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 <[email protected]> Kernel
  22  *              Probes initial implementation (includes suggestions from
  23  *              Rusty Russell).
  24  * 2004-Aug     Updated by Prasanna S Panchamukhi <[email protected]> with
  25  *              hlists and exceptions notifier as suggested by Andi Kleen.
  26  * 2004-July    Suparna Bhattacharya <[email protected]> added jumper probes
  27  *              interface to access function arguments.
  28  * 2004-Sep     Prasanna S Panchamukhi <[email protected]> Changed Kprobes
  29  *              exceptions notifier to be first on the priority list.
  30  * 2005-May     Hien Nguyen <[email protected]>, Jim Keniston
  31  *              <[email protected]> and Prasanna S Panchamukhi
  32  *              <[email protected]> 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 #define OPTIMIZE_DELAY 5
 475
 476 /*
 477  * Optimize (replace a breakpoint with a jump) kprobes listed on
 478  * optimizing_list.
 479  */
 480 static __kprobes void do_optimize_kprobes(void)
 481 {
 482         /* Optimization never be done when disarmed */
 483         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 484             list_empty(&optimizing_list))
 485                 return;
 486
 487         /*
 488          * The optimization/unoptimization refers online_cpus via
 489          * stop_machine() and cpu-hotplug modifies online_cpus.
 490          * And same time, text_mutex will be held in cpu-hotplug and here.
 491          * This combination can cause a deadlock (cpu-hotplug try to lock
 492          * text_mutex but stop_machine can not be done because online_cpus
 493          * has been changed)
 494          * To avoid this deadlock, we need to call get_online_cpus()
 495          * for preventing cpu-hotplug outside of text_mutex locking.
 496          */
 497         get_online_cpus();
 498         mutex_lock(&text_mutex);
 499         arch_optimize_kprobes(&optimizing_list);
 500         mutex_unlock(&text_mutex);
 501         put_online_cpus();
 502 }
 503
 504 /*
 505  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 506  * if need) kprobes listed on unoptimizing_list.
 507  */
 508 static __kprobes void do_unoptimize_kprobes(struct list_head *free_list)
 509 {
 510         struct optimized_kprobe *op, *tmp;
 511
 512         /* Unoptimization must be done anytime */
 513         if (list_empty(&unoptimizing_list))
 514                 return;
 515
 516         /* Ditto to do_optimize_kprobes */
 517         get_online_cpus();
 518         mutex_lock(&text_mutex);
 519         arch_unoptimize_kprobes(&unoptimizing_list, free_list);
 520         /* Loop free_list for disarming */
 521         list_for_each_entry_safe(op, tmp, free_list, list) {
 522                 /* Disarm probes if marked disabled */
 523                 if (kprobe_disabled(&op->kp))
 524                         arch_disarm_kprobe(&op->kp);
 525                 if (kprobe_unused(&op->kp)) {
 526                         /*
 527                          * Remove unused probes from hash list. After waiting
 528                          * for synchronization, these probes are reclaimed.
 529                          * (reclaiming is done by do_free_cleaned_kprobes.)
 530                          */
 531                         hlist_del_rcu(&op->kp.hlist);
 532                 } else
 533                         list_del_init(&op->list);
 534         }
 535         mutex_unlock(&text_mutex);
 536         put_online_cpus();
 537 }
 538
 539 /* Reclaim all kprobes on the free_list */
 540 static __kprobes void do_free_cleaned_kprobes(struct list_head *free_list)
 541 {
 542         struct optimized_kprobe *op, *tmp;
 543
 544         list_for_each_entry_safe(op, tmp, free_list, list) {
 545                 BUG_ON(!kprobe_unused(&op->kp));
 546                 list_del_init(&op->list);
 547                 free_aggr_kprobe(&op->kp);
 548         }
 549 }
 550
 551 /* Start optimizer after OPTIMIZE_DELAY passed */
 552 static __kprobes void kick_kprobe_optimizer(void)
 553 {
 554         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 555 }
 556
 557 /* Kprobe jump optimizer */
 558 static __kprobes void kprobe_optimizer(struct work_struct *work)
 559 {
 560         LIST_HEAD(free_list);
 561
 562         mutex_lock(&kprobe_mutex);
 563         /* Lock modules while optimizing kprobes */
 564         mutex_lock(&module_mutex);
 565
 566         /*
 567          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 568          * kprobes before waiting for quiesence period.
 569          */
 570         do_unoptimize_kprobes(&free_list);
 571
 572         /*
 573          * Step 2: Wait for quiesence period to ensure all running interrupts
 574          * are done. Because optprobe may modify multiple instructions
 575          * there is a chance that Nth instruction is interrupted. In that
 576          * case, running interrupt can return to 2nd-Nth byte of jump
 577          * instruction. This wait is for avoiding it.
 578          */
 579         synchronize_sched();
 580
 581         /* Step 3: Optimize kprobes after quiesence period */
 582         do_optimize_kprobes();
 583
 584         /* Step 4: Free cleaned kprobes after quiesence period */
 585         do_free_cleaned_kprobes(&free_list);
 586
 587         mutex_unlock(&module_mutex);
 588         mutex_unlock(&kprobe_mutex);
 589
 590         /* Step 5: Kick optimizer again if needed */
 591         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 592                 kick_kprobe_optimizer();
 593 }
 594
 595 /* Wait for completing optimization and unoptimization */
 596 static __kprobes void wait_for_kprobe_optimizer(void)
 597 {
 598         mutex_lock(&kprobe_mutex);
 599
 600         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 601                 mutex_unlock(&kprobe_mutex);
 602
 603                 /* this will also make optimizing_work execute immmediately */
 604                 flush_delayed_work(&optimizing_work);
 605                 /* @optimizing_work might not have been queued yet, relax */
 606                 cpu_relax();
 607
 608                 mutex_lock(&kprobe_mutex);
 609         }
 610
 611         mutex_unlock(&kprobe_mutex);
 612 }
 613
 614 /* Optimize kprobe if p is ready to be optimized */
 615 static __kprobes void optimize_kprobe(struct kprobe *p)
 616 {
 617         struct optimized_kprobe *op;
 618
 619         /* Check if the kprobe is disabled or not ready for optimization. */
 620         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 621             (kprobe_disabled(p) || kprobes_all_disarmed))
 622                 return;
 623
 624         /* Both of break_handler and post_handler are not supported. */
 625         if (p->break_handler || p->post_handler)
 626                 return;
 627
 628         op = container_of(p, struct optimized_kprobe, kp);
 629
 630         /* Check there is no other kprobes at the optimized instructions */
 631         if (arch_check_optimized_kprobe(op) < 0)
 632                 return;
 633
 634         /* Check if it is already optimized. */
 635         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 636                 return;
 637         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 638
 639         if (!list_empty(&op->list))
 640                 /* This is under unoptimizing. Just dequeue the probe */
 641                 list_del_init(&op->list);
 642         else {
 643                 list_add(&op->list, &optimizing_list);
 644                 kick_kprobe_optimizer();
 645         }
 646 }
 647
 648 /* Short cut to direct unoptimizing */
 649 static __kprobes void force_unoptimize_kprobe(struct optimized_kprobe *op)
 650 {
 651         get_online_cpus();
 652         arch_unoptimize_kprobe(op);
 653         put_online_cpus();
 654         if (kprobe_disabled(&op->kp))
 655                 arch_disarm_kprobe(&op->kp);
 656 }
 657
 658 /* Unoptimize a kprobe if p is optimized */
 659 static __kprobes void unoptimize_kprobe(struct kprobe *p, bool force)
 660 {
 661         struct optimized_kprobe *op;
 662
 663         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 664                 return; /* This is not an optprobe nor optimized */
 665
 666         op = container_of(p, struct optimized_kprobe, kp);
 667         if (!kprobe_optimized(p)) {
 668                 /* Unoptimized or unoptimizing case */
 669                 if (force && !list_empty(&op->list)) {
 670                         /*
 671                          * Only if this is unoptimizing kprobe and forced,
 672                          * forcibly unoptimize it. (No need to unoptimize
 673                          * unoptimized kprobe again :)
 674                          */
 675                         list_del_init(&op->list);
 676                         force_unoptimize_kprobe(op);
 677                 }
 678                 return;
 679         }
 680
 681         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 682         if (!list_empty(&op->list)) {
 683                 /* Dequeue from the optimization queue */
 684                 list_del_init(&op->list);
 685                 return;
 686         }
 687         /* Optimized kprobe case */
 688         if (force)
 689                 /* Forcibly update the code: this is a special case */
 690                 force_unoptimize_kprobe(op);
 691         else {
 692                 list_add(&op->list, &unoptimizing_list);
 693                 kick_kprobe_optimizer();
 694         }
 695 }
 696
 697 /* Cancel unoptimizing for reusing */
 698 static void reuse_unused_kprobe(struct kprobe *ap)
 699 {
 700         struct optimized_kprobe *op;
 701
 702         BUG_ON(!kprobe_unused(ap));
 703         /*
 704          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 705          * there is still a relative jump) and disabled.
