1 /* 2 * Kernel Probes (KProbes) 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, write to the Free Software 16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 17 * 18 * Copyright (C) IBM Corporation, 2002, 2004 19 * 20 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel 21 * Probes initial implementation ( includes contributions from 22 * Rusty Russell). 23 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes 24 * interface to access function arguments. 25 * 2004-Nov Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port 26 * for PPC64 27 */ 28 29 #include <linux/kprobes.h> 30 #include <linux/ptrace.h> 31 #include <linux/preempt.h> 32 #include <linux/extable.h> 33 #include <linux/kdebug.h> 34 #include <linux/slab.h> 35 #include <asm/code-patching.h> 36 #include <asm/cacheflush.h> 37 #include <asm/sstep.h> 38 #include <asm/sections.h> 39 #include <linux/uaccess.h> 40 41 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL; 42 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); 43 44 struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}}; 45 46 bool arch_within_kprobe_blacklist(unsigned long addr) 47 { 48 return (addr >= (unsigned long)__kprobes_text_start && 49 addr < (unsigned long)__kprobes_text_end) || 50 (addr >= (unsigned long)_stext && 51 addr < (unsigned long)__head_end); 52 } 53 54 kprobe_opcode_t *kprobe_lookup_name(const char *name, unsigned int offset) 55 { 56 kprobe_opcode_t *addr; 57 58 #ifdef PPC64_ELF_ABI_v2 59 /* PPC64 ABIv2 needs local entry point */ 60 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); 61 if (addr && !offset) { 62 #ifdef CONFIG_KPROBES_ON_FTRACE 63 unsigned long faddr; 64 /* 65 * Per livepatch.h, ftrace location is always within the first 66 * 16 bytes of a function on powerpc with -mprofile-kernel. 67 */ 68 faddr = ftrace_location_range((unsigned long)addr, 69 (unsigned long)addr + 16); 70 if (faddr) 71 addr = (kprobe_opcode_t *)faddr; 72 else 73 #endif 74 addr = (kprobe_opcode_t *)ppc_function_entry(addr); 75 } 76 #elif defined(PPC64_ELF_ABI_v1) 77 /* 78 * 64bit powerpc ABIv1 uses function descriptors: 79 * - Check for the dot variant of the symbol first. 80 * - If that fails, try looking up the symbol provided. 81 * 82 * This ensures we always get to the actual symbol and not 83 * the descriptor. 84 * 85 * Also handle <module:symbol> format. 86 */ 87 char dot_name[MODULE_NAME_LEN + 1 + KSYM_NAME_LEN]; 88 const char *modsym; 89 bool dot_appended = false; 90 if ((modsym = strchr(name, ':')) != NULL) { 91 modsym++; 92 if (*modsym != '\0' && *modsym != '.') { 93 /* Convert to <module:.symbol> */ 94 strncpy(dot_name, name, modsym - name); 95 dot_name[modsym - name] = '.'; 96 dot_name[modsym - name + 1] = '\0'; 97 strncat(dot_name, modsym, 98 sizeof(dot_name) - (modsym - name) - 2); 99 dot_appended = true; 100 } else { 101 dot_name[0] = '\0'; 102 strncat(dot_name, name, sizeof(dot_name) - 1); 103 } 104 } else if (name[0] != '.') { 105 dot_name[0] = '.'; 106 dot_name[1] = '\0'; 107 strncat(dot_name, name, KSYM_NAME_LEN - 2); 108 dot_appended = true; 109 } else { 110 dot_name[0] = '\0'; 111 strncat(dot_name, name, KSYM_NAME_LEN - 1); 112 } 113 addr = (kprobe_opcode_t *)kallsyms_lookup_name(dot_name); 114 if (!addr && dot_appended) { 115 /* Let's try the original non-dot symbol lookup */ 116 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); 117 } 118 #else 119 addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); 120 #endif 121 122 return addr; 123 } 124 125 int arch_prepare_kprobe(struct kprobe *p) 126 { 127 int ret = 0; 128 kprobe_opcode_t insn = *p->addr; 129 130 if ((unsigned long)p->addr & 0x03) { 131 printk("Attempt to register kprobe at an unaligned address\n"); 132 ret = -EINVAL; 133 } else if (IS_MTMSRD(insn) || IS_RFID(insn) || IS_RFI(insn)) { 134 printk("Cannot register a kprobe on rfi/rfid