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         mutex_lock(&kprobe_mutex);
 565         /* Lock modules while optimizing kprobes */
 566         mutex_lock(&module_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(&module_mutex);
 590         mutex_unlock(&kprobe_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         /* Impossible to optimize ftrace-based kprobe */
 763         if (kprobe_ftrace(p))
 764                 return;
 765
 766         /* For preparing optimization, jump_label_text_reserved() is called */
 767         jump_label_lock();
 768         mutex_lock(&text_mutex);
 769
 770         ap = alloc_aggr_kprobe(p);
 771         if (!ap)
 772                 goto out;
 773
 774         op = container_of(ap, struct optimized_kprobe, kp);
 775         if (!arch_prepared_optinsn(&op->optinsn)) {
 776                 /* If failed to setup optimizing, fallback to kprobe */
 777                 arch_remove_optimized_kprobe(op);
 778                 kfree(op);
 779                 goto out;
 780         }
 781
 782         init_aggr_kprobe(ap, p);
 783         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 784
 785 out:
 786         mutex_unlock(&text_mutex);
 787         jump_label_unlock();
 788 }
 789
 790 #ifdef CONFIG_SYSCTL
 791 /* This should be called with kprobe_mutex locked */
 792 static void __kprobes optimize_all_kprobes(void)
 793 {
 794         struct hlist_head *head;
 795         struct hlist_node *node;
 796         struct kprobe *p;
 797         unsigned int i;
 798
 799         /* If optimization is already allowed, just return */
 800         if (kprobes_allow_optimization)
 801                 return;
 802
 803         kprobes_allow_optimization = true;
 804         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 805                 head = &kprobe_table[i];
 806                 hlist_for_each_entry_rcu(p, node, head, hlist)
 807                         if (!kprobe_disabled(p))
 808                                 optimize_kprobe(p);
 809         }
 810         printk(KERN_INFO "Kprobes globally optimized\n");
 811 }
 812
 813 /* This should be called with kprobe_mutex locked */
 814 static void __kprobes unoptimize_all_kprobes(void)
 815 {
 816         struct hlist_head *head;
 817         struct hlist_node *node;
 818         struct kprobe *p;
 819         unsigned int i;
 820
 821         /* If optimization is already prohibited, just return */
 822         if (!kprobes_allow_optimization)
 823                 return;
 824
 825         kprobes_allow_optimization = false;
 826         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 827                 head = &kprobe_table[i];
 828                 hlist_for_each_entry_rcu(p, node, head, hlist) {
 829                         if (!kprobe_disabled(p))
 830                                 unoptimize_kprobe(p, false);
 831                 }
 832         }
 833         /* Wait for unoptimizing completion */
 834         wait_for_kprobe_optimizer();
 835         printk(KERN_INFO "Kprobes globally unoptimized\n");
 836 }
 837
 838 int sysctl_kprobes_optimization;
 839 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 840                                       void __user *buffer, size_t *length,
 841                                       loff_t *ppos)
 842 {
 843         int ret;
 844
 845         mutex_lock(&kprobe_mutex);
 846         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 847         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 848
 849         if (sysctl_kprobes_optimization)
 850                 optimize_all_kprobes();
 851         else
 852                 unoptimize_all_kprobes();
 853         mutex_unlock(&kprobe_mutex);
 854
 855         return ret;
 856 }
 857 #endif /* CONFIG_SYSCTL */
 858
 859 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 860 static void __kprobes __arm_kprobe(struct kprobe *p)
 861 {
 862         struct kprobe *_p;
 863
 864         /* Check collision with other optimized kprobes */
 865         _p = get_optimized_kprobe((unsigned long)p->addr);
 866         if (unlikely(_p))
 867                 /* Fallback to unoptimized kprobe */
 868                 unoptimize_kprobe(_p, true);
 869
 870         arch_arm_kprobe(p);
 871         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 872 }
 873
 874 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 875 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
 876 {
 877         struct kprobe *_p;
 878
 879         unoptimize_kprobe(p, false);    /* Try to unoptimize */
 880
 881         if (!kprobe_queued(p)) {
 882                 arch_disarm_kprobe(p);
 883                 /* If another kprobe was blocked, optimize it. */
 884                 _p = get_optimized_kprobe((unsigned long)p->addr);
 885                 if (unlikely(_p) && reopt)
 886                         optimize_kprobe(_p);
 887         }
 888         /* TODO: reoptimize others after unoptimized this probe */
 889 }
 890
 891 #else /* !CONFIG_OPTPROBES */
 892
 893 #define optimize_kprobe(p)                      do {} while (0)
 894 #define unoptimize_kprobe(p, f)                 do {} while (0)
 895 #define kill_optimized_kprobe(p)                do {} while (0)
 896 #define prepare_optimized_kprobe(p)             do {} while (0)
 897 #define try_to_optimize_kprobe(p)               do {} while (0)
 898 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 899 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 900 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 901 #define wait_for_kprobe_optimizer()             do {} while (0)
 902
 903 /* There should be no unused kprobes can be reused without optimization */
 904 static void reuse_unused_kprobe(struct kprobe *ap)
 905 {
 906         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 907         BUG_ON(kprobe_unused(ap));
 908 }
 909
 910 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 911 {
 912         arch_remove_kprobe(p);
 913         kfree(p);
 914 }
 915
 916 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 917 {
 918         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 919 }
 920 #endif /* CONFIG_OPTPROBES */
 921
 922 #ifdef KPROBES_CAN_USE_FTRACE
 923 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 924         .func = kprobe_ftrace_handler,
 925         .flags = FTRACE_OPS_FL_SAVE_REGS,
 926 };
 927 static int kprobe_ftrace_enabled;
 928
 929 /* Must ensure p->addr is really on ftrace */
 930 static int __kprobes prepare_kprobe(struct kprobe *p)
 931 {
 932         if (!kprobe_ftrace(p))
 933                 return arch_prepare_kprobe(p);
 934
 935         return arch_prepare_kprobe_ftrace(p);
 936 }
 937
 938 /* Caller must lock kprobe_mutex */
 939 static void __kprobes arm_kprobe_ftrace(struct kprobe *p)
 940 {
 941         int ret;
 942
 943         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 944                                    (unsigned long)p->addr, 0, 0);
 945         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 946         kprobe_ftrace_enabled++;
 947         if (kprobe_ftrace_enabled == 1) {
 948                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 949                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 950         }
 951 }
 952
 953 /* Caller must lock kprobe_mutex */
 954 static void __kprobes disarm_kprobe_ftrace(struct kprobe *p)
 955 {
 956         int ret;
 957
 958         kprobe_ftrace_enabled--;
 959         if (kprobe_ftrace_enabled == 0) {
 960                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 961                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 962         }
 963         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 964                            (unsigned long)p->addr, 1, 0);
 965         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 966 }
 967 #else   /* !KPROBES_CAN_USE_FTRACE */
 968 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 969 #define arm_kprobe_ftrace(p)    do {} while (0)
 970 #define disarm_kprobe_ftrace(p) do {} while (0)
 971 #endif
 972
 973 /* Arm a kprobe with text_mutex */
 974 static void __kprobes arm_kprobe(struct kprobe *kp)
 975 {
 976         if (unlikely(kprobe_ftrace(kp))) {
 977                 arm_kprobe_ftrace(kp);
 978                 return;
 979         }
 980         /*
 981          * Here, since __arm_kprobe() doesn't use stop_machine(),
 982          * this doesn't cause deadlock on text_mutex. So, we don't
 983          * need get_online_cpus().
