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/module.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/kdebug.h>
  46
  47 #include <asm-generic/sections.h>
  48 #include <asm/cacheflush.h>
  49 #include <asm/errno.h>
  50 #include <asm/uaccess.h>
  51
  52 #define KPROBE_HASH_BITS 6
  53 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  54
  55
  56 /*
  57  * Some oddball architectures like 64bit powerpc have function descriptors
  58  * so this must be overridable.
  59  */
  60 #ifndef kprobe_lookup_name
  61 #define kprobe_lookup_name(name, addr) \
  62         addr = ((kprobe_opcode_t *)(kallsyms_lookup_name(name)))
  63 #endif
  64
  65 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  66 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  67
  68 /* NOTE: change this value only with kprobe_mutex held */
  69 static bool kprobe_enabled;
  70
  71 DEFINE_MUTEX(kprobe_mutex);             /* Protects kprobe_table */
  72 DEFINE_SPINLOCK(kretprobe_lock);        /* Protects kretprobe_inst_table */
  73 static DEFINE_PER_CPU(struct kprobe *, kprobe_instance) = NULL;
  74
  75 #ifdef __ARCH_WANT_KPROBES_INSN_SLOT
  76 /*
  77  * kprobe->ainsn.insn points to the copy of the instruction to be
  78  * single-stepped. x86_64, POWER4 and above have no-exec support and
  79  * stepping on the instruction on a vmalloced/kmalloced/data page
  80  * is a recipe for disaster
  81  */
  82 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
  83
  84 struct kprobe_insn_page {
  85         struct hlist_node hlist;
  86         kprobe_opcode_t *insns;         /* Page of instruction slots */
  87         char slot_used[INSNS_PER_PAGE];
  88         int nused;
  89         int ngarbage;
  90 };
  91
  92 enum kprobe_slot_state {
  93         SLOT_CLEAN = 0,
  94         SLOT_DIRTY = 1,
  95         SLOT_USED = 2,
  96 };
  97
  98 static struct hlist_head kprobe_insn_pages;
  99 static int kprobe_garbage_slots;
 100 static int collect_garbage_slots(void);
 101
 102 static int __kprobes check_safety(void)
 103 {
 104         int ret = 0;
 105 #if defined(CONFIG_PREEMPT) && defined(CONFIG_PM)
 106         ret = freeze_processes();
 107         if (ret == 0) {
 108                 struct task_struct *p, *q;
 109                 do_each_thread(p, q) {
 110                         if (p != current && p->state == TASK_RUNNING &&
 111                             p->pid != 0) {
 112                                 printk("Check failed: %s is running\n",p->comm);
 113                                 ret = -1;
 114                                 goto loop_end;
 115                         }
 116                 } while_each_thread(p, q);
 117         }
 118 loop_end:
 119         thaw_processes();
 120 #else
 121         synchronize_sched();
 122 #endif
 123         return ret;
 124 }
 125
 126 /**
 127  * get_insn_slot() - Find a slot on an executable page for an instruction.
 128  * We allocate an executable page if there's no room on existing ones.
 129  */
 130 kprobe_opcode_t __kprobes *get_insn_slot(void)
 131 {
 132         struct kprobe_insn_page *kip;
 133         struct hlist_node *pos;
 134
 135  retry:
 136         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 137                 if (kip->nused < INSNS_PER_PAGE) {
 138                         int i;
 139                         for (i = 0; i < INSNS_PER_PAGE; i++) {
 140                                 if (kip->slot_used[i] == SLOT_CLEAN) {
 141                                         kip->slot_used[i] = SLOT_USED;
 142                                         kip->nused++;
 143                                         return kip->insns + (i * MAX_INSN_SIZE);
 144                                 }
 145                         }
 146                         /* Surprise!  No unused slots.  Fix kip->nused. */
 147                         kip->nused = INSNS_PER_PAGE;
 148                 }
 149         }
 150
 151         /* If there are any garbage slots, collect it and try again. */
 152         if (kprobe_garbage_slots && collect_garbage_slots() == 0) {
 153                 goto retry;
 154         }
 155         /* All out of space.  Need to allocate a new page. Use slot 0. */
 156         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
 157         if (!kip)
 158                 return NULL;
 159
 160         /*
 161          * Use module_alloc so this page is within +/- 2GB of where the
 162          * kernel image and loaded module images reside. This is required
 163          * so x86_64 can correctly handle the %rip-relative fixups.