 706          */
 707         op = container_of(ap, struct optimized_kprobe, kp);
 708         if (unlikely(list_empty(&op->list)))
 709                 printk(KERN_WARNING "Warning: found a stray unused "
 710                         "aggrprobe@%p\n", ap->addr);
 711         /* Enable the probe again */
 712         ap->flags &= ~KPROBE_FLAG_DISABLED;
 713         /* Optimize it again (remove from op->list) */
 714         BUG_ON(!kprobe_optready(ap));
 715         optimize_kprobe(ap);
 716 }
 717
 718 /* Remove optimized instructions */
 719 static void __kprobes kill_optimized_kprobe(struct kprobe *p)
 720 {
 721         struct optimized_kprobe *op;
 722
 723         op = container_of(p, struct optimized_kprobe, kp);
 724         if (!list_empty(&op->list))
 725                 /* Dequeue from the (un)optimization queue */
 726                 list_del_init(&op->list);
 727
 728         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 729         /* Don't touch the code, because it is already freed. */
 730         arch_remove_optimized_kprobe(op);
 731 }
 732
 733 /* Try to prepare optimized instructions */
 734 static __kprobes void prepare_optimized_kprobe(struct kprobe *p)
 735 {
 736         struct optimized_kprobe *op;
 737
 738         op = container_of(p, struct optimized_kprobe, kp);
 739         arch_prepare_optimized_kprobe(op);
 740 }
 741
 742 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 743 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 744 {
 745         struct optimized_kprobe *op;
 746
 747         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 748         if (!op)
 749                 return NULL;
 750
 751         INIT_LIST_HEAD(&op->list);
 752         op->kp.addr = p->addr;
 753         arch_prepare_optimized_kprobe(op);
 754
 755         return &op->kp;
 756 }
 757
 758 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 759
 760 /*
 761  * Prepare an optimized_kprobe and optimize it
 762  * NOTE: p must be a normal registered kprobe
 763  */
 764 static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
 765 {
 766         struct kprobe *ap;
 767         struct optimized_kprobe *op;
 768
 769         /* Impossible to optimize ftrace-based kprobe */
 770         if (kprobe_ftrace(p))
 771                 return;
 772
 773         /* For preparing optimization, jump_label_text_reserved() is called */
 774         jump_label_lock();
 775         mutex_lock(&text_mutex);
 776
 777         ap = alloc_aggr_kprobe(p);
 778         if (!ap)
 779                 goto out;
 780
 781         op = container_of(ap, struct optimized_kprobe, kp);
 782         if (!arch_prepared_optinsn(&op->optinsn)) {
 783                 /* If failed to setup optimizing, fallback to kprobe */
 784                 arch_remove_optimized_kprobe(op);
 785                 kfree(op);
 786                 goto out;
 787         }
 788
 789         init_aggr_kprobe(ap, p);
 790         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 791
 792 out:
 793         mutex_unlock(&text_mutex);
 794         jump_label_unlock();
 795 }
 796
 797 #ifdef CONFIG_SYSCTL
 798 /* This should be called with kprobe_mutex locked */
 799 static void __kprobes optimize_all_kprobes(void)
 800 {
 801         struct hlist_head *head;
 802         struct hlist_node *node;
 803         struct kprobe *p;
 804         unsigned int i;
 805
 806         /* If optimization is already allowed, just return */
 807         if (kprobes_allow_optimization)
 808                 return;
 809
 810         kprobes_allow_optimization = true;
 811         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 812                 head = &kprobe_table[i];
 813                 hlist_for_each_entry_rcu(p, node, head, hlist)
 814                         if (!kprobe_disabled(p))
 815                                 optimize_kprobe(p);
 816         }
 817         printk(KERN_INFO "Kprobes globally optimized\n");
 818 }
 819
 820 /* This should be called with kprobe_mutex locked */
 821 static void __kprobes unoptimize_all_kprobes(void)
 822 {
 823         struct hlist_head *head;
 824         struct hlist_node *node;
 825         struct kprobe *p;
 826         unsigned int i;
 827
 828         /* If optimization is already prohibited, just return */
 829         if (!kprobes_allow_optimization)
 830                 return;
 831
 832         kprobes_allow_optimization = false;
 833         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 834                 head = &kprobe_table[i];
 835                 hlist_for_each_entry_rcu(p, node, head, hlist) {
 836                         if (!kprobe_disabled(p))
 837                                 unoptimize_kprobe(p, false);
 838                 }
 839         }
 840         /* Wait for unoptimizing completion */
 841         wait_for_kprobe_optimizer();
 842         printk(KERN_INFO "Kprobes globally unoptimized\n");
 843 }
 844
 845 int sysctl_kprobes_optimization;
 846 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 847                                       void __user *buffer, size_t *length,
 848                                       loff_t *ppos)
 849 {
 850         int ret;
 851
 852         mutex_lock(&kprobe_mutex);
 853         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 854         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 855
 856         if (sysctl_kprobes_optimization)
 857                 optimize_all_kprobes();
 858         else
 859                 unoptimize_all_kprobes();
 860         mutex_unlock(&kprobe_mutex);
 861
 862         return ret;
 863 }
 864 #endif /* CONFIG_SYSCTL */
 865
 866 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 867 static void __kprobes __arm_kprobe(struct kprobe *p)
 868 {
 869         struct kprobe *_p;
 870
 871         /* Check collision with other optimized kprobes */
 872         _p = get_optimized_kprobe((unsigned long)p->addr);
 873         if (unlikely(_p))
 874                 /* Fallback to unoptimized kprobe */
 875                 unoptimize_kprobe(_p, true);
 876
 877         arch_arm_kprobe(p);
 878         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 879 }
 880
 881 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 882 static void __kprobes __disarm_kprobe(struct kprobe *p, bool reopt)
 883 {
 884         struct kprobe *_p;
 885
 886         unoptimize_kprobe(p, false);    /* Try to unoptimize */
 887
 888         if (!kprobe_queued(p)) {
 889                 arch_disarm_kprobe(p);
 890                 /* If another kprobe was blocked, optimize it. */
 891                 _p = get_optimized_kprobe((unsigned long)p->addr);
 892                 if (unlikely(_p) && reopt)
 893                         optimize_kprobe(_p);
 894         }
 895         /* TODO: reoptimize others after unoptimized this probe */
 896 }
 897
 898 #else /* !CONFIG_OPTPROBES */
 899
 900 #define optimize_kprobe(p)                      do {} while (0)
 901 #define unoptimize_kprobe(p, f)                 do {} while (0)
 902 #define kill_optimized_kprobe(p)                do {} while (0)
 903 #define prepare_optimized_kprobe(p)             do {} while (0)
 904 #define try_to_optimize_kprobe(p)               do {} while (0)
 905 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 906 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 907 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 908 #define wait_for_kprobe_optimizer()             do {} while (0)
 909
 910 /* There should be no unused kprobes can be reused without optimization */
 911 static void reuse_unused_kprobe(struct kprobe *ap)
 912 {
 913         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 914         BUG_ON(kprobe_unused(ap));
 915 }
 916
 917 static __kprobes void free_aggr_kprobe(struct kprobe *p)
 918 {
 919         arch_remove_kprobe(p);
 920         kfree(p);
 921 }
 922
 923 static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 924 {