 984          */
 985         mutex_lock(&text_mutex);
 986         __arm_kprobe(kp);
 987         mutex_unlock(&text_mutex);
 988 }
 989
 990 /* Disarm a kprobe with text_mutex */
 991 static void __kprobes disarm_kprobe(struct kprobe *kp, bool reopt)
 992 {
 993         if (unlikely(kprobe_ftrace(kp))) {
 994                 disarm_kprobe_ftrace(kp);
 995                 return;
 996         }
 997         /* Ditto */
 998         mutex_lock(&text_mutex);
 999         __disarm_kprobe(kp, reopt);
1000         mutex_unlock(&text_mutex);
1001 }
1002
1003 /*
1004  * Aggregate handlers for multiple kprobes support - these handlers
1005  * take care of invoking the individual kprobe handlers on p->list
1006  */
1007 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1008 {
1009         struct kprobe *kp;
1010
1011         list_for_each_entry_rcu(kp, &p->list, list) {
1012                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1013                         set_kprobe_instance(kp);
1014                         if (kp->pre_handler(kp, regs))
1015                                 return 1;
1016                 }
1017                 reset_kprobe_instance();
1018         }
1019         return 0;
1020 }
1021
1022 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1023                                         unsigned long flags)
1024 {
1025         struct kprobe *kp;
1026
1027         list_for_each_entry_rcu(kp, &p->list, list) {
1028                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1029                         set_kprobe_instance(kp);
1030                         kp->post_handler(kp, regs, flags);
1031                         reset_kprobe_instance();
1032                 }
1033         }
1034 }
1035
1036 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1037                                         int trapnr)
1038 {
1039         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1040
1041         /*
1042          * if we faulted "during" the execution of a user specified
1043          * probe handler, invoke just that probe's fault handler
1044          */
1045         if (cur && cur->fault_handler) {
1046                 if (cur->fault_handler(cur, regs, trapnr))
1047                         return 1;
1048         }
1049         return 0;
1050 }
1051
1052 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1053 {
1054         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1055         int ret = 0;
1056
1057         if (cur && cur->break_handler) {
1058                 if (cur->break_handler(cur, regs))
1059                         ret = 1;
1060         }
1061         reset_kprobe_instance();
1062         return ret;
1063 }
1064
1065 /* Walks the list and increments nmissed count for multiprobe case */
1066 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1067 {
1068         struct kprobe *kp;
1069         if (!kprobe_aggrprobe(p)) {
1070                 p->nmissed++;
1071         } else {
1072                 list_for_each_entry_rcu(kp, &p->list, list)
1073                         kp->nmissed++;
1074         }
1075         return;
1076 }
1077
1078 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1079                                 struct hlist_head *head)
1080 {
1081         struct kretprobe *rp = ri->rp;
1082
1083         /* remove rp inst off the rprobe_inst_table */
1084         hlist_del(&ri->hlist);
1085         INIT_HLIST_NODE(&ri->hlist);
1086         if (likely(rp)) {
1087                 raw_spin_lock(&rp->lock);
1088                 hlist_add_head(&ri->hlist, &rp->free_instances);
1089                 raw_spin_unlock(&rp->lock);
1090         } else
1091                 /* Unregistering */
1092                 hlist_add_head(&ri->hlist, head);
1093 }
1094
1095 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1096                          struct hlist_head **head, unsigned long *flags)
1097 __acquires(hlist_lock)
1098 {
1099         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1100         raw_spinlock_t *hlist_lock;
1101
1102         *head = &kretprobe_inst_table[hash];
1103         hlist_lock = kretprobe_table_lock_ptr(hash);
1104         raw_spin_lock_irqsave(hlist_lock, *flags);
1105 }
1106
1107 static void __kprobes kretprobe_table_lock(unsigned long hash,
1108         unsigned long *flags)
1109 __acquires(hlist_lock)
1110 {
1111         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1112         raw_spin_lock_irqsave(hlist_lock, *flags);
1113 }
1114
1115 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1116         unsigned long *flags)
1117 __releases(hlist_lock)
1118 {
1119         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1120         raw_spinlock_t *hlist_lock;
1121
1122         hlist_lock = kretprobe_table_lock_ptr(hash);
1123         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1124 }
1125
1126 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1127        unsigned long *flags)
1128 __releases(hlist_lock)
1129 {
1130         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1131         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1132 }
1133
1134 /*
1135  * This function is called from finish_task_switch when task tk becomes dead,
1136  * so that we can recycle any function-return probe instances associated
1137  * with this task. These left over instances represent probed functions
1138  * that have been called but will never return.
