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-Oct Jim Keniston <jkenisto@us.ibm.com> and Prasanna S Panchamukhi 26 * <prasanna@in.ibm.com> adapted for x86_64 from i386. 27 * 2005-Mar Roland McGrath <roland@redhat.com> 28 * Fixed to handle %rip-relative addressing mode correctly. 29 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston 30 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi 31 * <prasanna@in.ibm.com> added function-return probes. 32 * 2005-May Rusty Lynch <rusty.lynch@intel.com> 33 * Added function return probes functionality 34 * 2006-Feb Masami Hiramatsu <hiramatu@sdl.hitachi.co.jp> added 35 * kprobe-booster and kretprobe-booster for i386. 36 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com> added kprobe-booster 37 * and kretprobe-booster for x86-64 38 * 2007-Dec Masami Hiramatsu <mhiramat@redhat.com>, Arjan van de Ven 39 * <arjan@infradead.org> and Jim Keniston <jkenisto@us.ibm.com> 40 * unified x86 kprobes code. 41 */ 42 #include <linux/kprobes.h> 43 #include <linux/ptrace.h> 44 #include <linux/string.h> 45 #include <linux/slab.h> 46 #include <linux/hardirq.h> 47 #include <linux/preempt.h> 48 #include <linux/sched/debug.h> 49 #include <linux/extable.h> 50 #include <linux/kdebug.h> 51 #include <linux/kallsyms.h> 52 #include <linux/ftrace.h> 53 #include <linux/frame.h> 54 #include <linux/kasan.h> 55 #include <linux/moduleloader.h> 56 57 #include <asm/text-patching.h> 58 #include <asm/cacheflush.h> 59 #include <asm/desc.h> 60 #include <asm/pgtable.h> 61 #include <linux/uaccess.h> 62 #include <asm/alternative.h> 63 #include <asm/insn.h> 64 #include <asm/debugreg.h> 65 #include <asm/set_memory.h> 66 67 #include "common.h" 68 69 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL; 70 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); 71 72 #define stack_addr(regs) ((unsigned long *)kernel_stack_pointer(regs)) 73 74 #define W(row, b0, b1, b2, b3, b4, b5, b6, b7, b8, b9, ba, bb, bc, bd, be, bf)\ 75 (((b0##UL << 0x0)|(b1##UL << 0x1)|(b2##UL << 0x2)|(b3##UL << 0x3) | \ 76 (b4##UL << 0x4)|(b5##UL << 0x5)|(b6##UL << 0x6)|(b7##UL << 0x7) | \ 77 (b8##UL << 0x8)|(b9##UL << 0x9)|(ba##UL << 0xa)|(bb##UL << 0xb) | \ 78 (bc##UL << 0xc)|(bd##UL << 0xd)|(be##UL << 0xe)|(bf##UL << 0xf)) \ 79 << (row % 32)) 80 /* 81 * Undefined/reserved opcodes, conditional jump, Opcode Extension 82 * Groups, and some special opcodes can not boost. 83 * This is non-const and volatile to keep gcc from statically 84 * optimizing it out, as variable_test_bit makes gcc think only 85 * *(unsigned long*) is used. 86 */ 87 static volatile u32 twobyte_is_boostable[256 / 32] = { 88 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 89 /* ---------------------------------------------- */ 90 W(0x00, 0, 0, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 0, 0) | /* 00 */ 91 W(0x10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1) , /* 10 */ 92 W(0x20, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 20 */ 93 W(0x30, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 30 */ 94 W(0x40, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) | /* 40 */ 95 W(0x50, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) , /* 50 */ 96 W(0x60, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 1) | /* 60 */ 97 W(0x70, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1) , /* 70 */ 98 W(0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0) | /* 80 */ 99 W(0x90, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) , /* 90 */ 100 W(0xa0, 1, 1, 0, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* a0 */ 101 W(0xb0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 1, 1) , /* b0 */ 102 W(0xc0, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1) | /* c0 */ 103 W(0xd0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) , /* d0 */ 104 W(0xe0, 0, 1, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 1, 0, 1) | /* e0 */ 105 W(0xf0, 0, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0) /* f0 */ 106 /* ----------------------------------------------- */ 107 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */ 108 }; 109 #undef W 110 111 struct kretprobe_blackpoint kretprobe_blacklist[] = { 112 {"__switch_to", }, /* This function switches only current task, but 113 doesn't switch kernel stack.