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/spinlock.h>
  36 #include <linux/hash.h>
  37 #include <linux/init.h>
  38 #include <linux/module.h>
  39 #include <linux/moduleloader.h>
  40 #include <asm-generic/sections.h>
  41 #include <asm/cacheflush.h>
  42 #include <asm/errno.h>
  43 #include <asm/kdebug.h>
  44
  45 #define KPROBE_HASH_BITS 6
  46 #define KPROBE_TABLE_SIZE (1 << KPROBE_HASH_BITS)
  47
  48 static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];
  49 static struct hlist_head kretprobe_inst_table[KPROBE_TABLE_SIZE];
  50
  51 unsigned int kprobe_cpu = NR_CPUS;
  52 static DEFINE_SPINLOCK(kprobe_lock);
  53 static struct kprobe *curr_kprobe;
  54
  55 /*
  56  * kprobe->ainsn.insn points to the copy of the instruction to be
  57  * single-stepped. x86_64, POWER4 and above have no-exec support and
  58  * stepping on the instruction on a vmalloced/kmalloced/data page
  59  * is a recipe for disaster
  60  */
  61 #define INSNS_PER_PAGE  (PAGE_SIZE/(MAX_INSN_SIZE * sizeof(kprobe_opcode_t)))
  62
  63 struct kprobe_insn_page {
  64         struct hlist_node hlist;
  65         kprobe_opcode_t *insns;         /* Page of instruction slots */
  66         char slot_used[INSNS_PER_PAGE];
  67         int nused;
  68 };
  69
  70 static struct hlist_head kprobe_insn_pages;
  71
  72 /**
  73  * get_insn_slot() - Find a slot on an executable page for an instruction.
  74  * We allocate an executable page if there's no room on existing ones.
  75  */
  76 kprobe_opcode_t __kprobes *get_insn_slot(void)
  77 {
  78         struct kprobe_insn_page *kip;
  79         struct hlist_node *pos;
  80
  81         hlist_for_each(pos, &kprobe_insn_pages) {
  82                 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
  83                 if (kip->nused < INSNS_PER_PAGE) {
  84                         int i;
  85                         for (i = 0; i < INSNS_PER_PAGE; i++) {
  86                                 if (!kip->slot_used[i]) {
  87                                         kip->slot_used[i] = 1;
  88                                         kip->nused++;
  89                                         return kip->insns + (i * MAX_INSN_SIZE);
  90                                 }
  91                         }
  92                         /* Surprise!  No unused slots.  Fix kip->nused. */
  93                         kip->nused = INSNS_PER_PAGE;
  94                 }
  95         }
  96
  97         /* All out of space.  Need to allocate a new page. Use slot 0.*/
  98         kip = kmalloc(sizeof(struct kprobe_insn_page), GFP_KERNEL);
  99         if (!kip) {
 100                 return NULL;
 101         }
 102
 103         /*
 104          * Use module_alloc so this page is within +/- 2GB of where the
 105          * kernel image and loaded module images reside. This is required
 106          * so x86_64 can correctly handle the %rip-relative fixups.
 107          */
 108         kip->insns = module_alloc(PAGE_SIZE);
 109         if (!kip->insns) {
 110                 kfree(kip);
 111                 return NULL;
 112         }
 113         INIT_HLIST_NODE(&kip->hlist);
 114         hlist_add_head(&kip->hlist, &kprobe_insn_pages);
 115         memset(kip->slot_used, 0, INSNS_PER_PAGE);
 116         kip->slot_used[0] = 1;
 117         kip->nused = 1;
 118         return kip->insns;
 119 }
 120
 121 void __kprobes free_insn_slot(kprobe_opcode_t *slot)
 122 {
 123         struct kprobe_insn_page *kip;
 124         struct hlist_node *pos;
 125
 126         hlist_for_each(pos, &kprobe_insn_pages) {
 127                 kip = hlist_entry(pos, struct kprobe_insn_page, hlist);
 128                 if (kip->insns <= slot &&
 129                     slot < kip->insns + (INSNS_PER_PAGE * MAX_INSN_SIZE)) {
 130                         int i = (slot - kip->insns) / MAX_INSN_SIZE;
 131                         kip->slot_used[i] = 0;
 132                         kip->nused--;
 133                         if (kip->nused == 0) {
 134                                 /*
 135                                  * Page is no longer in use.  Free it unless
 136                                  * it's the last one.  We keep the last one
 137                                  * so as not to have to set it up again the
 138                                  * next time somebody inserts a probe.