or mtmsr[d]\n"); 135 ret = -EINVAL; 136 } 137 138 /* insn must be on a special executable page on ppc64. This is 139 * not explicitly required on ppc32 (right now), but it doesn't hurt */ 140 if (!ret) { 141 p->ainsn.insn = get_insn_slot(); 142 if (!p->ainsn.insn) 143 ret = -ENOMEM; 144 } 145 146 if (!ret) { 147 memcpy(p->ainsn.insn, p->addr, 148 MAX_INSN_SIZE * sizeof(kprobe_opcode_t)); 149 p->opcode = *p->addr; 150 flush_icache_range((unsigned long)p->ainsn.insn, 151 (unsigned long)p->ainsn.insn + sizeof(kprobe_opcode_t)); 152 } 153 154 p->ainsn.boostable = 0; 155 return ret; 156 } 157 NOKPROBE_SYMBOL(arch_prepare_kprobe); 158 159 void arch_arm_kprobe(struct kprobe *p) 160 { 161 *p->addr = BREAKPOINT_INSTRUCTION; 162 flush_icache_range((unsigned long) p->addr, 163 (unsigned long) p->addr + sizeof(kprobe_opcode_t)); 164 } 165 NOKPROBE_SYMBOL(arch_arm_kprobe); 166 167 void arch_disarm_kprobe(struct kprobe *p) 168 { 169 *p->addr = p->opcode; 170 flush_icache_range((unsigned long) p->addr, 171 (unsigned long) p->addr + sizeof(kprobe_opcode_t)); 172 } 173 NOKPROBE_SYMBOL(arch_disarm_kprobe); 174 175 void arch_remove_kprobe(struct kprobe *p) 176 { 177 if (p->ainsn.insn) { 178 free_insn_slot(p->ainsn.insn, 0); 179 p->ainsn.insn = NULL; 180 } 181 } 182 NOKPROBE_SYMBOL(arch_remove_kprobe); 183 184 static nokprobe_inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs) 185 { 186 enable_single_step(regs); 187 188 /* 189 * On powerpc we should single step on the original 190 * instruction even if the probed insn is a trap 191 * variant as values in regs could play a part in 192 * if the trap is taken or not 193 */ 194 regs->nip = (unsigned long)p->ainsn.insn; 195 } 196 197 static nokprobe_inline void save_previous_kprobe(struct kprobe_ctlblk *kcb) 198 { 199 kcb->prev_kprobe.kp = kprobe_running(); 200 kcb->prev_kprobe.status = kcb->kprobe_status; 201 kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr; 202 } 203 204 static nokprobe_inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb) 205 { 206 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp); 207 kcb->kprobe_status = kcb->prev_kprobe.status; 208 kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr; 209 } 210 211 static nokprobe_inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs, 212 struct kprobe_ctlblk *kcb) 213 { 214 __this_cpu_write(current_kprobe, p); 215 kcb->kprobe_saved_msr = regs->msr; 216 } 217 218 bool arch_function_offset_within_entry(unsigned long offset) 219 { 220 #ifdef PPC64_ELF_ABI_v2 221 #ifdef CONFIG_KPROBES_ON_FTRACE 222 return offset <= 16; 223 #else 224 return offset <= 8; 225 #endif 226 #else 227 return !offset; 228 #endif 229 } 230 231 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs) 232 { 233 ri->ret_addr = (kprobe_opcode_t *)regs->link; 234 235 /* Replace the return addr with trampoline addr */ 236 regs->link = (unsigned long)kretprobe_trampoline; 237 } 238 NOKPROBE_SYMBOL(arch_prepare_kretprobe); 239 240 int try_to_emulate(struct kprobe *p, struct pt_regs *regs) 241 { 242 int ret; 243 unsigned int insn = *p->ainsn.insn; 244 245 /* regs->nip is also adjusted if emulate_step returns 1 */ 246 ret = emulate_step(regs, insn); 247 if (ret > 0) { 248 /* 249 * Once this instruction has been boosted 250 * successfully, set the boostable flag 251 */ 252 if (unlikely(p->ainsn.boostable == 0)) 253 p->ainsn.boostable = 1; 254 } else if (ret < 0) { 255 /* 256 * We don't allow kprobes on mtmsr(d)/rfi(d), etc. 257 * So, we should never get here... but, its still 258 * good to catch them, just in case... 259 */ 260 printk("Can't step on instruction %x\n", insn); 261 BUG(); 262 } else if (ret == 0) 263 /* This instruction can't be boosted */ 264 p->ainsn.boostable = -1; 265 266 return ret; 267 } 268 NOKPROBE_SYMBOL(try_to_emulate); 269 270 int kprobe_handler(struct pt_regs *regs) 271 { 272 