 925         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 926 }
 927 #endif /* CONFIG_OPTPROBES */
 928
 929 #ifdef CONFIG_KPROBES_ON_FTRACE
 930 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 931         .func = kprobe_ftrace_handler,
 932         .flags = FTRACE_OPS_FL_SAVE_REGS,
 933 };
 934 static int kprobe_ftrace_enabled;
 935
 936 /* Must ensure p->addr is really on ftrace */
 937 static int __kprobes prepare_kprobe(struct kprobe *p)
 938 {
 939         if (!kprobe_ftrace(p))
 940                 return arch_prepare_kprobe(p);
 941
 942         return arch_prepare_kprobe_ftrace(p);
 943 }
 944
 945 /* Caller must lock kprobe_mutex */
 946 static void __kprobes arm_kprobe_ftrace(struct kprobe *p)
 947 {
 948         int ret;
 949
 950         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 951                                    (unsigned long)p->addr, 0, 0);
 952         WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 953         kprobe_ftrace_enabled++;
 954         if (kprobe_ftrace_enabled == 1) {
 955                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 956                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 957         }
 958 }
 959
 960 /* Caller must lock kprobe_mutex */
 961 static void __kprobes disarm_kprobe_ftrace(struct kprobe *p)
 962 {
 963         int ret;
 964
 965         kprobe_ftrace_enabled--;
 966         if (kprobe_ftrace_enabled == 0) {
 967                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
 968                 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
 969         }
 970         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 971                            (unsigned long)p->addr, 1, 0);
 972         WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
 973 }
 974 #else   /* !CONFIG_KPROBES_ON_FTRACE */
 975 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
 976 #define arm_kprobe_ftrace(p)    do {} while (0)
 977 #define disarm_kprobe_ftrace(p) do {} while (0)
 978 #endif
 979
 980 /* Arm a kprobe with text_mutex */
 981 static void __kprobes arm_kprobe(struct kprobe *kp)
 982 {
 983         if (unlikely(kprobe_ftrace(kp))) {
 984                 arm_kprobe_ftrace(kp);
 985                 return;
 986         }
 987         /*
 988          * Here, since __arm_kprobe() doesn't use stop_machine(),
 989          * this doesn't cause deadlock on text_mutex. So, we don't
 990          * need get_online_cpus().
 991          */
 992         mutex_lock(&text_mutex);
 993         __arm_kprobe(kp);
 994         mutex_unlock(&text_mutex);
 995 }
 996
 997 /* Disarm a kprobe with text_mutex */
 998 static void __kprobes disarm_kprobe(struct kprobe *kp, bool reopt)
 999 {
1000         if (unlikely(kprobe_ftrace(kp))) {
1001                 disarm_kprobe_ftrace(kp);
1002                 return;
1003         }
1004         /* Ditto */
1005         mutex_lock(&text_mutex);
1006         __disarm_kprobe(kp, reopt);
1007         mutex_unlock(&text_mutex);
1008 }
1009
1010 /*
1011  * Aggregate handlers for multiple kprobes support - these handlers
1012  * take care of invoking the individual kprobe handlers on p->list
1013  */
1014 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1015 {
1016         struct kprobe *kp;
1017
1018         list_for_each_entry_rcu(kp, &p->list, list) {
1019                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1020                         set_kprobe_instance(kp);
1021                         if (kp->pre_handler(kp, regs))
1022                                 return 1;
1023                 }
1024                 reset_kprobe_instance();
1025         }
1026         return 0;
1027 }
1028
1029 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1030                                         unsigned long flags)
1031 {
1032         struct kprobe *kp;
1033
1034         list_for_each_entry_rcu(kp, &p->list, list) {
1035                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1036                         set_kprobe_instance(kp);
1037                         kp->post_handler(kp, regs, flags);
1038                         reset_kprobe_instance();
1039                 }
1040         }
1041 }
1042
1043 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1044                                         int trapnr)
1045 {
1046         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1047
1048         /*
1049          * if we faulted "during" the execution of a user specified
1050          * probe handler, invoke just that probe's fault handler
1051          */
1052         if (cur && cur->fault_handler) {
1053                 if (cur->fault_handler(cur, regs, trapnr))
1054                         return 1;
1055         }
1056         return 0;
1057 }
1058
1059 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
1060 {
1061         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1062         int ret = 0;
1063
1064         if (cur && cur->break_handler) {
1065                 if (cur->break_handler(cur, regs))
1066                         ret = 1;
1067         }
1068         reset_kprobe_instance();
1069         return ret;
1070 }
1071
1072 /* Walks the list and increments nmissed count for multiprobe case */
1073 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
1074 {
1075         struct kprobe *kp;
1076         if (!kprobe_aggrprobe(p)) {
1077                 p->nmissed++;
1078         } else {
1079                 list_for_each_entry_rcu(kp, &p->list, list)
1080                         kp->nmissed++;
1081         }
1082         return;
1083 }
1084
1085 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
1086                                 struct hlist_head *head)
1087 {
1088         struct kretprobe *rp = ri->rp;
1089
1090         /* remove rp inst off the rprobe_inst_table */
1091         hlist_del(&ri->hlist);
1092         INIT_HLIST_NODE(&ri->hlist);
1093         if (likely(rp)) {
1094                 raw_spin_lock(&rp->lock);
1095                 hlist_add_head(&ri->hlist, &rp->free_instances);
1096                 raw_spin_unlock(&rp->lock);
1097         } else
1098                 /* Unregistering */
1099                 hlist_add_head(&ri->hlist, head);
1100 }
1101
1102 void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
1103                          struct hlist_head **head, unsigned long *flags)
1104 __acquires(hlist_lock)
1105 {
1106         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1107         raw_spinlock_t *hlist_lock;
1108
1109         *head = &kretprobe_inst_table[hash];
1110         hlist_lock = kretprobe_table_lock_ptr(hash);
1111         raw_spin_lock_irqsave(hlist_lock, *flags);
1112 }
1113
1114 static void __kprobes kretprobe_table_lock(unsigned long hash,
1115         unsigned long *flags)
1116 __acquires(hlist_lock)
1117 {
1118         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1119         raw_spin_lock_irqsave(hlist_lock, *flags);
1120 }
1121
1122 void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
1123         unsigned long *flags)
1124 __releases(hlist_lock)
1125 {
1126         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1127         raw_spinlock_t *hlist_lock;
1128
1129         hlist_lock = kretprobe_table_lock_ptr(hash);
1130         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1131 }
1132
1133 static void __kprobes kretprobe_table_unlock(unsigned long hash,
1134        unsigned long *flags)
1135 __releases(hlist_lock)
1136 {
1137         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1138         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1139 }
1140
1141 /*
1142  * This function is called from finish_task_switch when task tk becomes dead,
1143  * so that we can recycle any function-return probe instances associated
1144  * with this task. These left over instances represent probed functions
1145  * that have been called but will never return.
