
1167 {
1168         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1169         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1170 }
1171 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1172
1173 /*
1174  * This function is called from finish_task_switch when task tk becomes dead,
1175  * so that we can recycle any function-return probe instances associated
1176  * with this task. These left over instances represent probed functions
1177  * that have been called but will never return.
1178  */
1179 void kprobe_flush_task(struct task_struct *tk)
1180 {
1181         struct kretprobe_instance *ri;
1182         struct hlist_head *head, empty_rp;
1183         struct hlist_node *tmp;
1184         unsigned long hash, flags = 0;
1185
1186         if (unlikely(!kprobes_initialized))
1187                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1188                 return;
1189
1190         INIT_HLIST_HEAD(&empty_rp);
1191         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1192         head = &kretprobe_inst_table[hash];
1193         kretprobe_table_lock(hash, &flags);
1194         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1195                 if (ri->task == tk)
1196                         recycle_rp_inst(ri, &empty_rp);
1197         }
1198         kretprobe_table_unlock(hash, &flags);
1199         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1200                 hlist_del(&ri->hlist);
1201                 kfree(ri);
1202         }
1203 }
1204 NOKPROBE_SYMBOL(kprobe_flush_task);
1205
1206 static inline void free_rp_inst(struct kretprobe *rp)
1207 {
1208         struct kretprobe_instance *ri;
1209         struct hlist_node *next;
1210
1211         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1212                 hlist_del(&ri->hlist);
1213                 kfree(ri);
1214         }
1215 }
1216
1217 static void cleanup_rp_inst(struct kretprobe *rp)
1218 {
1219         unsigned long flags, hash;
1220         struct kretprobe_instance *ri;
1221         struct hlist_node *next;
1222         struct hlist_head *head;
1223
1224         /* No race here */
1225         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1226                 kretprobe_table_lock(hash, &flags);
1227                 head = &kretprobe_inst_table[hash];
1228                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1229                         if (ri->rp == rp)
1230                                 ri->rp = NULL;
1231                 }
1232                 kretprobe_table_unlock(hash, &flags);
1233         }
1234         free_rp_inst(rp);
1235 }
1236 NOKPROBE_SYMBOL(cleanup_rp_inst);
1237
1238 /*
1239 * Add the new probe to ap->list. Fail if this is the
1240 * second jprobe at the address - two jprobes can't coexist
1241 */
1242 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1243 {
1244         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1245
1246         if (p->break_handler || p->post_handler)
1247                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1248
1249         if (p->break_handler) {
1250                 if (ap->break_handler)
1251                         return -EEXIST;
1252                 list_add_tail_rcu(&p->list, &ap->list);
1253                 ap->break_handler = aggr_break_handler;
1254         } else
1255                 list_add_rcu(&p->list, &ap->list);
1256         if (p->post_handler && !ap->post_handler)
1257                 ap->post_handler = aggr_post_handler;
1258
1259         return 0;
1260 }
1261
1262 /*
1263  * Fill in the required fields of the "manager kprobe". Replace the
1264  * earlier kprobe in the hlist with the manager kprobe
1265  */
1266 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1267 {
1268         /* Copy p's insn slot to ap */
1269         copy_kprobe(p, ap);
1270         flush_insn_slot(ap);
1271         ap->addr = p->addr;
1272         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1273         ap->pre_handler = aggr_pre_handler;
1274         ap->fault_handler = aggr_fault_handler;
1275         /* We don't care the kprobe which has gone. */
1276         if (p->post_handler && !kprobe_gone(p))
1277                 ap->post_handler = aggr_post_handler;
1278         if (p->break_handler && !kprobe_gone(p))
1279                 ap->break_handler = aggr_break_handler;
1280
1281         INIT_LIST_HEAD(&ap->list);
1282         INIT_HLIST_NODE(&ap->hlist);
1283
1284         list_add_rcu(&p->list, &ap->list);
1285         hlist_replace_rcu(&p->hlist, &ap->hlist);
1286 }
1287
1288 /*
1289  * This is the second or subsequent kprobe at the address - handle
1290  * the intricacies
1291  */
1292 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1293 {
1294         int ret = 0;
1295         struct kprobe *ap = orig_p;
1296
1297         /* For preparing optimization, jump_label_text_reserved() is called */
1298         jump_label_lock();
1299         /*
1300          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1301          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1302          */
1303         get_online_cpus();
1304         mutex_lock(&text_mutex);
1305
1306         if (!kprobe_aggrprobe(orig_p)) {
1307                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1308                 ap = alloc_aggr_kprobe(orig_p);
1309                 if (!ap) {
1310                         ret = -ENOMEM;
1311                         goto out;
1312                 }
1313                 init_aggr_kprobe(ap, orig_p);
1314         } else if (kprobe_unused(ap))
1315                 /* This probe is going to die. Rescue it */
1316                 reuse_unused_kprobe(ap);
1317
1318         if (kprobe_gone(ap)) {
1319                 /*
1320                  * Attempting to insert new probe at the same location that
1321                  * had a probe in the module vaddr area which already
1322                  * freed. So, the instruction slot has already been
1323                  * released. We need a new slot for the new probe.
1324                  */
1325                 ret = arch_prepare_kprobe(ap);
1326                 if (ret)
1327                         /*
1328                          * Even if fail to allocate new slot, don't need to
1329                          * free aggr_probe. It will be used next time, or
1330                          * freed by unregister_kprobe.