struct kprobe *p; 273 int ret = 0; 274 unsigned int *addr = (unsigned int *)regs->nip; 275 struct kprobe_ctlblk *kcb; 276 277 if (user_mode(regs)) 278 return 0; 279 280 /* 281 * We don't want to be preempted for the entire 282 * duration of kprobe processing 283 */ 284 preempt_disable(); 285 kcb = get_kprobe_ctlblk(); 286 287 /* Check we're not actually recursing */ 288 if (kprobe_running()) { 289 p = get_kprobe(addr); 290 if (p) { 291 kprobe_opcode_t insn = *p->ainsn.insn; 292 if (kcb->kprobe_status == KPROBE_HIT_SS && 293 is_trap(insn)) { 294 /* Turn off 'trace' bits */ 295 regs->msr &= ~MSR_SINGLESTEP; 296 regs->msr |= kcb->kprobe_saved_msr; 297 goto no_kprobe; 298 } 299 /* We have reentered the kprobe_handler(), since 300 * another probe was hit while within the handler. 301 * We here save the original kprobes variables and 302 * just single step on the instruction of the new probe 303 * without calling any user handlers. 304 */ 305 save_previous_kprobe(kcb); 306 set_current_kprobe(p, regs, kcb); 307 kprobes_inc_nmissed_count(p); 308 prepare_singlestep(p, regs); 309 kcb->kprobe_status = KPROBE_REENTER; 310 if (p->ainsn.boostable >= 0) { 311 ret = try_to_emulate(p, regs); 312 313 if (ret > 0) { 314 restore_previous_kprobe(kcb); 315 return 1; 316 } 317 } 318 return 1; 319 } else { 320 if (*addr != BREAKPOINT_INSTRUCTION) { 321 /* If trap variant, then it belongs not to us */ 322 kprobe_opcode_t cur_insn = *addr; 323 if (is_trap(cur_insn)) 324 goto no_kprobe; 325 /* The breakpoint instruction was removed by 326 * another cpu right after we hit, no further 327 * handling of this interrupt is appropriate 328 */ 329 ret = 1; 330 goto no_kprobe; 331 } 332 p = __this_cpu_read(current_kprobe); 333 if (p->break_handler && p->break_handler(p, regs)) { 334 if (!skip_singlestep(p, regs, kcb)) 335 goto ss_probe; 336 ret = 1; 337 } 338 } 339 goto no_kprobe; 340 } 341 342 p = get_kprobe(addr); 343 if (!p) { 344 if (*addr != BREAKPOINT_INSTRUCTION) { 345 /* 346 * PowerPC has multiple variants of the "trap" 347 * instruction. If the current instruction is a 348 * trap variant, it could belong to someone else 349 */ 350 kprobe_opcode_t cur_insn = *addr; 351 if (is_trap(cur_insn)) 352 goto no_kprobe; 353 /* 354 * The breakpoint instruction was removed right 355 * after we hit it. Another cpu has removed 356 * either a probepoint or a debugger breakpoint 357 * at this address. In either case, no further 358 * handling of this interrupt is appropriate. 359 */ 360 ret = 1; 361 } 362 /* Not one of ours: let kernel handle it */ 363 goto no_kprobe; 364 } 365 366 kcb->kprobe_status = KPROBE_HIT_ACTIVE; 367 set_current_kprobe(p, regs, kcb); 368 if (p->pre_handler && p->pre_handler(p, regs)) 369 /* handler has already set things up, so skip ss setup */ 370 return 1; 371 372 ss_probe: 373 if (p->ainsn.boostable >= 0) { 374 ret = try_to_emulate(p, regs); 375 376 if (ret > 0) { 377 if (p->post_handler) 378 p->post_handler(p, regs, 0); 379 380 kcb->kprobe_status = KPROBE_HIT_SSDONE; 381 reset_current_kprobe(); 382 preempt_enable_no_resched(); 383 return 1; 384 } 385 } 386 prepare_singlestep(p, regs); 387 kcb->kprobe_status = KPROBE_HIT_SS; 388 return 1; 389 390 no_kprobe: 391 preempt_enable_no_resched(); 392 return ret; 393 } 394 NOKPROBE_SYMBOL(kprobe_handler); 395 396 /* 397 * Function return probe trampoline: 398 * - init_kprobes() establishes a probepoint here 399 * - When the probed function returns, this probe 400 * causes the handlers to fire 401 */ 402 asm(".global kretprobe_trampoline\n" 403 ".type kretprobe_trampoline, @function\n" 404 "kretprobe_trampoline:\n" 405 "nop\n" 406 "blr\n" 407 ".size kretprobe_trampoline, .