1146  */
1147 void __kprobes kprobe_flush_task(struct task_struct *tk)
1148 {
1149         struct kretprobe_instance *ri;
1150         struct hlist_head *head, empty_rp;
1151         struct hlist_node *node, *tmp;
1152         unsigned long hash, flags = 0;
1153
1154         if (unlikely(!kprobes_initialized))
1155                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1156                 return;
1157
1158         INIT_HLIST_HEAD(&empty_rp);
1159         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1160         head = &kretprobe_inst_table[hash];
1161         kretprobe_table_lock(hash, &flags);
1162         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
1163                 if (ri->task == tk)
1164                         recycle_rp_inst(ri, &empty_rp);
1165         }
1166         kretprobe_table_unlock(hash, &flags);
1167         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
1168                 hlist_del(&ri->hlist);
1169                 kfree(ri);
1170         }
1171 }
1172
1173 static inline void free_rp_inst(struct kretprobe *rp)
1174 {
1175         struct kretprobe_instance *ri;
1176         struct hlist_node *pos, *next;
1177
1178         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, hlist) {
1179                 hlist_del(&ri->hlist);
1180                 kfree(ri);
1181         }
1182 }
1183
1184 static void __kprobes cleanup_rp_inst(struct kretprobe *rp)
1185 {
1186         unsigned long flags, hash;
1187         struct kretprobe_instance *ri;
1188         struct hlist_node *pos, *next;
1189         struct hlist_head *head;
1190
1191         /* No race here */
1192         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1193                 kretprobe_table_lock(hash, &flags);
1194                 head = &kretprobe_inst_table[hash];
1195                 hlist_for_each_entry_safe(ri, pos, next, head, hlist) {
1196                         if (ri->rp == rp)
1197                                 ri->rp = NULL;
1198                 }
1199                 kretprobe_table_unlock(hash, &flags);
1200         }
1201         free_rp_inst(rp);
1202 }
1203
1204 /*
1205 * Add the new probe to ap->list. Fail if this is the
1206 * second jprobe at the address - two jprobes can't coexist
1207 */
1208 static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1209 {
1210         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1211
1212         if (p->break_handler || p->post_handler)
1213                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1214
1215         if (p->break_handler) {
1216                 if (ap->break_handler)
1217                         return -EEXIST;
1218                 list_add_tail_rcu(&p->list, &ap->list);
1219                 ap->break_handler = aggr_break_handler;
1220         } else
1221                 list_add_rcu(&p->list, &ap->list);
1222         if (p->post_handler && !ap->post_handler)
1223                 ap->post_handler = aggr_post_handler;
1224
1225         return 0;
1226 }
1227
1228 /*
1229  * Fill in the required fields of the "manager kprobe". Replace the
1230  * earlier kprobe in the hlist with the manager kprobe
1231  */
1232 static void __kprobes init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1233 {
1234         /* Copy p's insn slot to ap */
1235         copy_kprobe(p, ap);
1236         flush_insn_slot(ap);
1237         ap->addr = p->addr;
1238         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1239         ap->pre_handler = aggr_pre_handler;
1240         ap->fault_handler = aggr_fault_handler;
1241         /* We don't care the kprobe which has gone. */
1242         if (p->post_handler && !kprobe_gone(p))
1243                 ap->post_handler = aggr_post_handler;
1244         if (p->break_handler && !kprobe_gone(p))
1245                 ap->break_handler = aggr_break_handler;
1246
1247         INIT_LIST_HEAD(&ap->list);
1248         INIT_HLIST_NODE(&ap->hlist);
1249
1250         list_add_rcu(&p->list, &ap->list);
1251         hlist_replace_rcu(&p->hlist, &ap->hlist);
1252 }
1253
1254 /*
1255  * This is the second or subsequent kprobe at the address - handle
1256  * the intricacies
1257  */
1258 static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1259                                           struct kprobe *p)
1260 {
1261         int ret = 0;
1262         struct kprobe *ap = orig_p;
1263
1264         /* For preparing optimization, jump_label_text_reserved() is called */
1265         jump_label_lock();
1266         /*
1267          * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1268          * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1269          */
1270         get_online_cpus();
1271         mutex_lock(&text_mutex);
1272
1273         if (!kprobe_aggrprobe(orig_p)) {
1274                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1275                 ap = alloc_aggr_kprobe(orig_p);
1276                 if (!ap) {
1277                         ret = -ENOMEM;
1278                         goto out;
1279                 }
1280                 init_aggr_kprobe(ap, orig_p);
1281         } else if (kprobe_unused(ap))
1282                 /* This probe is going to die. Rescue it */
1283                 reuse_unused_kprobe(ap);
1284
1285         if (kprobe_gone(ap)) {
1286                 /*
1287                  * Attempting to insert new probe at the same location that
1288                  * had a probe in the module vaddr area which already
1289                  * freed. So, the instruction slot has already been
1290                  * released. We need a new slot for the new probe.
1291                  */
1292                 ret = arch_prepare_kprobe(ap);
1293                 if (ret)
1294                         /*
1295                          * Even if fail to allocate new slot, don't need to
1296                          * free aggr_probe. It will be used next time, or
1297                          * freed by unregister_kprobe.
1298                          */
1299                         goto out;
1300
1301                 /* Prepare optimized instructions if possible. */
1302                 prepare_optimized_kprobe(ap);
1303
1304                 /*
1305                  * Clear gone flag to prevent allocating new slot again, and
1306                  * set disabled flag because it is not armed yet.
1307                  */
1308                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1309                             | KPROBE_FLAG_DISABLED;
1310         }
1311
1312         /* Copy ap's insn slot to p */
1313         copy_kprobe(ap, p);
1314         ret = add_new_kprobe(ap, p);
1315
1316 out:
1317         mutex_unlock(&text_mutex);
1318         put_online_cpus();
1319         jump_label_unlock();
1320
1321         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1322                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1323                 if (!kprobes_all_disarmed)
1324                         /* Arm the breakpoint again. */
1325                         arm_kprobe(ap);
1326         }
1327         return ret;
1328 }
1329
1330 static int __kprobes in_kprobes_functions(unsigned long addr)
1331 {
1332         struct kprobe_blackpoint *kb;
1333
1334         if (addr >= (unsigned long)__kprobes_text_start &&
1335             addr < (unsigned long)__kprobes_text_end)
1336                 return -EINVAL;
1337         /*
1338          * If there exists a kprobe_blacklist, verify and
1339          * fail any probe registration in the prohibited area
1340          */
1341         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
1342                 if (kb->start_addr) {
1343                         if (addr >= kb->start_addr &&
1344                             addr < (kb->start_addr + kb->range))
1345                                 return -EINVAL;
1346                 }
1347         }
1348         return 0;
1349 }
1350
1351 /*
1352  * If we have a symbol_name argument, look it up and add the offset field
1353  * to it. This way, we can specify a relative address to a symbol.
1354  * This returns encoded errors if it fails to look up symbol or invalid
1355  * combination of parameters.
1356  */
1357 static kprobe_opcode_t __kprobes *kprobe_addr(struct kprobe *p)
1358 {
1359         kprobe_opcode_t *addr = p->addr;
1360
1361         if ((p->symbol_name && p->addr) ||
1362             (!p->symbol_name && !p->addr))
1363                 goto invalid;
1364
1365         if (p->symbol_name) {
1366                 kprobe_lookup_name(p->symbol_name, addr);
1367                 if (!addr)
1368                         return ERR_PTR(-ENOENT);
1369         }
1370
1371         addr = (kprobe_opcode_t *)(((char *)addr) + p->offset);
1372         if (addr)
1373                 return addr;
1374
1375 invalid:
1376         return ERR_PTR(-EINVAL);
1377 }
1378
1379 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1380 static struct kprobe * __kprobes __get_valid_kprobe(struct kprobe *p)
1381 {
1382         struct kprobe *ap, *list_p;
1383
1384         ap = get_kprobe(p->addr);
1385         if (unlikely(!ap))
1386                 return NULL;
1387
1388         if (p != ap) {
1389                 list_for_each_entry_rcu(list_p, &ap->list, list)
1390                         if (list_p == p)
1391                         /* kprobe p is a valid probe */
1392                                 goto valid;
1393                 return NULL;
1394         }
1395 valid:
1396         return ap;
1397 }
1398
1399 /* Return error if the kprobe is being re-registered */
1400 static inline int check_kprobe_rereg(struct kprobe *p)
1401 {
1402         int ret = 0;
1403
1404         mutex_lock(&kprobe_mutex);
1405         if (__get_valid_kprobe(p))
1406                 ret = -EINVAL;
1407         mutex_unlock(&kprobe_mutex);
1408
1409         return ret;
1410 }
1411
1412 static __kprobes int check_kprobe_address_safe(struct kprobe *p,
1413                                                struct module **probed_mod)
1414 {
1415         int ret = 0;
1416         unsigned long ftrace_addr;
1417
1418         /*
1419          * If the address is located on a ftrace nop, set the
1420          * breakpoint to the following instruction.