 139                                  */
 140                                 hlist_del(&kip->hlist);
 141                                 if (hlist_empty(&kprobe_insn_pages)) {
 142                                         INIT_HLIST_NODE(&kip->hlist);
 143                                         hlist_add_head(&kip->hlist,
 144                                                 &kprobe_insn_pages);
 145                                 } else {
 146                                         module_free(NULL, kip->insns);
 147                                         kfree(kip);
 148                                 }
 149                         }
 150                         return;
 151                 }
 152         }
 153 }
 154
 155 /* Locks kprobe: irqs must be disabled */
 156 void __kprobes lock_kprobes(void)
 157 {
 158         unsigned long flags = 0;
 159
 160         /* Avoiding local interrupts to happen right after we take the kprobe_lock
 161          * and before we get a chance to update kprobe_cpu, this to prevent
 162          * deadlock when we have a kprobe on ISR routine and a kprobe on task
 163          * routine
 164          */
 165         local_irq_save(flags);
 166
 167         spin_lock(&kprobe_lock);
 168         kprobe_cpu = smp_processor_id();
 169
 170         local_irq_restore(flags);
 171 }
 172
 173 void __kprobes unlock_kprobes(void)
 174 {
 175         unsigned long flags = 0;
 176
 177         /* Avoiding local interrupts to happen right after we update
 178          * kprobe_cpu and before we get a a chance to release kprobe_lock,
 179          * this to prevent deadlock when we have a kprobe on ISR routine and
 180          * a kprobe on task routine
 181          */
 182         local_irq_save(flags);
 183
 184         kprobe_cpu = NR_CPUS;
 185         spin_unlock(&kprobe_lock);
 186
 187         local_irq_restore(flags);
 188 }
 189
 190 /* You have to be holding the kprobe_lock */
 191 struct kprobe __kprobes *get_kprobe(void *addr)
 192 {
 193         struct hlist_head *head;
 194         struct hlist_node *node;
 195
 196         head = &kprobe_table[hash_ptr(addr, KPROBE_HASH_BITS)];
 197         hlist_for_each(node, head) {
 198                 struct kprobe *p = hlist_entry(node, struct kprobe, hlist);
 199                 if (p->addr == addr)
 200                         return p;
 201         }
 202         return NULL;
 203 }
 204
 205 /*
 206  * Aggregate handlers for multiple kprobes support - these handlers
 207  * take care of invoking the individual kprobe handlers on p->list
 208  */
 209 static int __kprobes aggr_pre_handler(struct kprobe *p, struct pt_regs *regs)
 210 {
 211         struct kprobe *kp;
 212
 213         list_for_each_entry(kp, &p->list, list) {
 214                 if (kp->pre_handler) {
 215                         curr_kprobe = kp;
 216                         if (kp->pre_handler(kp, regs))
 217                                 return 1;
 218                 }
 219                 curr_kprobe = NULL;
 220         }
 221         return 0;
 222 }
 223
 224 static void __kprobes aggr_post_handler(struct kprobe *p, struct pt_regs *regs,
 225                                         unsigned long flags)
 226 {
 227         struct kprobe *kp;
 228
 229         list_for_each_entry(kp, &p->list, list) {
 230                 if (kp->post_handler) {
 231                         curr_kprobe = kp;
 232                         kp->post_handler(kp, regs, flags);
 233                         curr_kprobe = NULL;
 234                 }
 235         }
 236         return;
 237 }
 238
 239 static int __kprobes aggr_fault_handler(struct kprobe *p, struct pt_regs *regs,
 240                                         int trapnr)
 241 {
 242         /*
 243          * if we faulted "during" the execution of a user specified
 244          * probe handler, invoke just that probe's fault handler
 245          */
 246         if (curr_kprobe && curr_kprobe->fault_handler) {
 247                 if (curr_kprobe->fault_handler(curr_kprobe, regs, trapnr))
 248                         return 1;
 249         }
 250         return 0;
 251 }
 252
 253 static int __kprobes aggr_break_handler(struct kprobe *p, struct pt_regs *regs)
 254 {
 255         struct kprobe *kp = curr_kprobe;
 256         if (curr_kprobe && kp->break_handler) {