-kretprobe_trampoline\n"); 408 409 /* 410 * Called when the probe at kretprobe trampoline is hit 411 */ 412 static int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs) 413 { 414 struct kretprobe_instance *ri = NULL; 415 struct hlist_head *head, empty_rp; 416 struct hlist_node *tmp; 417 unsigned long flags, orig_ret_address = 0; 418 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline; 419 420 INIT_HLIST_HEAD(&empty_rp); 421 kretprobe_hash_lock(current, &head, &flags); 422 423 /* 424 * It is possible to have multiple instances associated with a given 425 * task either because an multiple functions in the call path 426 * have a return probe installed on them, and/or more than one return 427 * return probe was registered for a target function. 428 * 429 * We can handle this because: 430 * - instances are always inserted at the head of the list 431 * - when multiple return probes are registered for the same 432 * function, the first instance's ret_addr will point to the 433 * real return address, and all the rest will point to 434 * kretprobe_trampoline 435 */ 436 hlist_for_each_entry_safe(ri, tmp, head, hlist) { 437 if (ri->task != current) 438 /* another task is sharing our hash bucket */ 439 continue; 440 441 if (ri->rp && ri->rp->handler) 442 ri->rp->handler(ri, regs); 443 444 orig_ret_address = (unsigned long)ri->ret_addr; 445 recycle_rp_inst(ri, &empty_rp); 446 447 if (orig_ret_address != trampoline_address) 448 /* 449 * This is the real return address. Any other 450 * instances associated with this task are for 451 * other calls deeper on the call stack 452 */ 453 break; 454 } 455 456 kretprobe_assert(ri, orig_ret_address, trampoline_address); 457 regs->nip = orig_ret_address; 458 /* 459 * Make LR point to the orig_ret_address. 460 * When the 'nop' inside the kretprobe_trampoline 461 * is optimized, we can do a 'blr' after executing the 462 * detour buffer code. 463 */ 464 regs->link = orig_ret_address; 465 466 reset_current_kprobe(); 467 kretprobe_hash_unlock(current, &flags); 468 preempt_enable_no_resched(); 469 470 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { 471 hlist_del(&ri->hlist); 472 kfree(ri); 473 } 474 /* 475 * By returning a non-zero value, we are telling 476 * kprobe_handler() that we don't want the post_handler 477 * to run (and have re-enabled preemption) 478 */ 479 return 1; 480 } 481 NOKPROBE_SYMBOL(trampoline_probe_handler); 482 483 /* 484 * Called after single-stepping. p->addr is the address of the 485 * instruction whose first byte has been replaced by the "breakpoint" 486 * instruction. To avoid the SMP problems that can occur when we 487 * temporarily put back the original opcode to single-step, we 488 * single-stepped a copy of the instruction. The address of this 489 * copy is p->ainsn.insn. 490 */ 491 int kprobe_post_handler(struct pt_regs *regs) 492 { 493 struct kprobe *cur = kprobe_running(); 494 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 495 496 if (!cur || user_mode(regs)) 497 return 0; 498 499 /* make sure we got here for instruction we have a kprobe on */ 500 if (((unsigned long)cur->ainsn.insn + 4) != regs->nip) 501 return 0; 502 503 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) { 504 kcb->kprobe_status = KPROBE_HIT_SSDONE; 505 cur->post_handler(cur, regs, 0); 506 } 507 508 /* Adjust nip to after the single-stepped instruction */ 509 regs->nip = (unsigned long)cur->addr + 4; 510 regs->msr |= kcb->kprobe_saved_msr; 511 512 /*Restore back the original saved kprobes variables and continue. */ 513 if (kcb->kprobe_status == KPROBE_REENTER) { 514 restore_previous_kprobe(kcb); 515 goto out; 516 } 517 reset_current_kprobe(); 518 out: 519 preempt_enable_no_resched(); 520 521 /* 522 * if somebody else is singlestepping across a probe point, msr 523 * will have DE/SE set, in which case, continue the remaining processing 524 * of do_debug, as if this is not a probe hit. 525 */ 526 if (regs->msr & MSR_SINGLESTEP) 527 return 0; 528 529 return 1; 530 } 531 NOKPROBE_SYMBOL(kprobe_post_handler); 532 533 int kprobe_fault_handler(struct pt_regs *regs, int trapnr) 534 { 535 struct kprobe *cur = kprobe_running(); 536 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 537 const struct exception_table_entry *entry; 538 539 switch(kcb->kprobe_status) { 540 case KPROBE_HIT_SS: 541 case KPROBE_REENTER: 542 /* 543 * We are here because the instruction being single 544 * stepped caused a page fault. We reset the current 545 * kprobe and the nip points back to the probe address 546 * and allow the page fault handler to continue as a 547 * normal page fault. 