1421          */
1422         ftrace_addr = ftrace_location((unsigned long)p->addr);
1423         if (ftrace_addr) {
1424 #ifdef CONFIG_KPROBES_ON_FTRACE
1425                 /* Given address is not on the instruction boundary */
1426                 if ((unsigned long)p->addr != ftrace_addr)
1427                         return -EILSEQ;
1428                 p->flags |= KPROBE_FLAG_FTRACE;
1429 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1430                 return -EINVAL;
1431 #endif
1432         }
1433
1434         jump_label_lock();
1435         preempt_disable();
1436
1437         /* Ensure it is not in reserved area nor out of text */
1438         if (!kernel_text_address((unsigned long) p->addr) ||
1439             in_kprobes_functions((unsigned long) p->addr) ||
1440             jump_label_text_reserved(p->addr, p->addr)) {
1441                 ret = -EINVAL;
1442                 goto out;
1443         }
1444
1445         /* Check if are we probing a module */
1446         *probed_mod = __module_text_address((unsigned long) p->addr);
1447         if (*probed_mod) {
1448                 /*
1449                  * We must hold a refcount of the probed module while updating
1450                  * its code to prohibit unexpected unloading.
1451                  */
1452                 if (unlikely(!try_module_get(*probed_mod))) {
1453                         ret = -ENOENT;
1454                         goto out;
1455                 }
1456
1457                 /*
1458                  * If the module freed .init.text, we couldn't insert
1459                  * kprobes in there.
1460                  */
1461                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1462                     (*probed_mod)->state != MODULE_STATE_COMING) {
1463                         module_put(*probed_mod);
1464                         *probed_mod = NULL;
1465                         ret = -ENOENT;
1466                 }
1467         }
1468 out:
1469         preempt_enable();
1470         jump_label_unlock();
1471
1472         return ret;
1473 }
1474
1475 int __kprobes register_kprobe(struct kprobe *p)
1476 {
1477         int ret;
1478         struct kprobe *old_p;
1479         struct module *probed_mod;
1480         kprobe_opcode_t *addr;
1481
1482         /* Adjust probe address from symbol */
1483         addr = kprobe_addr(p);
1484         if (IS_ERR(addr))
1485                 return PTR_ERR(addr);
1486         p->addr = addr;
1487
1488         ret = check_kprobe_rereg(p);
1489         if (ret)
1490                 return ret;
1491
1492         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1493         p->flags &= KPROBE_FLAG_DISABLED;
1494         p->nmissed = 0;
1495         INIT_LIST_HEAD(&p->list);
1496
1497         ret = check_kprobe_address_safe(p, &probed_mod);
1498         if (ret)
1499                 return ret;
1500
1501         mutex_lock(&kprobe_mutex);
1502
1503         old_p = get_kprobe(p->addr);
1504         if (old_p) {
1505                 /* Since this may unoptimize old_p, locking text_mutex. */
1506                 ret = register_aggr_kprobe(old_p, p);
1507                 goto out;
1508         }
1509
1510         mutex_lock(&text_mutex);        /* Avoiding text modification */
1511         ret = prepare_kprobe(p);
1512         mutex_unlock(&text_mutex);
1513         if (ret)
1514                 goto out;
1515
1516         INIT_HLIST_NODE(&p->hlist);
1517         hlist_add_head_rcu(&p->hlist,
1518                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1519
1520         if (!kprobes_all_disarmed && !kprobe_disabled(p))
1521                 arm_kprobe(p);
1522
1523         /* Try to optimize kprobe */
1524         try_to_optimize_kprobe(p);
1525
1526 out:
1527         mutex_unlock(&kprobe_mutex);
1528
1529         if (probed_mod)
1530                 module_put(probed_mod);
1531
1532         return ret;
1533 }
1534 EXPORT_SYMBOL_GPL(register_kprobe);
1535
1536 /* Check if all probes on the aggrprobe are disabled */
1537 static int __kprobes aggr_kprobe_disabled(struct kprobe *ap)
1538 {
1539         struct kprobe *kp;
1540
1541         list_for_each_entry_rcu(kp, &ap->list, list)
1542                 if (!kprobe_disabled(kp))
1543                         /*
1544                          * There is an active probe on the list.
1545                          * We can't disable this ap.
1546                          */
1547                         return 0;
1548
1549         return 1;
1550 }
1551
1552 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1553 static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1554 {
1555         struct kprobe *orig_p;
1556
1557         /* Get an original kprobe for return */
1558         orig_p = __get_valid_kprobe(p);
1559         if (unlikely(orig_p == NULL))
1560                 return NULL;
1561
1562         if (!kprobe_disabled(p)) {
1563                 /* Disable probe if it is a child probe */
1564                 if (p != orig_p)
1565                         p->flags |= KPROBE_FLAG_DISABLED;
1566
1567                 /* Try to disarm and disable this/parent probe */
1568                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1569                         disarm_kprobe(orig_p, true);
1570                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1571                 }
1572         }
1573
1574         return orig_p;
1575 }
1576
1577 /*
1578  * Unregister a kprobe without a scheduler synchronization.
1579  */
1580 static int __kprobes __unregister_kprobe_top(struct kprobe *p)
1581 {
1582         struct kprobe *ap, *list_p;
1583
1584         /* Disable kprobe. This will disarm it if needed. */
1585         ap = __disable_kprobe(p);
1586         if (ap == NULL)
1587                 return -EINVAL;
1588
1589         if (ap == p)
1590                 /*
1591                  * This probe is an independent(and non-optimized) kprobe
1592                  * (not an aggrprobe). Remove from the hash list.
1593                  */
1594                 goto disarmed;
1595
1596         /* Following process expects this probe is an aggrprobe */
1597         WARN_ON(!kprobe_aggrprobe(ap));
1598
1599         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1600                 /*
1601                  * !disarmed could be happen if the probe is under delayed
1602                  * unoptimizing.
1603                  */
1604                 goto disarmed;
1605         else {
1606                 /* If disabling probe has special handlers, update aggrprobe */
1607                 if (p->break_handler && !kprobe_gone(p))
1608                         ap->break_handler = NULL;
1609                 if (p->post_handler && !kprobe_gone(p)) {
1610                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1611                                 if ((list_p != p) && (list_p->post_handler))
1612                                         goto noclean;
1613                         }
1614                         ap->post_handler = NULL;
1615                 }
1616 noclean:
1617                 /*
1618                  * Remove from the aggrprobe: this path will do nothing in
1619                  * __unregister_kprobe_bottom().
1620                  */
1621                 list_del_rcu(&p->list);
1622                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1623                         /*
1624                          * Try to optimize this probe again, because post
1625                          * handler may have been changed.