1139  */
1140 void __kprobes kprobe_flush_task(struct task_struct *tk)
1141 {
1142         struct kretprobe_instance *ri;
1143         struct hlist_head *head, empty_rp;
1144         struct hlist_node *node, *tmp;
1145         unsigned long hash, flags = 0;
1146
1147         if (unlikely(!kprobes_initialized))
1148                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1149                 return;
1150
1151         INIT_HLIST_HEAD(&empty_rp);
1152         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1153         head = &kretprobe_inst_table[hash];
1154         kretprobe_table_lock(hash, &flags);
1155         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1156                 if (ri->task == tk)
1157                         recycle_rp_inst(ri, &empty_rp);
1158         }
1159         kretprobe_table_unlock(hash, &flags);
1160         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1161                 hlist_del(&ri->hlist);
1162                 kfree(ri);
1163         }
1164 }
1165
1166 static inline void free_rp_inst(struct kretprobe *rp)
1167 {
1168         struct kretprobe_instance *ri;
1169         struct hlist_node *pos, *next;
1170
1171         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1172                 hlist_del(&ri->hlist);
1173                 kfree(ri);
1174         }
1175 }
1176
1177 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1178 {
1179         unsigned long flags, hash;
1180         struct kretprobe_instance *ri;
1181         struct hlist_node *pos, *next;
1182         struct hlist_head *head;
1183
1184         /* No race here */
1185         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1186                 kretprobe_table_lock(hash, &flags);
1187                 head = &kretprobe_inst_table[hash];
1188                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1189                         if (ri->rp == rp)
1190                                 ri->rp = NULL;
1191                 }
1192                 kretprobe_table_unlock(hash, &flags);
1193         }
1194         free_rp_inst(rp);
1195 }
1196
1197 /*
1198 * Add the new probe to ap->list. Fail if this is the
1199 * second jprobe at the address - two jprobes can't coexist
1200 */
1201 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1202 {
1203         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1204
1205         if (p->break_handler || p->post_handler)
1206                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1207
1208         if (p->break_handler) {
1209                 if (ap->break_handler)
1210                         return -EEXIST;
1211                 list_add_tail_rcu(&p->list, &ap->list);
1212                 ap->break_handler = aggr_break_handler;
1213         } else
1214                 list_add_rcu(&p->list, &ap->list);
1215         if (p->post_handler && !ap->post_handler)
1216                 ap->post_handler = aggr_post_handler;
1217
1218         return 0;
1219 }
1220
1221 /*
1222  * Fill in the required fields of the "manager kprobe". Replace the
1223  * earlier kprobe in the hlist with the manager kprobe
1224  */
1225 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1226 {
1227         /* Copy p's insn slot to ap */
1228         copy_kprobe(p, ap);
1229         flush_insn_slot(ap);
1230         ap->addr = p->addr;
1231         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1232         ap->pre_handler = aggr_pre_handler;
1233         ap->fault_handler = aggr_fault_handler;
1234         /* We don't care the kprobe which has gone. */
1235         if (p->post_handler && !kprobe_gone(p))
1236                 ap->post_handler = aggr_post_handler;
1237         if (p->break_handler && !kprobe_gone(p))
1238                 ap->break_handler = aggr_break_handler;
1239
1240         INIT_LIST_HEAD(&ap->list);
1241         INIT_HLIST_NODE(&ap->hlist);
1242
1243         list_add_rcu(&p->list, &ap->list);
1244         hlist_replace_rcu(&p->hlist, &ap->hlist);
1245 }
1246
1247 /*
1248  * This is the second or subsequent kprobe at the address - handle
1249  * the intricacies
1250  */
1251 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1252                                           struct kprobe *p)
1253 {
1254         int ret = 0;
1255         struct kprobe *ap = orig_p;
1256
1257         /* For preparing optimization, jump_label_text_reserved() is called */
1258         jump_label_lock();
1259         /*
1260          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1261          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1262          */
1263         get_online_cpus();
1264         mutex_lock(&text_mutex);
1265
1266         if (!kprobe_aggrprobe(orig_p)) {
1267                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1268                 ap = alloc_aggr_kprobe(orig_p);
1269                 if (!ap) {
1270                         ret = -ENOMEM;
1271                         goto out;
1272                 }
1273                 init_aggr_kprobe(ap, orig_p);
1274         } else if (kprobe_unused(ap))
1275                 /* This probe is going to die. Rescue it */
1276                 reuse_unused_kprobe(ap);
1277
1278         if (kprobe_gone(ap)) {
1279                 /*
1280                  * Attempting to insert new probe at the same location that
1281                  * had a probe in the module vaddr area which already
1282                  * freed. So, the instruction slot has already been
1283                  * released. We need a new slot for the new probe.
1284                  */
1285                 ret = arch_prepare_kprobe(ap);
1286                 if (ret)
1287                         /*
1288                          * Even if fail to allocate new slot, don't need to
1289                          * free aggr_probe. It will be used next time, or
1290                          * freed by unregister_kprobe.
1291                          */
1292                         goto out;
1293
1294                 /* Prepare optimized instructions if possible. */
1295                 prepare_optimized_kprobe(ap);
1296
1297                 /*
1298                  * Clear gone flag to prevent allocating new slot again, and
1299                  * set disabled flag because it is not armed yet.
1300                  */
1301                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1302                             | KPROBE_FLAG_DISABLED;
1303         }
1304
1305         /* Copy ap's insn slot to p */
1306         copy_kprobe(ap, p);
1307         ret = add_new_kprobe(ap, p);
1308
1309 out:
1310         mutex_unlock(&text_mutex);
1311         put_online_cpus();
1312         jump_label_unlock();
1313
1314         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1315                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1316                 if (!kprobes_all_disarmed)
1317                         /* Arm the breakpoint again. */
1318                         arm_kprobe(ap);
1319         }
1320         return ret;
1321 }
1322
1323 static int __kprobes in_kprobes_functions(unsigned long addr)
1324 {
1325         struct kprobe_blackpoint *kb;
1326
1327         if (addr >= (unsigned long)__kprobes_text_start &&
1328             addr < (unsigned long)__kprobes_text_end)
1329                 return -EINVAL;
1330         /*
1331          * If there exists a kprobe_blacklist, verify and
1332          * fail any probe registration in the prohibited area
1333          */
1334         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1335                 if (kb->start_addr) {
1336                         if (addr >= kb->start_addr &&
1337                             addr < (kb->start_addr + kb->range))
1338                                 return -EINVAL;
1339                 }
1340         }
1341         return 0;
1342 }
1343
1344 /*
1345  * If we have a symbol_name argument, look it up and add the offset field
1346  * to it. This way, we can specify a relative address to a symbol.
