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