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/sections.h>
  53 #include <asm/cacheflush.h>
  54 #include <asm/errno.h>
  55 #include <linux/uaccess.h>
  56
  57 #define KPROBE_HASH_BITS 6
  58 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  59
  60
  61 static int kprobes_initialized;
  62 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  63 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  64
  65 /* NOTE: change this value only with kprobe_mutex held */
  66 static bool kprobes_all_disarmed;
  67
  68 /* This protects kprobe_table and optimizing_list */
  69 static DEFINE_MUTEX(kprobe_mutex);
  70 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  71 static struct {
  72         raw_spinlock_t lock ____cacheline_aligned_in_smp;
  73 } kretprobe_table_locks[KPROBE_TABLE_SIZE];
  74
  75 kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
  76                                         unsigned int __unused)
  77 {
  78         return ((kprobe_opcode_t *)(kallsyms_lookup_name(name)));
  79 }
  80
  81 static raw_spinlock_t *kretprobe_table_lock_ptr(unsigned long hash)
  82 {
  83         return &(kretprobe_table_locks[hash].lock);
  84 }
  85
  86 /* Blacklist -- list of struct kprobe_blacklist_entry */
  87 static LIST_HEAD(kprobe_blacklist);
  88
  89 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  90 /*
  91  * kprobe->ainsn.insn points to the copy of the instruction to be
  92  * single-stepped. x86_64, POWER4 and above have no-exec support and
  93  * stepping on the instruction on a vmalloced/kmalloced/data page
  94  * is a recipe for disaster
  95  */
  96 struct kprobe_insn_page {
  97         struct list_head list;
  98         kprobe_opcode_t *insns;         /* Page of instruction slots */
  99         struct kprobe_insn_cache *cache;
 100         int nused;
 101         int ngarbage;
 102         char slot_used[];
 103 };
 104
 105 #define KPROBE_INSN_PAGE_SIZE(slots)                    \
 106         (offsetof(struct kprobe_insn_page, slot_used) + \
 107          (sizeof(char) * (slots)))
 108
 109 static int slots_per_page(struct kprobe_insn_cache *c)
 110 {
 111         return PAGE_SIZE/(c->insn_size * sizeof(kprobe_opcode_t));
 112 }
 113
 114 enum kprobe_slot_state {
 115         SLOT_CLEAN = 0,
 116         SLOT_DIRTY = 1,
 117         SLOT_USED = 2,
 118 };
 119
 120 void __weak *alloc_insn_page(void)
 121 {
 122         return module_alloc(PAGE_SIZE);
 123 }
 124
 125 void __weak free_insn_page(void *page)
 126 {
 127         module_memfree(page);
 128 }
 129
 130 struct kprobe_insn_cache kprobe_insn_slots = {
 131         .mutex = __MUTEX_INITIALIZER(kprobe_insn_slots.mutex),
 132         .alloc = alloc_insn_page,
 133         .free = free_insn_page,
 134         .pages = LIST_HEAD_INIT(kprobe_insn_slots.pages),
 135         .insn_size = MAX_INSN_SIZE,
 136         .nr_garbage = 0,
 137 };
 138 static int collect_garbage_slots(struct kprobe_insn_cache *c);
 139
 140 /**
 141  * __get_insn_slot() - Find a slot on an executable page for an instruction.
 142  * We allocate an executable page if there's no room on existing ones.
 143  */
 144 kprobe_opcode_t *__get_insn_slot(struct kprobe_insn_cache *c)
 145 {
 146         struct kprobe_insn_page *kip;
 147         kprobe_opcode_t *slot = NULL;
 148
 149         /* Since the slot array is not protected by rcu, we need a mutex */
 150         mutex_lock(&c->mutex);
 151  retry:
 152         rcu_read_lock();
 153         list_for_each_entry_rcu(kip, &c->pages, list) {
 154                 if (kip->nused < slots_per_page(c)) {
 155                         int i;
 156                         for (i = 0; i < slots_per_page(c); i++) {
 157                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 158                                         kip->slot_used[i] = SLOT_USED;
 159                                         kip->nused++;
 160                                         slot = kip->insns + (i * c->insn_size);
 161                                         rcu_read_unlock();
 162                                         goto out;
 163                                 }
 164                         }
 165                         /* kip->nused is broken. Fix it. */
 166                         kip->nused = slots_per_page(c);
 167                         WARN_ON(1);
 168                 }
 169         }
 170         rcu_read_unlock();
 171
 172         /* If there are any garbage slots, collect it and try again. */
 173         if (c->nr_garbage && collect_garbage_slots(c) == 0)
 174                 goto retry;
 175
 176         /* All out of space.  Need to allocate a new page. */
 177         kip = kmalloc(KPROBE_INSN_PAGE_SIZE(slots_per_page(c)), GFP_KERNEL);
 178         if (!kip)
 179                 goto out;
 180
 181         /*
 182          * Use module_alloc so this page is within +/- 2GB of where the
 183          * kernel image and loaded module images reside. This is required
 184          * so x86_64 can correctly handle the %rip-relative fixups.
 185          */
 186         kip->insns = c->alloc();
 187         if (!kip->insns) {
 188                 kfree(kip);
 189                 goto out;
 190         }
 191         INIT_LIST_HEAD(&kip->list);
 192         memset(kip->slot_used, SLOT_CLEAN, slots_per_page(c));
 193         kip->slot_used[0] = SLOT_USED;
 194         kip->nused = 1;
 195         kip->ngarbage = 0;
 196         kip->cache = c;
 197         list_add_rcu(&kip->list, &c->pages);
 198         slot = kip->insns;
 199 out:
 200         mutex_unlock(&c->mutex);
 201         return slot;
 202 }
 203
 204 /* Return 1 if all garbages are collected, otherwise 0. */
 205 static int collect_one_slot(struct kprobe_insn_page *kip, int idx)
 206 {
 207         kip->slot_used[idx] = SLOT_CLEAN;
 208         kip->nused--;
 209         if (kip->nused == 0) {
 210                 /*
 211                  * Page is no longer in use.  Free it unless
 212                  * it's the last one.  We keep the last one
 213                  * so as not to have to set it up again the
 214                  * next time somebody inserts a probe.
 215                  */
 216                 if (!list_is_singular(&kip->list)) {
 217                         list_del_rcu(&kip->list);
 218                         synchronize_rcu();
 219                         kip->cache->free(kip->insns);
 220                         kfree(kip);
 221                 }
 222                 return 1;
 223         }
 224         return 0;
 225 }
 226
 227 static int collect_garbage_slots(struct kprobe_insn_cache *c)
 228 {
 229         struct kprobe_insn_page *kip, *next;
 230
 231         /* Ensure no-one is interrupted on the garbages */
 232         synchronize_sched();
 233
 234         list_for_each_entry_safe(kip, next, &c->pages, list) {
 235                 int i;
 236                 if (kip->ngarbage == 0)
 237                         continue;
 238                 kip->ngarbage = 0;      /* we will collect all garbages */
 239                 for (i = 0; i < slots_per_page(c); i++) {
 240                         if (kip->slot_used[i] == SLOT_DIRTY && collect_one_slot(kip, i))
 241                                 break;
 242                 }
 243         }
 244         c->nr_garbage = 0;
 245         return 0;
 246 }
 247
 248 void __free_insn_slot(struct kprobe_insn_cache *c,
 249                       kprobe_opcode_t *slot, int dirty)
 250 {
 251         struct kprobe_insn_page *kip;
 252         long idx;
 253
 254         mutex_lock(&c->mutex);
 255         rcu_read_lock();
 256         list_for_each_entry_rcu(kip, &c->pages, list) {
 257                 idx = ((long)slot - (long)kip->insns) /
 258                         (c->insn_size * sizeof(kprobe_opcode_t));
 259                 if (idx >= 0 && idx < slots_per_page(c))
 260                         goto out;
 261         }
 262         /* Could not find this slot. */
 263         WARN_ON(1);
 264         kip = NULL;
 265 out:
 266         rcu_read_unlock();
 267         /* Mark and sweep: this may sleep */
 268         if (kip) {
 269                 /* Check double free */
 270                 WARN_ON(kip->slot_used[idx] != SLOT_USED);
 271                 if (dirty) {
 272                         kip->slot_used[idx] = SLOT_DIRTY;
 273                         kip->ngarbage++;
 274                         if (++c->nr_garbage > slots_per_page(c))
 275                                 collect_garbage_slots(c);
 276                 } else {
 277                         collect_one_slot(kip, idx);
 278                 }
 279         }
 280         mutex_unlock(&c->mutex);
 281 }
 282
 283 /*
 284  * Check given address is on the page of kprobe instruction slots.
 285  * This will be used for checking whether the address on a stack
 286  * is on a text area or not.
 287  */
 288 bool __is_insn_slot_addr(struct kprobe_insn_cache *c, unsigned long addr)
 289 {
 290         struct kprobe_insn_page *kip;
 291         bool ret = false;
 292
 293         rcu_read_lock();
 294         list_for_each_entry_rcu(kip, &c->pages, list) {
 295                 if (addr >= (unsigned long)kip->insns &&
 296                     addr < (unsigned long)kip->insns + PAGE_SIZE) {
 297                         ret = true;
 298                         break;
 299                 }
 300         }
 301         rcu_read_unlock();
 302
 303         return ret;
 304 }
 305
 306 #ifdef CONFIG_OPTPROBES
 307 /* For optimized_kprobe buffer */
 308 struct kprobe_insn_cache kprobe_optinsn_slots = {
 309         .mutex = __MUTEX_INITIALIZER(kprobe_optinsn_slots.mutex),
 310         .alloc = alloc_insn_page,
 311         .free = free_insn_page,
 312         .pages = LIST_HEAD_INIT(kprobe_optinsn_slots.pages),
 313         /* .insn_size is initialized later */
 314         .nr_garbage = 0,
 315 };
 316 #endif
 317 #endif
 318
 319 /* We have preemption disabled.. so it is safe to use __ versions */
 320 static inline void set_kprobe_instance(struct kprobe *kp)
 321 {
 322         __this_cpu_write(kprobe_instance, kp);
 323 }
 324
 325 static inline void reset_kprobe_instance(void)
 326 {
 327         __this_cpu_write(kprobe_instance, NULL);
 328 }
 329
 330 /*
 331  * This routine is called either:
 332  *      - under the kprobe_mutex - during kprobe_[un]register()
 333  *                              OR
 334  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 335  */
 336 struct kprobe *get_kprobe(void *addr)
 337 {
 338         struct hlist_head *head;
 339         struct kprobe *p;
 340
 341         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 342         hlist_for_each_entry_rcu(p, head, hlist) {
 343                 if (p->addr == addr)
 344                         return p;
 345         }
 346
 347         return NULL;
 348 }
 349 NOKPROBE_SYMBOL(get_kprobe);
 350
 351 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs);
 352
 353 /* Return true if the kprobe is an aggregator */
 354 static inline int kprobe_aggrprobe(struct kprobe *p)
 355 {
 356         return p->pre_handler == aggr_pre_handler;
 357 }
 358
 359 /* Return true(!0) if the kprobe is unused */
 360 static inline int kprobe_unused(struct kprobe *p)
 361 {
 362         return kprobe_aggrprobe(p) && kprobe_disabled(p) &&
 363                list_empty(&p->list);
 364 }
 365
 366 /*
 367  * Keep all fields in the kprobe consistent
 368  */
 369 static inline void copy_kprobe(struct kprobe *ap, struct kprobe *p)
 370 {
 371         memcpy(&p->opcode, &ap->opcode, sizeof(kprobe_opcode_t));
 372         memcpy(&p->ainsn, &ap->ainsn, sizeof(struct arch_specific_insn));
 373 }
 374
 375 #ifdef CONFIG_OPTPROBES
 376 /* NOTE: change this value only with kprobe_mutex held */
 377 static bool kprobes_allow_optimization;
 378
 379 /*
 380  * Call all pre_handler on the list, but ignores its return value.