1331                          */
1332                         goto out;
1333
1334                 /* Prepare optimized instructions if possible. */
1335                 prepare_optimized_kprobe(ap);
1336
1337                 /*
1338                  * Clear gone flag to prevent allocating new slot again, and
1339                  * set disabled flag because it is not armed yet.
1340                  */
1341                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1342                             | KPROBE_FLAG_DISABLED;
1343         }
1344
1345         /* Copy ap's insn slot to p */
1346         copy_kprobe(ap, p);
1347         ret = add_new_kprobe(ap, p);
1348
1349 out:
1350         mutex_unlock(&text_mutex);
1351         put_online_cpus();
1352         jump_label_unlock();
1353
1354         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1355                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1356                 if (!kprobes_all_disarmed)
1357                         /* Arm the breakpoint again. */
1358                         arm_kprobe(ap);
1359         }
1360         return ret;
1361 }
1362
1363 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1364 {
1365         /* The __kprobes marked functions and entry code must not be probed */
1366         return addr >= (unsigned long)__kprobes_text_start &&
1367                addr < (unsigned long)__kprobes_text_end;
1368 }
1369
1370 bool within_kprobe_blacklist(unsigned long addr)
1371 {
1372         struct kprobe_blacklist_entry *ent;
1373
1374         if (arch_within_kprobe_blacklist(addr))
1375                 return true;
1376         /*
1377          * If there exists a kprobe_blacklist, verify and
1378          * fail any probe registration in the prohibited area
1379          */
1380         list_for_each_entry(ent, &kprobe_blacklist, list) {
1381                 if (addr >= ent->start_addr && addr < ent->end_addr)
1382                         return true;
1383         }
1384
1385         return false;
1386 }
1387
1388 /*
1389  * If we have a symbol_name argument, look it up and add the offset field
1390  * to it. This way, we can specify a relative address to a symbol.
1391  * This returns encoded errors if it fails to look up symbol or invalid
1392  * combination of parameters.
1393  */
1394 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1395 {
1396         kprobe_opcode_t *addr = p->addr;
1397
1398         if ((p->symbol_name && p->addr) ||
1399             (!p->symbol_name && !p->addr))
1400                 goto invalid;
1401
1402         if (p->symbol_name) {
1403                 kprobe_lookup_name(p->symbol_name, addr);
1404                 if (!addr)
1405                         return ERR_PTR(-ENOENT);
1406         }
1407
1408         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1409         if (addr)
1410                 return addr;
1411
1412 invalid:
1413         return ERR_PTR(-EINVAL);
1414 }
1415
1416 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1417 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1418 {
1419         struct kprobe *ap, *list_p;
1420
1421         ap = get_kprobe(p->addr);
1422         if (unlikely(!ap))
1423                 return NULL;
1424
1425         if (p != ap) {
1426                 list_for_each_entry_rcu(list_p, &ap->list, list)
1427                         if (list_p == p)
1428                         /* kprobe p is a valid probe */
1429                                 goto valid;
1430                 return NULL;
1431         }
1432 valid:
1433         return ap;
1434 }
1435
1436 /* Return error if the kprobe is being re-registered */
1437 static inline int check_kprobe_rereg(struct kprobe *p)
1438 {
1439         int ret = 0;
1440
1441         mutex_lock(&kprobe_mutex);
1442         if (__get_valid_kprobe(p))
1443                 ret = -EINVAL;
1444         mutex_unlock(&kprobe_mutex);
1445
1446         return ret;
1447 }
1448
1449 int __weak arch_check_ftrace_location(struct kprobe *p)
1450 {
1451         unsigned long ftrace_addr;
1452
1453         ftrace_addr = ftrace_location((unsigned long)p->addr);
1454         if (ftrace_addr) {
1455 #ifdef CONFIG_KPROBES_ON_FTRACE
1456                 /* Given address is not on the instruction boundary */
1457                 if ((unsigned long)p->addr != ftrace_addr)
1458                         return -EILSEQ;
1459                 p->flags |= KPROBE_FLAG_FTRACE;
1460 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1461                 return -EINVAL;
1462 #endif
1463         }
1464         return 0;
1465 }
1466
1467 static int check_kprobe_address_safe(struct kprobe *p,
1468                                      struct module **probed_mod)
1469 {
1470         int ret;
1471
1472         ret = arch_check_ftrace_location(p);
1473         if (ret)
1474                 return ret;
1475         jump_label_lock();
1476         preempt_disable();
1477
1478         /* Ensure it is not in reserved area nor out of text */
1479         if (!kernel_text_address((unsigned long) p->addr) ||
1480             within_kprobe_blacklist((unsigned long) p->addr) ||
1481             jump_label_text_reserved(p->addr, p->addr)) {
1482                 ret = -EINVAL;
1483                 goto out;
1484         }
1485
1486         /* Check if are we probing a module */
1487         *probed_mod = __module_text_address((unsigned long) p->addr);
1488         if (*probed_mod) {
1489                 /*
1490                  * We must hold a refcount of the probed module while updating
1491                  * its code to prohibit unexpected unloading.
1492                  */
1493                 if (unlikely(!try_module_get(*probed_mod))) {
1494                         ret = -ENOENT;
1495                         goto out;
1496                 }
1497
1498                 /*
1499                  * If the module freed .init.text, we couldn't insert
1500                  * kprobes in there.
