| 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
| 2 | /* |
| 3 | * Kernel Probes (KProbes) |
| 4 | * |
| 5 | * Copyright (C) IBM Corporation, 2002, 2004 |
| 6 | * |
| 7 | * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel |
| 8 | * Probes initial implementation ( includes contributions from |
| 9 | * Rusty Russell). |
| 10 | * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes |
| 11 | * interface to access function arguments. |
| 12 | * 2004-Nov Ananth N Mavinakayanahalli <ananth@in.ibm.com> kprobes port |
| 13 | * for PPC64 |
| 14 | */ |
| 15 | |
| 16 | #include <linux/kprobes.h> |
| 17 | #include <linux/ptrace.h> |
| 18 | #include <linux/preempt.h> |
| 19 | #include <linux/extable.h> |
| 20 | #include <linux/kdebug.h> |
| 21 | #include <linux/slab.h> |
| 22 | #include <linux/set_memory.h> |
| 23 | #include <linux/execmem.h> |
| 24 | #include <asm/text-patching.h> |
| 25 | #include <asm/cacheflush.h> |
| 26 | #include <asm/sstep.h> |
| 27 | #include <asm/sections.h> |
| 28 | #include <asm/inst.h> |
| 29 | #include <linux/uaccess.h> |
| 30 | |
| 31 | DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL; |
| 32 | DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); |
| 33 | |
| 34 | struct kretprobe_blackpoint kretprobe_blacklist[] = {{NULL, NULL}}; |
| 35 | |
| 36 | bool arch_within_kprobe_blacklist(unsigned long addr) |
| 37 | { |
| 38 | return (addr >= (unsigned long)__kprobes_text_start && |
| 39 | addr < (unsigned long)__kprobes_text_end) || |
| 40 | (addr >= (unsigned long)_stext && |
| 41 | addr < (unsigned long)__head_end); |
| 42 | } |
| 43 | |
| 44 | kprobe_opcode_t *kprobe_lookup_name(const char *name, unsigned int offset) |
| 45 | { |
| 46 | kprobe_opcode_t *addr = NULL; |
| 47 | |
| 48 | #ifdef CONFIG_PPC64_ELF_ABI_V2 |
| 49 | /* PPC64 ABIv2 needs local entry point */ |
| 50 | addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); |
| 51 | if (addr && !offset) { |
| 52 | #ifdef CONFIG_KPROBES_ON_FTRACE |
| 53 | unsigned long faddr; |
| 54 | /* |
| 55 | * Per livepatch.h, ftrace location is always within the first |
| 56 | * 16 bytes of a function on powerpc with -mprofile-kernel. |
| 57 | */ |
| 58 | faddr = ftrace_location_range((unsigned long)addr, |
| 59 | (unsigned long)addr + 16); |
| 60 | if (faddr) |
| 61 | addr = (kprobe_opcode_t *)faddr; |
| 62 | else |
| 63 | #endif |
| 64 | addr = (kprobe_opcode_t *)ppc_function_entry(addr); |
| 65 | } |
| 66 | #elif defined(CONFIG_PPC64_ELF_ABI_V1) |
| 67 | /* |
| 68 | * 64bit powerpc ABIv1 uses function descriptors: |
| 69 | * - Check for the dot variant of the symbol first. |
| 70 | * - If that fails, try looking up the symbol provided. |
| 71 | * |
| 72 | * This ensures we always get to the actual symbol and not |
| 73 | * the descriptor. |
| 74 | * |
| 75 | * Also handle <module:symbol> format. |
| 76 | */ |
| 77 | char dot_name[MODULE_NAME_LEN + 1 + KSYM_NAME_LEN]; |
| 78 | bool dot_appended = false; |
| 79 | const char *c; |
| 80 | ssize_t ret = 0; |
| 81 | int len = 0; |
| 82 | |