*/ 114 {NULL, NULL} /* Terminator */ 115 }; 116 117 const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist); 118 119 static nokprobe_inline void 120 __synthesize_relative_insn(void *dest, void *from, void *to, u8 op) 121 { 122 struct __arch_relative_insn { 123 u8 op; 124 s32 raddr; 125 } __packed *insn; 126 127 insn = (struct __arch_relative_insn *)dest; 128 insn->raddr = (s32)((long)(to) - ((long)(from) + 5)); 129 insn->op = op; 130 } 131 132 /* Insert a jump instruction at address 'from', which jumps to address 'to'.*/ 133 void synthesize_reljump(void *dest, void *from, void *to) 134 { 135 __synthesize_relative_insn(dest, from, to, RELATIVEJUMP_OPCODE); 136 } 137 NOKPROBE_SYMBOL(synthesize_reljump); 138 139 /* Insert a call instruction at address 'from', which calls address 'to'.*/ 140 void synthesize_relcall(void *dest, void *from, void *to) 141 { 142 __synthesize_relative_insn(dest, from, to, RELATIVECALL_OPCODE); 143 } 144 NOKPROBE_SYMBOL(synthesize_relcall); 145 146 /* 147 * Skip the prefixes of the instruction. 148 */ 149 static kprobe_opcode_t *skip_prefixes(kprobe_opcode_t *insn) 150 { 151 insn_attr_t attr; 152 153 attr = inat_get_opcode_attribute((insn_byte_t)*insn); 154 while (inat_is_legacy_prefix(attr)) { 155 insn++; 156 attr = inat_get_opcode_attribute((insn_byte_t)*insn); 157 } 158 #ifdef CONFIG_X86_64 159 if (inat_is_rex_prefix(attr)) 160 insn++; 161 #endif 162 return insn; 163 } 164 NOKPROBE_SYMBOL(skip_prefixes); 165 166 /* 167 * Returns non-zero if INSN is boostable. 168 * RIP relative instructions are adjusted at copying time in 64 bits mode 169 */ 170 int can_boost(struct insn *insn, void *addr) 171 { 172 kprobe_opcode_t opcode; 173 174 if (search_exception_tables((unsigned long)addr)) 175 return 0; /* Page fault may occur on this address. */ 176 177 /* 2nd-byte opcode */ 178 if (insn->opcode.nbytes == 2) 179 return test_bit(insn->opcode.bytes[1], 180 (unsigned long *)twobyte_is_boostable); 181 182 if (insn->opcode.nbytes != 1) 183 return 0; 184 185 /* Can't boost Address-size override prefix */ 186 if (unlikely(inat_is_address_size_prefix(insn->attr))) 187 return 0; 188 189 opcode = insn->opcode.bytes[0]; 190 191 switch (opcode & 0xf0) { 192 case 0x60: 193 /* can't boost "bound" */ 194 return (opcode != 0x62); 195 case 0x70: 196 return 0; /* can't boost conditional jump */ 197 case 0x90: 198 return opcode != 0x9a; /* can't boost call far */ 199 case 0xc0: 200 /* can't boost software-interruptions */ 201 return (0xc1 < opcode && opcode < 0xcc) || opcode == 0xcf; 202 case 0xd0: 203 /* can boost AA* and XLAT */ 204 return (opcode == 0xd4 || opcode == 0xd5 || opcode == 0xd7); 205 case 0xe0: 206 /* can boost in/out and absolute jmps */ 207 return ((opcode & 0x04) || opcode == 0xea); 208 case 0xf0: 209 /* clear and set flags are boostable */ 210 return (opcode == 0xf5 || (0xf7 < opcode && opcode < 0xfe)); 211 default: 212 /* CS override prefix and call are not boostable */ 213 return (opcode != 0x2e && opcode != 0x9a); 214 } 215 } 216 217 static unsigned long 218 __recover_probed_insn(kprobe_opcode_t *buf, unsigned long addr) 219 { 220 struct kprobe *kp; 221 unsigned long faddr; 222 223 kp = get_kprobe((void *)addr); 224 faddr = ftrace_location(addr); 225 /* 226 * Addresses inside the ftrace location are refused by 227 * arch_check_ftrace_location(). Something went terribly wrong 228 * if such an address is checked here. 229 */ 230 if (WARN_ON(faddr && faddr != addr)) 231 return 0UL; 232 /* 233 * Use the current code if it is not modified by Kprobe 234 * and it cannot be modified by ftrace. 235 */ 236 if (!kp && !faddr) 237 return addr; 238 239 /* 240 * Basically, kp->ainsn.insn has an original instruction. 241 * However, RIP-relative instruction can not do single-stepping 242 * at different place, __copy_instruction() tweaks the displacement of 243 * that instruction. In that case, we can't recover the instruction 244 * from the kp->ainsn.insn. 245 * 246 * On the other hand, in case on normal Kprobe, kp->opcode has a copy 247 * of the first byte of the probed instruction, which is overwritten 248 * by int3. And the instruction at kp->addr is not modified by kprobes 249 * except for the first byte, we can recover the original instruction 250 * from it and kp->opcode. 