 381  * This must be called from arch-dep optimized caller.
 382  */
 383 void opt_pre_handler(struct kprobe *p, struct pt_regs *regs)
 384 {
 385         struct kprobe *kp;
 386
 387         list_for_each_entry_rcu(kp, &p->list, list) {
 388                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
 389                         set_kprobe_instance(kp);
 390                         kp->pre_handler(kp, regs);
 391                 }
 392                 reset_kprobe_instance();
 393         }
 394 }
 395 NOKPROBE_SYMBOL(opt_pre_handler);
 396
 397 /* Free optimized instructions and optimized_kprobe */
 398 static void free_aggr_kprobe(struct kprobe *p)
 399 {
 400         struct optimized_kprobe *op;
 401
 402         op = container_of(p, struct optimized_kprobe, kp);
 403         arch_remove_optimized_kprobe(op);
 404         arch_remove_kprobe(p);
 405         kfree(op);
 406 }
 407
 408 /* Return true(!0) if the kprobe is ready for optimization. */
 409 static inline int kprobe_optready(struct kprobe *p)
 410 {
 411         struct optimized_kprobe *op;
 412
 413         if (kprobe_aggrprobe(p)) {
 414                 op = container_of(p, struct optimized_kprobe, kp);
 415                 return arch_prepared_optinsn(&op->optinsn);
 416         }
 417
 418         return 0;
 419 }
 420
 421 /* Return true(!0) if the kprobe is disarmed. Note: p must be on hash list */
 422 static inline int kprobe_disarmed(struct kprobe *p)
 423 {
 424         struct optimized_kprobe *op;
 425
 426         /* If kprobe is not aggr/opt probe, just return kprobe is disabled */
 427         if (!kprobe_aggrprobe(p))
 428                 return kprobe_disabled(p);
 429
 430         op = container_of(p, struct optimized_kprobe, kp);
 431
 432         return kprobe_disabled(p) && list_empty(&op->list);
 433 }
 434
 435 /* Return true(!0) if the probe is queued on (un)optimizing lists */
 436 static int kprobe_queued(struct kprobe *p)
 437 {
 438         struct optimized_kprobe *op;
 439
 440         if (kprobe_aggrprobe(p)) {
 441                 op = container_of(p, struct optimized_kprobe, kp);
 442                 if (!list_empty(&op->list))
 443                         return 1;
 444         }
 445         return 0;
 446 }
 447
 448 /*
 449  * Return an optimized kprobe whose optimizing code replaces
 450  * instructions including addr (exclude breakpoint).
 451  */
 452 static struct kprobe *get_optimized_kprobe(unsigned long addr)
 453 {
 454         int i;
 455         struct kprobe *p = NULL;
 456         struct optimized_kprobe *op;
 457
 458         /* Don't check i == 0, since that is a breakpoint case. */
 459         for (i = 1; !p && i < MAX_OPTIMIZED_LENGTH; i++)
 460                 p = get_kprobe((void *)(addr - i));
 461
 462         if (p && kprobe_optready(p)) {
 463                 op = container_of(p, struct optimized_kprobe, kp);
 464                 if (arch_within_optimized_kprobe(op, addr))
 465                         return p;
 466         }
 467
 468         return NULL;
 469 }
 470
 471 /* Optimization staging list, protected by kprobe_mutex */
 472 static LIST_HEAD(optimizing_list);
 473 static LIST_HEAD(unoptimizing_list);
 474 static LIST_HEAD(freeing_list);
 475
 476 static void kprobe_optimizer(struct work_struct *work);
 477 static DECLARE_DELAYED_WORK(optimizing_work, kprobe_optimizer);
 478 #define OPTIMIZE_DELAY 5
 479
 480 /*
 481  * Optimize (replace a breakpoint with a jump) kprobes listed on
 482  * optimizing_list.
 483  */
 484 static void do_optimize_kprobes(void)
 485 {
 486         /*
 487          * The optimization/unoptimization refers online_cpus via
 488          * stop_machine() and cpu-hotplug modifies online_cpus.
 489          * And same time, text_mutex will be held in cpu-hotplug and here.
 490          * This combination can cause a deadlock (cpu-hotplug try to lock
 491          * text_mutex but stop_machine can not be done because online_cpus
 492          * has been changed)
 493          * To avoid this deadlock, caller must have locked cpu hotplug
 494          * for preventing cpu-hotplug outside of text_mutex locking.
 495          */
 496         lockdep_assert_cpus_held();
 497
 498         /* Optimization never be done when disarmed */
 499         if (kprobes_all_disarmed || !kprobes_allow_optimization ||
 500             list_empty(&optimizing_list))
 501                 return;
 502
 503         mutex_lock(&text_mutex);
 504         arch_optimize_kprobes(&optimizing_list);
 505         mutex_unlock(&text_mutex);
 506 }
 507
 508 /*
 509  * Unoptimize (replace a jump with a breakpoint and remove the breakpoint
 510  * if need) kprobes listed on unoptimizing_list.
 511  */
 512 static void do_unoptimize_kprobes(void)
 513 {
 514         struct optimized_kprobe *op, *tmp;
 515
 516         /* See comment in do_optimize_kprobes() */
 517         lockdep_assert_cpus_held();
 518
 519         /* Unoptimization must be done anytime */
 520         if (list_empty(&unoptimizing_list))
 521                 return;
 522
 523         mutex_lock(&text_mutex);
 524         arch_unoptimize_kprobes(&unoptimizing_list, &freeing_list);
 525         /* Loop free_list for disarming */
 526         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 527                 /* Disarm probes if marked disabled */
 528                 if (kprobe_disabled(&op->kp))
 529                         arch_disarm_kprobe(&op->kp);
 530                 if (kprobe_unused(&op->kp)) {
 531                         /*
 532                          * Remove unused probes from hash list. After waiting
 533                          * for synchronization, these probes are reclaimed.
 534                          * (reclaiming is done by do_free_cleaned_kprobes.)
 535                          */
 536                         hlist_del_rcu(&op->kp.hlist);
 537                 } else
 538                         list_del_init(&op->list);
 539         }
 540         mutex_unlock(&text_mutex);
 541 }
 542
 543 /* Reclaim all kprobes on the free_list */
 544 static void do_free_cleaned_kprobes(void)
 545 {
 546         struct optimized_kprobe *op, *tmp;
 547
 548         list_for_each_entry_safe(op, tmp, &freeing_list, list) {
 549                 BUG_ON(!kprobe_unused(&op->kp));
 550                 list_del_init(&op->list);
 551                 free_aggr_kprobe(&op->kp);
 552         }
 553 }
 554
 555 /* Start optimizer after OPTIMIZE_DELAY passed */
 556 static void kick_kprobe_optimizer(void)
 557 {
 558         schedule_delayed_work(&optimizing_work, OPTIMIZE_DELAY);
 559 }
 560
 561 /* Kprobe jump optimizer */
 562 static void kprobe_optimizer(struct work_struct *work)
 563 {
 564         mutex_lock(&kprobe_mutex);
 565         cpus_read_lock();
 566         /* Lock modules while optimizing kprobes */
 567         mutex_lock(&module_mutex);
 568
 569         /*
 570          * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
 571          * kprobes before waiting for quiesence period.
 572          */
 573         do_unoptimize_kprobes();
 574
 575         /*
 576          * Step 2: Wait for quiesence period to ensure all potentially
 577          * preempted tasks to have normally scheduled. Because optprobe
 578          * may modify multiple instructions, there is a chance that Nth
 579          * instruction is preempted. In that case, such tasks can return
 580          * to 2nd-Nth byte of jump instruction. This wait is for avoiding it.
 581          * Note that on non-preemptive kernel, this is transparently converted
 582          * to synchronoze_sched() to wait for all interrupts to have completed.
 583          */
 584         synchronize_rcu_tasks();
 585
 586         /* Step 3: Optimize kprobes after quiesence period */
 587         do_optimize_kprobes();
 588
 589         /* Step 4: Free cleaned kprobes after quiesence period */
 590         do_free_cleaned_kprobes();
 591
 592         mutex_unlock(&module_mutex);
 593         cpus_read_unlock();
 594         mutex_unlock(&kprobe_mutex);
 595
 596         /* Step 5: Kick optimizer again if needed */
 597         if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
 598                 kick_kprobe_optimizer();
 599 }
 600
 601 /* Wait for completing optimization and unoptimization */
 602 void wait_for_kprobe_optimizer(void)
 603 {
 604         mutex_lock(&kprobe_mutex);
 605
 606         while (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) {
 607                 mutex_unlock(&kprobe_mutex);
 608
 609                 /* this will also make optimizing_work execute immmediately */
 610                 flush_delayed_work(&optimizing_work);
 611                 /* @optimizing_work might not have been queued yet, relax */
 612                 cpu_relax();
 613
 614                 mutex_lock(&kprobe_mutex);
 615         }
 616
 617         mutex_unlock(&kprobe_mutex);
 618 }
 619
 620 /* Optimize kprobe if p is ready to be optimized */
 621 static void optimize_kprobe(struct kprobe *p)
 622 {
 623         struct optimized_kprobe *op;
 624
 625         /* Check if the kprobe is disabled or not ready for optimization. */
 626         if (!kprobe_optready(p) || !kprobes_allow_optimization ||
 627             (kprobe_disabled(p) || kprobes_all_disarmed))
 628                 return;
 629
 630         /* kprobes with post_handler can not be optimized */
 631         if (p->post_handler)
 632                 return;
 633
 634         op = container_of(p, struct optimized_kprobe, kp);
 635
 636         /* Check there is no other kprobes at the optimized instructions */
 637         if (arch_check_optimized_kprobe(op) < 0)
 638                 return;
 639
 640         /* Check if it is already optimized. */
 641         if (op->kp.flags & KPROBE_FLAG_OPTIMIZED)
 642                 return;
 643         op->kp.flags |= KPROBE_FLAG_OPTIMIZED;
 644
 645         if (!list_empty(&op->list))
 646                 /* This is under unoptimizing. Just dequeue the probe */
 647                 list_del_init(&op->list);
 648         else {
 649                 list_add(&op->list, &optimizing_list);
 650                 kick_kprobe_optimizer();
 651         }
 652 }
 653
 654 /* Short cut to direct unoptimizing */
 655 static void force_unoptimize_kprobe(struct optimized_kprobe *op)
 656 {
 657         lockdep_assert_cpus_held();
 658         arch_unoptimize_kprobe(op);
 659         if (kprobe_disabled(&op->kp))
 660                 arch_disarm_kprobe(&op->kp);
 661 }
 662
 663 /* Unoptimize a kprobe if p is optimized */
 664 static void unoptimize_kprobe(struct kprobe *p, bool force)
 665 {
 666         struct optimized_kprobe *op;
 667
 668         if (!kprobe_aggrprobe(p) || kprobe_disarmed(p))
 669                 return; /* This is not an optprobe nor optimized */
 670
 671         op = container_of(p, struct optimized_kprobe, kp);
 672         if (!kprobe_optimized(p)) {
 673                 /* Unoptimized or unoptimizing case */
 674                 if (force && !list_empty(&op->list)) {
 675                         /*
 676                          * Only if this is unoptimizing kprobe and forced,
 677                          * forcibly unoptimize it. (No need to unoptimize
 678                          * unoptimized kprobe again :)
 679                          */
 680                         list_del_init(&op->list);
 681                         force_unoptimize_kprobe(op);
 682                 }
 683                 return;
 684         }
 685
 686         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 687         if (!list_empty(&op->list)) {
 688                 /* Dequeue from the optimization queue */
 689                 list_del_init(&op->list);
 690                 return;
 691         }
 692         /* Optimized kprobe case */
 693         if (force)
 694                 /* Forcibly update the code: this is a special case */
 695                 force_unoptimize_kprobe(op);
 696         else {
 697                 list_add(&op->list, &unoptimizing_list);
 698                 kick_kprobe_optimizer();
 699         }
 700 }
 701
 702 /* Cancel unoptimizing for reusing */
 703 static void reuse_unused_kprobe(struct kprobe *ap)
 704 {
 705         struct optimized_kprobe *op;
 706
 707         BUG_ON(!kprobe_unused(ap));
 708         /*
 709          * Unused kprobe MUST be on the way of delayed unoptimizing (means
 710          * there is still a relative jump) and disabled.