| 83 | if ((c = strnchr(name, MODULE_NAME_LEN, ':')) != NULL) { |
| 84 | c++; |
| 85 | len = c - name; |
| 86 | memcpy(dot_name, name, len); |
| 87 | } else |
| 88 | c = name; |
| 89 | |
| 90 | if (*c != '\0' && *c != '.') { |
| 91 | dot_name[len++] = '.'; |
| 92 | dot_appended = true; |
| 93 | } |
| 94 | ret = strscpy(dot_name + len, c, KSYM_NAME_LEN); |
| 95 | if (ret > 0) |
| 96 | addr = (kprobe_opcode_t *)kallsyms_lookup_name(dot_name); |
| 97 | |
| 98 | /* Fallback to the original non-dot symbol lookup */ |
| 99 | if (!addr && dot_appended) |
| 100 | addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); |
| 101 | #else |
| 102 | addr = (kprobe_opcode_t *)kallsyms_lookup_name(name); |
| 103 | #endif |
| 104 | |
| 105 | return addr; |
| 106 | } |
| 107 | |
| 108 | static bool arch_kprobe_on_func_entry(unsigned long addr, unsigned long offset) |
| 109 | { |
| 110 | unsigned long ip = ftrace_location(ip: addr); |
| 111 | |
| 112 | if (ip) |
| 113 | return offset <= (ip - addr); |
| 114 | if (IS_ENABLED(CONFIG_PPC64_ELF_ABI_V2) && !IS_ENABLED(CONFIG_PPC_KERNEL_PCREL)) |
| 115 | return offset <= 8; |
| 116 | return !offset; |
| 117 | } |
| 118 | |
| 119 | /* XXX try and fold the magic of kprobe_lookup_name() in this */ |
| 120 | kprobe_opcode_t *arch_adjust_kprobe_addr(unsigned long addr, unsigned long offset, |
| 121 | bool *on_func_entry) |
| 122 | { |
| 123 | *on_func_entry = arch_kprobe_on_func_entry(addr, offset); |
| 124 | return (kprobe_opcode_t *)(addr + offset); |
| 125 | } |
| 126 | |
| 127 | int arch_prepare_kprobe(struct kprobe *p) |
| 128 | { |
| 129 | int ret = 0; |
| 130 | struct kprobe *prev; |
| 131 | ppc_inst_t insn = ppc_inst_read(p->addr); |
| 132 | |
| 133 | if ((unsigned long)p->addr & 0x03) { |
| 134 | printk("Attempt to register kprobe at an unaligned address\n" ); |
| 135 | ret = -EINVAL; |
| 136 | } else if (!can_single_step(ppc_inst_val(insn))) { |
| 137 | printk("Cannot register a kprobe on instructions that can't be single stepped\n" ); |
| 138 | ret = -EINVAL; |
| 139 | } else if ((unsigned long)p->addr & ~PAGE_MASK && |
| 140 | ppc_inst_prefixed(ppc_inst_read(p->addr - 1))) { |
| 141 | printk("Cannot register a kprobe on the second word of prefixed instruction\n" ); |
| 142 | ret = -EINVAL; |
| 143 | } |
| 144 | prev = get_kprobe(addr: p->addr - 1); |
| 145 | |
| 146 | /* |
| 147 | * When prev is a ftrace-based kprobe, we don't have an insn, and it |
| 148 | * doesn't probe for prefixed instruction. |
| 149 | */ |
| 150 | if (prev && !kprobe_ftrace(p: prev) && |
| 151 | ppc_inst_prefixed(ppc_inst_read(prev->ainsn.insn))) { |
| 152 | printk("Cannot register a kprobe on the second word of prefixed instruction\n" ); |
| 153 | ret = -EINVAL; |
| 154 | } |
| 155 | |
| 156 | /* insn must be on a special executable page on ppc64. This is |
| 157 | * not explicitly required on ppc32 (right now), but it doesn't hurt */ |
| 158 | if (!ret) { |
| 159 | p->ainsn.insn = get_insn_slot(); |
| 160 | if (!p->ainsn.insn) |
| 161 | ret = -ENOMEM; |
| 162 | } |
| 163 | |
| 164 | if (!ret) { |