1347  * This returns encoded errors if it fails to look up symbol or invalid
1348  * combination of parameters.
1349  */
1350 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1351 {
1352         kprobe_opcode_t *addr = p->addr;
1353
1354         if ((p->symbol_name && p->addr) ||
1355             (!p->symbol_name && !p->addr))
1356                 goto invalid;
1357
1358         if (p->symbol_name) {
1359                 kprobe_lookup_name(p->symbol_name, addr);
1360                 if (!addr)
1361                         return ERR_PTR(-ENOENT);
1362         }
1363
1364         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1365         if (addr)
1366                 return addr;
1367
1368 invalid:
1369         return ERR_PTR(-EINVAL);
1370 }
1371
1372 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1373 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1374 {
1375         struct kprobe *ap, *list_p;
1376
1377         ap = get_kprobe(p->addr);
1378         if (unlikely(!ap))
1379                 return NULL;
1380
1381         if (p != ap) {
1382                 list_for_each_entry_rcu(list_p, &ap->list, list)
1383                         if (list_p == p)
1384                         /* kprobe p is a valid probe */
1385                                 goto valid;
1386                 return NULL;
1387         }
1388 valid:
1389         return ap;
1390 }
1391
1392 /* Return error if the kprobe is being re-registered */
1393 static inline int check_kprobe_rereg(struct kprobe *p)
1394 {
1395         int ret = 0;
1396
1397         mutex_lock(&kprobe_mutex);
1398         if (__get_valid_kprobe(p))
1399                 ret = -EINVAL;
1400         mutex_unlock(&kprobe_mutex);
1401
1402         return ret;
1403 }
1404
1405 static __kprobes int check_kprobe_address_safe(struct kprobe *p,
1406                                                struct module **probed_mod)
1407 {
1408         int ret = 0;
1409         unsigned long ftrace_addr;
1410
1411         /*
1412          * If the address is located on a ftrace nop, set the
1413          * breakpoint to the following instruction.
1414          */
1415         ftrace_addr = ftrace_location((unsigned long)p->addr);
1416         if (ftrace_addr) {
1417 #ifdef KPROBES_CAN_USE_FTRACE
1418                 /* Given address is not on the instruction boundary */
1419                 if ((unsigned long)p->addr != ftrace_addr)
1420                         return -EILSEQ;
1421                 p->flags |= KPROBE_FLAG_FTRACE;
1422 #else   /* !KPROBES_CAN_USE_FTRACE */
1423                 return -EINVAL;
1424 #endif
1425         }
1426
1427         jump_label_lock();
1428         preempt_disable();
1429
1430         /* Ensure it is not in reserved area nor out of text */
1431         if (!kernel_text_address((unsigned long) p->addr) ||
1432             in_kprobes_functions((unsigned long) p->addr) ||
1433             jump_label_text_reserved(p->addr, p->addr)) {
1434                 ret = -EINVAL;
1435                 goto out;
1436         }
1437
1438         /* Check if are we probing a module */
1439         *probed_mod = __module_text_address((unsigned long) p->addr);
1440         if (*probed_mod) {
1441                 /*
1442                  * We must hold a refcount of the probed module while updating
1443                  * its code to prohibit unexpected unloading.
1444                  */
1445                 if (unlikely(!try_module_get(*probed_mod))) {
1446                         ret = -ENOENT;
1447                         goto out;
1448                 }
1449
1450                 /*
1451                  * If the module freed .init.text, we couldn't insert
1452                  * kprobes in there.
1453                  */
1454                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1455                     (*probed_mod)->state != MODULE_STATE_COMING) {
1456                         module_put(*probed_mod);
1457                         *probed_mod = NULL;
1458                         ret = -ENOENT;
1459                 }
1460         }
1461 out:
1462         preempt_enable();
1463         jump_label_unlock();
1464
1465         return ret;
1466 }
1467
1468 int __kprobes register_kprobe(struct kprobe *p)
1469 {
1470         int ret;
1471         struct kprobe *old_p;
1472         struct module *probed_mod;
1473         kprobe_opcode_t *addr;
1474
1475         /* Adjust probe address from symbol */
1476         addr = kprobe_addr(p);
1477         if (IS_ERR(addr))
1478                 return PTR_ERR(addr);
1479         p->addr = addr;
1480
1481         ret = check_kprobe_rereg(p);
1482         if (ret)
1483                 return ret;
1484
1485         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1486         p->flags &= KPROBE_FLAG_DISABLED;
1487         p->nmissed = 0;
1488         INIT_LIST_HEAD(&p->list);
1489
1490         ret = check_kprobe_address_safe(p, &probed_mod);
1491         if (ret)
1492                 return ret;
1493
1494         mutex_lock(&kprobe_mutex);
1495
1496         old_p = get_kprobe(p->addr);
1497         if (old_p) {
1498                 /* Since this may unoptimize old_p, locking text_mutex. */
1499                 ret = register_aggr_kprobe(old_p, p);
1500                 goto out;
1501         }
1502
1503         mutex_lock(&text_mutex);        /* Avoiding text modification */
1504         ret = prepare_kprobe(p);
1505         mutex_unlock(&text_mutex);
1506         if (ret)
1507                 goto out;
1508
1509         INIT_HLIST_NODE(&p->hlist);
1510         hlist_add_head_rcu(&p->hlist,
1511                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1512
1513         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1514                 arm_kprobe(p);
1515
1516         /* Try to optimize kprobe */
1517         try_to_optimize_kprobe(p);
1518
1519 out:
1520         mutex_unlock(&kprobe_mutex);
1521
1522         if (probed_mod)
1523                 module_put(probed_mod);
1524
1525         return ret;
1526 }
1527 EXPORT_SYMBOL_GPL(register_kprobe);
1528
1529 /* Check if all probes on the aggrprobe are disabled */
1530 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1531 {
1532         struct kprobe *kp;
1533
1534         list_for_each_entry_rcu(kp, &ap->list, list)
1535                 if (!kprobe_disabled(kp))
1536                         /*
1537                          * There is an active probe on the list.