 711          */
 712         op = container_of(ap, struct optimized_kprobe, kp);
 713         WARN_ON_ONCE(list_empty(&op->list));
 714         /* Enable the probe again */
 715         ap->flags &= ~KPROBE_FLAG_DISABLED;
 716         /* Optimize it again (remove from op->list) */
 717         BUG_ON(!kprobe_optready(ap));
 718         optimize_kprobe(ap);
 719 }
 720
 721 /* Remove optimized instructions */
 722 static void kill_optimized_kprobe(struct kprobe *p)
 723 {
 724         struct optimized_kprobe *op;
 725
 726         op = container_of(p, struct optimized_kprobe, kp);
 727         if (!list_empty(&op->list))
 728                 /* Dequeue from the (un)optimization queue */
 729                 list_del_init(&op->list);
 730         op->kp.flags &= ~KPROBE_FLAG_OPTIMIZED;
 731
 732         if (kprobe_unused(p)) {
 733                 /* Enqueue if it is unused */
 734                 list_add(&op->list, &freeing_list);
 735                 /*
 736                  * Remove unused probes from the hash list. After waiting
 737                  * for synchronization, this probe is reclaimed.
 738                  * (reclaiming is done by do_free_cleaned_kprobes().)
 739                  */
 740                 hlist_del_rcu(&op->kp.hlist);
 741         }
 742
 743         /* Don't touch the code, because it is already freed. */
 744         arch_remove_optimized_kprobe(op);
 745 }
 746
 747 static inline
 748 void __prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
 749 {
 750         if (!kprobe_ftrace(p))
 751                 arch_prepare_optimized_kprobe(op, p);
 752 }
 753
 754 /* Try to prepare optimized instructions */
 755 static void prepare_optimized_kprobe(struct kprobe *p)
 756 {
 757         struct optimized_kprobe *op;
 758
 759         op = container_of(p, struct optimized_kprobe, kp);
 760         __prepare_optimized_kprobe(op, p);
 761 }
 762
 763 /* Allocate new optimized_kprobe and try to prepare optimized instructions */
 764 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 765 {
 766         struct optimized_kprobe *op;
 767
 768         op = kzalloc(sizeof(struct optimized_kprobe), GFP_KERNEL);
 769         if (!op)
 770                 return NULL;
 771
 772         INIT_LIST_HEAD(&op->list);
 773         op->kp.addr = p->addr;
 774         __prepare_optimized_kprobe(op, p);
 775
 776         return &op->kp;
 777 }
 778
 779 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p);
 780
 781 /*
 782  * Prepare an optimized_kprobe and optimize it
 783  * NOTE: p must be a normal registered kprobe
 784  */
 785 static void try_to_optimize_kprobe(struct kprobe *p)
 786 {
 787         struct kprobe *ap;
 788         struct optimized_kprobe *op;
 789
 790         /* Impossible to optimize ftrace-based kprobe */
 791         if (kprobe_ftrace(p))
 792                 return;
 793
 794         /* For preparing optimization, jump_label_text_reserved() is called */
 795         cpus_read_lock();
 796         jump_label_lock();
 797         mutex_lock(&text_mutex);
 798
 799         ap = alloc_aggr_kprobe(p);
 800         if (!ap)
 801                 goto out;
 802
 803         op = container_of(ap, struct optimized_kprobe, kp);
 804         if (!arch_prepared_optinsn(&op->optinsn)) {
 805                 /* If failed to setup optimizing, fallback to kprobe */
 806                 arch_remove_optimized_kprobe(op);
 807                 kfree(op);
 808                 goto out;
 809         }
 810
 811         init_aggr_kprobe(ap, p);
 812         optimize_kprobe(ap);    /* This just kicks optimizer thread */
 813
 814 out:
 815         mutex_unlock(&text_mutex);
 816         jump_label_unlock();
 817         cpus_read_unlock();
 818 }
 819
 820 #ifdef CONFIG_SYSCTL
 821 static void optimize_all_kprobes(void)
 822 {
 823         struct hlist_head *head;
 824         struct kprobe *p;
 825         unsigned int i;
 826
 827         mutex_lock(&kprobe_mutex);
 828         /* If optimization is already allowed, just return */
 829         if (kprobes_allow_optimization)
 830                 goto out;
 831
 832         cpus_read_lock();
 833         kprobes_allow_optimization = true;
 834         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 835                 head = &kprobe_table[i];
 836                 hlist_for_each_entry_rcu(p, head, hlist)
 837                         if (!kprobe_disabled(p))
 838                                 optimize_kprobe(p);
 839         }
 840         cpus_read_unlock();
 841         printk(KERN_INFO "Kprobes globally optimized\n");
 842 out:
 843         mutex_unlock(&kprobe_mutex);
 844 }
 845
 846 static void unoptimize_all_kprobes(void)
 847 {
 848         struct hlist_head *head;
 849         struct kprobe *p;
 850         unsigned int i;
 851
 852         mutex_lock(&kprobe_mutex);
 853         /* If optimization is already prohibited, just return */
 854         if (!kprobes_allow_optimization) {
 855                 mutex_unlock(&kprobe_mutex);
 856                 return;
 857         }
 858
 859         cpus_read_lock();
 860         kprobes_allow_optimization = false;
 861         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 862                 head = &kprobe_table[i];
 863                 hlist_for_each_entry_rcu(p, head, hlist) {
 864                         if (!kprobe_disabled(p))
 865                                 unoptimize_kprobe(p, false);
 866                 }
 867         }
 868         cpus_read_unlock();
 869         mutex_unlock(&kprobe_mutex);
 870
 871         /* Wait for unoptimizing completion */
 872         wait_for_kprobe_optimizer();
 873         printk(KERN_INFO "Kprobes globally unoptimized\n");
 874 }
 875
 876 static DEFINE_MUTEX(kprobe_sysctl_mutex);
 877 int sysctl_kprobes_optimization;
 878 int proc_kprobes_optimization_handler(struct ctl_table *table, int write,
 879                                       void __user *buffer, size_t *length,
 880                                       loff_t *ppos)
 881 {
 882         int ret;
 883
 884         mutex_lock(&kprobe_sysctl_mutex);
 885         sysctl_kprobes_optimization = kprobes_allow_optimization ? 1 : 0;
 886         ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
 887
 888         if (sysctl_kprobes_optimization)
 889                 optimize_all_kprobes();
 890         else
 891                 unoptimize_all_kprobes();
 892         mutex_unlock(&kprobe_sysctl_mutex);
 893
 894         return ret;
 895 }
 896 #endif /* CONFIG_SYSCTL */
 897
 898 /* Put a breakpoint for a probe. Must be called with text_mutex locked */
 899 static void __arm_kprobe(struct kprobe *p)
 900 {
 901         struct kprobe *_p;
 902
 903         /* Check collision with other optimized kprobes */
 904         _p = get_optimized_kprobe((unsigned long)p->addr);
 905         if (unlikely(_p))
 906                 /* Fallback to unoptimized kprobe */
 907                 unoptimize_kprobe(_p, true);
 908
 909         arch_arm_kprobe(p);
 910         optimize_kprobe(p);     /* Try to optimize (add kprobe to a list) */
 911 }
 912
 913 /* Remove the breakpoint of a probe. Must be called with text_mutex locked */
 914 static void __disarm_kprobe(struct kprobe *p, bool reopt)
 915 {
 916         struct kprobe *_p;
 917
 918         /* Try to unoptimize */
 919         unoptimize_kprobe(p, kprobes_all_disarmed);
 920
 921         if (!kprobe_queued(p)) {
 922                 arch_disarm_kprobe(p);
 923                 /* If another kprobe was blocked, optimize it. */
 924                 _p = get_optimized_kprobe((unsigned long)p->addr);
 925                 if (unlikely(_p) && reopt)
 926                         optimize_kprobe(_p);
 927         }
 928         /* TODO: reoptimize others after unoptimized this probe */
 929 }
 930
 931 #else /* !CONFIG_OPTPROBES */
 932
 933 #define optimize_kprobe(p)                      do {} while (0)
 934 #define unoptimize_kprobe(p, f)                 do {} while (0)
 935 #define kill_optimized_kprobe(p)                do {} while (0)
 936 #define prepare_optimized_kprobe(p)             do {} while (0)
 937 #define try_to_optimize_kprobe(p)               do {} while (0)
 938 #define __arm_kprobe(p)                         arch_arm_kprobe(p)
 939 #define __disarm_kprobe(p, o)                   arch_disarm_kprobe(p)
 940 #define kprobe_disarmed(p)                      kprobe_disabled(p)
 941 #define wait_for_kprobe_optimizer()             do {} while (0)
 942
 943 /* There should be no unused kprobes can be reused without optimization */
 944 static void reuse_unused_kprobe(struct kprobe *ap)
 945 {
 946         printk(KERN_ERR "Error: There should be no unused kprobe here.\n");
 947         BUG_ON(kprobe_unused(ap));
 948 }
 949
 950 static void free_aggr_kprobe(struct kprobe *p)
 951 {
 952         arch_remove_kprobe(p);
 953         kfree(p);
 954 }
 955
 956 static struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
 957 {
 958         return kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 959 }
 960 #endif /* CONFIG_OPTPROBES */
 961
 962 #ifdef CONFIG_KPROBES_ON_FTRACE
 963 static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
 964         .func = kprobe_ftrace_handler,
 965         .flags = FTRACE_OPS_FL_SAVE_REGS | FTRACE_OPS_FL_IPMODIFY,
 966 };
 967 static int kprobe_ftrace_enabled;
 968
 969 /* Must ensure p->addr is really on ftrace */
 970 static int prepare_kprobe(struct kprobe *p)
 971 {
 972         if (!kprobe_ftrace(p))
 973                 return arch_prepare_kprobe(p);
 974
 975         return arch_prepare_kprobe_ftrace(p);
 976 }
 977
 978 /* Caller must lock kprobe_mutex */
 979 static int arm_kprobe_ftrace(struct kprobe *p)
 980 {
 981         int ret = 0;
 982
 983         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
 984                                    (unsigned long)p->addr, 0, 0);
 985         if (ret) {
 986                 pr_debug("Failed to arm kprobe-ftrace at %pS (%d)\n",
 987                          p->addr, ret);
 988                 return ret;
 989         }
 990
 991         if (kprobe_ftrace_enabled == 0) {
 992                 ret = register_ftrace_function(&kprobe_ftrace_ops);
 993                 if (ret) {
 994                         pr_debug("Failed to init kprobe-ftrace (%d)\n", ret);
 995                         goto err_ftrace;
 996                 }
 997         }
 998
 999         kprobe_ftrace_enabled++;
1000         return ret;
1001
1002 err_ftrace:
1003         /*
1004          * Note: Since kprobe_ftrace_ops has IPMODIFY set, and ftrace requires a
1005          * non-empty filter_hash for IPMODIFY ops, we're safe from an accidental
1006          * empty filter_hash which would undesirably trace all functions.