1501                  */
1502                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1503                     (*probed_mod)->state != MODULE_STATE_COMING) {
1504                         module_put(*probed_mod);
1505                         *probed_mod = NULL;
1506                         ret = -ENOENT;
1507                 }
1508         }
1509 out:
1510         preempt_enable();
1511         jump_label_unlock();
1512
1513         return ret;
1514 }
1515
1516 int register_kprobe(struct kprobe *p)
1517 {
1518         int ret;
1519         struct kprobe *old_p;
1520         struct module *probed_mod;
1521         kprobe_opcode_t *addr;
1522
1523         /* Adjust probe address from symbol */
1524         addr = kprobe_addr(p);
1525         if (IS_ERR(addr))
1526                 return PTR_ERR(addr);
1527         p->addr = addr;
1528
1529         ret = check_kprobe_rereg(p);
1530         if (ret)
1531                 return ret;
1532
1533         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1534         p->flags &= KPROBE_FLAG_DISABLED;
1535         p->nmissed = 0;
1536         INIT_LIST_HEAD(&p->list);
1537
1538         ret = check_kprobe_address_safe(p, &probed_mod);
1539         if (ret)
1540                 return ret;
1541
1542         mutex_lock(&kprobe_mutex);
1543
1544         old_p = get_kprobe(p->addr);
1545         if (old_p) {
1546                 /* Since this may unoptimize old_p, locking text_mutex. */
1547                 ret = register_aggr_kprobe(old_p, p);
1548                 goto out;
1549         }
1550
1551         mutex_lock(&text_mutex);        /* Avoiding text modification */
1552         ret = prepare_kprobe(p);
1553         mutex_unlock(&text_mutex);
1554         if (ret)
1555                 goto out;
1556
1557         INIT_HLIST_NODE(&p->hlist);
1558         hlist_add_head_rcu(&p->hlist,
1559                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1560
1561         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1562                 arm_kprobe(p);
1563
1564         /* Try to optimize kprobe */
1565         try_to_optimize_kprobe(p);
1566
1567 out:
1568         mutex_unlock(&kprobe_mutex);
1569
1570         if (probed_mod)
1571                 module_put(probed_mod);
1572
1573         return ret;
1574 }
1575 EXPORT_SYMBOL_GPL(register_kprobe);
1576
1577 /* Check if all probes on the aggrprobe are disabled */
1578 static int aggr_kprobe_disabled(struct kprobe *ap)
1579 {
1580         struct kprobe *kp;
1581
1582         list_for_each_entry_rcu(kp, &ap->list, list)
1583                 if (!kprobe_disabled(kp))
1584                         /*
1585                          * There is an active probe on the list.
1586                          * We can't disable this ap.
1587                          */
1588                         return 0;
1589
1590         return 1;
1591 }
1592
1593 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1594 static struct kprobe *__disable_kprobe(struct kprobe *p)
1595 {
1596         struct kprobe *orig_p;
1597
1598         /* Get an original kprobe for return */
1599         orig_p = __get_valid_kprobe(p);
1600         if (unlikely(orig_p == NULL))
1601                 return NULL;
1602
1603         if (!kprobe_disabled(p)) {
1604                 /* Disable probe if it is a child probe */
1605                 if (p != orig_p)
1606                         p->flags |= KPROBE_FLAG_DISABLED;
1607
1608                 /* Try to disarm and disable this/parent probe */
1609                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1610                         /*
1611                          * If kprobes_all_disarmed is set, orig_p
1612                          * should have already been disarmed, so
1613                          * skip unneed disarming process.
1614                          */
1615                         if (!kprobes_all_disarmed)
1616                                 disarm_kprobe(orig_p, true);
1617                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1618                 }
1619         }
1620
1621         return orig_p;
1622 }
1623
1624 /*
1625  * Unregister a kprobe without a scheduler synchronization.
1626  */
1627 static int __unregister_kprobe_top(struct kprobe *p)
1628 {
1629         struct kprobe *ap, *list_p;
1630
1631         /* Disable kprobe. This will disarm it if needed. */
1632         ap = __disable_kprobe(p);
1633         if (ap == NULL)
1634                 return -EINVAL;
1635
1636         if (ap == p)
1637                 /*
1638                  * This probe is an independent(and non-optimized) kprobe
1639                  * (not an aggrprobe). Remove from the hash list.
1640                  */
1641                 goto disarmed;
1642
1643         /* Following process expects this probe is an aggrprobe */
1644         WARN_ON(!kprobe_aggrprobe(ap));
1645
1646         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1647                 /*
1648                  * !disarmed could be happen if the probe is under delayed
1649                  * unoptimizing.
1650                  */
1651                 goto disarmed;
1652         else {
1653                 /* If disabling probe has special handlers, update aggrprobe */
1654                 if (p->break_handler && !kprobe_gone(p))
1655                         ap->break_handler = NULL;
1656                 if (p->post_handler && !kprobe_gone(p)) {
1657                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1658                                 if ((list_p != p) && (list_p->post_handler))
1659                                         goto noclean;
1660                         }
1661                         ap->post_handler = NULL;
1662                 }
1663 noclean:
1664                 /*
1665                  * Remove from the aggrprobe: this path will do nothing in
1666                  * __unregister_kprobe_bottom().
1667                  */
1668                 list_del_rcu(&p->list);
1669                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1670                         /*
1671                          * Try to optimize this probe again, because post
1672                          * handler may have been changed.