1538                          * We can't disable this ap.
1539                          */
1540                         return 0;
1541
1542         return 1;
1543 }
1544
1545 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1546 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1547 {
1548         struct kprobe *orig_p;
1549
1550         /* Get an original kprobe for return */
1551         orig_p = __get_valid_kprobe(p);
1552         if (unlikely(orig_p == NULL))
1553                 return NULL;
1554
1555         if (!kprobe_disabled(p)) {
1556                 /* Disable probe if it is a child probe */
1557                 if (p != orig_p)
1558                         p->flags |= KPROBE_FLAG_DISABLED;
1559
1560                 /* Try to disarm and disable this/parent probe */
1561                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1562                         disarm_kprobe(orig_p, true);
1563                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1564                 }
1565         }
1566
1567         return orig_p;
1568 }
1569
1570 /*
1571  * Unregister a kprobe without a scheduler synchronization.
1572  */
1573 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1574 {
1575         struct kprobe *ap, *list_p;
1576
1577         /* Disable kprobe. This will disarm it if needed. */
1578         ap = __disable_kprobe(p);
1579         if (ap == NULL)
1580                 return -EINVAL;
1581
1582         if (ap == p)
1583                 /*
1584                  * This probe is an independent(and non-optimized) kprobe
1585                  * (not an aggrprobe). Remove from the hash list.
1586                  */
1587                 goto disarmed;
1588
1589         /* Following process expects this probe is an aggrprobe */
1590         WARN_ON(!kprobe_aggrprobe(ap));
1591
1592         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1593                 /*
1594                  * !disarmed could be happen if the probe is under delayed
1595                  * unoptimizing.
1596                  */
1597                 goto disarmed;
1598         else {
1599                 /* If disabling probe has special handlers, update aggrprobe */
1600                 if (p->break_handler && !kprobe_gone(p))
1601                         ap->break_handler = NULL;
1602                 if (p->post_handler && !kprobe_gone(p)) {
1603                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1604                                 if ((list_p != p) && (list_p->post_handler))
1605                                         goto noclean;
1606                         }
1607                         ap->post_handler = NULL;
1608                 }
1609 noclean:
1610                 /*
1611                  * Remove from the aggrprobe: this path will do nothing in
1612                  * __unregister_kprobe_bottom().
1613                  */
1614                 list_del_rcu(&p->list);
1615                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1616                         /*
1617                          * Try to optimize this probe again, because post
1618                          * handler may have been changed.
1619                          */
1620                         optimize_kprobe(ap);
1621         }
1622         return 0;
1623
1624 disarmed:
1625         BUG_ON(!kprobe_disarmed(ap));
1626         hlist_del_rcu(&ap->hlist);
1627         return 0;
1628 }
1629
1630 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1631 {
1632         struct kprobe *ap;
1633
1634         if (list_empty(&p->list))
1635                 /* This is an independent kprobe */
1636                 arch_remove_kprobe(p);
1637         else if (list_is_singular(&p->list)) {
1638                 /* This is the last child of an aggrprobe */
1639                 ap = list_entry(p->list.next, struct kprobe, list);
1640                 list_del(&p->list);
1641                 free_aggr_kprobe(ap);
1642         }
1643         /* Otherwise, do nothing. */
1644 }
1645
1646 int __kprobes register_kprobes(struct kprobe **kps, int num)
1647 {
1648         int i, ret = 0;
1649
1650         if (num <= 0)
1651                 return -EINVAL;
1652         for (i = 0; i < num; i++) {
1653                 ret = register_kprobe(kps[i]);
1654                 if (ret < 0) {
1655                         if (i > 0)
1656                                 unregister_kprobes(kps, i);
1657                         break;
1658                 }
1659         }
1660         return ret;
1661 }
1662 EXPORT_SYMBOL_GPL(register_kprobes);
1663
1664 void __kprobes unregister_kprobe(struct kprobe *p)
1665 {
1666         unregister_kprobes(&p, 1);
1667 }
1668 EXPORT_SYMBOL_GPL(unregister_kprobe);
1669
1670 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1671 {
1672         int i;
1673
1674         if (num <= 0)
1675                 return;
1676         mutex_lock(&kprobe_mutex);
1677         for (i = 0; i < num; i++)
1678                 if (__unregister_kprobe_top(kps[i]) < 0)
1679                         kps[i]->addr = NULL;
1680         mutex_unlock(&kprobe_mutex);
1681
1682         synchronize_sched();
1683         for (i = 0; i < num; i++)
1684                 if (kps[i]->addr)
1685                         __unregister_kprobe_bottom(kps[i]);
1686 }
1687 EXPORT_SYMBOL_GPL(unregister_kprobes);
1688
1689 static struct notifier_block kprobe_exceptions_nb = {
1690         .notifier_call = kprobe_exceptions_notify,
1691         .priority = 0x7fffffff /* we need to be notified first */
1692 };
1693
1694 unsigned long __weak arch_deref_entry_point(void *entry)
1695 {
1696         return (unsigned long)entry;
1697 }
1698
1699 int __kprobes register_jprobes(struct jprobe **jps, int num)
1700 {
1701         struct jprobe *jp;
1702         int ret = 0, i;
1703
1704         if (num <= 0)
1705                 return -EINVAL;
1706         for (i = 0; i < num; i++) {
1707                 unsigned long addr, offset;
1708                 jp = jps[i];
1709                 addr = arch_deref_entry_point(jp->entry);
1710
1711                 /* Verify probepoint is a function entry point */
1712                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1713                     offset == 0) {
1714                         jp->kp.pre_handler = setjmp_pre_handler;
1715                         jp->kp.break_handler = longjmp_break_handler;
1716                         ret = register_kprobe(&jp->kp);
1717                 } else
1718                         ret = -EINVAL;
1719
1720                 if (ret < 0) {
1721                         if (i > 0)
1722                                 unregister_jprobes(jps, i);
1723                         break;
1724                 }
1725         }
1726         return ret;
1727 }
1728 EXPORT_SYMBOL_GPL(register_jprobes);
1729
1730 int __kprobes register_jprobe(struct jprobe *jp)
1731 {
1732         return register_jprobes(&jp, 1);
1733 }
1734 EXPORT_SYMBOL_GPL(register_jprobe);
1735
1736 void __kprobes unregister_jprobe(struct jprobe *jp)
1737 {
1738         unregister_jprobes(&jp, 1);
1739 }
1740 EXPORT_SYMBOL_GPL(unregister_jprobe);
1741
1742 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1743 {
1744         int i;
1745
1746         if (num <= 0)
1747                 return;
1748         mutex_lock(&kprobe_mutex);
1749         for (i = 0; i < num; i++)
1750                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1751                         jps[i]->kp.addr = NULL;
1752         mutex_unlock(&kprobe_mutex);
1753
1754         synchronize_sched();
1755         for (i = 0; i < num; i++) {
1756                 if (jps[i]->kp.addr)
1757                         __unregister_kprobe_bottom(&jps[i]->kp);
1758         }
1759 }
1760 EXPORT_SYMBOL_GPL(unregister_jprobes);
1761
1762 #ifdef CONFIG_KRETPROBES
1763 /*
1764  * This kprobe pre_handler is registered with every kretprobe. When probe
1765  * hits it will set up the return probe.