 257                 if (kp->break_handler(kp, regs)) {
 258                         curr_kprobe = NULL;
 259                         return 1;
 260                 }
 261         }
 262         curr_kprobe = NULL;
 263         return 0;
 264 }
 265
 266 struct kretprobe_instance __kprobes *get_free_rp_inst(struct kretprobe *rp)
 267 {
 268         struct hlist_node *node;
 269         struct kretprobe_instance *ri;
 270         hlist_for_each_entry(ri, node, &rp->free_instances, uflist)
 271                 return ri;
 272         return NULL;
 273 }
 274
 275 static struct kretprobe_instance __kprobes *get_used_rp_inst(struct kretprobe
 276                                                               *rp)
 277 {
 278         struct hlist_node *node;
 279         struct kretprobe_instance *ri;
 280         hlist_for_each_entry(ri, node, &rp->used_instances, uflist)
 281                 return ri;
 282         return NULL;
 283 }
 284
 285 void __kprobes add_rp_inst(struct kretprobe_instance *ri)
 286 {
 287         /*
 288          * Remove rp inst off the free list -
 289          * Add it back when probed function returns
 290          */
 291         hlist_del(&ri->uflist);
 292
 293         /* Add rp inst onto table */
 294         INIT_HLIST_NODE(&ri->hlist);
 295         hlist_add_head(&ri->hlist,
 296                         &kretprobe_inst_table[hash_ptr(ri->task, KPROBE_HASH_BITS)]);
 297
 298         /* Also add this rp inst to the used list. */
 299         INIT_HLIST_NODE(&ri->uflist);
 300         hlist_add_head(&ri->uflist, &ri->rp->used_instances);
 301 }
 302
 303 void __kprobes recycle_rp_inst(struct kretprobe_instance *ri)
 304 {
 305         /* remove rp inst off the rprobe_inst_table */
 306         hlist_del(&ri->hlist);
 307         if (ri->rp) {
 308                 /* remove rp inst off the used list */
 309                 hlist_del(&ri->uflist);
 310                 /* put rp inst back onto the free list */
 311                 INIT_HLIST_NODE(&ri->uflist);
 312                 hlist_add_head(&ri->uflist, &ri->rp->free_instances);
 313         } else
 314                 /* Unregistering */
 315                 kfree(ri);
 316 }
 317
 318 struct hlist_head __kprobes *kretprobe_inst_table_head(struct task_struct *tsk)
 319 {
 320         return &kretprobe_inst_table[hash_ptr(tsk, KPROBE_HASH_BITS)];
 321 }
 322
 323 /*
 324  * This function is called from exit_thread or flush_thread when task tk's
 325  * stack is being recycled so that we can recycle any function-return probe
 326  * instances associated with this task. These left over instances represent
 327  * probed functions that have been called but will never return.
 328  */
 329 void __kprobes kprobe_flush_task(struct task_struct *tk)
 330 {
 331         struct kretprobe_instance *ri;
 332         struct hlist_head *head;
 333         struct hlist_node *node, *tmp;
 334         unsigned long flags = 0;
 335
 336         spin_lock_irqsave(&kprobe_lock, flags);
 337         head = kretprobe_inst_table_head(current);
 338         hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
 339                 if (ri->task == tk)
 340                         recycle_rp_inst(ri);
 341         }
 342         spin_unlock_irqrestore(&kprobe_lock, flags);
 343 }
 344
 345 /*
 346  * This kprobe pre_handler is registered with every kretprobe. When probe
 347  * hits it will set up the return probe.
 348  */
 349 static int __kprobes pre_handler_kretprobe(struct kprobe *p,
 350                                            struct pt_regs *regs)
 351 {
 352         struct kretprobe *rp = container_of(p, struct kretprobe, kp);
 353
 354         /*TODO: consider to only swap the RA after the last pre_handler fired */
 355         arch_prepare_kretprobe(rp, regs);
 356         return 0;
 357 }
 358
 359 static inline void free_rp_inst(struct kretprobe *rp)
 360 {
 361         struct kretprobe_instance *ri;
 362         while ((ri = get_free_rp_inst(rp)) != NULL) {
 363                 hlist_del(&ri->uflist);
 364                 kfree(ri);
 365         }
 366 }
 367
 368 /*
 369  * Keep all fields in the kprobe consistent
 370  */
 371 static inline void copy_kprobe(struct kprobe *old_p, struct kprobe *p)
 372 {
 373         memcpy(&p->opcode, &old_p->opcode, sizeof(kprobe_opcode_t));