251 * 252 * In case of Kprobes using ftrace, we do not have a copy of 253 * the original instruction. In fact, the ftrace location might 254 * be modified at anytime and even could be in an inconsistent state. 255 * Fortunately, we know that the original code is the ideal 5-byte 256 * long NOP. 257 */ 258 if (probe_kernel_read(buf, (void *)addr, 259 MAX_INSN_SIZE * sizeof(kprobe_opcode_t))) 260 return 0UL; 261 262 if (faddr) 263 memcpy(buf, ideal_nops[NOP_ATOMIC5], 5); 264 else 265 buf[0] = kp->opcode; 266 return (unsigned long)buf; 267 } 268 269 /* 270 * Recover the probed instruction at addr for further analysis. 271 * Caller must lock kprobes by kprobe_mutex, or disable preemption 272 * for preventing to release referencing kprobes. 273 * Returns zero if the instruction can not get recovered (or access failed). 274 */ 275 unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr) 276 { 277 unsigned long __addr; 278 279 __addr = __recover_optprobed_insn(buf, addr); 280 if (__addr != addr) 281 return __addr; 282 283 return __recover_probed_insn(buf, addr); 284 } 285 286 /* Check if paddr is at an instruction boundary */ 287 static int can_probe(unsigned long paddr) 288 { 289 unsigned long addr, __addr, offset = 0; 290 struct insn insn; 291 kprobe_opcode_t buf[MAX_INSN_SIZE]; 292 293 if (!kallsyms_lookup_size_offset(paddr, NULL, &offset)) 294 return 0; 295 296 /* Decode instructions */ 297 addr = paddr - offset; 298 while (addr < paddr) { 299 /* 300 * Check if the instruction has been modified by another 301 * kprobe, in which case we replace the breakpoint by the 302 * original instruction in our buffer. 303 * Also, jump optimization will change the breakpoint to 304 * relative-jump. Since the relative-jump itself is 305 * normally used, we just go through if there is no kprobe. 306 */ 307 __addr = recover_probed_instruction(buf, addr); 308 if (!__addr) 309 return 0; 310 kernel_insn_init(&insn, (void *)__addr, MAX_INSN_SIZE); 311 insn_get_length(&insn); 312 313 /* 314 * Another debugging subsystem might insert this breakpoint. 315 * In that case, we can't recover it. 316 */ 317 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) 318 return 0; 319 addr += insn.length; 320 } 321 322 return (addr == paddr); 323 } 324 325 /* 326 * Returns non-zero if opcode modifies the interrupt flag. 327 */ 328 static int is_IF_modifier(kprobe_opcode_t *insn) 329 { 330 /* Skip prefixes */ 331 insn = skip_prefixes(insn); 332 333 switch (*insn) { 334 case 0xfa: /* cli */ 335 case 0xfb: /* sti */ 336 case 0xcf: /* iret/iretd */ 337 case 0x9d: /* popf/popfd */ 338 return 1; 339 } 340 341 return 0; 342 } 343 344 /* 345 * Copy an instruction with recovering modified instruction by kprobes 346 * and adjust the displacement if the instruction uses the %rip-relative 347 * addressing mode. Note that since @real will be the final place of copied 348 * instruction, displacement must be adjust by @real, not @dest. 349 * This returns the length of copied instruction, or 0 if it has an error. 350 */ 351 int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn) 352 { 353 kprobe_opcode_t buf[MAX_INSN_SIZE]; 354 unsigned long recovered_insn = 355 recover_probed_instruction(buf, (unsigned long)src); 356 357 if (!recovered_insn || !insn) 358 return 0; 359 360 /* This can access kernel text if given address is not recovered */ 361 if (probe_kernel_read(dest, (void *)recovered_insn, MAX_INSN_SIZE)) 362 return 0; 363 364 kernel_insn_init(insn, dest, MAX_INSN_SIZE); 365 insn_get_length(insn); 366 367 /* Another subsystem puts a breakpoint, failed to recover */ 368 if (insn->opcode.bytes[0] == BREAKPOINT_INSTRUCTION) 369 return 0; 370 371 /* We should not singlestep on the exception masking instructions */ 372 if (insn_masking_exception(insn)) 373 return 0; 374 375 #ifdef CONFIG_X86_64 376 /* Only x86_64 has RIP relative instructions */ 377 if (insn_rip_relative(insn)) { 378 s64 newdisp; 379 u8 *disp; 380 /* 381 * The copied instruction uses the %rip-relative addressing 382 * mode. Adjust the displacement for the difference between 383 * the original location of this instruction and the location 384 * of the copy that will actually be run. The tricky bit here 385 * is making sure that the sign extension happens correctly in 386 * this calculation, since we need a signed 32-bit result to 387 * be sign-extended to 64 bits when it's added to the %rip 388 * value and yield the same 64-bit result that the sign- 389 * extension of the original signed 32-bit displacement would 390 * have given. 