1673                          */
1674                         optimize_kprobe(ap);
1675         }
1676         return 0;
1677
1678 disarmed:
1679         BUG_ON(!kprobe_disarmed(ap));
1680         hlist_del_rcu(&ap->hlist);
1681         return 0;
1682 }
1683
1684 static void __unregister_kprobe_bottom(struct kprobe *p)
1685 {
1686         struct kprobe *ap;
1687
1688         if (list_empty(&p->list))
1689                 /* This is an independent kprobe */
1690                 arch_remove_kprobe(p);
1691         else if (list_is_singular(&p->list)) {
1692                 /* This is the last child of an aggrprobe */
1693                 ap = list_entry(p->list.next, struct kprobe, list);
1694                 list_del(&p->list);
1695                 free_aggr_kprobe(ap);
1696         }
1697         /* Otherwise, do nothing. */
1698 }
1699
1700 int register_kprobes(struct kprobe **kps, int num)
1701 {
1702         int i, ret = 0;
1703
1704         if (num <= 0)
1705                 return -EINVAL;
1706         for (i = 0; i < num; i++) {
1707                 ret = register_kprobe(kps[i]);
1708                 if (ret < 0) {
1709                         if (i > 0)
1710                                 unregister_kprobes(kps, i);
1711                         break;
1712                 }
1713         }
1714         return ret;
1715 }
1716 EXPORT_SYMBOL_GPL(register_kprobes);
1717
1718 void unregister_kprobe(struct kprobe *p)
1719 {
1720         unregister_kprobes(&p, 1);
1721 }
1722 EXPORT_SYMBOL_GPL(unregister_kprobe);
1723
1724 void unregister_kprobes(struct kprobe **kps, int num)
1725 {
1726         int i;
1727
1728         if (num <= 0)
1729                 return;
1730         mutex_lock(&kprobe_mutex);
1731         for (i = 0; i < num; i++)
1732                 if (__unregister_kprobe_top(kps[i]) < 0)
1733                         kps[i]->addr = NULL;
1734         mutex_unlock(&kprobe_mutex);
1735
1736         synchronize_sched();
1737         for (i = 0; i < num; i++)
1738                 if (kps[i]->addr)
1739                         __unregister_kprobe_bottom(kps[i]);
1740 }
1741 EXPORT_SYMBOL_GPL(unregister_kprobes);
1742
1743 int __weak __kprobes kprobe_exceptions_notify(struct notifier_block *self,
1744                                               unsigned long val, void *data)
1745 {
1746         return NOTIFY_DONE;
1747 }
1748
1749 static struct notifier_block kprobe_exceptions_nb = {
1750         .notifier_call = kprobe_exceptions_notify,
1751         .priority = 0x7fffffff /* we need to be notified first */
1752 };
1753
1754 unsigned long __weak arch_deref_entry_point(void *entry)
1755 {
1756         return (unsigned long)entry;
1757 }
1758
1759 int register_jprobes(struct jprobe **jps, int num)
1760 {
1761         struct jprobe *jp;
1762         int ret = 0, i;
1763
1764         if (num <= 0)
1765                 return -EINVAL;
1766         for (i = 0; i < num; i++) {
1767                 unsigned long addr, offset;
1768                 jp = jps[i];
1769                 addr = arch_deref_entry_point(jp->entry);
1770
1771                 /* Verify probepoint is a function entry point */
1772                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1773                     offset == 0) {
1774                         jp->kp.pre_handler = setjmp_pre_handler;
1775                         jp->kp.break_handler = longjmp_break_handler;
1776                         ret = register_kprobe(&jp->kp);
1777                 } else
1778                         ret = -EINVAL;
1779
1780                 if (ret < 0) {
1781                         if (i > 0)
1782                                 unregister_jprobes(jps, i);
1783                         break;
1784                 }
1785         }
1786         return ret;
1787 }
1788 EXPORT_SYMBOL_GPL(register_jprobes);
1789
1790 int register_jprobe(struct jprobe *jp)
1791 {
1792         return register_jprobes(&jp, 1);
1793 }
1794 EXPORT_SYMBOL_GPL(register_jprobe);
1795
1796 void unregister_jprobe(struct jprobe *jp)
1797 {
1798         unregister_jprobes(&jp, 1);
1799 }
1800 EXPORT_SYMBOL_GPL(unregister_jprobe);
1801
1802 void unregister_jprobes(struct jprobe **jps, int num)
1803 {
1804         int i;
1805
1806         if (num <= 0)
1807                 return;
1808         mutex_lock(&kprobe_mutex);
1809         for (i = 0; i < num; i++)
1810                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1811                         jps[i]->kp.addr = NULL;
1812         mutex_unlock(&kprobe_mutex);
1813
1814         synchronize_sched();
1815         for (i = 0; i < num; i++) {
1816                 if (jps[i]->kp.addr)
1817                         __unregister_kprobe_bottom(&jps[i]->kp);
1818         }
1819 }
1820 EXPORT_SYMBOL_GPL(unregister_jprobes);
1821
1822 #ifdef CONFIG_KRETPROBES
1823 /*
1824  * This kprobe pre_handler is registered with every kretprobe. When probe
1825  * hits it will set up the return probe.