 164          */
 165         kip->insns = module_alloc(PAGE_SIZE);
 166         if (!kip->insns) {
 167                 kfree(kip);
 168                 return NULL;
 169         }
 170         INIT_HLIST_NODE(&kip->hlist);
 171         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
 172         memset(kip->slot_used, SLOT_CLEAN, INSNS_PER_PAGE);
 173         kip->slot_used[0] = SLOT_USED;
 174         kip->nused = 1;
 175         kip->ngarbage = 0;
 176         return kip->insns;
 177 }
 178
 179 /* Return 1 if all garbages are collected, otherwise 0. */
 180 static int __kprobes collect_one_slot(struct kprobe_insn_page *kip, int idx)
 181 {
 182         kip->slot_used[idx] = SLOT_CLEAN;
 183         kip->nused--;
 184         if (kip->nused == 0) {
 185                 /*
 186                  * Page is no longer in use.  Free it unless
 187                  * it's the last one.  We keep the last one
 188                  * so as not to have to set it up again the
 189                  * next time somebody inserts a probe.
 190                  */
 191                 hlist_del(&kip->hlist);
 192                 if (hlist_empty(&kprobe_insn_pages)) {
 193                         INIT_HLIST_NODE(&kip->hlist);
 194                         hlist_add_head(&kip->hlist,
 195                                        &kprobe_insn_pages);
 196                 } else {
 197                         module_free(NULL, kip->insns);
 198                         kfree(kip);
 199                 }
 200                 return 1;
 201         }
 202         return 0;
 203 }
 204
 205 static int __kprobes collect_garbage_slots(void)
 206 {
 207         struct kprobe_insn_page *kip;
 208         struct hlist_node *pos, *next;
 209
 210         /* Ensure no-one is preepmted on the garbages */
 211         if (check_safety() != 0)
 212                 return -EAGAIN;
 213
 214         hlist_for_each_entry_safe(kip, pos, next, &kprobe_insn_pages, hlist) {
 215                 int i;
 216                 if (kip->ngarbage == 0)
 217                         continue;
 218                 kip->ngarbage = 0;      /* we will collect all garbages */
 219                 for (i = 0; i < INSNS_PER_PAGE; i++) {
 220                         if (kip->slot_used[i] == SLOT_DIRTY &&
 221                             collect_one_slot(kip, i))
 222                                 break;
 223                 }
 224         }
 225         kprobe_garbage_slots = 0;
 226         return 0;
 227 }
 228
 229 void __kprobes free_insn_slot(kprobe_opcode_t * slot, int dirty)
 230 {
 231         struct kprobe_insn_page *kip;
 232         struct hlist_node *pos;
 233
 234         hlist_for_each_entry(kip, pos, &kprobe_insn_pages, hlist) {
 235                 if (kip->insns <= slot &&
 236                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 237                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 238                         if (dirty) {
 239                                 kip->slot_used[i] = SLOT_DIRTY;
 240                                 kip->ngarbage++;
 241                         } else {
 242                                 collect_one_slot(kip, i);
 243                         }
 244                         break;
 245                 }
 246         }
 247
 248         if (dirty && ++kprobe_garbage_slots > INSNS_PER_PAGE)
 249                 collect_garbage_slots();
 250 }
 251 #endif
 252
 253 /* We have preemption disabled.. so it is safe to use __ versions */
 254 static inline void set_kprobe_instance(struct kprobe *kp)
 255 {
 256         __get_cpu_var(kprobe_instance) = kp;
 257 }
 258
 259 static inline void reset_kprobe_instance(void)
 260 {
 261         __get_cpu_var(kprobe_instance) = NULL;
 262 }
 263
 264 /*
 265  * This routine is called either:
 266  *      - under the kprobe_mutex - during kprobe_[un]register()
 267  *                              OR
 268  *      - with preemption disabled - from arch/xxx/kernel/kprobes.c
 269  */
 270 struct kprobe __kprobes *get_kprobe(void *addr)
 271 {
 272         struct hlist_head *head;
 273         struct hlist_node *node;
 274         struct kprobe *p;
 275
 276         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 277         hlist_for_each_entry_rcu(p, node, head, hlist) {
 278                 if (p->addr == addr)
 279                         return p;
 280         }
 281         return NULL;
 282 }
 283
 284 /*
 285  * Aggregate handlers for multiple kprobes support - these handlers