1766  */
1767 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1768                                            struct pt_regs *regs)
1769 {
1770         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1771         unsigned long hash, flags = 0;
1772         struct kretprobe_instance *ri;
1773
1774         /*TODO: consider to only swap the RA after the last pre_handler fired */
1775         hash = hash_ptr(current, KPROBE_HASH_BITS);
1776         raw_spin_lock_irqsave(&rp->lock, flags);
1777         if (!hlist_empty(&rp->free_instances)) {
1778                 ri = hlist_entry(rp->free_instances.first,
1779                                 struct kretprobe_instance, hlist);
1780                 hlist_del(&ri->hlist);
1781                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1782
1783                 ri->rp = rp;
1784                 ri->task = current;
1785
1786                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1787                         raw_spin_lock_irqsave(&rp->lock, flags);
1788                         hlist_add_head(&ri->hlist, &rp->free_instances);
1789                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1790                         return 0;
1791                 }
1792
1793                 arch_prepare_kretprobe(ri, regs);
1794
1795                 /* XXX(hch): why is there no hlist_move_head? */
1796                 INIT_HLIST_NODE(&ri->hlist);
1797                 kretprobe_table_lock(hash, &flags);
1798                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1799                 kretprobe_table_unlock(hash, &flags);
1800         } else {
1801                 rp->nmissed++;
1802                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1803         }
1804         return 0;
1805 }
1806
1807 int __kprobes register_kretprobe(struct kretprobe *rp)
1808 {
1809         int ret = 0;
1810         struct kretprobe_instance *inst;
1811         int i;
1812         void *addr;
1813
1814         if (kretprobe_blacklist_size) {
1815                 addr = kprobe_addr(&rp->kp);
1816                 if (IS_ERR(addr))
1817                         return PTR_ERR(addr);
1818
1819                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1820                         if (kretprobe_blacklist[i].addr == addr)
1821                                 return -EINVAL;
1822                 }
1823         }
1824
1825         rp->kp.pre_handler = pre_handler_kretprobe;
1826         rp->kp.post_handler = NULL;
1827         rp->kp.fault_handler = NULL;
1828         rp->kp.break_handler = NULL;
1829
1830         /* Pre-allocate memory for max kretprobe instances */
1831         if (rp->maxactive <= 0) {
1832 #ifdef CONFIG_PREEMPT
1833                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1834 #else
1835                 rp->maxactive = num_possible_cpus();
1836 #endif
1837         }
1838         raw_spin_lock_init(&rp->lock);
1839         INIT_HLIST_HEAD(&rp->free_instances);
1840         for (i = 0; i < rp->maxactive; i++) {
1841                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1842                                rp->data_size, GFP_KERNEL);
1843                 if (inst == NULL) {
1844                         free_rp_inst(rp);
1845                         return -ENOMEM;
1846                 }
1847                 INIT_HLIST_NODE(&inst->hlist);
1848                 hlist_add_head(&inst->hlist, &rp->free_instances);
1849         }
1850
1851         rp->nmissed = 0;
1852         /* Establish function entry probe point */
1853         ret = register_kprobe(&rp->kp);
1854         if (ret != 0)
1855                 free_rp_inst(rp);
1856         return ret;
1857 }
1858 EXPORT_SYMBOL_GPL(register_kretprobe);
1859
1860 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1861 {
1862         int ret = 0, i;
1863
1864         if (num <= 0)
1865                 return -EINVAL;
1866         for (i = 0; i < num; i++) {
1867                 ret = register_kretprobe(rps[i]);
1868                 if (ret < 0) {
1869                         if (i > 0)
1870                                 unregister_kretprobes(rps, i);
1871                         break;
1872                 }
1873         }
1874         return ret;
1875 }
1876 EXPORT_SYMBOL_GPL(register_kretprobes);
1877
1878 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1879 {
1880         unregister_kretprobes(&rp, 1);
1881 }
1882 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1883
1884 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1885 {
1886         int i;
1887
1888         if (num <= 0)
1889                 return;
1890         mutex_lock(&kprobe_mutex);
1891         for (i = 0; i < num; i++)
1892                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1893                         rps[i]->kp.addr = NULL;
1894         mutex_unlock(&kprobe_mutex);
1895
1896         synchronize_sched();
1897         for (i = 0; i < num; i++) {
1898                 if (rps[i]->kp.addr) {
1899                         __unregister_kprobe_bottom(&rps[i]->kp);
1900                         cleanup_rp_inst(rps[i]);
1901                 }
1902         }
1903 }
1904 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1905
1906 #else /* CONFIG_KRETPROBES */
1907 int __kprobes register_kretprobe(struct kretprobe *rp)
1908 {
1909         return -ENOSYS;
1910 }
1911 EXPORT_SYMBOL_GPL(register_kretprobe);
1912
1913 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1914 {
1915         return -ENOSYS;
1916 }
1917 EXPORT_SYMBOL_GPL(register_kretprobes);
1918
1919 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1920 {
1921 }
1922 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1923
1924 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1925 {
1926 }
1927 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1928
1929 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1930                                            struct pt_regs *regs)
1931 {
1932         return 0;
1933 }
1934
1935 #endif /* CONFIG_KRETPROBES */
1936
1937 /* Set the kprobe gone and remove its instruction buffer. */
1938 static void __kprobes kill_kprobe(struct kprobe *p)
1939 {
1940         struct kprobe *kp;
1941
1942         p->flags |= KPROBE_FLAG_GONE;
1943         if (kprobe_aggrprobe(p)) {
1944                 /*
1945                  * If this is an aggr_kprobe, we have to list all the
1946                  * chained probes and mark them GONE.