1626                          */
1627                         optimize_kprobe(ap);
1628         }
1629         return 0;
1630
1631 disarmed:
1632         BUG_ON(!kprobe_disarmed(ap));
1633         hlist_del_rcu(&ap->hlist);
1634         return 0;
1635 }
1636
1637 static void __kprobes __unregister_kprobe_bottom(struct kprobe *p)
1638 {
1639         struct kprobe *ap;
1640
1641         if (list_empty(&p->list))
1642                 /* This is an independent kprobe */
1643                 arch_remove_kprobe(p);
1644         else if (list_is_singular(&p->list)) {
1645                 /* This is the last child of an aggrprobe */
1646                 ap = list_entry(p->list.next, struct kprobe, list);
1647                 list_del(&p->list);
1648                 free_aggr_kprobe(ap);
1649         }
1650         /* Otherwise, do nothing. */
1651 }
1652
1653 int __kprobes register_kprobes(struct kprobe **kps, int num)
1654 {
1655         int i, ret = 0;
1656
1657         if (num <= 0)
1658                 return -EINVAL;
1659         for (i = 0; i < num; i++) {
1660                 ret = register_kprobe(kps[i]);
1661                 if (ret < 0) {
1662                         if (i > 0)
1663                                 unregister_kprobes(kps, i);
1664                         break;
1665                 }
1666         }
1667         return ret;
1668 }
1669 EXPORT_SYMBOL_GPL(register_kprobes);
1670
1671 void __kprobes unregister_kprobe(struct kprobe *p)
1672 {
1673         unregister_kprobes(&p, 1);
1674 }
1675 EXPORT_SYMBOL_GPL(unregister_kprobe);
1676
1677 void __kprobes unregister_kprobes(struct kprobe **kps, int num)
1678 {
1679         int i;
1680
1681         if (num <= 0)
1682                 return;
1683         mutex_lock(&kprobe_mutex);
1684         for (i = 0; i < num; i++)
1685                 if (__unregister_kprobe_top(kps[i]) < 0)
1686                         kps[i]->addr = NULL;
1687         mutex_unlock(&kprobe_mutex);
1688
1689         synchronize_sched();
1690         for (i = 0; i < num; i++)
1691                 if (kps[i]->addr)
1692                         __unregister_kprobe_bottom(kps[i]);
1693 }
1694 EXPORT_SYMBOL_GPL(unregister_kprobes);
1695
1696 static struct notifier_block kprobe_exceptions_nb = {
1697         .notifier_call = kprobe_exceptions_notify,
1698         .priority = 0x7fffffff /* we need to be notified first */
1699 };
1700
1701 unsigned long __weak arch_deref_entry_point(void *entry)
1702 {
1703         return (unsigned long)entry;
1704 }
1705
1706 int __kprobes register_jprobes(struct jprobe **jps, int num)
1707 {
1708         struct jprobe *jp;
1709         int ret = 0, i;
1710
1711         if (num <= 0)
1712                 return -EINVAL;
1713         for (i = 0; i < num; i++) {
1714                 unsigned long addr, offset;
1715                 jp = jps[i];
1716                 addr = arch_deref_entry_point(jp->entry);
1717
1718                 /* Verify probepoint is a function entry point */
1719                 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1720                     offset == 0) {
1721                         jp->kp.pre_handler = setjmp_pre_handler;
1722                         jp->kp.break_handler = longjmp_break_handler;
1723                         ret = register_kprobe(&jp->kp);
1724                 } else
1725                         ret = -EINVAL;
1726
1727                 if (ret < 0) {
1728                         if (i > 0)
1729                                 unregister_jprobes(jps, i);
1730                         break;
1731                 }
1732         }
1733         return ret;
1734 }
1735 EXPORT_SYMBOL_GPL(register_jprobes);
1736
1737 int __kprobes register_jprobe(struct jprobe *jp)
1738 {
1739         return register_jprobes(&jp, 1);
1740 }
1741 EXPORT_SYMBOL_GPL(register_jprobe);
1742
1743 void __kprobes unregister_jprobe(struct jprobe *jp)
1744 {
1745         unregister_jprobes(&jp, 1);
1746 }
1747 EXPORT_SYMBOL_GPL(unregister_jprobe);
1748
1749 void __kprobes unregister_jprobes(struct jprobe **jps, int num)
1750 {
1751         int i;
1752
1753         if (num <= 0)
1754                 return;
1755         mutex_lock(&kprobe_mutex);
1756         for (i = 0; i < num; i++)
1757                 if (__unregister_kprobe_top(&jps[i]->kp) < 0)
1758                         jps[i]->kp.addr = NULL;
1759         mutex_unlock(&kprobe_mutex);
1760
1761         synchronize_sched();
1762         for (i = 0; i < num; i++) {
1763                 if (jps[i]->kp.addr)
1764                         __unregister_kprobe_bottom(&jps[i]->kp);
1765         }
1766 }
1767 EXPORT_SYMBOL_GPL(unregister_jprobes);
1768
1769 #ifdef CONFIG_KRETPROBES
1770 /*
1771  * This kprobe pre_handler is registered with every kretprobe. When probe
1772  * hits it will set up the return probe.
1773  */
1774 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1775                                            struct pt_regs *regs)
1776 {
1777         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1778         unsigned long hash, flags = 0;
1779         struct kretprobe_instance *ri;
1780
1781         /*TODO: consider to only swap the RA after the last pre_handler fired */
1782         hash = hash_ptr(current, KPROBE_HASH_BITS);
1783         raw_spin_lock_irqsave(&rp->lock, flags);
1784         if (!hlist_empty(&rp->free_instances)) {
1785                 ri = hlist_entry(rp->free_instances.first,
1786                                 struct kretprobe_instance, hlist);
1787                 hlist_del(&ri->hlist);
1788                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1789
1790                 ri->rp = rp;
1791                 ri->task = current;
1792
1793                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1794                         raw_spin_lock_irqsave(&rp->lock, flags);
1795                         hlist_add_head(&ri->hlist, &rp->free_instances);
1796                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1797                         return 0;
1798                 }
1799
1800                 arch_prepare_kretprobe(ri, regs);
1801
1802                 /* XXX(hch): why is there no hlist_move_head? */
1803                 INIT_HLIST_NODE(&ri->hlist);
1804                 kretprobe_table_lock(hash, &flags);
1805                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1806                 kretprobe_table_unlock(hash, &flags);
1807         } else {
1808                 rp->nmissed++;
1809                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1810         }
1811         return 0;
1812 }
1813
1814 int __kprobes register_kretprobe(struct kretprobe *rp)
1815 {
1816         int ret = 0;
1817         struct kretprobe_instance *inst;
1818         int i;
1819         void *addr;
1820
1821         if (kretprobe_blacklist_size) {
1822                 addr = kprobe_addr(&rp->kp);
1823                 if (IS_ERR(addr))
1824                         return PTR_ERR(addr);
1825
1826                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1827                         if (kretprobe_blacklist[i].addr == addr)
1828                                 return -EINVAL;
1829                 }
1830         }
1831
1832         rp->kp.pre_handler = pre_handler_kretprobe;
1833         rp->kp.post_handler = NULL;
1834         rp->kp.fault_handler = NULL;
1835         rp->kp.break_handler = NULL;
1836
1837         /* Pre-allocate memory for max kretprobe instances */
1838         if (rp->maxactive <= 0) {
1839 #ifdef CONFIG_PREEMPT
1840                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1841 #else
1842                 rp->maxactive = num_possible_cpus();
1843 #endif
1844         }
1845         raw_spin_lock_init(&rp->lock);
1846         INIT_HLIST_HEAD(&rp->free_instances);
1847         for (i = 0; i < rp->maxactive; i++) {
1848                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1849                                rp->data_size, GFP_KERNEL);
1850                 if (inst == NULL) {
1851                         free_rp_inst(rp);
1852                         return -ENOMEM;
1853                 }
1854                 INIT_HLIST_NODE(&inst->hlist);
1855                 hlist_add_head(&inst->hlist, &rp->free_instances);
1856         }
1857
1858         rp->nmissed = 0;
1859         /* Establish function entry probe point */
1860         ret = register_kprobe(&rp->kp);
1861         if (ret != 0)
1862                 free_rp_inst(rp);
1863         return ret;
1864 }
1865 EXPORT_SYMBOL_GPL(register_kretprobe);
1866
1867 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1868 {
1869         int ret = 0, i;
1870
1871         if (num <= 0)
1872                 return -EINVAL;
1873         for (i = 0; i < num; i++) {
1874                 ret = register_kretprobe(rps[i]);
1875                 if (ret < 0) {
1876                         if (i > 0)
1877                                 unregister_kretprobes(rps, i);
1878                         break;
1879                 }
1880         }
1881         return ret;
1882 }
1883 EXPORT_SYMBOL_GPL(register_kretprobes);
1884
1885 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1886 {
1887         unregister_kretprobes(&rp, 1);
1888 }
1889 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1890
1891 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1892 {
1893         int i;
1894
1895         if (num <= 0)
1896                 return;
1897         mutex_lock(&kprobe_mutex);
1898         for (i = 0; i < num; i++)
1899                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1900                         rps[i]->kp.addr = NULL;
1901         mutex_unlock(&kprobe_mutex);
1902
1903         synchronize_sched();
1904         for (i = 0; i < num; i++) {
1905                 if (rps[i]->kp.addr) {
1906                         __unregister_kprobe_bottom(&rps[i]->kp);
1907                         cleanup_rp_inst(rps[i]);
1908                 }
1909         }
1910 }
1911 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1912
1913 #else /* CONFIG_KRETPROBES */
1914 int __kprobes register_kretprobe(struct kretprobe *rp)
1915 {
1916         return -ENOSYS;
1917 }
1918 EXPORT_SYMBOL_GPL(register_kretprobe);
1919
1920 int __kprobes register_kretprobes(struct kretprobe **rps, int num)
1921 {
1922         return -ENOSYS;
1923 }
1924 EXPORT_SYMBOL_GPL(register_kretprobes);
1925
1926 void __kprobes unregister_kretprobe(struct kretprobe *rp)
1927 {
1928 }
1929 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1930
1931 void __kprobes unregister_kretprobes(struct kretprobe **rps, int num)
1932 {
1933 }
1934 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1935
1936 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
1937                                            struct pt_regs *regs)
1938 {
1939         return 0;
1940 }
1941
1942 #endif /* CONFIG_KRETPROBES */
1943
1944 /* Set the kprobe gone and remove its instruction buffer. */
1945 static void __kprobes kill_kprobe(struct kprobe *p)
1946 {
1947         struct kprobe *kp;
1948
1949         p->flags |= KPROBE_FLAG_GONE;
1950         if (kprobe_aggrprobe(p)) {
1951                 /*
1952                  * If this is an aggr_kprobe, we have to list all the
1953                  * chained probes and mark them GONE.