548 */ 549 regs->nip = (unsigned long)cur->addr; 550 regs->msr &= ~MSR_SINGLESTEP; /* Turn off 'trace' bits */ 551 regs->msr |= kcb->kprobe_saved_msr; 552 if (kcb->kprobe_status == KPROBE_REENTER) 553 restore_previous_kprobe(kcb); 554 else 555 reset_current_kprobe(); 556 preempt_enable_no_resched(); 557 break; 558 case KPROBE_HIT_ACTIVE: 559 case KPROBE_HIT_SSDONE: 560 /* 561 * We increment the nmissed count for accounting, 562 * we can also use npre/npostfault count for accounting 563 * these specific fault cases. 564 */ 565 kprobes_inc_nmissed_count(cur); 566 567 /* 568 * We come here because instructions in the pre/post 569 * handler caused the page_fault, this could happen 570 * if handler tries to access user space by 571 * copy_from_user(), get_user() etc. Let the 572 * user-specified handler try to fix it first. 573 */ 574 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr)) 575 return 1; 576 577 /* 578 * In case the user-specified fault handler returned 579 * zero, try to fix up. 580 */ 581 if ((entry = search_exception_tables(regs->nip)) != NULL) { 582 regs->nip = extable_fixup(entry); 583 return 1; 584 } 585 586 /* 587 * fixup_exception() could not handle it, 588 * Let do_page_fault() fix it. 589 */ 590 break; 591 default: 592 break; 593 } 594 return 0; 595 } 596 NOKPROBE_SYMBOL(kprobe_fault_handler); 597 598 unsigned long arch_deref_entry_point(void *entry) 599 { 600 return ppc_global_function_entry(entry); 601 } 602 NOKPROBE_SYMBOL(arch_deref_entry_point); 603 604 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs) 605 { 606 struct jprobe *jp = container_of(p, struct jprobe, kp); 607 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 608 609 memcpy(&kcb->jprobe_saved_regs, regs, sizeof(struct pt_regs)); 610 611 /* setup return addr to the jprobe handler routine */ 612 regs->nip = arch_deref_entry_point(jp->entry); 613 #ifdef PPC64_ELF_ABI_v2 614 regs->gpr[12] = (unsigned long)jp->entry; 615 #elif defined(PPC64_ELF_ABI_v1) 616 regs->gpr[2] = (unsigned long)(((func_descr_t *)jp->entry)->toc); 617 #endif 618 619 return 1; 620 } 621 NOKPROBE_SYMBOL(setjmp_pre_handler); 622 623 void __used jprobe_return(void) 624 { 625 asm volatile("trap" ::: "memory"); 626 } 627 NOKPROBE_SYMBOL(jprobe_return); 628 629 static void __used jprobe_return_end(void) 630 { 631 } 632 NOKPROBE_SYMBOL(jprobe_return_end); 633 634 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs) 635 { 636 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 637 638 /* 639 * FIXME - we should ideally be validating that we got here 'cos 640 * of the "trap" in jprobe_return() above, before restoring the 641 * saved regs... 642 */ 643 memcpy(regs, &kcb->jprobe_saved_regs, sizeof(struct pt_regs)); 644 preempt_enable_no_resched(); 645 return 1; 646 } 647 NOKPROBE_SYMBOL(longjmp_break_handler); 648 649 static struct kprobe trampoline_p = { 650 .addr = (kprobe_opcode_t *) &kretprobe_trampoline, 651 .pre_handler = trampoline_probe_handler 652 }; 653 654 int __init arch_init_kprobes(void) 655 { 656 return register_kprobe(&trampoline_p); 657 } 658 659 int arch_trampoline_kprobe(struct kprobe *p) 660 { 661 if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline) 662 return 1; 663 664 return 0; 665 } 666 NOKPROBE_SYMBOL(arch_trampoline_kprobe); 667