391 */ 392 newdisp = (u8 *) src + (s64) insn->displacement.value 393 - (u8 *) real; 394 if ((s64) (s32) newdisp != newdisp) { 395 pr_err("Kprobes error: new displacement does not fit into s32 (%llx)\n", newdisp); 396 return 0; 397 } 398 disp = (u8 *) dest + insn_offset_displacement(insn); 399 *(s32 *) disp = (s32) newdisp; 400 } 401 #endif 402 return insn->length; 403 } 404 405 /* Prepare reljump right after instruction to boost */ 406 static int prepare_boost(kprobe_opcode_t *buf, struct kprobe *p, 407 struct insn *insn) 408 { 409 int len = insn->length; 410 411 if (can_boost(insn, p->addr) && 412 MAX_INSN_SIZE - len >= RELATIVEJUMP_SIZE) { 413 /* 414 * These instructions can be executed directly if it 415 * jumps back to correct address. 416 */ 417 synthesize_reljump(buf + len, p->ainsn.insn + len, 418 p->addr + insn->length); 419 len += RELATIVEJUMP_SIZE; 420 p->ainsn.boostable = true; 421 } else { 422 p->ainsn.boostable = false; 423 } 424 425 return len; 426 } 427 428 /* Make page to RO mode when allocate it */ 429 void *alloc_insn_page(void) 430 { 431 void *page; 432 433 page = module_alloc(PAGE_SIZE); 434 if (page) 435 set_memory_ro((unsigned long)page & PAGE_MASK, 1); 436 437 return page; 438 } 439 440 /* Recover page to RW mode before releasing it */ 441 void free_insn_page(void *page) 442 { 443 set_memory_nx((unsigned long)page & PAGE_MASK, 1); 444 set_memory_rw((unsigned long)page & PAGE_MASK, 1); 445 module_memfree(page); 446 } 447 448 static int arch_copy_kprobe(struct kprobe *p) 449 { 450 struct insn insn; 451 kprobe_opcode_t buf[MAX_INSN_SIZE]; 452 int len; 453 454 /* Copy an instruction with recovering if other optprobe modifies it.*/ 455 len = __copy_instruction(buf, p->addr, p->ainsn.insn, &insn); 456 if (!len) 457 return -EINVAL; 458 459 /* 460 * __copy_instruction can modify the displacement of the instruction, 461 * but it doesn't affect boostable check. 462 */ 463 len = prepare_boost(buf, p, &insn); 464 465 /* Check whether the instruction modifies Interrupt Flag or not */ 466 p->ainsn.if_modifier = is_IF_modifier(buf); 467 468 /* Also, displacement change doesn't affect the first byte */ 469 p->opcode = buf[0]; 470 471 /* OK, write back the instruction(s) into ROX insn buffer */ 472 text_poke(p->ainsn.insn, buf, len); 473 474 return 0; 475 } 476 477 int arch_prepare_kprobe(struct kprobe *p) 478 { 479 int ret; 480 481 if (alternatives_text_reserved(p->addr, p->addr)) 482 return -EINVAL; 483 484 if (!can_probe((unsigned long)p->addr)) 485 return -EILSEQ; 486 /* insn: must be on special executable page on x86. */ 487 p->ainsn.insn = get_insn_slot(); 488 if (!p->ainsn.insn) 489 return -ENOMEM; 490 491 ret = arch_copy_kprobe(p); 492 if (ret) { 493 free_insn_slot(p->ainsn.insn, 0); 494 p->ainsn.insn = NULL; 495 } 496 497 return ret; 498 } 499 500 void arch_arm_kprobe(struct kprobe *p) 501 { 502 text_poke(p->addr, ((unsigned char []){BREAKPOINT_INSTRUCTION}), 1); 503 } 504 505 void arch_disarm_kprobe(struct kprobe *p) 506 { 507 text_poke(p->addr, &p->opcode, 1); 508 } 509 510 void arch_remove_kprobe(struct kprobe *p) 511 { 512 if (p->ainsn.insn) { 513 free_insn_slot(p->ainsn.insn, p->ainsn.boostable); 514 p->ainsn.insn = NULL; 515 } 516 } 517 518 static nokprobe_inline void 519 save_previous_kprobe(struct kprobe_ctlblk *kcb) 520 { 521 kcb->prev_kprobe.kp = kprobe_running(); 522 kcb->prev_kprobe.status = kcb->kprobe_status; 523 kcb->prev_kprobe.old_flags = kcb->kprobe_old_flags; 524 kcb->prev_kprobe.saved_flags = kcb->kprobe_saved_flags; 525 } 526 527 static nokprobe_inline void 528 restore_previous_kprobe(struct kprobe_ctlblk *kcb) 529 { 530 __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp); 531 kcb->kprobe_status = kcb->prev_kprobe.status; 532 kcb->kprobe_old_flags = kcb->prev_kprobe.old_flags; 533 kcb->kprobe_saved_flags = kcb->prev_kprobe.saved_flags; 534 } 535 536 static