1007          */
1008         ftrace_set_filter_ip(&kprobe_ftrace_ops, (unsigned long)p->addr, 1, 0);
1009         return ret;
1010 }
1011
1012 /* Caller must lock kprobe_mutex */
1013 static int disarm_kprobe_ftrace(struct kprobe *p)
1014 {
1015         int ret = 0;
1016
1017         if (kprobe_ftrace_enabled == 1) {
1018                 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
1019                 if (WARN(ret < 0, "Failed to unregister kprobe-ftrace (%d)\n", ret))
1020                         return ret;
1021         }
1022
1023         kprobe_ftrace_enabled--;
1024
1025         ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
1026                            (unsigned long)p->addr, 1, 0);
1027         WARN_ONCE(ret < 0, "Failed to disarm kprobe-ftrace at %pS (%d)\n",
1028                   p->addr, ret);
1029         return ret;
1030 }
1031 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1032 #define prepare_kprobe(p)       arch_prepare_kprobe(p)
1033 #define arm_kprobe_ftrace(p)    (-ENODEV)
1034 #define disarm_kprobe_ftrace(p) (-ENODEV)
1035 #endif
1036
1037 /* Arm a kprobe with text_mutex */
1038 static int arm_kprobe(struct kprobe *kp)
1039 {
1040         if (unlikely(kprobe_ftrace(kp)))
1041                 return arm_kprobe_ftrace(kp);
1042
1043         cpus_read_lock();
1044         mutex_lock(&text_mutex);
1045         __arm_kprobe(kp);
1046         mutex_unlock(&text_mutex);
1047         cpus_read_unlock();
1048
1049         return 0;
1050 }
1051
1052 /* Disarm a kprobe with text_mutex */
1053 static int disarm_kprobe(struct kprobe *kp, bool reopt)
1054 {
1055         if (unlikely(kprobe_ftrace(kp)))
1056                 return disarm_kprobe_ftrace(kp);
1057
1058         cpus_read_lock();
1059         mutex_lock(&text_mutex);
1060         __disarm_kprobe(kp, reopt);
1061         mutex_unlock(&text_mutex);
1062         cpus_read_unlock();
1063
1064         return 0;
1065 }
1066
1067 /*
1068  * Aggregate handlers for multiple kprobes support - these handlers
1069  * take care of invoking the individual kprobe handlers on p->list
1070  */
1071 static int aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
1072 {
1073         struct kprobe *kp;
1074
1075         list_for_each_entry_rcu(kp, &p->list, list) {
1076                 if (kp->pre_handler && likely(!kprobe_disabled(kp))) {
1077                         set_kprobe_instance(kp);
1078                         if (kp->pre_handler(kp, regs))
1079                                 return 1;
1080                 }
1081                 reset_kprobe_instance();
1082         }
1083         return 0;
1084 }
1085 NOKPROBE_SYMBOL(aggr_pre_handler);
1086
1087 static void aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
1088                               unsigned long flags)
1089 {
1090         struct kprobe *kp;
1091
1092         list_for_each_entry_rcu(kp, &p->list, list) {
1093                 if (kp->post_handler && likely(!kprobe_disabled(kp))) {
1094                         set_kprobe_instance(kp);
1095                         kp->post_handler(kp, regs, flags);
1096                         reset_kprobe_instance();
1097                 }
1098         }
1099 }
1100 NOKPROBE_SYMBOL(aggr_post_handler);
1101
1102 static int aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
1103                               int trapnr)
1104 {
1105         struct kprobe *cur = __this_cpu_read(kprobe_instance);
1106
1107         /*
1108          * if we faulted "during" the execution of a user specified
1109          * probe handler, invoke just that probe's fault handler
1110          */
1111         if (cur && cur->fault_handler) {
1112                 if (cur->fault_handler(cur, regs, trapnr))
1113                         return 1;
1114         }
1115         return 0;
1116 }
1117 NOKPROBE_SYMBOL(aggr_fault_handler);
1118
1119 /* Walks the list and increments nmissed count for multiprobe case */
1120 void kprobes_inc_nmissed_count(struct kprobe *p)
1121 {
1122         struct kprobe *kp;
1123         if (!kprobe_aggrprobe(p)) {
1124                 p->nmissed++;
1125         } else {
1126                 list_for_each_entry_rcu(kp, &p->list, list)
1127                         kp->nmissed++;
1128         }
1129         return;
1130 }
1131 NOKPROBE_SYMBOL(kprobes_inc_nmissed_count);
1132
1133 void recycle_rp_inst(struct kretprobe_instance *ri,
1134                      struct hlist_head *head)
1135 {
1136         struct kretprobe *rp = ri->rp;
1137
1138         /* remove rp inst off the rprobe_inst_table */
1139         hlist_del(&ri->hlist);
1140         INIT_HLIST_NODE(&ri->hlist);
1141         if (likely(rp)) {
1142                 raw_spin_lock(&rp->lock);
1143                 hlist_add_head(&ri->hlist, &rp->free_instances);
1144                 raw_spin_unlock(&rp->lock);
1145         } else
1146                 /* Unregistering */
1147                 hlist_add_head(&ri->hlist, head);
1148 }
1149 NOKPROBE_SYMBOL(recycle_rp_inst);
1150
1151 void kretprobe_hash_lock(struct task_struct *tsk,
1152                          struct hlist_head **head, unsigned long *flags)
1153 __acquires(hlist_lock)
1154 {
1155         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1156         raw_spinlock_t *hlist_lock;
1157
1158         *head = &kretprobe_inst_table[hash];
1159         hlist_lock = kretprobe_table_lock_ptr(hash);
1160         raw_spin_lock_irqsave(hlist_lock, *flags);
1161 }
1162 NOKPROBE_SYMBOL(kretprobe_hash_lock);
1163
1164 static void kretprobe_table_lock(unsigned long hash,
1165                                  unsigned long *flags)
1166 __acquires(hlist_lock)
1167 {
1168         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1169         raw_spin_lock_irqsave(hlist_lock, *flags);
1170 }
1171 NOKPROBE_SYMBOL(kretprobe_table_lock);
1172
1173 void kretprobe_hash_unlock(struct task_struct *tsk,
1174                            unsigned long *flags)
1175 __releases(hlist_lock)
1176 {
1177         unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
1178         raw_spinlock_t *hlist_lock;
1179
1180         hlist_lock = kretprobe_table_lock_ptr(hash);
1181         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1182 }
1183 NOKPROBE_SYMBOL(kretprobe_hash_unlock);
1184
1185 static void kretprobe_table_unlock(unsigned long hash,
1186                                    unsigned long *flags)
1187 __releases(hlist_lock)
1188 {
1189         raw_spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
1190         raw_spin_unlock_irqrestore(hlist_lock, *flags);
1191 }
1192 NOKPROBE_SYMBOL(kretprobe_table_unlock);
1193
1194 /*
1195  * This function is called from finish_task_switch when task tk becomes dead,
1196  * so that we can recycle any function-return probe instances associated
1197  * with this task. These left over instances represent probed functions
1198  * that have been called but will never return.
