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