 286  * take care of invoking the individual kprobe handlers on p->list
 287  */
 288 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 289 {
 290         struct kprobe *kp;
 291
 292         list_for_each_entry_rcu(kp, &p->list, list) {
 293                 if (kp->pre_handler) {
 294                         set_kprobe_instance(kp);
 295                         if (kp->pre_handler(kp, regs))
 296                                 return 1;
 297                 }
 298                 reset_kprobe_instance();
 299         }
 300         return 0;
 301 }
 302
 303 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 304                                         unsigned long flags)
 305 {
 306         struct kprobe *kp;
 307
 308         list_for_each_entry_rcu(kp, &p->list, list) {
 309                 if (kp->post_handler) {
 310                         set_kprobe_instance(kp);
 311                         kp->post_handler(kp, regs, flags);
 312                         reset_kprobe_instance();
 313                 }
 314         }
 315 }
 316
 317 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 318                                         int trapnr)
 319 {
 320         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 321
 322         /*
 323          * if we faulted "during" the execution of a user specified
 324          * probe handler, invoke just that probe's fault handler
 325          */
 326         if (cur && cur->fault_handler) {
 327                 if (cur->fault_handler(cur, regs, trapnr))
 328                         return 1;
 329         }
 330         return 0;
 331 }
 332
 333 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 334 {
 335         struct kprobe *cur = __get_cpu_var(kprobe_instance);
 336         int ret = 0;
 337
 338         if (cur && cur->break_handler) {
 339                 if (cur->break_handler(cur, regs))
 340                         ret = 1;
 341         }
 342         reset_kprobe_instance();
 343         return ret;
 344 }
 345
 346 /* Walks the list and increments nmissed count for multiprobe case */
 347 void __kprobes kprobes_inc_nmissed_count(struct kprobe *p)
 348 {
 349         struct kprobe *kp;
 350         if (p->pre_handler != aggr_pre_handler) {
 351                 p->nmissed++;
 352         } else {
 353                 list_for_each_entry_rcu(kp, &p->list, list)
 354                         kp->nmissed++;
 355         }
 356         return;
 357 }
 358
 359 /* Called with kretprobe_lock held */
 360 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
 361                                 struct hlist_head *head)
 362 {
 363         /* remove rp inst off the rprobe_inst_table */
 364         hlist_del(&ri->hlist);
 365         if (ri->rp) {
 366                 /* remove rp inst off the used list */
 367                 hlist_del(&ri->uflist);
 368                 /* put rp inst back onto the free list */
 369                 INIT_HLIST_NODE(&ri->uflist);
 370                 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
 371         } else
 372                 /* Unregistering */
 373                 hlist_add_head(&ri->hlist, head);
 374 }
 375
 376 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
 377 {
 378         return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
 379 }
 380
 381 /*
 382  * This function is called from finish_task_switch when task tk becomes dead,
 383  * so that we can recycle any function-return probe instances associated
 384  * with this task. These left over instances represent probed functions
 385  * that have been called but will never return.
 386  */
 387 void __kprobes kprobe_flush_task(struct task_struct *tk)
 388 {
 389         struct kretprobe_instance *ri;
 390         struct hlist_head *head, empty_rp;
 391         struct hlist_node *node, *tmp;
 392         unsigned long flags = 0;
 393
 394         INIT_HLIST_HEAD(&empty_rp);
 395         spin_lock_irqsave(&kretprobe_lock, flags);
 396         head = kretprobe_inst_table_head(tk);
 397         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 398                 if (ri->task == tk)
 399                         recycle_rp_inst(ri, &empty_rp);
 400         }
 401         spin_unlock_irqrestore(&kretprobe_lock, flags);
 402
 403         hlist_for_each_entry_safe(ri, node, tmp, &empty_rp, hlist) {
 404                 hlist_del(&ri->hlist);
 405                 kfree(ri);
 406         }
 407 }
 408
 409 static inline void free_rp_inst(struct kretprobe *rp)
 410 {
 411         struct kretprobe_instance *ri;