1199  */
1200 void kprobe_flush_task(struct task_struct *tk)
1201 {
1202         struct kretprobe_instance *ri;
1203         struct hlist_head *head, empty_rp;
1204         struct hlist_node *tmp;
1205         unsigned long hash, flags = 0;
1206
1207         if (unlikely(!kprobes_initialized))
1208                 /* Early boot.  kretprobe_table_locks not yet initialized. */
1209                 return;
1210
1211         INIT_HLIST_HEAD(&empty_rp);
1212         hash = hash_ptr(tk, KPROBE_HASH_BITS);
1213         head = &kretprobe_inst_table[hash];
1214         kretprobe_table_lock(hash, &flags);
1215         hlist_for_each_entry_safe(ri, tmp, head, hlist) {
1216                 if (ri->task == tk)
1217                         recycle_rp_inst(ri, &empty_rp);
1218         }
1219         kretprobe_table_unlock(hash, &flags);
1220         hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) {
1221                 hlist_del(&ri->hlist);
1222                 kfree(ri);
1223         }
1224 }
1225 NOKPROBE_SYMBOL(kprobe_flush_task);
1226
1227 static inline void free_rp_inst(struct kretprobe *rp)
1228 {
1229         struct kretprobe_instance *ri;
1230         struct hlist_node *next;
1231
1232         hlist_for_each_entry_safe(ri, next, &rp->free_instances, hlist) {
1233                 hlist_del(&ri->hlist);
1234                 kfree(ri);
1235         }
1236 }
1237
1238 static void cleanup_rp_inst(struct kretprobe *rp)
1239 {
1240         unsigned long flags, hash;
1241         struct kretprobe_instance *ri;
1242         struct hlist_node *next;
1243         struct hlist_head *head;
1244
1245         /* No race here */
1246         for (hash = 0; hash < KPROBE_TABLE_SIZE; hash++) {
1247                 kretprobe_table_lock(hash, &flags);
1248                 head = &kretprobe_inst_table[hash];
1249                 hlist_for_each_entry_safe(ri, next, head, hlist) {
1250                         if (ri->rp == rp)
1251                                 ri->rp = NULL;
1252                 }
1253                 kretprobe_table_unlock(hash, &flags);
1254         }
1255         free_rp_inst(rp);
1256 }
1257 NOKPROBE_SYMBOL(cleanup_rp_inst);
1258
1259 /* Add the new probe to ap->list */
1260 static int add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1261 {
1262         BUG_ON(kprobe_gone(ap) || kprobe_gone(p));
1263
1264         if (p->post_handler)
1265                 unoptimize_kprobe(ap, true);    /* Fall back to normal kprobe */
1266
1267         list_add_rcu(&p->list, &ap->list);
1268         if (p->post_handler && !ap->post_handler)
1269                 ap->post_handler = aggr_post_handler;
1270
1271         return 0;
1272 }
1273
1274 /*
1275  * Fill in the required fields of the "manager kprobe". Replace the
1276  * earlier kprobe in the hlist with the manager kprobe
1277  */
1278 static void init_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
1279 {
1280         /* Copy p's insn slot to ap */
1281         copy_kprobe(p, ap);
1282         flush_insn_slot(ap);
1283         ap->addr = p->addr;
1284         ap->flags = p->flags & ~KPROBE_FLAG_OPTIMIZED;
1285         ap->pre_handler = aggr_pre_handler;
1286         ap->fault_handler = aggr_fault_handler;
1287         /* We don't care the kprobe which has gone. */
1288         if (p->post_handler && !kprobe_gone(p))
1289                 ap->post_handler = aggr_post_handler;
1290
1291         INIT_LIST_HEAD(&ap->list);
1292         INIT_HLIST_NODE(&ap->hlist);
1293
1294         list_add_rcu(&p->list, &ap->list);
1295         hlist_replace_rcu(&p->hlist, &ap->hlist);
1296 }
1297
1298 /*
1299  * This is the second or subsequent kprobe at the address - handle
1300  * the intricacies
1301  */
1302 static int register_aggr_kprobe(struct kprobe *orig_p, struct kprobe *p)
1303 {
1304         int ret = 0;
1305         struct kprobe *ap = orig_p;
1306
1307         cpus_read_lock();
1308
1309         /* For preparing optimization, jump_label_text_reserved() is called */
1310         jump_label_lock();
1311         mutex_lock(&text_mutex);
1312
1313         if (!kprobe_aggrprobe(orig_p)) {
1314                 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1315                 ap = alloc_aggr_kprobe(orig_p);
1316                 if (!ap) {
1317                         ret = -ENOMEM;
1318                         goto out;
1319                 }
1320                 init_aggr_kprobe(ap, orig_p);
1321         } else if (kprobe_unused(ap))
1322                 /* This probe is going to die. Rescue it */
1323                 reuse_unused_kprobe(ap);
1324
1325         if (kprobe_gone(ap)) {
1326                 /*
1327                  * Attempting to insert new probe at the same location that
1328                  * had a probe in the module vaddr area which already
1329                  * freed. So, the instruction slot has already been
1330                  * released. We need a new slot for the new probe.
1331                  */
1332                 ret = arch_prepare_kprobe(ap);
1333                 if (ret)
1334                         /*
1335                          * Even if fail to allocate new slot, don't need to
1336                          * free aggr_probe. It will be used next time, or
1337                          * freed by unregister_kprobe.
1338                          */
1339                         goto out;
1340
1341                 /* Prepare optimized instructions if possible. */
1342                 prepare_optimized_kprobe(ap);
1343
1344                 /*
1345                  * Clear gone flag to prevent allocating new slot again, and
1346                  * set disabled flag because it is not armed yet.
1347                  */
1348                 ap->flags = (ap->flags & ~KPROBE_FLAG_GONE)
1349                             | KPROBE_FLAG_DISABLED;
1350         }
1351
1352         /* Copy ap's insn slot to p */
1353         copy_kprobe(ap, p);
1354         ret = add_new_kprobe(ap, p);
1355
1356 out:
1357         mutex_unlock(&text_mutex);
1358         jump_label_unlock();
1359         cpus_read_unlock();
1360
1361         if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1362                 ap->flags &= ~KPROBE_FLAG_DISABLED;
1363                 if (!kprobes_all_disarmed) {
1364                         /* Arm the breakpoint again. */
1365                         ret = arm_kprobe(ap);
1366                         if (ret) {
1367                                 ap->flags |= KPROBE_FLAG_DISABLED;
1368                                 list_del_rcu(&p->list);
1369                                 synchronize_sched();
1370                         }
1371                 }
1372         }
1373         return ret;
1374 }
1375
1376 bool __weak arch_within_kprobe_blacklist(unsigned long addr)
1377 {
1378         /* The __kprobes marked functions and entry code must not be probed */
1379         return addr >= (unsigned long)__kprobes_text_start &&
1380                addr < (unsigned long)__kprobes_text_end;
1381 }
1382
1383 bool within_kprobe_blacklist(unsigned long addr)
1384 {
1385         struct kprobe_blacklist_entry *ent;
1386
1387         if (arch_within_kprobe_blacklist(addr))
1388                 return true;
1389         /*
1390          * If there exists a kprobe_blacklist, verify and
1391          * fail any probe registration in the prohibited area
1392          */
1393         list_for_each_entry(ent, &kprobe_blacklist, list) {
1394                 if (addr >= ent->start_addr && addr < ent->end_addr)
1395                         return true;
1396         }
1397
1398         return false;
1399 }
1400
1401 /*
1402  * If we have a symbol_name argument, look it up and add the offset field
1403  * to it. This way, we can specify a relative address to a symbol.
1404  * This returns encoded errors if it fails to look up symbol or invalid
1405  * combination of parameters.
1406  */
1407 static kprobe_opcode_t *_kprobe_addr(kprobe_opcode_t *addr,
1408                         const char *symbol_name, unsigned int offset)
1409 {
1410         if ((symbol_name && addr) || (!symbol_name && !addr))
1411                 goto invalid;
1412
1413         if (symbol_name) {
1414                 addr = kprobe_lookup_name(symbol_name, offset);
1415                 if (!addr)
1416                         return ERR_PTR(-ENOENT);
1417         }
1418
1419         addr = (kprobe_opcode_t *)(((char *)addr) + offset);
1420         if (addr)
1421                 return addr;
1422
1423 invalid:
1424         return ERR_PTR(-EINVAL);
1425 }
1426
1427 static kprobe_opcode_t *kprobe_addr(struct kprobe *p)
1428 {
1429         return _kprobe_addr(p->addr, p->symbol_name, p->offset);
1430 }
1431
1432 /* Check passed kprobe is valid and return kprobe in kprobe_table. */
1433 static struct kprobe *__get_valid_kprobe(struct kprobe *p)
1434 {
1435         struct kprobe *ap, *list_p;
1436
1437         ap = get_kprobe(p->addr);
1438         if (unlikely(!ap))
1439                 return NULL;
1440
1441         if (p != ap) {
1442                 list_for_each_entry_rcu(list_p, &ap->list, list)
1443                         if (list_p == p)
1444                         /* kprobe p is a valid probe */
1445                                 goto valid;
1446                 return NULL;
1447         }
1448 valid:
1449         return ap;
1450 }
1451
1452 /* Return error if the kprobe is being re-registered */
1453 static inline int check_kprobe_rereg(struct kprobe *p)
1454 {
1455         int ret = 0;
1456
1457         mutex_lock(&kprobe_mutex);
1458         if (__get_valid_kprobe(p))
1459                 ret = -EINVAL;
1460         mutex_unlock(&kprobe_mutex);
1461
1462         return ret;
1463 }
1464
1465 int __weak arch_check_ftrace_location(struct kprobe *p)
1466 {
1467         unsigned long ftrace_addr;
1468
1469         ftrace_addr = ftrace_location((unsigned long)p->addr);
1470         if (ftrace_addr) {
1471 #ifdef CONFIG_KPROBES_ON_FTRACE
1472                 /* Given address is not on the instruction boundary */
1473                 if ((unsigned long)p->addr != ftrace_addr)
1474                         return -EILSEQ;
1475                 p->flags |= KPROBE_FLAG_FTRACE;
1476 #else   /* !CONFIG_KPROBES_ON_FTRACE */
1477                 return -EINVAL;
1478 #endif
1479         }
1480         return 0;
1481 }
1482
1483 static int check_kprobe_address_safe(struct kprobe *p,
1484                                      struct module **probed_mod)
1485 {
1486         int ret;
1487
1488         ret = arch_check_ftrace_location(p);
1489         if (ret)
1490                 return ret;
1491         jump_label_lock();
1492         preempt_disable();
1493
1494         /* Ensure it is not in reserved area nor out of text */
1495         if (!kernel_text_address((unsigned long) p->addr) ||
1496             within_kprobe_blacklist((unsigned long) p->addr) ||
1497             jump_label_text_reserved(p->addr, p->addr)) {
1498                 ret = -EINVAL;
1499                 goto out;
1500         }
1501
1502         /* Check if are we probing a module */
1503         *probed_mod = __module_text_address((unsigned long) p->addr);
1504         if (*probed_mod) {
1505                 /*
1506                  * We must hold a refcount of the probed module while updating
1507                  * its code to prohibit unexpected unloading.
1508                  */
1509                 if (unlikely(!try_module_get(*probed_mod))) {
1510                         ret = -ENOENT;
1511                         goto out;
1512                 }
1513
1514                 /*
1515                  * If the module freed .init.text, we couldn't insert
1516                  * kprobes in there.
1517                  */
1518                 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1519                     (*probed_mod)->state != MODULE_STATE_COMING) {
1520                         module_put(*probed_mod);
1521                         *probed_mod = NULL;
1522                         ret = -ENOENT;
1523                 }
1524         }
1525 out:
1526         preempt_enable();
1527         jump_label_unlock();
1528
1529         return ret;
1530 }
1531
1532 int register_kprobe(struct kprobe *p)
1533 {
1534         int ret;
1535         struct kprobe *old_p;
1536         struct module *probed_mod;
1537         kprobe_opcode_t *addr;
1538
1539         /* Adjust probe address from symbol */
1540         addr = kprobe_addr(p);
1541         if (IS_ERR(addr))
1542                 return PTR_ERR(addr);
1543         p->addr = addr;
1544
1545         ret = check_kprobe_rereg(p);
1546         if (ret)
1547                 return ret;
1548
1549         /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1550         p->flags &= KPROBE_FLAG_DISABLED;
1551         p->nmissed = 0;
1552         INIT_LIST_HEAD(&p->list);
1553
1554         ret = check_kprobe_address_safe(p, &probed_mod);
1555         if (ret)
1556                 return ret;
1557
1558         mutex_lock(&kprobe_mutex);
1559
1560         old_p = get_kprobe(p->addr);
1561         if (old_p) {
1562                 /* Since this may unoptimize old_p, locking text_mutex. */
1563                 ret = register_aggr_kprobe(old_p, p);
1564                 goto out;
1565         }
1566
1567         cpus_read_lock();
1568         /* Prevent text modification */
1569         mutex_lock(&text_mutex);
1570         ret = prepare_kprobe(p);
1571         mutex_unlock(&text_mutex);
1572         cpus_read_unlock();
1573         if (ret)
1574                 goto out;
1575
1576         INIT_HLIST_NODE(&p->hlist);
1577         hlist_add_head_rcu(&p->hlist,
1578                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1579
1580         if (!kprobes_all_disarmed && !kprobe_disabled(p)) {
1581                 ret = arm_kprobe(p);
1582                 if (ret) {
1583                         hlist_del_rcu(&p->hlist);
1584                         synchronize_sched();
1585                         goto out;
1586                 }
1587         }
1588
1589         /* Try to optimize kprobe */
1590         try_to_optimize_kprobe(p);
1591 out:
1592         mutex_unlock(&kprobe_mutex);
1593
1594         if (probed_mod)
1595                 module_put(probed_mod);
1596
1597         return ret;
1598 }
1599 EXPORT_SYMBOL_GPL(register_kprobe);
1600
1601 /* Check if all probes on the aggrprobe are disabled */
1602 static int aggr_kprobe_disabled(struct kprobe *ap)
1603 {
1604         struct kprobe *kp;
1605
1606         list_for_each_entry_rcu(kp, &ap->list, list)
1607                 if (!kprobe_disabled(kp))
1608                         /*
1609                          * There is an active probe on the list.