nokprobe_inline void 537 set_current_kprobe(struct kprobe *p, struct pt_regs *regs, 538 struct kprobe_ctlblk *kcb) 539 { 540 __this_cpu_write(current_kprobe, p); 541 kcb->kprobe_saved_flags = kcb->kprobe_old_flags 542 = (regs->flags & (X86_EFLAGS_TF | X86_EFLAGS_IF)); 543 if (p->ainsn.if_modifier) 544 kcb->kprobe_saved_flags &= ~X86_EFLAGS_IF; 545 } 546 547 static nokprobe_inline void clear_btf(void) 548 { 549 if (test_thread_flag(TIF_BLOCKSTEP)) { 550 unsigned long debugctl = get_debugctlmsr(); 551 552 debugctl &= ~DEBUGCTLMSR_BTF; 553 update_debugctlmsr(debugctl); 554 } 555 } 556 557 static nokprobe_inline void restore_btf(void) 558 { 559 if (test_thread_flag(TIF_BLOCKSTEP)) { 560 unsigned long debugctl = get_debugctlmsr(); 561 562 debugctl |= DEBUGCTLMSR_BTF; 563 update_debugctlmsr(debugctl); 564 } 565 } 566 567 void arch_prepare_kretprobe(struct kretprobe_instance *ri, struct pt_regs *regs) 568 { 569 unsigned long *sara = stack_addr(regs); 570 571 ri->ret_addr = (kprobe_opcode_t *) *sara; 572 573 /* Replace the return addr with trampoline addr */ 574 *sara = (unsigned long) &kretprobe_trampoline; 575 } 576 NOKPROBE_SYMBOL(arch_prepare_kretprobe); 577 578 static void setup_singlestep(struct kprobe *p, struct pt_regs *regs, 579 struct kprobe_ctlblk *kcb, int reenter) 580 { 581 if (setup_detour_execution(p, regs, reenter)) 582 return; 583 584 #if !defined(CONFIG_PREEMPT) 585 if (p->ainsn.boostable && !p->post_handler) { 586 /* Boost up -- we can execute copied instructions directly */ 587 if (!reenter) 588 reset_current_kprobe(); 589 /* 590 * Reentering boosted probe doesn't reset current_kprobe, 591 * nor set current_kprobe, because it doesn't use single 592 * stepping. 593 */ 594 regs->ip = (unsigned long)p->ainsn.insn; 595 return; 596 } 597 #endif 598 if (reenter) { 599 save_previous_kprobe(kcb); 600 set_current_kprobe(p, regs, kcb); 601 kcb->kprobe_status = KPROBE_REENTER; 602 } else 603 kcb->kprobe_status = KPROBE_HIT_SS; 604 /* Prepare real single stepping */ 605 clear_btf(); 606 regs->flags |= X86_EFLAGS_TF; 607 regs->flags &= ~X86_EFLAGS_IF; 608 /* single step inline if the instruction is an int3 */ 609 if (p->opcode == BREAKPOINT_INSTRUCTION) 610 regs->ip = (unsigned long)p->addr; 611 else 612 regs->ip = (unsigned long)p->ainsn.insn; 613 } 614 NOKPROBE_SYMBOL(setup_singlestep); 615 616 /* 617 * We have reentered the kprobe_handler(), since another probe was hit while 618 * within the handler. We save the original kprobes variables and just single 619 * step on the instruction of the new probe without calling any user handlers. 620 */ 621 static int reenter_kprobe(struct kprobe *p, struct pt_regs *regs, 622 struct kprobe_ctlblk *kcb) 623 { 624 switch (kcb->kprobe_status) { 625 case KPROBE_HIT_SSDONE: 626 case KPROBE_HIT_ACTIVE: 627 case KPROBE_HIT_SS: 628 kprobes_inc_nmissed_count(p); 629 setup_singlestep(p, regs, kcb, 1); 630 break; 631 case KPROBE_REENTER: 632 /* A probe has been hit in the codepath leading up to, or just 633 * after, single-stepping of a probed instruction. This entire 634 * codepath should strictly reside in .kprobes.text section. 635 * Raise a BUG or we'll continue in an endless reentering loop 636 * and eventually a stack overflow. 637 */ 638 pr_err("Unrecoverable kprobe detected.\n"); 639 dump_kprobe(p); 640 BUG(); 641 default: 642 /* impossible cases */ 643 WARN_ON(1); 644 return 0; 645 } 646 647 return 1; 648 } 649 NOKPROBE_SYMBOL(reenter_kprobe); 650 651 /* 652 * Interrupts are disabled on entry as trap3 is an interrupt gate and they 653 * remain disabled throughout this function. 654 */ 655 int kprobe_int3_handler(struct pt_regs *regs) 656 { 657 kprobe_opcode_t *addr; 658 struct kprobe *p; 659 struct kprobe_ctlblk *kcb; 660 661 if (user_mode(regs)) 662 return 0; 663 664 addr = (kprobe_opcode_t *)(regs->ip - sizeof(kprobe_opcode_t)); 665 /* 666 * We don't want to be preempted for the entire duration of kprobe 667 * processing. Since int3 and debug trap disables irqs and we clear 668 * IF while singlestepping, it must be no preemptible. 