| 165 | patch_instruction(p->ainsn.insn, insn); |
| 166 | p->opcode = ppc_inst_val(insn); |
| 167 | } |
| 168 | |
| 169 | p->ainsn.boostable = 0; |
| 170 | return ret; |
| 171 | } |
| 172 | NOKPROBE_SYMBOL(arch_prepare_kprobe); |
| 173 | |
| 174 | void arch_arm_kprobe(struct kprobe *p) |
| 175 | { |
| 176 | WARN_ON_ONCE(patch_instruction(p->addr, ppc_inst(BREAKPOINT_INSTRUCTION))); |
| 177 | } |
| 178 | NOKPROBE_SYMBOL(arch_arm_kprobe); |
| 179 | |
| 180 | void arch_disarm_kprobe(struct kprobe *p) |
| 181 | { |
| 182 | WARN_ON_ONCE(patch_instruction(p->addr, ppc_inst(p->opcode))); |
| 183 | } |
| 184 | NOKPROBE_SYMBOL(arch_disarm_kprobe); |
| 185 | |
| 186 | void arch_remove_kprobe(struct kprobe *p) |
| 187 | { |
| 188 | if (p->ainsn.insn) { |
| 189 | free_insn_slot(slot: p->ainsn.insn, dirty: 0); |
| 190 | p->ainsn.insn = NULL; |
| 191 | } |
| 192 | } |
| 193 | NOKPROBE_SYMBOL(arch_remove_kprobe); |
| 194 | |
| 195 | static nokprobe_inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs) |
| 196 | { |
| 197 | enable_single_step(regs); |
| 198 | |
| 199 | /* |
| 200 | * On powerpc we should single step on the original |
| 201 | * instruction even if the probed insn is a trap |
| 202 | * variant as values in regs could play a part in |
| 203 | * if the trap is taken or not |
| 204 | */ |
| 205 | regs_set_return_ip(regs, (unsigned long)p->ainsn.insn); |
| 206 | } |
| 207 | |
| 208 | static nokprobe_inline void save_previous_kprobe(struct kprobe_ctlblk *kcb) |
| 209 | { |
| 210 | kcb->prev_kprobe.kp = kprobe_running(); |
| 211 | kcb->prev_kprobe.status = kcb->kprobe_status; |
| 212 | kcb->prev_kprobe.saved_msr = kcb->kprobe_saved_msr; |
| 213 | } |
| 214 | |
| 215 | static nokprobe_inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb) |
| 216 | { |
| 217 | __this_cpu_write(current_kprobe, kcb->prev_kprobe.kp); |
| 218 | kcb->kprobe_status = kcb->prev_kprobe.status; |
| 219 | kcb->kprobe_saved_msr = kcb->prev_kprobe.saved_msr; |
| 220 | } |
| 221 | |
| 222 | static nokprobe_inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs, |
| 223 | struct kprobe_ctlblk *kcb) |
| 224 | { |
| 225 | __this_cpu_write(current_kprobe, p); |
| 226 | kcb->kprobe_saved_msr = regs->msr; |
| 227 | } |
| 228 | |
| 229 | static int try_to_emulate(struct kprobe *p, struct pt_regs *regs) |
| 230 | { |
| 231 | int ret; |
| 232 | ppc_inst_t insn = ppc_inst_read(p->ainsn.insn); |
| 233 | |
| 234 | /* regs->nip is also adjusted if emulate_step returns 1 */ |
| 235 | ret = emulate_step(regs, insn); |
| 236 | if (ret > 0) { |
| 237 | /* |
| 238 | * Once this instruction has been boosted |
| 239 | * successfully, set the boostable flag |
| 240 | */ |
| 241 | if (unlikely(p->ainsn.boostable == 0)) |
| 242 | p->ainsn.boostable = 1; |
| 243 | } else if (ret < 0) { |
| 244 | /* |
| 245 | * We don't allow kprobes on mtmsr(d)/rfi(d), etc. |
| 246 | * So, we should never get here... but, its still |
| 247 | * good to catch them, just in case... |
| 248 | */ |
| 249 | printk("Can't step on instruction %08lx\n" , ppc_inst_as_ulong(insn)); |