1610                          * We can't disable this ap.
1611                          */
1612                         return 0;
1613
1614         return 1;
1615 }
1616
1617 /* Disable one kprobe: Make sure called under kprobe_mutex is locked */
1618 static struct kprobe *__disable_kprobe(struct kprobe *p)
1619 {
1620         struct kprobe *orig_p;
1621         int ret;
1622
1623         /* Get an original kprobe for return */
1624         orig_p = __get_valid_kprobe(p);
1625         if (unlikely(orig_p == NULL))
1626                 return ERR_PTR(-EINVAL);
1627
1628         if (!kprobe_disabled(p)) {
1629                 /* Disable probe if it is a child probe */
1630                 if (p != orig_p)
1631                         p->flags |= KPROBE_FLAG_DISABLED;
1632
1633                 /* Try to disarm and disable this/parent probe */
1634                 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1635                         /*
1636                          * If kprobes_all_disarmed is set, orig_p
1637                          * should have already been disarmed, so
1638                          * skip unneed disarming process.
1639                          */
1640                         if (!kprobes_all_disarmed) {
1641                                 ret = disarm_kprobe(orig_p, true);
1642                                 if (ret) {
1643                                         p->flags &= ~KPROBE_FLAG_DISABLED;
1644                                         return ERR_PTR(ret);
1645                                 }
1646                         }
1647                         orig_p->flags |= KPROBE_FLAG_DISABLED;
1648                 }
1649         }
1650
1651         return orig_p;
1652 }
1653
1654 /*
1655  * Unregister a kprobe without a scheduler synchronization.
1656  */
1657 static int __unregister_kprobe_top(struct kprobe *p)
1658 {
1659         struct kprobe *ap, *list_p;
1660
1661         /* Disable kprobe. This will disarm it if needed. */
1662         ap = __disable_kprobe(p);
1663         if (IS_ERR(ap))
1664                 return PTR_ERR(ap);
1665
1666         if (ap == p)
1667                 /*
1668                  * This probe is an independent(and non-optimized) kprobe
1669                  * (not an aggrprobe). Remove from the hash list.
1670                  */
1671                 goto disarmed;
1672
1673         /* Following process expects this probe is an aggrprobe */
1674         WARN_ON(!kprobe_aggrprobe(ap));
1675
1676         if (list_is_singular(&ap->list) && kprobe_disarmed(ap))
1677                 /*
1678                  * !disarmed could be happen if the probe is under delayed
1679                  * unoptimizing.
1680                  */
1681                 goto disarmed;
1682         else {
1683                 /* If disabling probe has special handlers, update aggrprobe */
1684                 if (p->post_handler && !kprobe_gone(p)) {
1685                         list_for_each_entry_rcu(list_p, &ap->list, list) {
1686                                 if ((list_p != p) && (list_p->post_handler))
1687                                         goto noclean;
1688                         }
1689                         ap->post_handler = NULL;
1690                 }
1691 noclean:
1692                 /*
1693                  * Remove from the aggrprobe: this path will do nothing in
1694                  * __unregister_kprobe_bottom().
1695                  */
1696                 list_del_rcu(&p->list);
1697                 if (!kprobe_disabled(ap) && !kprobes_all_disarmed)
1698                         /*
1699                          * Try to optimize this probe again, because post
1700                          * handler may have been changed.
1701                          */
1702                         optimize_kprobe(ap);
1703         }
1704         return 0;
1705
1706 disarmed:
1707         BUG_ON(!kprobe_disarmed(ap));
1708         hlist_del_rcu(&ap->hlist);
1709         return 0;
1710 }
1711
1712 static void __unregister_kprobe_bottom(struct kprobe *p)
1713 {
1714         struct kprobe *ap;
1715
1716         if (list_empty(&p->list))
1717                 /* This is an independent kprobe */
1718                 arch_remove_kprobe(p);
1719         else if (list_is_singular(&p->list)) {
1720                 /* This is the last child of an aggrprobe */
1721                 ap = list_entry(p->list.next, struct kprobe, list);
1722                 list_del(&p->list);
1723                 free_aggr_kprobe(ap);
1724         }
1725         /* Otherwise, do nothing. */
1726 }
1727
1728 int register_kprobes(struct kprobe **kps, int num)
1729 {
1730         int i, ret = 0;
1731
1732         if (num <= 0)
1733                 return -EINVAL;
1734         for (i = 0; i < num; i++) {
1735                 ret = register_kprobe(kps[i]);
1736                 if (ret < 0) {
1737                         if (i > 0)
1738                                 unregister_kprobes(kps, i);
1739                         break;
1740                 }
1741         }
1742         return ret;
1743 }
1744 EXPORT_SYMBOL_GPL(register_kprobes);
1745
1746 void unregister_kprobe(struct kprobe *p)
1747 {
1748         unregister_kprobes(&p, 1);
1749 }
1750 EXPORT_SYMBOL_GPL(unregister_kprobe);
1751
1752 void unregister_kprobes(struct kprobe **kps, int num)
1753 {
1754         int i;
1755
1756         if (num <= 0)
1757                 return;
1758         mutex_lock(&kprobe_mutex);
1759         for (i = 0; i < num; i++)
1760                 if (__unregister_kprobe_top(kps[i]) < 0)
1761                         kps[i]->addr = NULL;
1762         mutex_unlock(&kprobe_mutex);
1763
1764         synchronize_sched();
1765         for (i = 0; i < num; i++)
1766                 if (kps[i]->addr)
1767                         __unregister_kprobe_bottom(kps[i]);
1768 }
1769 EXPORT_SYMBOL_GPL(unregister_kprobes);
1770
1771 int __weak kprobe_exceptions_notify(struct notifier_block *self,
1772                                         unsigned long val, void *data)
1773 {
1774         return NOTIFY_DONE;
1775 }
1776 NOKPROBE_SYMBOL(kprobe_exceptions_notify);
1777
1778 static struct notifier_block kprobe_exceptions_nb = {
1779         .notifier_call = kprobe_exceptions_notify,
1780         .priority = 0x7fffffff /* we need to be notified first */
1781 };
1782
1783 unsigned long __weak arch_deref_entry_point(void *entry)
1784 {
1785         return (unsigned long)entry;
1786 }
1787
1788 #ifdef CONFIG_KRETPROBES
1789 /*
1790  * This kprobe pre_handler is registered with every kretprobe. When probe
1791  * hits it will set up the return probe.
1792  */
1793 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1794 {
1795         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
1796         unsigned long hash, flags = 0;
1797         struct kretprobe_instance *ri;
1798
1799         /*
1800          * To avoid deadlocks, prohibit return probing in NMI contexts,
1801          * just skip the probe and increase the (inexact) 'nmissed'
1802          * statistical counter, so that the user is informed that
1803          * something happened:
1804          */
1805         if (unlikely(in_nmi())) {
1806                 rp->nmissed++;
1807                 return 0;
1808         }
1809
1810         /* TODO: consider to only swap the RA after the last pre_handler fired */
1811         hash = hash_ptr(current, KPROBE_HASH_BITS);
1812         raw_spin_lock_irqsave(&rp->lock, flags);
1813         if (!hlist_empty(&rp->free_instances)) {
1814                 ri = hlist_entry(rp->free_instances.first,
1815                                 struct kretprobe_instance, hlist);
1816                 hlist_del(&ri->hlist);
1817                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1818
1819                 ri->rp = rp;
1820                 ri->task = current;
1821
1822                 if (rp->entry_handler && rp->entry_handler(ri, regs)) {
1823                         raw_spin_lock_irqsave(&rp->lock, flags);
1824                         hlist_add_head(&ri->hlist, &rp->free_instances);
1825                         raw_spin_unlock_irqrestore(&rp->lock, flags);
1826                         return 0;
1827                 }
1828
1829                 arch_prepare_kretprobe(ri, regs);
1830
1831                 /* XXX(hch): why is there no hlist_move_head? */
1832                 INIT_HLIST_NODE(&ri->hlist);
1833                 kretprobe_table_lock(hash, &flags);
1834                 hlist_add_head(&ri->hlist, &kretprobe_inst_table[hash]);
1835                 kretprobe_table_unlock(hash, &flags);
1836         } else {
1837                 rp->nmissed++;
1838                 raw_spin_unlock_irqrestore(&rp->lock, flags);
1839         }
1840         return 0;
1841 }
1842 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1843
1844 bool __weak arch_kprobe_on_func_entry(unsigned long offset)
1845 {
1846         return !offset;
1847 }
1848
1849 bool kprobe_on_func_entry(kprobe_opcode_t *addr, const char *sym, unsigned long offset)
1850 {
1851         kprobe_opcode_t *kp_addr = _kprobe_addr(addr, sym, offset);
1852
1853         if (IS_ERR(kp_addr))
1854                 return false;
1855
1856         if (!kallsyms_lookup_size_offset((unsigned long)kp_addr, NULL, &offset) ||
1857                                                 !arch_kprobe_on_func_entry(offset))
1858                 return false;
1859
1860         return true;
1861 }
1862
1863 int register_kretprobe(struct kretprobe *rp)
1864 {
1865         int ret = 0;
1866         struct kretprobe_instance *inst;
1867         int i;
1868         void *addr;
1869
1870         if (!kprobe_on_func_entry(rp->kp.addr, rp->kp.symbol_name, rp->kp.offset))
1871                 return -EINVAL;
1872
1873         if (kretprobe_blacklist_size) {
1874                 addr = kprobe_addr(&rp->kp);
1875                 if (IS_ERR(addr))
1876                         return PTR_ERR(addr);