669 */ 670 671 kcb = get_kprobe_ctlblk(); 672 p = get_kprobe(addr); 673 674 if (p) { 675 if (kprobe_running()) { 676 if (reenter_kprobe(p, regs, kcb)) 677 return 1; 678 } else { 679 set_current_kprobe(p, regs, kcb); 680 kcb->kprobe_status = KPROBE_HIT_ACTIVE; 681 682 /* 683 * If we have no pre-handler or it returned 0, we 684 * continue with normal processing. If we have a 685 * pre-handler and it returned non-zero, that means 686 * user handler setup registers to exit to another 687 * instruction, we must skip the single stepping. 688 */ 689 if (!p->pre_handler || !p->pre_handler(p, regs)) 690 setup_singlestep(p, regs, kcb, 0); 691 else 692 reset_current_kprobe(); 693 return 1; 694 } 695 } else if (*addr != BREAKPOINT_INSTRUCTION) { 696 /* 697 * The breakpoint instruction was removed right 698 * after we hit it. Another cpu has removed 699 * either a probepoint or a debugger breakpoint 700 * at this address. In either case, no further 701 * handling of this interrupt is appropriate. 702 * Back up over the (now missing) int3 and run 703 * the original instruction. 704 */ 705 regs->ip = (unsigned long)addr; 706 return 1; 707 } /* else: not a kprobe fault; let the kernel handle it */ 708 709 return 0; 710 } 711 NOKPROBE_SYMBOL(kprobe_int3_handler); 712 713 /* 714 * When a retprobed function returns, this code saves registers and 715 * calls trampoline_handler() runs, which calls the kretprobe's handler. 716 */ 717 asm( 718 ".global kretprobe_trampoline\n" 719 ".type kretprobe_trampoline, @function\n" 720 "kretprobe_trampoline:\n" 721 #ifdef CONFIG_X86_64 722 /* We don't bother saving the ss register */ 723 " pushq %rsp\n" 724 " pushfq\n" 725 SAVE_REGS_STRING 726 " movq %rsp, %rdi\n" 727 " call trampoline_handler\n" 728 /* Replace saved sp with true return address. */ 729 " movq %rax, 152(%rsp)\n" 730 RESTORE_REGS_STRING 731 " popfq\n" 732 #else 733 " pushf\n" 734 SAVE_REGS_STRING 735 " movl %esp, %eax\n" 736 " call trampoline_handler\n" 737 /* Move flags to cs */ 738 " movl 56(%esp), %edx\n" 739 " movl %edx, 52(%esp)\n" 740 /* Replace saved flags with true return address. */ 741 " movl %eax, 56(%esp)\n" 742 RESTORE_REGS_STRING 743 " popf\n" 744 #endif 745 " ret\n" 746 ".size kretprobe_trampoline, .-kretprobe_trampoline\n" 747 ); 748 NOKPROBE_SYMBOL(kretprobe_trampoline); 749 STACK_FRAME_NON_STANDARD(kretprobe_trampoline); 750 751 /* 752 * Called from kretprobe_trampoline 753 */ 754 __visible __used void *trampoline_handler(struct pt_regs *regs) 755 { 756 struct kretprobe_instance *ri = NULL; 757 struct hlist_head *head, empty_rp; 758 struct hlist_node *tmp; 759 unsigned long flags, orig_ret_address = 0; 760 unsigned long trampoline_address = (unsigned long)&kretprobe_trampoline; 761 kprobe_opcode_t *correct_ret_addr = NULL; 762 763 INIT_HLIST_HEAD(&empty_rp); 764 kretprobe_hash_lock(current, &head, &flags); 765 /* fixup registers */ 766 #ifdef CONFIG_X86_64 767 regs->cs = __KERNEL_CS; 768 #else 769 regs->cs = __KERNEL_CS | get_kernel_rpl(); 770 regs->gs = 0; 771 #endif 772 regs->ip = trampoline_address; 773 regs->orig_ax = ~0UL; 774 775 /* 776 * It is possible to have multiple instances associated with a given 777 * task either because multiple functions in the call path have 778 * return probes installed on them, and/or more than one 779 * return probe was registered for a target function. 780 * 781 * We can handle this because: 782 * - instances are always pushed into the head of the list 783 * - when multiple return probes are registered for the same 784 * function, the (chronologically) first instance's ret_addr 785 * will be the real return address, and all the rest will 786 * point to kretprobe_trampoline. 787 */ 788 hlist_for_each_entry(ri, head, hlist) { 789 if (ri->task != current) 790 /* another task is sharing our hash bucket */ 791 continue; 792 793 orig_ret_address = (unsigned long)ri->ret_addr; 794 795 if (orig_ret_address != trampoline_address) 796 /* 797 * This is the real return address. Any other 798 * instances associated with this task are for 799 * other calls deeper on the call stack 800 */ 801 break; 802 } 803 804 kretprobe_assert(ri, orig_ret_address, trampoline_address); 805 806 correct_ret_addr = ri->ret_addr; 807 hlist_for_each_entry_safe(ri, tmp, head, hlist) { 808 if (ri->task != current) 809 /* another task is sharing our hash bucket */ 810 continue; 811 812 orig_ret_address = (unsigned long)ri->ret_addr; 813 if (ri->rp && ri->rp->handler) { 814 __this_cpu_write(current_kprobe, &ri->rp->kp); 815 get_kprobe_ctlblk()->kprobe_status = KPROBE_HIT_ACTIVE; 816 ri->ret_addr = correct_ret_addr; 817 ri->rp->handler(ri, regs); 818 __this_cpu_write(current_kprobe, NULL); 819 } 820 821 recycle_rp_inst(ri, &empty_rp); 822 823 if (orig_ret_address != trampoline_address) 824 /* 825 * This is the real return address. Any other 826 * instances associated with this task are for 827 * other calls deeper on the call stack 828 */ 829 break; 830 } 831 832 kretprobe_hash_unlock(current, &flags); 833 834 hlist_for_each_entry_safe(ri, tmp, &empty_rp, hlist) { 835 hlist_del(&ri->hlist); 836 kfree(ri); 837 } 838 return (void *)orig_ret_address; 839 } 840 NOKPROBE_SYMBOL(trampoline_handler); 841 842 /* 843 * Called after single-stepping. p->addr is the address of the 844 * instruction whose first byte has been replaced by the "int 3" 845 * instruction. To avoid the SMP problems that can occur when we 846 * temporarily put back the original opcode to single-step, we 847 * single-stepped a copy of the instruction. The address of this 848 * copy is p->ainsn.insn. 849 * 850 * This function prepares to return from the post-single-step 851 * interrupt. We have to fix up the stack as follows: 852 * 853 * 0) Except in the case of absolute or indirect jump or call instructions, 854 * the new ip is relative to the copied instruction. We need to make 855 * it relative to the original instruction. 856 * 857 * 1) If the single-stepped instruction was pushfl, then the TF and IF 858 * flags are set in the just-pushed flags, and may need to be cleared. 859 * 860 * 2) If the single-stepped instruction was a call, the return address 861 * that is atop the stack is the address following the copied instruction. 862 * We need to make it the address following the original instruction. 863 * 864 * If this is the first time we've single-stepped the instruction at 865 * this probepoint, and the instruction is boostable, boost it: add a 866 * jump instruction after the copied instruction, that jumps to the next 867 * instruction after the probepoint. 868 */ 869 static void resume_execution(struct kprobe *p, struct pt_regs *regs, 870 struct kprobe_ctlblk *kcb) 871 { 872 unsigned long *tos = stack_addr(regs); 873 unsigned long copy_ip = (unsigned long)p->ainsn.insn; 874 unsigned long orig_ip = (unsigned long)p->addr; 875 kprobe_opcode_t *insn = p->ainsn.insn; 876 877 /* Skip prefixes */ 878 insn = skip_prefixes(insn); 879 880 regs->flags &= ~X86_EFLAGS_TF; 881 switch (*insn) { 882 case 0x9c: /* pushfl */ 883 *tos &= ~(X86_EFLAGS_TF | X86_EFLAGS_IF); 884 *tos |= kcb->kprobe_old_flags; 885 break; 886 case 0xc2: /* iret/ret/lret */ 887 case 0xc3: 888 case 0xca: 889 case 0xcb: 890 case 0xcf: 891 case 0xea: /* jmp absolute -- ip is correct */ 892 /* ip is already adjusted, no more changes required */ 893 p->ainsn.boostable = true; 894 goto no_change; 895 case 0xe8: /* call relative - Fix return addr */ 896 *tos = orig_ip + (*tos - copy_ip); 897 break; 898 #ifdef CONFIG_X86_32 899 case 0x9a: /* call absolute -- same as call absolute, indirect */ 900 *tos = orig_ip + (*tos - copy_ip); 901 goto no_change; 902 #endif 903 case 0xff: 904 if ((insn[1] & 0x30) == 0x10) { 905 /* 906 * call absolute, indirect 907 * Fix return addr; ip is correct. 908 * But this is not boostable 909 */ 910 *tos = orig_ip + (*tos - copy_ip); 911 goto no_change; 912 } else if (((insn[1] & 0x31) == 0x20) || 913 ((insn[1] & 0x31) == 0x21)) { 914 /* 915 * jmp near and far, absolute indirect 916 * ip is correct. And this is boostable 917 */ 918 p->ainsn.boostable = true; 919 goto no_change; 920 } 921 default: 922 break; 923 } 924 925 regs->ip += orig_ip - copy_ip; 926 927 no_change: 928 restore_btf(); 929 } 930 NOKPROBE_SYMBOL(resume_execution); 931 932 /* 933 * Interrupts are disabled on entry as trap1 is an interrupt gate and they 934 * remain disabled throughout this function. 