1947                  */
1948                 list_for_each_entry_rcu(kp, &p->list, list)
1949                         kp->flags |= KPROBE_FLAG_GONE;
1950                 p->post_handler = NULL;
1951                 p->break_handler = NULL;
1952                 kill_optimized_kprobe(p);
1953         }
1954         /*
1955          * Here, we can remove insn_slot safely, because no thread calls
1956          * the original probed function (which will be freed soon) any more.
1957          */
1958         arch_remove_kprobe(p);
1959 }
1960
1961 /* Disable one kprobe */
1962 int __kprobes disable_kprobe(struct kprobe *kp)
1963 {
1964         int ret = 0;
1965
1966         mutex_lock(&kprobe_mutex);
1967
1968         /* Disable this kprobe */
1969         if (__disable_kprobe(kp) == NULL)
1970                 ret = -EINVAL;
1971
1972         mutex_unlock(&kprobe_mutex);
1973         return ret;
1974 }
1975 EXPORT_SYMBOL_GPL(disable_kprobe);
1976
1977 /* Enable one kprobe */
1978 int __kprobes enable_kprobe(struct kprobe *kp)
1979 {
1980         int ret = 0;
1981         struct kprobe *p;
1982
1983         mutex_lock(&kprobe_mutex);
1984
1985         /* Check whether specified probe is valid. */
1986         p = __get_valid_kprobe(kp);
1987         if (unlikely(p == NULL)) {
1988                 ret = -EINVAL;
1989                 goto out;
1990         }
1991
1992         if (kprobe_gone(kp)) {
1993                 /* This kprobe has gone, we couldn't enable it. */
1994                 ret = -EINVAL;
1995                 goto out;
1996         }
1997
1998         if (p != kp)
1999                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2000
2001         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2002                 p->flags &= ~KPROBE_FLAG_DISABLED;
2003                 arm_kprobe(p);
2004         }
2005 out:
2006         mutex_unlock(&kprobe_mutex);
2007         return ret;
2008 }
2009 EXPORT_SYMBOL_GPL(enable_kprobe);
2010
2011 void __kprobes dump_kprobe(struct kprobe *kp)
2012 {
2013         printk(KERN_WARNING "Dumping kprobe:\n");
2014         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2015                kp->symbol_name, kp->addr, kp->offset);
2016 }
2017
2018 /* Module notifier call back, checking kprobes on the module */
2019 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
2020                                              unsigned long val, void *data)
2021 {
2022         struct module *mod = data;
2023         struct hlist_head *head;
2024         struct hlist_node *node;
2025         struct kprobe *p;
2026         unsigned int i;
2027         int checkcore = (val == MODULE_STATE_GOING);
2028
2029         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2030                 return NOTIFY_DONE;
2031
2032         /*
2033          * When MODULE_STATE_GOING was notified, both of module .text and
2034          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2035          * notified, only .init.text section would be freed. We need to
2036          * disable kprobes which have been inserted in the sections.
2037          */
2038         mutex_lock(&kprobe_mutex);
2039         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2040                 head = &kprobe_table[i];
2041                 hlist_for_each_entry_rcu(p, node, head, hlist)
2042                         if (within_module_init((unsigned long)p->addr, mod) ||
2043                             (checkcore &&
2044                              within_module_core((unsigned long)p->addr, mod))) {
2045                                 /*
2046                                  * The vaddr this probe is installed will soon
2047                                  * be vfreed buy not synced to disk. Hence,
2048                                  * disarming the breakpoint isn't needed.
2049                                  */
2050                                 kill_kprobe(p);
2051                         }
2052         }
2053         mutex_unlock(&kprobe_mutex);
2054         return NOTIFY_DONE;
2055 }
2056
2057 static struct notifier_block kprobe_module_nb = {
2058         .notifier_call = kprobes_module_callback,
2059         .priority = 0
2060 };
2061
2062 static int __init init_kprobes(void)
2063 {
2064         int i, err = 0;
2065         unsigned long offset = 0, size = 0;
2066         char *modname, namebuf[128];
2067         const char *symbol_name;
2068         void *addr;
2069         struct kprobe_blackpoint *kb;
2070
2071         /* FIXME allocate the probe table, currently defined statically */
2072         /* initialize all list heads */
2073         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2074                 INIT_HLIST_HEAD(&kprobe_table[i]);
2075                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2076                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2077         }
2078
2079         /*
2080          * Lookup and populate the kprobe_blacklist.
2081          *
2082          * Unlike the kretprobe blacklist, we'll need to determine
2083          * the range of addresses that belong to the said functions,
2084          * since a kprobe need not necessarily be at the beginning
2085          * of a function.