 412         struct hlist_node *pos, *next;
 413
 414         hlist_for_each_entry_safe(ri, pos, next, &rp->free_instances, uflist) {
 415                 hlist_del(&ri->uflist);
 416                 kfree(ri);
 417         }
 418 }
 419
 420 /*
 421  * Keep all fields in the kprobe consistent
 422  */
 423 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 424 {
 425         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 426         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 427 }
 428
 429 /*
 430 * Add the new probe to old_p->list. Fail if this is the
 431 * second jprobe at the address - two jprobes can't coexist
 432 */
 433 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
 434 {
 435         if (p->break_handler) {
 436                 if (old_p->break_handler)
 437                         return -EEXIST;
 438                 list_add_tail_rcu(&p->list, &old_p->list);
 439                 old_p->break_handler = aggr_break_handler;
 440         } else
 441                 list_add_rcu(&p->list, &old_p->list);
 442         if (p->post_handler && !old_p->post_handler)
 443                 old_p->post_handler = aggr_post_handler;
 444         return 0;
 445 }
 446
 447 /*
 448  * Fill in the required fields of the "manager kprobe". Replace the
 449  * earlier kprobe in the hlist with the manager kprobe
 450  */
 451 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 452 {
 453         copy_kprobe(p, ap);
 454         flush_insn_slot(ap);
 455         ap->addr = p->addr;
 456         ap->pre_handler = aggr_pre_handler;
 457         ap->fault_handler = aggr_fault_handler;
 458         if (p->post_handler)
 459                 ap->post_handler = aggr_post_handler;
 460         if (p->break_handler)
 461                 ap->break_handler = aggr_break_handler;
 462
 463         INIT_LIST_HEAD(&ap->list);
 464         list_add_rcu(&p->list, &ap->list);
 465
 466         hlist_replace_rcu(&p->hlist, &ap->hlist);
 467 }
 468
 469 /*
 470  * This is the second or subsequent kprobe at the address - handle
 471  * the intricacies
 472  */
 473 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 474                                           struct kprobe *p)
 475 {
 476         int ret = 0;
 477         struct kprobe *ap;
 478
 479         if (old_p->pre_handler == aggr_pre_handler) {
 480                 copy_kprobe(old_p, p);
 481                 ret = add_new_kprobe(old_p, p);
 482         } else {
 483                 ap = kzalloc(sizeof(struct kprobe), GFP_KERNEL);
 484                 if (!ap)
 485                         return -ENOMEM;
 486                 add_aggr_kprobe(ap, old_p);
 487                 copy_kprobe(ap, p);
 488                 ret = add_new_kprobe(ap, p);
 489         }
 490         return ret;
 491 }
 492
 493 static int __kprobes in_kprobes_functions(unsigned long addr)
 494 {
 495         if (addr >= (unsigned long)__kprobes_text_start &&
 496             addr < (unsigned long)__kprobes_text_end)
 497                 return -EINVAL;
 498         return 0;
 499 }
 500
 501 static int __kprobes __register_kprobe(struct kprobe *p,
 502         unsigned long called_from)
 503 {
 504         int ret = 0;
 505         struct kprobe *old_p;
 506         struct module *probed_mod;
 507
 508         /*
 509          * If we have a symbol_name argument look it up,
 510          * and add it to the address.  That way the addr
 511          * field can either be global or relative to a symbol.
 512          */
 513         if (p->symbol_name) {
 514                 if (p->addr)
 515                         return -EINVAL;
 516                 kprobe_lookup_name(p->symbol_name, p->addr);
 517         }
 518
 519         if (!p->addr)
 520                 return -EINVAL;
 521         p->addr = (kprobe_opcode_t *)(((char *)p->addr)+ p->offset);
 522
 523         if (!kernel_text_address((unsigned long) p->addr) ||
 524             in_kprobes_functions((unsigned long) p->addr))
 525                 return -EINVAL;
 526
 527         p->mod_refcounted = 0;
 528
 529         /*
 530          * Check if are we probing a module.
 531          */
 532         probed_mod = module_text_address((unsigned long) p->addr);
 533         if (probed_mod) {
 534                 struct module *calling_mod = module_text_address(called_from);
 535                 /*
 536                  * We must allow modules to probe themself and in this case
 537                  * avoid incrementing the module refcount, so as to allow
 538                  * unloading of self probing modules.