| 250 | BUG(); |
| 251 | } else { |
| 252 | /* |
| 253 | * If we haven't previously emulated this instruction, then it |
| 254 | * can't be boosted. Note it down so we don't try to do so again. |
| 255 | * |
| 256 | * If, however, we had emulated this instruction in the past, |
| 257 | * then this is just an error with the current run (for |
| 258 | * instance, exceptions due to a load/store). We return 0 so |
| 259 | * that this is now single-stepped, but continue to try |
| 260 | * emulating it in subsequent probe hits. |
| 261 | */ |
| 262 | if (unlikely(p->ainsn.boostable != 1)) |
| 263 | p->ainsn.boostable = -1; |
| 264 | } |
| 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 | if (!IS_ENABLED(CONFIG_BOOKE) && |
| 281 | (!(regs->msr & MSR_IR) || !(regs->msr & MSR_DR))) |
| 282 | return 0; |
| 283 | |
| 284 | /* |
| 285 | * We don't want to be preempted for the entire |
| 286 | * duration of kprobe processing |
| 287 | */ |
| 288 | preempt_disable(); |
| 289 | kcb = get_kprobe_ctlblk(); |
| 290 | |
| 291 | p = get_kprobe(addr); |
| 292 | if (!p) { |
| 293 | unsigned int instr; |
| 294 | |
| 295 | if (get_kernel_nofault(instr, addr)) |
| 296 | goto no_kprobe; |
| 297 | |
| 298 | if (instr != BREAKPOINT_INSTRUCTION) { |
| 299 | /* |
| 300 | * PowerPC has multiple variants of the "trap" |
| 301 | * instruction. If the current instruction is a |
| 302 | * trap variant, it could belong to someone else |
| 303 | */ |
| 304 | if (is_trap(instr)) |
| 305 | goto no_kprobe; |
| 306 | /* |
| 307 | * The breakpoint instruction was removed right |
| 308 | * after we hit it. Another cpu has removed |
| 309 | * either a probepoint or a debugger breakpoint |
| 310 | * at this address. In either case, no further |
| 311 | * handling of this interrupt is appropriate. |
| 312 | */ |
| 313 | ret = 1; |
| 314 | } |
| 315 | /* Not one of ours: let kernel handle it */ |
| 316 | goto no_kprobe; |
| 317 | } |
| 318 | |
| 319 | /* Check we're not actually recursing */ |
| 320 | if (kprobe_running()) { |
| 321 | kprobe_opcode_t insn = *p->ainsn.insn; |
| 322 | if (kcb->kprobe_status == KPROBE_HIT_SS && is_trap(insn)) { |
| 323 | /* Turn off 'trace' bits */ |
| 324 | regs_set_return_msr(regs, |
| 325 | (regs->msr & ~MSR_SINGLESTEP) | |
| 326 | kcb->kprobe_saved_msr); |
| 327 | goto no_kprobe; |
| 328 | } |
| 329 | |
| 330 | /* |
| 331 | * We have reentered the kprobe_handler(), since another probe |
| 332 | * was hit while within the handler. We here save the original |
| 333 | * kprobes variables and just single step on the instruction of |
| 334 | * the new probe without calling any user handlers. |
| 335 | */ |
| 336 | save_previous_kprobe(kcb); |
| 337 | set_current_kprobe(p, regs, kcb); |
| 338 | kprobes_inc_nmissed_count(p); |
| 339 | kcb->kprobe_status = KPROBE_REENTER; |
| 340 | if (p->ainsn.boostable >= 0) { |
| 341 | ret = try_to_emulate(p, regs); |
| 342 | |
| 343 | if (ret > 0) { |
| 344 | restore_previous_kprobe(kcb); |
| 345 | preempt_enable(); |
| 346 | return 1; |
| 347 | } |
| 348 | } |