 374         memcpy(&p->ainsn, &old_p->ainsn, sizeof(struct arch_specific_insn));
 375 }
 376
 377 /*
 378 * Add the new probe to old_p->list. Fail if this is the
 379 * second jprobe at the address - two jprobes can't coexist
 380 */
 381 static int __kprobes add_new_kprobe(struct kprobe *old_p, struct kprobe *p)
 382 {
 383         struct kprobe *kp;
 384
 385         if (p->break_handler) {
 386                 list_for_each_entry(kp, &old_p->list, list) {
 387                         if (kp->break_handler)
 388                                 return -EEXIST;
 389                 }
 390                 list_add_tail(&p->list, &old_p->list);
 391         } else
 392                 list_add(&p->list, &old_p->list);
 393         return 0;
 394 }
 395
 396 /*
 397  * Fill in the required fields of the "manager kprobe". Replace the
 398  * earlier kprobe in the hlist with the manager kprobe
 399  */
 400 static inline void add_aggr_kprobe(struct kprobe *ap, struct kprobe *p)
 401 {
 402         copy_kprobe(p, ap);
 403         ap->addr = p->addr;
 404         ap->pre_handler = aggr_pre_handler;
 405         ap->post_handler = aggr_post_handler;
 406         ap->fault_handler = aggr_fault_handler;
 407         ap->break_handler = aggr_break_handler;
 408
 409         INIT_LIST_HEAD(&ap->list);
 410         list_add(&p->list, &ap->list);
 411
 412         INIT_HLIST_NODE(&ap->hlist);
 413         hlist_del(&p->hlist);
 414         hlist_add_head(&ap->hlist,
 415                 &kprobe_table[hash_ptr(ap->addr, KPROBE_HASH_BITS)]);
 416 }
 417
 418 /*
 419  * This is the second or subsequent kprobe at the address - handle
 420  * the intricacies
 421  * TODO: Move kcalloc outside the spinlock
 422  */
 423 static int __kprobes register_aggr_kprobe(struct kprobe *old_p,
 424                                           struct kprobe *p)
 425 {
 426         int ret = 0;
 427         struct kprobe *ap;
 428
 429         if (old_p->pre_handler == aggr_pre_handler) {
 430                 copy_kprobe(old_p, p);
 431                 ret = add_new_kprobe(old_p, p);
 432         } else {
 433                 ap = kcalloc(1, sizeof(struct kprobe), GFP_ATOMIC);
 434                 if (!ap)
 435                         return -ENOMEM;
 436                 add_aggr_kprobe(ap, old_p);
 437                 copy_kprobe(ap, p);
 438                 ret = add_new_kprobe(ap, p);
 439         }
 440         return ret;
 441 }
 442
 443 /* kprobe removal house-keeping routines */
 444 static inline void cleanup_kprobe(struct kprobe *p, unsigned long flags)
 445 {
 446         arch_disarm_kprobe(p);
 447         hlist_del(&p->hlist);
 448         spin_unlock_irqrestore(&kprobe_lock, flags);
 449         arch_remove_kprobe(p);
 450 }
 451
 452 static inline void cleanup_aggr_kprobe(struct kprobe *old_p,
 453                 struct kprobe *p, unsigned long flags)
 454 {
 455         list_del(&p->list);
 456         if (list_empty(&old_p->list)) {
 457                 cleanup_kprobe(old_p, flags);
 458                 kfree(old_p);
 459         } else
 460                 spin_unlock_irqrestore(&kprobe_lock, flags);
 461 }
 462
 463 static int __kprobes in_kprobes_functions(unsigned long addr)
 464 {
 465         if (addr >= (unsigned long)__kprobes_text_start
 466                 && addr < (unsigned long)__kprobes_text_end)
 467                 return -EINVAL;
 468         return 0;
 469 }
 470
 471 int __kprobes register_kprobe(struct kprobe *p)
 472 {
 473         int ret = 0;
 474         unsigned long flags = 0;
 475         struct kprobe *old_p;
 476
 477         if ((ret = in_kprobes_functions((unsigned long) p->addr)) != 0)
 478                 return ret;
 479         if ((ret = arch_prepare_kprobe(p)) != 0)
 480                 goto rm_kprobe;
 481
 482         spin_lock_irqsave(&kprobe_lock, flags);
 483         old_p = get_kprobe(p->addr);
 484         p->nmissed = 0;
 485         if (old_p) {
 486                 ret = register_aggr_kprobe(old_p, p);
 487                 goto out;
 488         }
 489
 490         arch_copy_kprobe(p);
 491         INIT_HLIST_NODE(&p->hlist);
 492         hlist_add_head(&p->hlist,
 493                        &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
 494
 495         arch_arm_kprobe(p);
 496
 497 out:
 498         spin_unlock_irqrestore(&kprobe_lock, flags);
 499 rm_kprobe:
 500         if (ret == -EEXIST)
 501                 arch_remove_kprobe(p);
 502         return ret;
 503 }
 504
 505 void __kprobes unregister_kprobe(struct kprobe *p)
 506 {
 507         unsigned long flags;
 508         struct kprobe *old_p;
 509
 510         spin_lock_irqsave(&kprobe_lock, flags);
 511         old_p = get_kprobe(p->addr);
 512         if (old_p) {
 513                 if (old_p->pre_handler == aggr_pre_handler)
 514                         cleanup_aggr_kprobe(old_p, p, flags);
 515                 else
 516                         cleanup_kprobe(p, flags);
 517         } else
 518                 spin_unlock_irqrestore(&kprobe_lock, flags);
 519 }
 520
 521 static struct notifier_block kprobe_exceptions_nb = {
 522         .notifier_call = kprobe_exceptions_notify,
 523         .priority = 0x7fffffff /* we need to notified first */
 524 };
 525
 526 int __kprobes register_jprobe(struct jprobe *jp)
 527 {
 528         /* Todo: Verify probepoint is a function entry point */
 529         jp->kp.pre_handler = setjmp_pre_handler;
 530         jp->kp.break_handler = longjmp_break_handler;
 531
 532         return register_kprobe(&jp->kp);
 533 }
 534
 535 void __kprobes unregister_jprobe(struct jprobe *jp)
 536 {
 537         unregister_kprobe(&jp->kp);
 538 }
 539
 540 #ifdef ARCH_SUPPORTS_KRETPROBES
 541
 542 int __kprobes register_kretprobe(struct kretprobe *rp)
 543 {
 544         int ret = 0;
 545         struct kretprobe_instance *inst;
 546         int i;
 547
 548         rp->kp.pre_handler = pre_handler_kretprobe;
 549
 550         /* Pre-allocate memory for max kretprobe instances */
 551         if (rp->maxactive <= 0) {
 552 #ifdef CONFIG_PREEMPT
 553                 rp->maxactive = max(10, 2 * NR_CPUS);
 554 #else
 555                 rp->maxactive = NR_CPUS;
 556 #endif
 557         }
 558         INIT_HLIST_HEAD(&rp->used_instances);
 559         INIT_HLIST_HEAD(&rp->free_instances);
 560         for (i = 0; i < rp->maxactive; i++) {
 561                 inst = kmalloc(sizeof(struct kretprobe_instance), GFP_KERNEL);
 562                 if (inst == NULL) {
 563                         free_rp_inst(rp);
 564                         return -ENOMEM;
 565                 }
 566                 INIT_HLIST_NODE(&inst->uflist);
 567                 hlist_add_head(&inst->uflist, &rp->free_instances);
 568         }
 569
 570         rp->nmissed = 0;
 571         /* Establish function entry probe point */
 572         if ((ret = register_kprobe(&rp->kp)) != 0)
 573                 free_rp_inst(rp);
 574         return ret;
 575 }
 576
 577 #else /* ARCH_SUPPORTS_KRETPROBES */
 578
 579 int __kprobes register_kretprobe(struct kretprobe *rp)
 580 {
 581         return -ENOSYS;
 582 }
 583
 584 #endif /* ARCH_SUPPORTS_KRETPROBES */
 585
 586 void __kprobes unregister_kretprobe(struct kretprobe *rp)
 587 {
 588         unsigned long flags;
 589         struct kretprobe_instance *ri;
 590
 591         unregister_kprobe(&rp->kp);
 592         /* No race here */
 593         spin_lock_irqsave(&kprobe_lock, flags);
 594         free_rp_inst(rp);
 595         while ((ri = get_used_rp_inst(rp)) != NULL) {
 596                 ri->rp = NULL;
 597                 hlist_del(&ri->uflist);
 598         }
 599         spin_unlock_irqrestore(&kprobe_lock, flags);
 600 }
 601
 602 static int __init init_kprobes(void)
 603 {
 604         int i, err = 0;
 605
 606         /* FIXME allocate the probe table, currently defined statically */
 607         /* initialize all list heads */
 608         for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
 609                 INIT_HLIST_HEAD(&kprobe_table[i]);
 610                 INIT_HLIST_HEAD(&kretprobe_inst_table[i]);
 611         }
 612
 613         err = arch_init_kprobes();
 614         if (!err)
 615                 err = register_die_notifier(&kprobe_exceptions_nb);
 616
 617         return err;
 618 }
 619
 620 __initcall(init_kprobes);
 621
 622 EXPORT_SYMBOL_GPL(register_kprobe);
 623 EXPORT_SYMBOL_GPL(unregister_kprobe);
 624 EXPORT_SYMBOL_GPL(register_jprobe);
 625 EXPORT_SYMBOL_GPL(unregister_jprobe);
 626 EXPORT_SYMBOL_GPL(jprobe_return);
 627 EXPORT_SYMBOL_GPL(register_kretprobe);
 628 EXPORT_SYMBOL_GPL(unregister_kretprobe);
 629