 539                  */
 540                 if (calling_mod && calling_mod != probed_mod) {
 541                         if (unlikely(!try_module_get(probed_mod)))
 542                                 return -EINVAL;
 543                         p->mod_refcounted = 1;
 544                 } else
 545                         probed_mod = NULL;
 546         }
 547
 548         p->nmissed = 0;
 549         mutex_lock(&kprobe_mutex);
 550         old_p = get_kprobe(p->addr);
 551         if (old_p) {
 552                 ret = register_aggr_kprobe(old_p, p);
 553                 goto out;
 554         }
 555
 556         ret = arch_prepare_kprobe(p);
 557         if (ret)
 558                 goto out;
 559
 560         INIT_HLIST_NODE(&p->hlist);
 561         hlist_add_head_rcu(&p->hlist,
 562                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 563
 564         if (kprobe_enabled)
 565                 arch_arm_kprobe(p);
 566
 567 out:
 568         mutex_unlock(&kprobe_mutex);
 569
 570         if (ret && probed_mod)
 571                 module_put(probed_mod);
 572         return ret;
 573 }
 574
 575 int __kprobes register_kprobe(struct kprobe *p)
 576 {
 577         return __register_kprobe(p, (unsigned long)__builtin_return_address(0));
 578 }
 579
 580 void __kprobes unregister_kprobe(struct kprobe *p)
 581 {
 582         struct module *mod;
 583         struct kprobe *old_p, *list_p;
 584         int cleanup_p;
 585
 586         mutex_lock(&kprobe_mutex);
 587         old_p = get_kprobe(p->addr);
 588         if (unlikely(!old_p)) {
 589                 mutex_unlock(&kprobe_mutex);
 590                 return;
 591         }
 592         if (p != old_p) {
 593                 list_for_each_entry_rcu(list_p, &old_p->list, list)
 594                         if (list_p == p)
 595                         /* kprobe p is a valid probe */
 596                                 goto valid_p;
 597                 mutex_unlock(&kprobe_mutex);
 598                 return;
 599         }
 600 valid_p:
 601         if (old_p == p ||
 602             (old_p->pre_handler == aggr_pre_handler &&
 603              p->list.next == &old_p->list && p->list.prev == &old_p->list)) {
 604                 /*
 605                  * Only probe on the hash list. Disarm only if kprobes are
 606                  * enabled - otherwise, the breakpoint would already have
 607                  * been removed. We save on flushing icache.
 608                  */
 609                 if (kprobe_enabled)
 610                         arch_disarm_kprobe(p);
 611                 hlist_del_rcu(&old_p->hlist);
 612                 cleanup_p = 1;
 613         } else {
 614                 list_del_rcu(&p->list);
 615                 cleanup_p = 0;
 616         }
 617
 618         mutex_unlock(&kprobe_mutex);
 619
 620         synchronize_sched();
 621         if (p->mod_refcounted) {
 622                 mod = module_text_address((unsigned long)p->addr);
 623                 if (mod)
 624                         module_put(mod);
 625         }
 626
 627         if (cleanup_p) {
 628                 if (p != old_p) {
 629                         list_del_rcu(&p->list);
 630                         kfree(old_p);
 631                 }
 632                 arch_remove_kprobe(p);
 633         } else {
 634                 mutex_lock(&kprobe_mutex);
 635                 if (p->break_handler)
 636                         old_p->break_handler = NULL;
 637                 if (p->post_handler){
 638                         list_for_each_entry_rcu(list_p, &old_p->list, list){
 639                                 if (list_p->post_handler){
 640                                         cleanup_p = 2;
 641                                         break;
 642                                 }
 643                         }
 644                         if (cleanup_p == 0)
 645                                 old_p->post_handler = NULL;
 646                 }
 647                 mutex_unlock(&kprobe_mutex);
 648         }
 649 }
 650
 651 static struct notifier_block kprobe_exceptions_nb = {
 652         .notifier_call = kprobe_exceptions_notify,
 653         .priority = 0x7fffffff /* we need to be notified first */
 654 };
 655
 656 unsigned long __weak arch_deref_entry_point(void *entry)
 657 {
 658         return (unsigned long)entry;
 659 }
 660
 661 int __kprobes register_jprobe(struct jprobe *jp)
 662 {
 663         unsigned long addr = arch_deref_entry_point(jp->entry);
 664
 665         if (!kernel_text_address(addr))
 666                 return -EINVAL;
 667
 668         /* Todo: Verify probepoint is a function entry point */
 669         jp->kp.pre_handler = setjmp_pre_handler;
 670         jp->kp.break_handler = longjmp_break_handler;
 671
 672         return __register_kprobe(&jp->kp,
 673                 (unsigned long)__builtin_return_address(0));
 674 }
 675
 676 void __kprobes unregister_jprobe(struct jprobe *jp)
 677 {
 678         unregister_kprobe(&jp->kp);
 679 }
 680
 681 #ifdef ARCH_SUPPORTS_KRETPROBES
 682
 683 /*
 684  * This kprobe pre_handler is registered with every kretprobe. When probe
 685  * hits it will set up the return probe.
