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