1877
1878                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
1879                         if (kretprobe_blacklist[i].addr == addr)
1880                                 return -EINVAL;
1881                 }
1882         }
1883
1884         rp->kp.pre_handler = pre_handler_kretprobe;
1885         rp->kp.post_handler = NULL;
1886         rp->kp.fault_handler = NULL;
1887
1888         /* Pre-allocate memory for max kretprobe instances */
1889         if (rp->maxactive <= 0) {
1890 #ifdef CONFIG_PREEMPT
1891                 rp->maxactive = max_t(unsigned int, 10, 2*num_possible_cpus());
1892 #else
1893                 rp->maxactive = num_possible_cpus();
1894 #endif
1895         }
1896         raw_spin_lock_init(&rp->lock);
1897         INIT_HLIST_HEAD(&rp->free_instances);
1898         for (i = 0; i < rp->maxactive; i++) {
1899                 inst = kmalloc(sizeof(struct kretprobe_instance) +
1900                                rp->data_size, GFP_KERNEL);
1901                 if (inst == NULL) {
1902                         free_rp_inst(rp);
1903                         return -ENOMEM;
1904                 }
1905                 INIT_HLIST_NODE(&inst->hlist);
1906                 hlist_add_head(&inst->hlist, &rp->free_instances);
1907         }
1908
1909         rp->nmissed = 0;
1910         /* Establish function entry probe point */
1911         ret = register_kprobe(&rp->kp);
1912         if (ret != 0)
1913                 free_rp_inst(rp);
1914         return ret;
1915 }
1916 EXPORT_SYMBOL_GPL(register_kretprobe);
1917
1918 int register_kretprobes(struct kretprobe **rps, int num)
1919 {
1920         int ret = 0, i;
1921
1922         if (num <= 0)
1923                 return -EINVAL;
1924         for (i = 0; i < num; i++) {
1925                 ret = register_kretprobe(rps[i]);
1926                 if (ret < 0) {
1927                         if (i > 0)
1928                                 unregister_kretprobes(rps, i);
1929                         break;
1930                 }
1931         }
1932         return ret;
1933 }
1934 EXPORT_SYMBOL_GPL(register_kretprobes);
1935
1936 void unregister_kretprobe(struct kretprobe *rp)
1937 {
1938         unregister_kretprobes(&rp, 1);
1939 }
1940 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1941
1942 void unregister_kretprobes(struct kretprobe **rps, int num)
1943 {
1944         int i;
1945
1946         if (num <= 0)
1947                 return;
1948         mutex_lock(&kprobe_mutex);
1949         for (i = 0; i < num; i++)
1950                 if (__unregister_kprobe_top(&rps[i]->kp) < 0)
1951                         rps[i]->kp.addr = NULL;
1952         mutex_unlock(&kprobe_mutex);
1953
1954         synchronize_sched();
1955         for (i = 0; i < num; i++) {
1956                 if (rps[i]->kp.addr) {
1957                         __unregister_kprobe_bottom(&rps[i]->kp);
1958                         cleanup_rp_inst(rps[i]);
1959                 }
1960         }
1961 }
1962 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1963
1964 #else /* CONFIG_KRETPROBES */
1965 int register_kretprobe(struct kretprobe *rp)
1966 {
1967         return -ENOSYS;
1968 }
1969 EXPORT_SYMBOL_GPL(register_kretprobe);
1970
1971 int register_kretprobes(struct kretprobe **rps, int num)
1972 {
1973         return -ENOSYS;
1974 }
1975 EXPORT_SYMBOL_GPL(register_kretprobes);
1976
1977 void unregister_kretprobe(struct kretprobe *rp)
1978 {
1979 }
1980 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1981
1982 void unregister_kretprobes(struct kretprobe **rps, int num)
1983 {
1984 }
1985 EXPORT_SYMBOL_GPL(unregister_kretprobes);
1986
1987 static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs)
1988 {
1989         return 0;
1990 }
1991 NOKPROBE_SYMBOL(pre_handler_kretprobe);
1992
1993 #endif /* CONFIG_KRETPROBES */
1994
1995 /* Set the kprobe gone and remove its instruction buffer. */
1996 static void kill_kprobe(struct kprobe *p)
1997 {
1998         struct kprobe *kp;
1999
2000         p->flags |= KPROBE_FLAG_GONE;
2001         if (kprobe_aggrprobe(p)) {
2002                 /*
2003                  * If this is an aggr_kprobe, we have to list all the
2004                  * chained probes and mark them GONE.
2005                  */
2006                 list_for_each_entry_rcu(kp, &p->list, list)
2007                         kp->flags |= KPROBE_FLAG_GONE;
2008                 p->post_handler = NULL;
2009                 kill_optimized_kprobe(p);
2010         }
2011         /*
2012          * Here, we can remove insn_slot safely, because no thread calls
2013          * the original probed function (which will be freed soon) any more.
2014          */
2015         arch_remove_kprobe(p);
2016 }
2017
2018 /* Disable one kprobe */
2019 int disable_kprobe(struct kprobe *kp)
2020 {
2021         int ret = 0;
2022         struct kprobe *p;
2023
2024         mutex_lock(&kprobe_mutex);
2025
2026         /* Disable this kprobe */
2027         p = __disable_kprobe(kp);
2028         if (IS_ERR(p))
2029                 ret = PTR_ERR(p);
2030
2031         mutex_unlock(&kprobe_mutex);
2032         return ret;
2033 }
2034 EXPORT_SYMBOL_GPL(disable_kprobe);
2035
2036 /* Enable one kprobe */
2037 int enable_kprobe(struct kprobe *kp)
2038 {
2039         int ret = 0;
2040         struct kprobe *p;
2041
2042         mutex_lock(&kprobe_mutex);
2043
2044         /* Check whether specified probe is valid. */
2045         p = __get_valid_kprobe(kp);
2046         if (unlikely(p == NULL)) {
2047                 ret = -EINVAL;
2048                 goto out;
2049         }
2050
2051         if (kprobe_gone(kp)) {
2052                 /* This kprobe has gone, we couldn't enable it. */
2053                 ret = -EINVAL;
2054                 goto out;
2055         }
2056
2057         if (p != kp)
2058                 kp->flags &= ~KPROBE_FLAG_DISABLED;
2059
2060         if (!kprobes_all_disarmed && kprobe_disabled(p)) {
2061                 p->flags &= ~KPROBE_FLAG_DISABLED;
2062                 ret = arm_kprobe(p);
2063                 if (ret)
2064                         p->flags |= KPROBE_FLAG_DISABLED;
2065         }
2066 out:
2067         mutex_unlock(&kprobe_mutex);
2068         return ret;
2069 }
2070 EXPORT_SYMBOL_GPL(enable_kprobe);
2071
2072 /* Caller must NOT call this in usual path. This is only for critical case */
2073 void dump_kprobe(struct kprobe *kp)
2074 {
2075         pr_err("Dumping kprobe:\n");
2076         pr_err("Name: %s\nOffset: %x\nAddress: %pS\n",
2077                kp->symbol_name, kp->offset, kp->addr);
2078 }
2079 NOKPROBE_SYMBOL(dump_kprobe);
2080
2081 /*
2082  * Lookup and populate the kprobe_blacklist.
2083  *
2084  * Unlike the kretprobe blacklist, we'll need to determine
2085  * the range of addresses that belong to the said functions,
2086  * since a kprobe need not necessarily be at the beginning
2087  * of a function.
2088  */
2089 static int __init populate_kprobe_blacklist(unsigned long *start,
2090                                              unsigned long *end)
2091 {
2092         unsigned long *iter;
2093         struct kprobe_blacklist_entry *ent;
2094         unsigned long entry, offset = 0, size = 0;
2095
2096         for (iter = start; iter < end; iter++) {
2097                 entry = arch_deref_entry_point((void *)*iter);
2098
2099                 if (!kernel_text_address(entry) ||
2100                     !kallsyms_lookup_size_offset(entry, &size, &offset))
2101                         continue;
2102
2103                 ent = kmalloc(sizeof(*ent), GFP_KERNEL);
2104                 if (!ent)
2105                         return -ENOMEM;
2106                 ent->start_addr = entry;
2107                 ent->end_addr = entry + size;
2108                 INIT_LIST_HEAD(&ent->list);
2109                 list_add_tail(&ent->list, &kprobe_blacklist);
2110         }
2111         return 0;
2112 }
2113
2114 /* Module notifier call back, checking kprobes on the module */
2115 static int kprobes_module_callback(struct notifier_block *nb,
2116                                    unsigned long val, void *data)
2117 {
2118         struct module *mod = data;
2119         struct hlist_head *head;
2120         struct kprobe *p;
2121         unsigned int i;
2122         int checkcore = (val == MODULE_STATE_GOING);
2123
2124         if (val != MODULE_STATE_GOING && val != MODULE_STATE_LIVE)
2125                 return NOTIFY_DONE;
2126
2127         /*
2128          * When MODULE_STATE_GOING was notified, both of module .text and
2129          * .init.text sections would be freed. When MODULE_STATE_LIVE was
2130          * notified, only .init.text section would be freed. We need to
2131          * disable kprobes which have been inserted in the sections.
2132          */
2133         mutex_lock(&kprobe_mutex);
2134         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2135                 head = &kprobe_table[i];
2136                 hlist_for_each_entry_rcu(p, head, hlist)
2137                         if (within_module_init((unsigned long)p->addr, mod) ||
2138                             (checkcore &&
2139                              within_module_core((unsigned long)p->addr, mod))) {
2140                                 /*
2141                                  * The vaddr this probe is installed will soon
2142                                  * be vfreed buy not synced to disk. Hence,
2143                                  * disarming the breakpoint isn't needed.
2144                                  *
2145                                  * Note, this will also move any optimized probes
2146                                  * that are pending to be removed from their
2147                                  * corresponding lists to the freeing_list and
2148                                  * will not be touched by the delayed
2149                                  * kprobe_optimizer work handler.