 686  */
 687 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 688                                            struct pt_regs *regs)
 689 {
 690         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 691         unsigned long flags = 0;
 692
 693         /*TODO: consider to only swap the RA after the last pre_handler fired */
 694         spin_lock_irqsave(&kretprobe_lock, flags);
 695         if (!hlist_empty(&rp->free_instances)) {
 696                 struct kretprobe_instance *ri;
 697
 698                 ri = hlist_entry(rp->free_instances.first,
 699                                  struct kretprobe_instance, uflist);
 700                 ri->rp = rp;
 701                 ri->task = current;
 702                 arch_prepare_kretprobe(ri, regs);
 703
 704                 /* XXX(hch): why is there no hlist_move_head? */
 705                 hlist_del(&ri->uflist);
 706                 hlist_add_head(&ri->uflist, &ri->rp->used_instances);
 707                 hlist_add_head(&ri->hlist, kretprobe_inst_table_head(ri->task));
 708         } else
 709                 rp->nmissed++;
 710         spin_unlock_irqrestore(&kretprobe_lock, flags);
 711         return 0;
 712 }
 713
 714 int __kprobes register_kretprobe(struct kretprobe *rp)
 715 {
 716         int ret = 0;
 717         struct kretprobe_instance *inst;
 718         int i;
 719         void *addr = rp->kp.addr;
 720
 721         if (kretprobe_blacklist_size) {
 722                 if (addr == NULL)
 723                         kprobe_lookup_name(rp->kp.symbol_name, addr);
 724                 addr += rp->kp.offset;
 725
 726                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
 727                         if (kretprobe_blacklist[i].addr == addr)
 728                                 return -EINVAL;
 729                 }
 730         }
 731
 732         rp->kp.pre_handler = pre_handler_kretprobe;
 733         rp->kp.post_handler = NULL;
 734         rp->kp.fault_handler = NULL;
 735         rp->kp.break_handler = NULL;
 736
 737         /* Pre-allocate memory for max kretprobe instances */
 738         if (rp->maxactive <= 0) {
 739 #ifdef CONFIG_PREEMPT
 740                 rp->maxactive = max(10, 2 * NR_CPUS);
 741 #else
 742                 rp->maxactive = NR_CPUS;
 743 #endif
 744         }
 745         INIT_HLIST_HEAD(&rp->used_instances);
 746         INIT_HLIST_HEAD(&rp->free_instances);
 747         for (i = 0; i < rp->maxactive; i++) {
 748                 inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL);
 749                 if (inst == NULL) {
 750                         free_rp_inst(rp);
 751                         return -ENOMEM;
 752                 }
 753                 INIT_HLIST_NODE(&inst->uflist);
 754                 hlist_add_head(&inst->uflist, &rp->free_instances);
 755         }
 756
 757         rp->nmissed = 0;
 758         /* Establish function entry probe point */
 759         if ((ret = __register_kprobe(&rp->kp,
 760                 (unsigned long)__builtin_return_address(0))) != 0)
 761                 free_rp_inst(rp);
 762         return ret;
 763 }
 764
 765 #else /* ARCH_SUPPORTS_KRETPROBES */
 766
 767 int __kprobes register_kretprobe(struct kretprobe *rp)
 768 {
 769         return -ENOSYS;
 770 }
 771
 772 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 773                                            struct pt_regs *regs)
 774 {
 775         return 0;
 776 }
 777
 778 #endif /* ARCH_SUPPORTS_KRETPROBES */
 779
 780 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 781 {
 782         unsigned long flags;
 783         struct kretprobe_instance *ri;
 784         struct hlist_node *pos, *next;
 785
 786         unregister_kprobe(&rp->kp);
 787
 788         /* No race here */
 789         spin_lock_irqsave(&kretprobe_lock, flags);
 790         hlist_for_each_entry_safe(ri, pos, next, &rp->used_instances, uflist) {
 791                 ri->rp = NULL;
 792                 hlist_del(&ri->uflist);
 793         }
 794         spin_unlock_irqrestore(&kretprobe_lock, flags);
 795         free_rp_inst(rp);
 796 }
 797
 798 static int __init init_kprobes(void)
 799 {
 800         int i, err = 0;
 801
 802         /* FIXME allocate the probe table, currently defined statically */
 803         /* initialize all list heads */