1826  */
1827 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1828 {
1829         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1830         unsigned long hash, flags = 0;
1831         struct kretprobe_instance *ri;
1832
1833         /*
1834          * To avoid deadlocks, prohibit return probing in NMI contexts,
1835          * just skip the probe and increase the (inexact) 'nmissed'
1836          * statistical counter, so that the user is informed that
1837          * something happened:
1838          */
1839         if (unlikely(in_nmi())) {
1840                 rp->nmissed++;
1841                 return 0;
1842         }
1843
1844         /* TODO: consider to only swap the RA after the last pre_handler fired */
1845         hash = hash_ptr(current, KPROBE_HASH_BITS);
1846         raw_spin_lock_irqsave(&rp->lock, flags);
1847         if (!hlist_empty(&rp->free_instances)) {
1848                 ri = hlist_entry(rp->free_instances.first,
1849                                 struct kretprobe_instance, hlist);
1850                 hlist_del(&ri->hlist);
1851                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1852
1853                 ri->rp = rp;
1854                 ri->task = current;
1855
1856                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1857                         raw_spin_lock_irqsave(&rp->lock, flags);
1858                         hlist_add_head(&ri->hlist, &rp->free_instances);
1859                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1860                         return 0;
1861                 }
1862
1863                 arch_prepare_kretprobe(ri, regs);
1864
1865                 /* XXX(hch): why is there no hlist_move_head? */
1866                 INIT_HLIST_NODE(&ri->hlist);
1867                 kretprobe_table_lock(hash, &flags);
1868                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1869                 kretprobe_table_unlock(hash, &flags);
1870         } else {
1871                 rp->nmissed++;
1872                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1873         }
1874         return 0;
1875 }
1876 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1877
1878 int register_kretprobe(struct kretprobe *rp)
1879 {
1880         int ret = 0;
1881         struct kretprobe_instance *inst;
1882         int i;
1883         void *addr;
1884
1885         if (kretprobe_blacklist_size) {
1886                 addr = kprobe_addr(&rp->kp);
1887                 if (IS_ERR(addr))
1888                         return PTR_ERR(addr);
1889
1890                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1891                         if (kretprobe_blacklist[i].addr == addr)
1892                                 return -EINVAL;
1893                 }
1894         }
1895
1896         rp->kp.pre_handler = pre_handler_kretprobe;
1897         rp->kp.post_handler = NULL;
1898         rp->kp.fault_handler = NULL;
1899         rp->kp.break_handler = NULL;
1900
1901         /* Pre-allocate memory for max kretprobe instances */
1902         if (rp->maxactive <= 0) {
1903 #ifdef CONFIG_PREEMPT
1904                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1905 #else
1906                 rp->maxactive = num_possible_cpus();
1907 #endif
1908         }
1909         raw_spin_lock_init(&rp->lock);
1910         INIT_HLIST_HEAD(&rp->free_instances);
1911         for (i = 0; i < rp->maxactive; i++) {
1912                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1913                                rp->data_size, GFP_KERNEL);
1914                 if (inst == NULL) {
1915                         free_rp_inst(rp);
1916                         return -ENOMEM;
1917                 }
1918                 INIT_HLIST_NODE(&inst->hlist);
1919                 hlist_add_head(&inst->hlist, &rp->free_instances);
1920         }
1921
1922         rp->nmissed = 0;
1923         /* Establish function entry probe point */
1924         ret = register_kprobe(&rp->kp);
1925         if (ret != 0)
1926                 free_rp_inst(rp);
1927         return ret;
1928 }
1929 EXPORT_SYMBOL_GPL(register_kretprobe);
1930
1931 int register_kretprobes(struct kretprobe **rps, int num)
1932 {
1933         int ret = 0, i;
1934
1935         if (num <= 0)
1936                 return -EINVAL;
1937         for (i = 0; i < num; i++) {
1938                 ret = register_kretprobe(rps[i]);
1939                 if (ret < 0) {
1940                         if (i > 0)
1941                                 unregister_kretprobes(rps, i);
1942                         break;
1943                 }
1944         }
1945         return ret;
1946 }
1947 EXPORT_SYMBOL_GPL(register_kretprobes);
1948
1949 void unregister_kretprobe(struct kretprobe *rp)
1950 {
1951         unregister_kretprobes(&rp, 1);
1952 }
1953 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1954
1955 void unregister_kretprobes(struct kretprobe **rps, int num)
1956 {
1957         int i;
1958
1959         if (num <= 0)
1960                 return;
1961         mutex_lock(&kprobe_mutex);
1962         for (i = 0; i < num; i++)
1963                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1964                         rps[i]->kp.addr = NULL;
1965         mutex_unlock(&kprobe_mutex);
1966
1967         synchronize_sched();
1968         for (i = 0; i < num; i++) {
1969                 if (rps[i]->kp.addr) {
1970                         __unregister_kprobe_bottom(&rps[i]->kp);
1971                         cleanup_rp_inst(rps[i]);
1972                 }
1973         }
1974 }
1975 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1976
1977 #else /* CONFIG_KRETPROBES */
1978 int register_kretprobe(struct kretprobe *rp)
1979 {
1980         return -ENOSYS;
1981 }
1982 EXPORT_SYMBOL_GPL(register_kretprobe);
1983
1984 int register_kretprobes(struct kretprobe **rps, int num)
1985 {
1986         return -ENOSYS;
1987 }
1988 EXPORT_SYMBOL_GPL(register_kretprobes);
1989
1990 void unregister_kretprobe(struct kretprobe *rp)
1991 {
1992 }
1993 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1994
1995 void unregister_kretprobes(struct kretprobe **rps, int num)
1996 {
1997 }
1998 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1999
2000 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
2001 {
2002         return 0;
2003 }
2004 NOKPROBE_SYMBOL(pre_handler_kretprobe);
2005
2006 #endif /* CONFIG_KRETPROBES */
2007
2008 /* Set the kprobe gone and remove its instruction buffer. */
2009 static void kill_kprobe(struct kprobe *p)
2010 {
2011         struct kprobe *kp;
2012
2013         p->flags |= KPROBE_FLAG_GONE;
2014         if (kprobe_aggrprobe(p)) {
2015                 /*
2016                  * If this is an aggr_kprobe, we have to list all the
2017                  * chained probes and mark them GONE.