2150                                  */
2151                                 kill_kprobe(p);
2152                         }
2153         }
2154         mutex_unlock(&kprobe_mutex);
2155         return NOTIFY_DONE;
2156 }
2157
2158 static struct notifier_block kprobe_module_nb = {
2159         .notifier_call = kprobes_module_callback,
2160         .priority = 0
2161 };
2162
2163 /* Markers of _kprobe_blacklist section */
2164 extern unsigned long __start_kprobe_blacklist[];
2165 extern unsigned long __stop_kprobe_blacklist[];
2166
2167 static int __init init_kprobes(void)
2168 {
2169         int i, err = 0;
2170
2171         /* FIXME allocate the probe table, currently defined statically */
2172         /* initialize all list heads */
2173         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2174                 INIT_HLIST_HEAD(&kprobe_table[i]);
2175                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
2176                 raw_spin_lock_init(&(kretprobe_table_locks[i].lock));
2177         }
2178
2179         err = populate_kprobe_blacklist(__start_kprobe_blacklist,
2180                                         __stop_kprobe_blacklist);
2181         if (err) {
2182                 pr_err("kprobes: failed to populate blacklist: %d\n", err);
2183                 pr_err("Please take care of using kprobes.\n");
2184         }
2185
2186         if (kretprobe_blacklist_size) {
2187                 /* lookup the function address from its name */
2188                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
2189                         kretprobe_blacklist[i].addr =
2190                                 kprobe_lookup_name(kretprobe_blacklist[i].name, 0);
2191                         if (!kretprobe_blacklist[i].addr)
2192                                 printk("kretprobe: lookup failed: %s\n",
2193                                        kretprobe_blacklist[i].name);
2194                 }
2195         }
2196
2197 #if defined(CONFIG_OPTPROBES)
2198 #if defined(__ARCH_WANT_KPROBES_INSN_SLOT)
2199         /* Init kprobe_optinsn_slots */
2200         kprobe_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
2201 #endif
2202         /* By default, kprobes can be optimized */
2203         kprobes_allow_optimization = true;
2204 #endif
2205
2206         /* By default, kprobes are armed */
2207         kprobes_all_disarmed = false;
2208
2209         err = arch_init_kprobes();
2210         if (!err)
2211                 err = register_die_notifier(&kprobe_exceptions_nb);
2212         if (!err)
2213                 err = register_module_notifier(&kprobe_module_nb);
2214
2215         kprobes_initialized = (err == 0);
2216
2217         if (!err)
2218                 init_test_probes();
2219         return err;
2220 }
2221
2222 #ifdef CONFIG_DEBUG_FS
2223 static void report_probe(struct seq_file *pi, struct kprobe *p,
2224                 const char *sym, int offset, char *modname, struct kprobe *pp)
2225 {
2226         char *kprobe_type;
2227         void *addr = p->addr;
2228
2229         if (p->pre_handler == pre_handler_kretprobe)
2230                 kprobe_type = "r";
2231         else
2232                 kprobe_type = "k";
2233
2234         if (!kallsyms_show_value())
2235                 addr = NULL;
2236
2237         if (sym)
2238                 seq_printf(pi, "%px  %s  %s+0x%x  %s ",
2239                         addr, kprobe_type, sym, offset,
2240                         (modname ? modname : " "));
2241         else    /* try to use %pS */
2242                 seq_printf(pi, "%px  %s  %pS ",
2243                         addr, kprobe_type, p->addr);
2244
2245         if (!pp)
2246                 pp = p;
2247         seq_printf(pi, "%s%s%s%s\n",
2248                 (kprobe_gone(p) ? "[GONE]" : ""),
2249                 ((kprobe_disabled(p) && !kprobe_gone(p)) ?  "[DISABLED]" : ""),
2250                 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2251                 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2252 }
2253
2254 static void *kprobe_seq_start(struct seq_file *f, loff_t *pos)
2255 {
2256         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
2257 }
2258
2259 static void *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
2260 {
2261         (*pos)++;
2262         if (*pos >= KPROBE_TABLE_SIZE)
2263                 return NULL;
2264         return pos;
2265 }
2266
2267 static void kprobe_seq_stop(struct seq_file *f, void *v)
2268 {
2269         /* Nothing to do */
2270 }
2271
2272 static int show_kprobe_addr(struct seq_file *pi, void *v)
2273 {
2274         struct hlist_head *head;
2275         struct kprobe *p, *kp;
2276         const char *sym = NULL;
2277         unsigned int i = *(loff_t *) v;
2278         unsigned long offset = 0;
2279         char *modname, namebuf[KSYM_NAME_LEN];
2280
2281         head = &kprobe_table[i];
2282         preempt_disable();
2283         hlist_for_each_entry_rcu(p, head, hlist) {
2284                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
2285                                         &offset, &modname, namebuf);
2286                 if (kprobe_aggrprobe(p)) {
2287                         list_for_each_entry_rcu(kp, &p->list, list)
2288                                 report_probe(pi, kp, sym, offset, modname, p);
2289                 } else
2290                         report_probe(pi, p, sym, offset, modname, NULL);
2291         }
2292         preempt_enable();
2293         return 0;
2294 }
2295
2296 static const struct seq_operations kprobes_seq_ops = {
2297         .start = kprobe_seq_start,
2298         .next  = kprobe_seq_next,
2299         .stop  = kprobe_seq_stop,
2300         .show  = show_kprobe_addr
2301 };
2302
2303 static int kprobes_open(struct inode *inode, struct file *filp)
2304 {
2305         return seq_open(filp, &kprobes_seq_ops);
2306 }
2307
2308 static const struct file_operations debugfs_kprobes_operations = {
2309         .open           = kprobes_open,
2310         .read           = seq_read,
2311         .llseek         = seq_lseek,
2312         .release        = seq_release,
2313 };
2314
2315 /* kprobes/blacklist -- shows which functions can not be probed */
2316 static void *kprobe_blacklist_seq_start(struct seq_file *m, loff_t *pos)
2317 {
2318         return seq_list_start(&kprobe_blacklist, *pos);
2319 }
2320
2321 static void *kprobe_blacklist_seq_next(struct seq_file *m, void *v, loff_t *pos)
2322 {
2323         return seq_list_next(v, &kprobe_blacklist, pos);
2324 }
2325
2326 static int kprobe_blacklist_seq_show(struct seq_file *m, void *v)
2327 {
2328         struct kprobe_blacklist_entry *ent =
2329                 list_entry(v, struct kprobe_blacklist_entry, list);
2330
2331         /*
2332          * If /proc/kallsyms is not showing kernel address, we won't
2333          * show them here either.
2334          */
2335         if (!kallsyms_show_value())
2336                 seq_printf(m, "0x%px-0x%px\t%ps\n", NULL, NULL,
2337                            (void *)ent->start_addr);
2338         else
2339                 seq_printf(m, "0x%px-0x%px\t%ps\n", (void *)ent->start_addr,
2340                            (void *)ent->end_addr, (void *)ent->start_addr);
2341         return 0;
2342 }
2343
2344 static const struct seq_operations kprobe_blacklist_seq_ops = {
2345         .start = kprobe_blacklist_seq_start,
2346         .next  = kprobe_blacklist_seq_next,
2347         .stop  = kprobe_seq_stop,       /* Reuse void function */
2348         .show  = kprobe_blacklist_seq_show,
2349 };
2350
2351 static int kprobe_blacklist_open(struct inode *inode, struct file *filp)
2352 {
2353         return seq_open(filp, &kprobe_blacklist_seq_ops);
2354 }
2355
2356 static const struct file_operations debugfs_kprobe_blacklist_ops = {
2357         .open           = kprobe_blacklist_open,
2358         .read           = seq_read,
2359         .llseek         = seq_lseek,
2360         .release        = seq_release,
2361 };
2362
2363 static int arm_all_kprobes(void)
2364 {
2365         struct hlist_head *head;
2366         struct kprobe *p;
2367         unsigned int i, total = 0, errors = 0;
2368         int err, ret = 0;
2369
2370         mutex_lock(&kprobe_mutex);
2371
2372         /* If kprobes are armed, just return */
2373         if (!kprobes_all_disarmed)
2374                 goto already_enabled;
2375
2376         /*
2377          * optimize_kprobe() called by arm_kprobe() checks
2378          * kprobes_all_disarmed, so set kprobes_all_disarmed before
2379          * arm_kprobe.
2380          */
2381         kprobes_all_disarmed = false;
2382         /* Arming kprobes doesn't optimize kprobe itself */
2383         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2384                 head = &kprobe_table[i];
2385                 /* Arm all kprobes on a best-effort basis */
2386                 hlist_for_each_entry_rcu(p, head, hlist) {
2387                         if (!kprobe_disabled(p)) {
2388                                 err = arm_kprobe(p);
2389                                 if (err)  {
2390                                         errors++;
2391                                         ret = err;
2392                                 }
2393                                 total++;
2394                         }
2395                 }
2396         }
2397
2398         if (errors)
2399                 pr_warn("Kprobes globally enabled, but failed to arm %d out of %d probes\n",
2400                         errors, total);
2401         else
2402                 pr_info("Kprobes globally enabled\n");
2403
2404 already_enabled:
2405         mutex_unlock(&kprobe_mutex);
2406         return ret;
2407 }
2408
2409 static int disarm_all_kprobes(void)
2410 {
2411         struct hlist_head *head;
2412         struct kprobe *p;
2413         unsigned int i, total = 0, errors = 0;
2414         int err, ret = 0;
2415
2416         mutex_lock(&kprobe_mutex);
2417
2418         /* If kprobes are already disarmed, just return */
2419         if (kprobes_all_disarmed) {
2420                 mutex_unlock(&kprobe_mutex);
2421                 return 0;
2422         }
2423
2424         kprobes_all_disarmed = true;
2425
2426         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2427                 head = &kprobe_table[i];
2428                 /* Disarm all kprobes on a best-effort basis */
2429                 hlist_for_each_entry_rcu(p, head, hlist) {
2430                         if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) {
2431                                 err = disarm_kprobe(p, false);
2432                                 if (err) {
2433                                         errors++;
2434                                         ret = err;
2435                                 }
2436                                 total++;
2437                         }
2438                 }
2439         }
2440
2441         if (errors)
2442                 pr_warn("Kprobes globally disabled, but failed to disarm %d out of %d probes\n",
2443                         errors, total);
2444         else
2445                 pr_info("Kprobes globally disabled\n");
2446
2447         mutex_unlock(&kprobe_mutex);
2448
2449         /* Wait for disarming all kprobes by optimizer */
2450         wait_for_kprobe_optimizer();
2451
2452         return ret;
2453 }
2454
2455 /*
2456  * XXX: The debugfs bool file interface doesn't allow for callbacks
2457  * when the bool state is switched. We can reuse that facility when
2458  * available
2459  */
2460 static ssize_t read_enabled_file_bool(struct file *file,
2461                char __user *user_buf, size_t count, loff_t *ppos)
2462 {
2463         char buf[3];
2464
2465         if (!kprobes_all_disarmed)
2466                 buf[0] = '1';
2467         else
2468                 buf[0] = '0';
2469         buf[1] = '\n';
2470         buf[2] = 0x00;
2471         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
2472 }
2473
2474 static ssize_t write_enabled_file_bool(struct file *file,
2475                const char __user *user_buf, size_t count, loff_t *ppos)
2476 {
2477         char buf[32];
2478         size_t buf_size;
2479         int ret = 0;
2480
2481         buf_size = min(count, (sizeof(buf)-1));
2482         if (copy_from_user(buf, user_buf, buf_size))
2483                 return -EFAULT;
2484
2485         buf[buf_size] = '\0';
2486         switch (buf[0]) {
2487         case 'y':
2488         case 'Y':
2489         case '1':
2490                 ret = arm_all_kprobes();
2491                 break;
2492         case 'n':
2493         case 'N':
2494         case '0':
2495                 ret = disarm_all_kprobes();
2496                 break;
2497         default:
2498                 return -EINVAL;
2499         }
2500
2501         if (ret)
2502                 return ret;
2503
2504         return count;
2505 }
2506
2507 static const struct file_operations fops_kp = {
2508         .read =         read_enabled_file_bool,
2509         .write =        write_enabled_file_bool,
2510         .llseek =       default_llseek,
2511 };
2512
2513 static int __init debugfs_kprobe_init(void)
2514 {
2515         struct dentry *dir, *file;
2516         unsigned int value = 1;
2517
2518         dir = debugfs_create_dir("kprobes", NULL);
2519         if (!dir)
2520                 return -ENOMEM;
2521
2522         file = debugfs_create_file("list", 0400, dir, NULL,
2523                                 &debugfs_kprobes_operations);
2524         if (!file)
2525                 goto error;
2526
2527         file = debugfs_create_file("enabled", 0600, dir,
2528                                         &value, &fops_kp);
2529         if (!file)
2530                 goto error;
2531
2532         file = debugfs_create_file("blacklist", 0400, dir, NULL,
2533                                 &debugfs_kprobe_blacklist_ops);
2534         if (!file)
2535                 goto error;
2536
2537         return 0;
2538
2539 error:
2540         debugfs_remove(dir);
2541         return -ENOMEM;
2542 }
2543
2544 late_initcall(debugfs_kprobe_init);
2545 #endif /* CONFIG_DEBUG_FS */
2546
2547 module_init(init_kprobes);