 804         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 805                 INIT_HLIST_HEAD(&kprobe_table[i]);
 806                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
 807         }
 808
 809         if (kretprobe_blacklist_size) {
 810                 /* lookup the function address from its name */
 811                 for (i = 0; kretprobe_blacklist[i].name != NULL; i++) {
 812                         kprobe_lookup_name(kretprobe_blacklist[i].name,
 813                                            kretprobe_blacklist[i].addr);
 814                         if (!kretprobe_blacklist[i].addr)
 815                                 printk("kretprobe: lookup failed: %s\n",
 816                                        kretprobe_blacklist[i].name);
 817                 }
 818         }
 819
 820         /* By default, kprobes are enabled */
 821         kprobe_enabled = true;
 822
 823         err = arch_init_kprobes();
 824         if (!err)
 825                 err = register_die_notifier(&kprobe_exceptions_nb);
 826
 827         return err;
 828 }
 829
 830 #ifdef CONFIG_DEBUG_FS
 831 static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
 832                 const char *sym, int offset,char *modname)
 833 {
 834         char *kprobe_type;
 835
 836         if (p->pre_handler == pre_handler_kretprobe)
 837                 kprobe_type = "r";
 838         else if (p->pre_handler == setjmp_pre_handler)
 839                 kprobe_type = "j";
 840         else
 841                 kprobe_type = "k";
 842         if (sym)
 843                 seq_printf(pi, "%p  %s  %s+0x%x  %s\n", p->addr, kprobe_type,
 844                         sym, offset, (modname ? modname : " "));
 845         else
 846                 seq_printf(pi, "%p  %s  %p\n", p->addr, kprobe_type, p->addr);
 847 }
 848
 849 static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
 850 {
 851         return (*pos < KPROBE_TABLE_SIZE) ? pos : NULL;
 852 }
 853
 854 static void __kprobes *kprobe_seq_next(struct seq_file *f, void *v, loff_t *pos)
 855 {
 856         (*pos)++;
 857         if (*pos >= KPROBE_TABLE_SIZE)
 858                 return NULL;
 859         return pos;
 860 }
 861
 862 static void __kprobes kprobe_seq_stop(struct seq_file *f, void *v)
 863 {
 864         /* Nothing to do */
 865 }
 866
 867 static int __kprobes show_kprobe_addr(struct seq_file *pi, void *v)
 868 {
 869         struct hlist_head *head;
 870         struct hlist_node *node;
 871         struct kprobe *p, *kp;
 872         const char *sym = NULL;
 873         unsigned int i = *(loff_t *) v;
 874         unsigned long offset = 0;
 875         char *modname, namebuf[128];
 876
 877         head = &kprobe_table[i];
 878         preempt_disable();
 879         hlist_for_each_entry_rcu(p, node, head, hlist) {
 880                 sym = kallsyms_lookup((unsigned long)p->addr, NULL,
 881                                         &offset, &modname, namebuf);
 882                 if (p->pre_handler == aggr_pre_handler) {
 883                         list_for_each_entry_rcu(kp, &p->list, list)
 884                                 report_probe(pi, kp, sym, offset, modname);
 885                 } else
 886                         report_probe(pi, p, sym, offset, modname);
 887         }
 888         preempt_enable();
 889         return 0;
 890 }
 891
 892 static struct seq_operations kprobes_seq_ops = {
 893         .start = kprobe_seq_start,
 894         .next  = kprobe_seq_next,
 895         .stop  = kprobe_seq_stop,
 896         .show  = show_kprobe_addr
 897 };
 898
 899 static int __kprobes kprobes_open(struct inode *inode, struct file *filp)
 900 {
 901         return seq_open(filp, &kprobes_seq_ops);
 902 }
 903
 904 static struct file_operations debugfs_kprobes_operations = {
 905         .open           = kprobes_open,
 906         .read           = seq_read,
 907         .llseek         = seq_lseek,
 908         .release        = seq_release,
 909 };
 910
 911 static void __kprobes enable_all_kprobes(void)
 912 {
 913         struct hlist_head *head;
 914         struct hlist_node *node;
 915         struct kprobe *p;
 916         unsigned int i;
 917
 918         mutex_lock(&kprobe_mutex);
 919
 920         /* If kprobes are already enabled, just return */
 921         if (kprobe_enabled)
 922                 goto already_enabled;
 923