2018                  */
2019                 list_for_each_entry_rcu(kp, &p->list, list)
2020                         kp->flags |= KPROBE_FLAG_GONE;
2021                 p->post_handler = NULL;
2022                 p->break_handler = NULL;
2023                 kill_optimized_kprobe(p);
2024         }
2025         /*
2026          * Here, we can remove insn_slot safely, because no thread calls
2027          * the original probed function (which will be freed soon) any more.
2028          */
2029         arch_remove_kprobe(p);
2030 }
2031
2032 /* Disable one kprobe */
2033 int disable_kprobe(struct kprobe *kp)
2034 {
2035         int ret = 0;
2036
2037         mutex_lock(&kprobe_mutex);
2038
2039         /* Disable this kprobe */
2040         if (__disable_kprobe(kp) == NULL)
2041                 ret = -EINVAL;
2042
2043         mutex_unlock(&kprobe_mutex);
2044         return ret;
2045 }
2046 EXPORT_SYMBOL_GPL(disable_kprobe);
2047
2048 /* Enable one kprobe */
2049 int enable_kprobe(struct kprobe *kp)
2050 {
2051         int ret = 0;
2052         struct kprobe *p;
2053
2054         mutex_lock(&kprobe_mutex);
2055
2056         /* Check whether specified probe is valid. */
2057         p = __get_valid_kprobe(kp);
2058         if (unlikely(p == NULL)) {
2059                 ret = -EINVAL;
2060                 goto out;
2061         }
2062
2063         if (kprobe_gone(kp)) {
2064                 /* This kprobe has gone, we couldn't enable it. */
2065                 ret = -EINVAL;
2066                 goto out;
2067         }
2068
2069         if (p != kp)
2070                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2071
2072         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2073                 p->flags &= ~KPROBE_FLAG_DISABLED;
2074                 arm_kprobe(p);
2075         }
2076 out:
2077         mutex_unlock(&kprobe_mutex);
2078         return ret;
2079 }
2080 EXPORT_SYMBOL_GPL(enable_kprobe);
2081
2082 void dump_kprobe(struct kprobe *kp)
2083 {
2084         printk(KERN_WARNING "Dumping kprobe:\n");
2085         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2086                kp->symbol_name, kp->addr, kp->offset);
2087 }
2088 NOKPROBE_SYMBOL(dump_kprobe);
2089
2090 /*
2091  * Lookup and populate the kprobe_blacklist.
2092  *
2093  * Unlike the kretprobe blacklist, we'll need to determine
2094  * the range of addresses that belong to the said functions,
2095  * since a kprobe need not necessarily be at the beginning
2096  * of a function.
2097  */
2098 static int __init populate_kprobe_blacklist(unsigned long *start,
2099                                              unsigned long *end)
2100 {
2101         unsigned long *iter;
2102         struct kprobe_blacklist_entry *ent;
2103         unsigned long entry, offset = 0, size = 0;
2104
2105         for (iter = start; iter < end; iter++) {
2106                 entry = arch_deref_entry_point((void *)*iter);
2107
2108                 if (!kernel_text_address(entry) ||
2109                     !kallsyms_lookup_size_offset(entry, &size, &offset)) {
2110                         pr_err("Failed to find blacklist at %p\n",
2111                                 (void *)entry);
2112                         continue;
2113                 }
2114
2115                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2116                 if (!ent)
2117                         return -ENOMEM;
2118                 ent->start_addr = entry;
2119                 ent->end_addr = entry + size;
2120                 INIT_LIST_HEAD(&ent->list);
2121                 list_add_tail(&ent->list, &kprobe_blacklist);
2122         }
2123         return 0;
2124 }
2125
2126 /* Module notifier call back, checking kprobes on the module */
2127 static int kprobes_module_callback(struct notifier_block *nb,
2128                                    unsigned long val, void *data)
2129 {
2130         struct module *mod = data;
2131         struct hlist_head *head;
2132         struct kprobe *p;
2133         unsigned int i;
2134         int checkcore = (val == MODULE_STATE_GOING);
2135
2136         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2137                 return NOTIFY_DONE;
2138
2139         /*
2140          * When MODULE_STATE_GOING was notified, both of module .text and
2141          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2142          * notified, only .init.text section would be freed. We need to
2143          * disable kprobes which have been inserted in the sections.
2144          */
2145         mutex_lock(&kprobe_mutex);
2146         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2147                 head = &kprobe_table[i];
2148                 hlist_for_each_entry_rcu(p, head, hlist)
2149                         if (within_module_init((unsigned long)p->addr, mod) ||
2150                             (checkcore &&
2151                              within_module_core((unsigned long)p->addr, mod))) {
2152                                 /*
2153                                  * The vaddr this probe is installed will soon
2154                                  * be vfreed buy not synced to disk. Hence,
2155                                  * disarming the breakpoint isn't needed.