935 */ 936 int kprobe_debug_handler(struct pt_regs *regs) 937 { 938 struct kprobe *cur = kprobe_running(); 939 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 940 941 if (!cur) 942 return 0; 943 944 resume_execution(cur, regs, kcb); 945 regs->flags |= kcb->kprobe_saved_flags; 946 947 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) { 948 kcb->kprobe_status = KPROBE_HIT_SSDONE; 949 cur->post_handler(cur, regs, 0); 950 } 951 952 /* Restore back the original saved kprobes variables and continue. */ 953 if (kcb->kprobe_status == KPROBE_REENTER) { 954 restore_previous_kprobe(kcb); 955 goto out; 956 } 957 reset_current_kprobe(); 958 out: 959 /* 960 * if somebody else is singlestepping across a probe point, flags 961 * will have TF set, in which case, continue the remaining processing 962 * of do_debug, as if this is not a probe hit. 963 */ 964 if (regs->flags & X86_EFLAGS_TF) 965 return 0; 966 967 return 1; 968 } 969 NOKPROBE_SYMBOL(kprobe_debug_handler); 970 971 int kprobe_fault_handler(struct pt_regs *regs, int trapnr) 972 { 973 struct kprobe *cur = kprobe_running(); 974 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); 975 976 if (unlikely(regs->ip == (unsigned long)cur->ainsn.insn)) { 977 /* This must happen on single-stepping */ 978 WARN_ON(kcb->kprobe_status != KPROBE_HIT_SS && 979 kcb->kprobe_status != KPROBE_REENTER); 980 /* 981 * We are here because the instruction being single 982 * stepped caused a page fault. We reset the current 983 * kprobe and the ip points back to the probe address 984 * and allow the page fault handler to continue as a 985 * normal page fault. 986 */ 987 regs->ip = (unsigned long)cur->addr; 988 /* 989 * Trap flag (TF) has been set here because this fault 990 * happened where the single stepping will be done. 991 * So clear it by resetting the current kprobe: 992 */ 993 regs->flags &= ~X86_EFLAGS_TF; 994 995 /* 996 * If the TF flag was set before the kprobe hit, 997 * don't touch it: 998 */ 999 regs->flags |= kcb->kprobe_old_flags; 1000 1001 if (kcb->kprobe_status == KPROBE_REENTER) 1002 restore_previous_kprobe(kcb); 1003 else 1004 reset_current_kprobe(); 1005 } else if (kcb->kprobe_status == KPROBE_HIT_ACTIVE || 1006 kcb->kprobe_status == KPROBE_HIT_SSDONE) { 1007 /* 1008 * We increment the nmissed count for accounting, 1009 * we can also use npre/npostfault count for accounting 1010 * these specific fault cases. 1011 */ 1012 kprobes_inc_nmissed_count(cur); 1013 1014 /* 1015 * We come here because instructions in the pre/post 1016 * handler caused the page_fault, this could happen 1017 * if handler tries to access user space by 1018 * copy_from_user(), get_user() etc. Let the 1019 * user-specified handler try to fix it first. 1020 */ 1021 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr)) 1022 return 1; 1023 1024 /* 1025 * In case the user-specified fault handler returned 1026 * zero, try to fix up. 1027 */ 1028 if (fixup_exception(regs, trapnr)) 1029 return 1; 1030 1031 /* 1032 * fixup routine could not handle it, 1033 * Let do_page_fault() fix it. 1034 */ 1035 } 1036 1037 return 0; 1038 } 1039 NOKPROBE_SYMBOL(kprobe_fault_handler); 1040 1041 /* 1042 * Wrapper routine for handling exceptions. 1043 */ 1044 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val, 1045 void *data) 1046 { 1047 struct die_args *args = data; 1048 int ret = NOTIFY_DONE; 1049 1050 if (args->regs && user_mode(args->regs)) 1051 return ret; 1052 1053 if (val == DIE_GPF) { 1054 /* 1055 * To be potentially processing a kprobe fault and to 1056 * trust the result from kprobe_running(), we have 1057 * be non-preemptible. 1058 */ 1059 if (!preemptible() && kprobe_running() && 1060 kprobe_fault_handler(args->regs, args->trapnr)) 1061 ret = NOTIFY_STOP; 1062 } 1063 return ret; 1064 } 1065 NOKPROBE_SYMBOL(kprobe_exceptions_notify); 1066 1067 bool arch_within_kprobe_blacklist(unsigned long addr) 1068 { 1069 bool is_in_entry_trampoline_section = false; 1070 1071 #ifdef CONFIG_X86_64 1072 is_in_entry_trampoline_section = 1073 (addr >= (unsigned long)__entry_trampoline_start && 1074 addr < (unsigned long)__entry_trampoline_end); 1075 #endif 1076 return (addr >= (unsigned long)__kprobes_text_start && 1077 addr < (unsigned long)__kprobes_text_end) || 1078 (addr >= (unsigned long)__entry_text_start && 1079 addr < (unsigned long)__entry_text_end) || 1080 is_in_entry_trampoline_section; 1081 } 1082 1083 int __init arch_init_kprobes(void) 1084 { 1085 return 0; 1086 } 1087 1088 int arch_trampoline_kprobe(struct kprobe *p) 1089 { 1090 return 0; 1091 } 1092