 924         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 925                 head = &kprobe_table[i];
 926                 hlist_for_each_entry_rcu(p, node, head, hlist)
 927                         arch_arm_kprobe(p);
 928         }
 929
 930         kprobe_enabled = true;
 931         printk(KERN_INFO "Kprobes globally enabled\n");
 932
 933 already_enabled:
 934         mutex_unlock(&kprobe_mutex);
 935         return;
 936 }
 937
 938 static void __kprobes disable_all_kprobes(void)
 939 {
 940         struct hlist_head *head;
 941         struct hlist_node *node;
 942         struct kprobe *p;
 943         unsigned int i;
 944
 945         mutex_lock(&kprobe_mutex);
 946
 947         /* If kprobes are already disabled, just return */
 948         if (!kprobe_enabled)
 949                 goto already_disabled;
 950
 951         kprobe_enabled = false;
 952         printk(KERN_INFO "Kprobes globally disabled\n");
 953         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 954                 head = &kprobe_table[i];
 955                 hlist_for_each_entry_rcu(p, node, head, hlist) {
 956                         if (!arch_trampoline_kprobe(p))
 957                                 arch_disarm_kprobe(p);
 958                 }
 959         }
 960
 961         mutex_unlock(&kprobe_mutex);
 962         /* Allow all currently running kprobes to complete */
 963         synchronize_sched();
 964         return;
 965
 966 already_disabled:
 967         mutex_unlock(&kprobe_mutex);
 968         return;
 969 }
 970
 971 /*
 972  * XXX: The debugfs bool file interface doesn't allow for callbacks
 973  * when the bool state is switched. We can reuse that facility when
 974  * available
 975  */
 976 static ssize_t read_enabled_file_bool(struct file *file,
 977                char __user *user_buf, size_t count, loff_t *ppos)
 978 {
 979         char buf[3];
 980
 981         if (kprobe_enabled)
 982                 buf[0] = '1';
 983         else
 984                 buf[0] = '0';
 985         buf[1] = '\n';
 986         buf[2] = 0x00;
 987         return simple_read_from_buffer(user_buf, count, ppos, buf, 2);
 988 }
 989
 990 static ssize_t write_enabled_file_bool(struct file *file,
 991                const char __user *user_buf, size_t count, loff_t *ppos)
 992 {
 993         char buf[32];
 994         int buf_size;
 995
 996         buf_size = min(count, (sizeof(buf)-1));
 997         if (copy_from_user(buf, user_buf, buf_size))
 998                 return -EFAULT;
 999
1000         switch (buf[0]) {
1001         case 'y':
1002         case 'Y':
1003         case '1':
1004                 enable_all_kprobes();
1005                 break;
1006         case 'n':
1007         case 'N':
1008         case '0':
1009                 disable_all_kprobes();
1010                 break;
1011         }
1012
1013         return count;
1014 }
1015
1016 static struct file_operations fops_kp = {
1017         .read =         read_enabled_file_bool,
1018         .write =        write_enabled_file_bool,
1019 };
1020
1021 static int __kprobes debugfs_kprobe_init(void)
1022 {
1023         struct dentry *dir, *file;
1024         unsigned int value = 1;
1025
1026         dir = debugfs_create_dir("kprobes", NULL);
1027         if (!dir)
1028                 return -ENOMEM;
1029
1030         file = debugfs_create_file("list", 0444, dir, NULL,
1031                                 &debugfs_kprobes_operations);
1032         if (!file) {
1033                 debugfs_remove(dir);
1034                 return -ENOMEM;
1035         }
1036
1037         file = debugfs_create_file("enabled", 0600, dir,
1038                                         &value, &fops_kp);
1039         if (!file) {
1040                 debugfs_remove(dir);
1041                 return -ENOMEM;
1042         }
1043
1044         return 0;
1045 }
1046
1047 late_initcall(debugfs_kprobe_init);
1048 #endif /* CONFIG_DEBUG_FS */
1049
1050 module_init(init_kprobes);
1051
1052 EXPORT_SYMBOL_GPL(register_kprobe);
1053 EXPORT_SYMBOL_GPL(unregister_kprobe);
1054 EXPORT_SYMBOL_GPL(register_jprobe);
1055 EXPORT_SYMBOL_GPL(unregister_jprobe);
1056 #ifdef CONFIG_KPROBES
1057 EXPORT_SYMBOL_GPL(jprobe_return);
1058 #endif
1059
1060 #ifdef CONFIG_KPROBES
1061 EXPORT_SYMBOL_GPL(register_kretprobe);
1062 EXPORT_SYMBOL_GPL(unregister_kretprobe);
1063 #endif