2156                                  */
2157                                 kill_kprobe(p);
2158                         }
2159         }
2160         mutex_unlock(&kprobe_mutex);
2161         return NOTIFY_DONE;
2162 }
2163
2164 static struct notifier_block kprobe_module_nb = {
2165         .notifier_call = kprobes_module_callback,
2166         .priority = 0
2167 };
2168
2169 /* Markers of _kprobe_blacklist section */
2170 extern unsigned long __start_kprobe_blacklist[];
2171 extern unsigned long __stop_kprobe_blacklist[];
2172
2173 static int __init init_kprobes(void)
2174 {
2175         int i, err = 0;
2176
2177         /* FIXME allocate the probe table, currently defined statically */
2178         /* initialize all list heads */
2179         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2180                 INIT_HLIST_HEAD(&kprobe_table[i]);
2181                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2182                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2183         }
2184
2185         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2186                                         __stop_kprobe_blacklist);
2187         if (err) {
2188                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2189                 pr_err("Please take care of using kprobes.\n");
2190         }
2191
2192         if (kretprobe_blacklist_size) {
2193                 /* lookup the function address from its name */
2194                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2195                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2196                                            kretprobe_blacklist[i].addr);
2197                         if (!kretprobe_blacklist[i].addr)
2198                                 printk("kretprobe: lookup failed: %s\n",
2199                                        kretprobe_blacklist[i].name);
2200                 }
2201         }
2202
2203 #if defined(CONFIG_OPTPROBES)
2204 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2205         /* Init kprobe_optinsn_slots */
2206         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2207 #endif
2208         /* By default, kprobes can be optimized */
2209         kprobes_allow_optimization = true;
2210 #endif
2211
2212         /* By default, kprobes are armed */
2213         kprobes_all_disarmed = false;
2214
2215         err = arch_init_kprobes();
2216         if (!err)
2217                 err = register_die_notifier(&kprobe_exceptions_nb);
2218         if (!err)
2219                 err = register_module_notifier(&kprobe_module_nb);
2220
2221         kprobes_initialized = (err == 0);
2222
2223         if (!err)
2224                 init_test_probes();
2225         return err;
2226 }
2227
2228 #ifdef CONFIG_DEBUG_FS
2229 static void report_probe(struct seq_file *pi, struct kprobe *p,
2230                 const char *sym, int offset, char *modname, struct kprobe *pp)
2231 {
2232         char *kprobe_type;
2233
2234         if (p->pre_handler == pre_handler_kretprobe)
2235                 kprobe_type = "r";
2236         else if (p->pre_handler == setjmp_pre_handler)
2237                 kprobe_type = "j";
2238         else
2239                 kprobe_type = "k";
2240
2241         if (sym)
2242                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2243                         p->addr, kprobe_type, sym, offset,
2244                         (modname ? modname : " "));
2245         else
2246                 seq_printf(pi, "%p  %s  %p ",
2247                         p->addr, kprobe_type, p->addr);
2248
2249         if (!pp)
2250                 pp = p;
2251         seq_printf(pi, "%s%s%s%s\n",
2252                 (kprobe_gone(p) ? "[GONE]" : ""),
2253                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2254                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2255                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2256 }
2257
2258 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2259 {
2260         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2261 }
2262
2263 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2264 {
2265         (*pos)++;
2266         if (*pos >= KPROBE_TABLE_SIZE)
2267                 return NULL;
2268         return pos;
2269 }
2270
2271 static void kprobe_seq_stop(struct seq_file *f, void *v)
2272 {
2273         /* Nothing to do */
2274 }
2275
2276 static int show_kprobe_addr(struct seq_file *pi, void *v)
2277 {
2278         struct hlist_head *head;
2279         struct kprobe *p, *kp;
2280         const char *sym = NULL;
2281         unsigned int i = *(loff_t *) v;
2282         unsigned long offset = 0;
2283         char *modname, namebuf[KSYM_NAME_LEN];
2284
2285         head = &kprobe_table[i];
2286         preempt_disable();
2287         hlist_for_each_entry_rcu(p, head, hlist) {
2288                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2289                                         &offset, &modname, namebuf);
2290                 if (kprobe_aggrprobe(p)) {
2291                         list_for_each_entry_rcu(kp, &p->list, list)
2292                                 report_probe(pi, kp, sym, offset, modname, p);
2293                 } else
2294                         report_probe(pi, p, sym, offset, modname, NULL);
2295         }
2296         preempt_enable();
2297         return 0;
2298 }
2299
2300 static const struct seq_operations kprobes_seq_ops = {
2301         .start = kprobe_seq_start,
2302         .next  = kprobe_seq_next,
2303         .stop  = kprobe_seq_stop,
2304         .show  = show_kprobe_addr
2305 };
2306
2307 static int kprobes_open(struct inode *inode, struct file *filp)
2308 {
2309         return seq_open(filp, &kprobes_seq_ops);
2310 }
2311
2312 static const struct file_operations debugfs_kprobes_operations = {
2313         .open           = kprobes_open,
2314         .read           = seq_read,
2315         .llseek         = seq_lseek,
2316         .release        = seq_release,
2317 };
2318
2319 /* kprobes/blacklist -- shows which functions can not be probed */
2320 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2321 {
2322         return seq_list_start(&kprobe_blacklist, *pos);
2323 }
2324
2325 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2326 {
2327         return seq_list_next(v, &kprobe_blacklist, pos);
2328 }
2329
2330 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2331 {
2332         struct kprobe_blacklist_entry *ent =
2333                 list_entry(v, struct kprobe_blacklist_entry, list);
2334
2335         seq_printf(m, "0x%p-0x%p\t%ps\n", (void *)ent->start_addr,
2336                    (void *)ent->end_addr, (void *)ent->start_addr);
2337         return 0;
2338 }
2339
2340 static const struct seq_operations kprobe_blacklist_seq_ops = {
2341         .start = kprobe_blacklist_seq_start,
2342         .next  = kprobe_blacklist_seq_next,
2343         .stop  = kprobe_seq_stop,       /* Reuse void function */
2344         .show  = kprobe_blacklist_seq_show,
2345 };
2346
2347 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2348 {
2349         return seq_open(filp, &kprobe_blacklist_seq_ops);
2350 }
2351
2352 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2353         .open           = kprobe_blacklist_open,
2354         .read           = seq_read,
2355         .llseek         = seq_lseek,
2356         .release        = seq_release,
2357 };
2358
2359 static void arm_all_kprobes(void)
2360 {
2361         struct hlist_head *head;
2362         struct kprobe *p;
2363         unsigned int i;
2364
2365         mutex_lock(&kprobe_mutex);
2366
2367         /* If kprobes are armed, just return */
2368         if (!kprobes_all_disarmed)
2369                 goto already_enabled;
2370
2371         /*
2372          * optimize_kprobe() called by arm_kprobe() checks
2373          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2374          * arm_kprobe.