1954                  */
1955                 list_for_each_entry_rcu(kp, &p->list, list)
1956                         kp->flags |= KPROBE_FLAG_GONE;
1957                 p->post_handler = NULL;
1958                 p->break_handler = NULL;
1959                 kill_optimized_kprobe(p);
1960         }
1961         /*
1962          * Here, we can remove insn_slot safely, because no thread calls
1963          * the original probed function (which will be freed soon) any more.
1964          */
1965         arch_remove_kprobe(p);
1966 }
1967
1968 /* Disable one kprobe */
1969 int __kprobes disable_kprobe(struct kprobe *kp)
1970 {
1971         int ret = 0;
1972
1973         mutex_lock(&kprobe_mutex);
1974
1975         /* Disable this kprobe */
1976         if (__disable_kprobe(kp) == NULL)
1977                 ret = -EINVAL;
1978
1979         mutex_unlock(&kprobe_mutex);
1980         return ret;
1981 }
1982 EXPORT_SYMBOL_GPL(disable_kprobe);
1983
1984 /* Enable one kprobe */
1985 int __kprobes enable_kprobe(struct kprobe *kp)
1986 {
1987         int ret = 0;
1988         struct kprobe *p;
1989
1990         mutex_lock(&kprobe_mutex);
1991
1992         /* Check whether specified probe is valid. */
1993         p = __get_valid_kprobe(kp);
1994         if (unlikely(p == NULL)) {
1995                 ret = -EINVAL;
1996                 goto out;
1997         }
1998
1999         if (kprobe_gone(kp)) {
2000                 /* This kprobe has gone, we couldn't enable it. */
2001                 ret = -EINVAL;
2002                 goto out;
2003         }
2004
2005         if (p != kp)
2006                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2007
2008         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2009                 p->flags &= ~KPROBE_FLAG_DISABLED;
2010                 arm_kprobe(p);
2011         }
2012 out:
2013         mutex_unlock(&kprobe_mutex);
2014         return ret;
2015 }
2016 EXPORT_SYMBOL_GPL(enable_kprobe);
2017
2018 void __kprobes dump_kprobe(struct kprobe *kp)
2019 {
2020         printk(KERN_WARNING "Dumping kprobe:\n");
2021         printk(KERN_WARNING "Name: %s\nAddress: %p\nOffset: %x\n",
2022                kp->symbol_name, kp->addr, kp->offset);
2023 }
2024
2025 /* Module notifier call back, checking kprobes on the module */
2026 static int __kprobes kprobes_module_callback(struct notifier_block *nb,
2027                                              unsigned long val, void *data)
2028 {
2029         struct module *mod = data;
2030         struct hlist_head *head;
2031         struct hlist_node *node;
2032         struct kprobe *p;
2033         unsigned int i;
2034         int checkcore = (val == MODULE_STATE_GOING);
2035
2036         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2037                 return NOTIFY_DONE;
2038
2039         /*
2040          * When MODULE_STATE_GOING was notified, both of module .text and
2041          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2042          * notified, only .init.text section would be freed. We need to
2043          * disable kprobes which have been inserted in the sections.
2044          */
2045         mutex_lock(&kprobe_mutex);
2046         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2047                 head = &kprobe_table[i];
2048                 hlist_for_each_entry_rcu(p, node, head, hlist)
2049                         if (within_module_init((unsigned long)p->addr, mod) ||
2050                             (checkcore &&
2051                              within_module_core((unsigned long)p->addr, mod))) {
2052                                 /*
2053                                  * The vaddr this probe is installed will soon
2054                                  * be vfreed buy not synced to disk. Hence,
2055                                  * disarming the breakpoint isn't needed.
2056                                  */
2057                                 kill_kprobe(p);
2058                         }
2059         }
2060         mutex_unlock(&kprobe_mutex);
2061         return NOTIFY_DONE;
2062 }
2063
2064 static struct notifier_block kprobe_module_nb = {
2065         .notifier_call = kprobes_module_callback,
2066         .priority = 0
2067 };
2068
2069 static int __init init_kprobes(void)
2070 {
2071         int i, err = 0;
2072         unsigned long offset = 0, size = 0;
2073         char *modname, namebuf[128];
2074         const char *symbol_name;
2075         void *addr;
2076         struct kprobe_blackpoint *kb;
2077
2078         /* FIXME allocate the probe table, currently defined statically */
2079         /* initialize all list heads */
2080         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2081                 INIT_HLIST_HEAD(&kprobe_table[i]);
2082                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2083                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2084         }
2085
2086         /*
2087          * Lookup and populate the kprobe_blacklist.
2088          *
2089          * Unlike the kretprobe blacklist, we'll need to determine
2090          * the range of addresses that belong to the said functions,
2091          * since a kprobe need not necessarily be at the beginning
2092          * of a function.