2086          */
2087         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
2088                 kprobe_lookup_name(kb->name, addr);
2089                 if (!addr)
2090                         continue;
2091
2092                 kb->start_addr = (unsigned long)addr;
2093                 symbol_name = kallsyms_lookup(kb->start_addr,
2094                                 &size, &offset, &modname, namebuf);
2095                 if (!symbol_name)
2096                         kb->range = 0;
2097                 else
2098                         kb->range = size;
2099         }
2100
2101         if (kretprobe_blacklist_size) {
2102                 /* lookup the function address from its name */
2103                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2104                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2105                                            kretprobe_blacklist[i].addr);
2106                         if (!kretprobe_blacklist[i].addr)
2107                                 printk("kretprobe: lookup failed: %s\n",
2108                                        kretprobe_blacklist[i].name);
2109                 }
2110         }
2111
2112 #if defined(CONFIG_OPTPROBES)
2113 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2114         /* Init kprobe_optinsn_slots */
2115         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2116 #endif
2117         /* By default, kprobes can be optimized */
2118         kprobes_allow_optimization = true;
2119 #endif
2120
2121         /* By default, kprobes are armed */
2122         kprobes_all_disarmed = false;
2123
2124         err = arch_init_kprobes();
2125         if (!err)
2126                 err = register_die_notifier(&kprobe_exceptions_nb);
2127         if (!err)
2128                 err = register_module_notifier(&kprobe_module_nb);
2129
2130         kprobes_initialized = (err == 0);
2131
2132         if (!err)
2133                 init_test_probes();
2134         return err;
2135 }
2136
2137 #ifdef CONFIG_DEBUG_FS
2138 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2139                 const char *sym, int offset, char *modname, struct kprobe *pp)
2140 {
2141         char *kprobe_type;
2142
2143         if (p->pre_handler == pre_handler_kretprobe)
2144                 kprobe_type = "r";
2145         else if (p->pre_handler == setjmp_pre_handler)
2146                 kprobe_type = "j";
2147         else
2148                 kprobe_type = "k";
2149
2150         if (sym)
2151                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2152                         p->addr, kprobe_type, sym, offset,
2153                         (modname ? modname : " "));
2154         else
2155                 seq_printf(pi, "%p  %s  %p ",
2156                         p->addr, kprobe_type, p->addr);
2157
2158         if (!pp)
2159                 pp = p;
2160         seq_printf(pi, "%s%s%s%s\n",
2161                 (kprobe_gone(p) ? "[GONE]" : ""),
2162                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2163                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2164                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2165 }
2166
2167 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2168 {
2169         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2170 }
2171
2172 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2173 {
2174         (*pos)++;
2175         if (*pos >= KPROBE_TABLE_SIZE)
2176                 return NULL;
2177         return pos;
2178 }
2179
2180 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2181 {
2182         /* Nothing to do */
2183 }
2184
2185 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2186 {
2187         struct hlist_head *head;
2188         struct hlist_node *node;
2189         struct kprobe *p, *kp;
2190         const char *sym = NULL;
2191         unsigned int i = *(loff_t *) v;
2192         unsigned long offset = 0;
2193         char *modname, namebuf[128];
2194
2195         head = &kprobe_table[i];
2196         preempt_disable();
2197         hlist_for_each_entry_rcu(p, node, head, hlist) {
2198                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2199                                         &offset, &modname, namebuf);
2200                 if (kprobe_aggrprobe(p)) {
2201                         list_for_each_entry_rcu(kp, &p->list, list)
2202                                 report_probe(pi, kp, sym, offset, modname, p);
2203                 } else
2204                         report_probe(pi, p, sym, offset, modname, NULL);
2205         }
2206         preempt_enable();
2207         return 0;
2208 }
2209
2210 static const struct seq_operations kprobes_seq_ops = {
2211         .start = kprobe_seq_start,
2212         .next  = kprobe_seq_next,
2213         .stop  = kprobe_seq_stop,
2214         .show  = show_kprobe_addr
2215 };
2216
2217 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2218 {
2219         return seq_open(filp, &kprobes_seq_ops);
2220 }
2221
2222 static const struct file_operations debugfs_kprobes_operations = {
2223         .open           = kprobes_open,
2224         .read           = seq_read,
2225         .llseek         = seq_lseek,
2226         .release        = seq_release,
2227 };
2228
2229 static void __kprobes arm_all_kprobes(void)
2230 {
2231         struct hlist_head *head;
2232         struct hlist_node *node;
2233         struct kprobe *p;
2234         unsigned int i;
2235
2236         mutex_lock(&kprobe_mutex);
2237
2238         /* If kprobes are armed, just return */
2239         if (!kprobes_all_disarmed)
2240                 goto already_enabled;
2241
2242         /* Arming kprobes doesn't optimize kprobe itself */
2243         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2244                 head = &kprobe_table[i];
2245                 hlist_for_each_entry_rcu(p, node, head, hlist)
2246                         if (!kprobe_disabled(p))
2247                                 arm_kprobe(p);
2248         }
2249
2250         kprobes_all_disarmed = false;
2251         printk(KERN_INFO "Kprobes globally enabled\n");
2252
2253 already_enabled:
2254         mutex_unlock(&kprobe_mutex);
2255         return;
2256 }
2257
2258 static void __kprobes disarm_all_kprobes(void)
2259 {
2260         struct hlist_head *head;
2261         struct hlist_node *node;
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, node, 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         switch (buf[0]) {
2319         case 'y':
2320         case 'Y':
2321         case '1':
2322                 arm_all_kprobes();
2323                 break;
2324         case 'n':
2325         case 'N':
2326         case '0':
2327                 disarm_all_kprobes();
2328                 break;
2329         }
2330
2331         return count;
2332 }
2333
2334 static const struct file_operations fops_kp = {
2335         .read =         read_enabled_file_bool,
2336         .write =        write_enabled_file_bool,
2337         .llseek =       default_llseek,
2338 };
2339
2340 static int __kprobes debugfs_kprobe_init(void)
2341 {
2342         struct dentry *dir, *file;
2343         unsigned int value = 1;
2344
2345         dir = debugfs_create_dir("kprobes", NULL);
2346         if (!dir)
2347                 return -ENOMEM;
2348
2349         file = debugfs_create_file("list", 0444, dir, NULL,
2350                                 &debugfs_kprobes_operations);
2351         if (!file) {
2352                 debugfs_remove(dir);
2353                 return -ENOMEM;
2354         }
2355
2356         file = debugfs_create_file("enabled", 0600, dir,
2357                                         &value, &fops_kp);
2358         if (!file) {
2359                 debugfs_remove(dir);
2360                 return -ENOMEM;
2361         }
2362
2363         return 0;
2364 }
2365
2366 late_initcall(debugfs_kprobe_init);
2367 #endif /* CONFIG_DEBUG_FS */
2368
2369 module_init(init_kprobes);
2370
2371 /* defined in arch/.../kernel/kprobes.c */
2372 EXPORT_SYMBOL_GPL(jprobe_return);