2375          */
2376         kprobes_all_disarmed = false;
2377         /* Arming kprobes doesn't optimize kprobe itself */
2378         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2379                 head = &kprobe_table[i];
2380                 hlist_for_each_entry_rcu(p, head, hlist)
2381                         if (!kprobe_disabled(p))
2382                                 arm_kprobe(p);
2383         }
2384
2385         printk(KERN_INFO "Kprobes globally enabled\n");
2386
2387 already_enabled:
2388         mutex_unlock(&kprobe_mutex);
2389         return;
2390 }
2391
2392 static void disarm_all_kprobes(void)
2393 {
2394         struct hlist_head *head;
2395         struct kprobe *p;
2396         unsigned int i;
2397
2398         mutex_lock(&kprobe_mutex);
2399
2400         /* If kprobes are already disarmed, just return */
2401         if (kprobes_all_disarmed) {
2402                 mutex_unlock(&kprobe_mutex);
2403                 return;
2404         }
2405
2406         kprobes_all_disarmed = true;
2407         printk(KERN_INFO "Kprobes globally disabled\n");
2408
2409         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2410                 head = &kprobe_table[i];
2411                 hlist_for_each_entry_rcu(p, head, hlist) {
2412                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2413                                 disarm_kprobe(p, false);
2414                 }
2415         }
2416         mutex_unlock(&kprobe_mutex);
2417
2418         /* Wait for disarming all kprobes by optimizer */
2419         wait_for_kprobe_optimizer();
2420 }
2421
2422 /*
2423  * XXX: The debugfs bool file interface doesn't allow for callbacks
2424  * when the bool state is switched. We can reuse that facility when
2425  * available
2426  */
2427 static ssize_t read_enabled_file_bool(struct file *file,
2428                char __user *user_buf, size_t count, loff_t *ppos)
2429 {
2430         char buf[3];
2431
2432         if (!kprobes_all_disarmed)
2433                 buf[0] = '1';
2434         else
2435                 buf[0] = '0';
2436         buf[1] = '\n';
2437         buf[2] = 0x00;
2438         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2439 }
2440
2441 static ssize_t write_enabled_file_bool(struct file *file,
2442                const char __user *user_buf, size_t count, loff_t *ppos)
2443 {
2444         char buf[32];
2445         size_t buf_size;
2446
2447         buf_size = min(count, (sizeof(buf)-1));
2448         if (copy_from_user(buf, user_buf, buf_size))
2449                 return -EFAULT;
2450
2451         buf[buf_size] = '\0';
2452         switch (buf[0]) {
2453         case 'y':
2454         case 'Y':
2455         case '1':
2456                 arm_all_kprobes();
2457                 break;
2458         case 'n':
2459         case 'N':
2460         case '0':
2461                 disarm_all_kprobes();
2462                 break;
2463         default:
2464                 return -EINVAL;
2465         }
2466
2467         return count;
2468 }
2469
2470 static const struct file_operations fops_kp = {
2471         .read =         read_enabled_file_bool,
2472         .write =        write_enabled_file_bool,
2473         .llseek =       default_llseek,
2474 };
2475
2476 static int __init debugfs_kprobe_init(void)
2477 {
2478         struct dentry *dir, *file;
2479         unsigned int value = 1;
2480
2481         dir = debugfs_create_dir("kprobes", NULL);
2482         if (!dir)
2483                 return -ENOMEM;
2484
2485         file = debugfs_create_file("list", 0444, dir, NULL,
2486                                 &debugfs_kprobes_operations);
2487         if (!file)
2488                 goto error;
2489
2490         file = debugfs_create_file("enabled", 0600, dir,
2491                                         &value, &fops_kp);
2492         if (!file)
2493                 goto error;
2494
2495         file = debugfs_create_file("blacklist", 0444, dir, NULL,
2496                                 &debugfs_kprobe_blacklist_ops);
2497         if (!file)
2498                 goto error;
2499
2500         return 0;
2501
2502 error:
2503         debugfs_remove(dir);
2504         return -ENOMEM;
2505 }
2506
2507 late_initcall(debugfs_kprobe_init);
2508 #endif /* CONFIG_DEBUG_FS */
2509
2510 module_init(init_kprobes);
2511
2512 /* defined in arch/.../kernel/kprobes.c */
2513 EXPORT_SYMBOL_GPL(jprobe_return);