2093          */
2094         for (kb = kprobe_blacklist; kb->name != NULL; kb++) {
2095                 kprobe_lookup_name(kb->name, addr);
2096                 if (!addr)
2097                         continue;
2098
2099                 kb->start_addr = (unsigned long)addr;
2100                 symbol_name = kallsyms_lookup(kb->start_addr,
2101                                 &size, &offset, &modname, namebuf);
2102                 if (!symbol_name)
2103                         kb->range = 0;
2104                 else
2105                         kb->range = size;
2106         }
2107
2108         if (kretprobe_blacklist_size) {
2109                 /* lookup the function address from its name */
2110                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2111                         kprobe_lookup_name(kretprobe_blacklist[i].name,
2112                                            kretprobe_blacklist[i].addr);
2113                         if (!kretprobe_blacklist[i].addr)
2114                                 printk("kretprobe: lookup failed: %s\n",
2115                                        kretprobe_blacklist[i].name);
2116                 }
2117         }
2118
2119 #if defined(CONFIG_OPTPROBES)
2120 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2121         /* Init kprobe_optinsn_slots */
2122         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2123 #endif
2124         /* By default, kprobes can be optimized */
2125         kprobes_allow_optimization = true;
2126 #endif
2127
2128         /* By default, kprobes are armed */
2129         kprobes_all_disarmed = false;
2130
2131         err = arch_init_kprobes();
2132         if (!err)
2133                 err = register_die_notifier(&kprobe_exceptions_nb);
2134         if (!err)
2135                 err = register_module_notifier(&kprobe_module_nb);
2136
2137         kprobes_initialized = (err == 0);
2138
2139         if (!err)
2140                 init_test_probes();
2141         return err;
2142 }
2143
2144 #ifdef CONFIG_DEBUG_FS
2145 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2146                 const char *sym, int offset, char *modname, struct kprobe *pp)
2147 {
2148         char *kprobe_type;
2149
2150         if (p->pre_handler == pre_handler_kretprobe)
2151                 kprobe_type = "r";
2152         else if (p->pre_handler == setjmp_pre_handler)
2153                 kprobe_type = "j";
2154         else
2155                 kprobe_type = "k";
2156
2157         if (sym)
2158                 seq_printf(pi, "%p  %s  %s+0x%x  %s ",
2159                         p->addr, kprobe_type, sym, offset,
2160                         (modname ? modname : " "));
2161         else
2162                 seq_printf(pi, "%p  %s  %p ",
2163                         p->addr, kprobe_type, p->addr);
2164
2165         if (!pp)
2166                 pp = p;
2167         seq_printf(pi, "%s%s%s%s\n",
2168                 (kprobe_gone(p) ? "[GONE]" : ""),
2169                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2170                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2171                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2172 }
2173
2174 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2175 {
2176         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2177 }
2178
2179 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2180 {
2181         (*pos)++;
2182         if (*pos >= KPROBE_TABLE_SIZE)
2183                 return NULL;
2184         return pos;
2185 }
2186
2187 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
2188 {
2189         /* Nothing to do */
2190 }
2191
2192 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
2193 {
2194         struct hlist_head *head;
2195         struct hlist_node *node;
2196         struct kprobe *p, *kp;
2197         const char *sym = NULL;
2198         unsigned int i = *(loff_t *) v;
2199         unsigned long offset = 0;
2200         char *modname, namebuf[128];
2201
2202         head = &kprobe_table[i];
2203         preempt_disable();
2204         hlist_for_each_entry_rcu(p, node, head, hlist) {
2205                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2206                                         &offset, &modname, namebuf);
2207                 if (kprobe_aggrprobe(p)) {
2208                         list_for_each_entry_rcu(kp, &p->list, list)
2209                                 report_probe(pi, kp, sym, offset, modname, p);
2210                 } else
2211                         report_probe(pi, p, sym, offset, modname, NULL);
2212         }
2213         preempt_enable();
2214         return 0;
2215 }
2216
2217 static const struct seq_operations kprobes_seq_ops = {
2218         .start = kprobe_seq_start,
2219         .next  = kprobe_seq_next,
2220         .stop  = kprobe_seq_stop,
2221         .show  = show_kprobe_addr
2222 };
2223
2224 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
2225 {
2226         return seq_open(filp, &kprobes_seq_ops);
2227 }
2228
2229 static const struct file_operations debugfs_kprobes_operations = {
2230         .open           = kprobes_open,
2231         .read           = seq_read,
2232         .llseek         = seq_lseek,
2233         .release        = seq_release,
2234 };
2235
2236 static void __kprobes arm_all_kprobes(void)
2237 {
2238         struct hlist_head *head;
2239         struct hlist_node *node;
2240         struct kprobe *p;
2241         unsigned int i;
2242
2243         mutex_lock(&kprobe_mutex);
2244
2245         /* If kprobes are armed, just return */
2246         if (!kprobes_all_disarmed)
2247                 goto already_enabled;
2248
2249         /* Arming kprobes doesn't optimize kprobe itself */
2250         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2251                 head = &kprobe_table[i];
2252                 hlist_for_each_entry_rcu(p, node, head, hlist)
2253                         if (!kprobe_disabled(p))
2254                                 arm_kprobe(p);
2255         }
2256
2257         kprobes_all_disarmed = false;
2258         printk(KERN_INFO "Kprobes globally enabled\n");
2259
2260 already_enabled:
2261         mutex_unlock(&kprobe_mutex);
2262         return;
2263 }
2264
2265 static void __kprobes disarm_all_kprobes(void)
2266 {
2267         struct hlist_head *head;
2268         struct hlist_node *node;
2269         struct kprobe *p;
2270         unsigned int i;
2271
2272         mutex_lock(&kprobe_mutex);
2273
2274         /* If kprobes are already disarmed, just return */
2275         if (kprobes_all_disarmed) {
2276                 mutex_unlock(&kprobe_mutex);
2277                 return;
2278         }
2279
2280         kprobes_all_disarmed = true;
2281         printk(KERN_INFO "Kprobes globally disabled\n");
2282
2283         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2284                 head = &kprobe_table[i];
2285                 hlist_for_each_entry_rcu(p, node, head, hlist) {
2286                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2287                                 disarm_kprobe(p, false);
2288                 }
2289         }
2290         mutex_unlock(&kprobe_mutex);
2291
2292         /* Wait for disarming all kprobes by optimizer */
2293         wait_for_kprobe_optimizer();
2294 }
2295
2296 /*
2297  * XXX: The debugfs bool file interface doesn't allow for callbacks
2298  * when the bool state is switched. We can reuse that facility when
2299  * available
2300  */
2301 static ssize_t read_enabled_file_bool(struct file *file,
2302                char __user *user_buf, size_t count, loff_t *ppos)
2303 {
2304         char buf[3];
2305
2306         if (!kprobes_all_disarmed)
2307                 buf[0] = '1';
2308         else
2309                 buf[0] = '0';
2310         buf[1] = '\n';
2311         buf[2] = 0x00;
2312         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2313 }
2314
2315 static ssize_t write_enabled_file_bool(struct file *file,
2316                const char __user *user_buf, size_t count, loff_t *ppos)
2317 {
2318         char buf[32];
2319         size_t buf_size;
2320
2321         buf_size = min(count, (sizeof(buf)-1));
2322         if (copy_from_user(buf, user_buf, buf_size))
2323                 return -EFAULT;
2324
2325         switch (buf[0]) {
2326         case 'y':
2327         case 'Y':
2328         case '1':
2329                 arm_all_kprobes();
2330                 break;
2331         case 'n':
2332         case 'N':
2333         case '0':
2334                 disarm_all_kprobes();
2335                 break;
2336         }
2337
2338         return count;
2339 }
2340
2341 static const struct file_operations fops_kp = {
2342         .read =         read_enabled_file_bool,
2343         .write =        write_enabled_file_bool,
2344         .llseek =       default_llseek,
2345 };
2346
2347 static int __kprobes debugfs_kprobe_init(void)
2348 {
2349         struct dentry *dir, *file;
2350         unsigned int value = 1;
2351
2352         dir = debugfs_create_dir("kprobes", NULL);
2353         if (!dir)
2354                 return -ENOMEM;
2355
2356         file = debugfs_create_file("list", 0444, dir, NULL,
2357                                 &debugfs_kprobes_operations);
2358         if (!file) {
2359                 debugfs_remove(dir);
2360                 return -ENOMEM;
2361         }
2362
2363         file = debugfs_create_file("enabled", 0600, dir,
2364                                         &value, &fops_kp);
2365         if (!file) {
2366                 debugfs_remove(dir);
2367                 return -ENOMEM;
2368         }
2369
2370         return 0;
2371 }
2372
2373 late_initcall(debugfs_kprobe_init);
2374 #endif /* CONFIG_DEBUG_FS */
2375
2376 module_init(init_kprobes);
2377
2378 /* defined in arch/.../kernel/kprobes.c */
2379 EXPORT_SYMBOL_GPL(jprobe_return);