| 349 | prepare_singlestep(p, regs); |
| 350 | return 1; |
| 351 | } |
| 352 | |
| 353 | kcb->kprobe_status = KPROBE_HIT_ACTIVE; |
| 354 | set_current_kprobe(p, regs, kcb); |
| 355 | if (p->pre_handler && p->pre_handler(p, regs)) { |
| 356 | /* handler changed execution path, so skip ss setup */ |
| 357 | reset_current_kprobe(); |
| 358 | preempt_enable(); |
| 359 | return 1; |
| 360 | } |
| 361 | |
| 362 | if (p->ainsn.boostable >= 0) { |
| 363 | ret = try_to_emulate(p, regs); |
| 364 | |
| 365 | if (ret > 0) { |
| 366 | if (p->post_handler) |
| 367 | p->post_handler(p, regs, 0); |
| 368 | |
| 369 | kcb->kprobe_status = KPROBE_HIT_SSDONE; |
| 370 | reset_current_kprobe(); |
| 371 | preempt_enable(); |
| 372 | return 1; |
| 373 | } |
| 374 | } |
| 375 | prepare_singlestep(p, regs); |
| 376 | kcb->kprobe_status = KPROBE_HIT_SS; |
| 377 | return 1; |
| 378 | |
| 379 | no_kprobe: |
| 380 | preempt_enable(); |
| 381 | return ret; |
| 382 | } |
| 383 | NOKPROBE_SYMBOL(kprobe_handler); |
| 384 | |
| 385 | /* |
| 386 | * Called after single-stepping. p->addr is the address of the |
| 387 | * instruction whose first byte has been replaced by the "breakpoint" |
| 388 | * instruction. To avoid the SMP problems that can occur when we |
| 389 | * temporarily put back the original opcode to single-step, we |
| 390 | * single-stepped a copy of the instruction. The address of this |
| 391 | * copy is p->ainsn.insn. |
| 392 | */ |
| 393 | int kprobe_post_handler(struct pt_regs *regs) |
| 394 | { |
| 395 | int len; |
| 396 | struct kprobe *cur = kprobe_running(); |
| 397 | struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); |
| 398 | |
| 399 | if (!cur || user_mode(regs)) |
| 400 | return 0; |
| 401 | |
| 402 | len = ppc_inst_len(ppc_inst_read(cur->ainsn.insn)); |
| 403 | /* make sure we got here for instruction we have a kprobe on */ |
| 404 | if (((unsigned long)cur->ainsn.insn + len) != regs->nip) |
| 405 | return 0; |
| 406 | |
| 407 | if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) { |
| 408 | kcb->kprobe_status = KPROBE_HIT_SSDONE; |
| 409 | cur->post_handler(cur, regs, 0); |
| 410 | } |
| 411 | |
| 412 | /* Adjust nip to after the single-stepped instruction */ |
| 413 | regs_set_return_ip(regs, (unsigned long)cur->addr + len); |
| 414 | regs_set_return_msr(regs, regs->msr | kcb->kprobe_saved_msr); |
| 415 | |
| 416 | /*Restore back the original saved kprobes variables and continue. */ |
| 417 | if (kcb->kprobe_status == KPROBE_REENTER) { |
| 418 | restore_previous_kprobe(kcb); |
| 419 | goto out; |
| 420 | } |
| 421 | reset_current_kprobe(); |
| 422 | out: |
| 423 | preempt_enable(); |
| 424 | |
| 425 | /* |
| 426 | * if somebody else is singlestepping across a probe point, msr |
| 427 | * will have DE/SE set, in which case, continue the remaining processing |
| 428 | * of do_debug, as if this is not a probe hit. |
| 429 | */ |
| 430 | if (regs->msr & MSR_SINGLESTEP) |
| 431 | return 0; |
| 432 | |
| 433 | return 1; |
| 434 | } |
| 435 | NOKPROBE_SYMBOL(kprobe_post_handler); |
| 436 | |
| 437 | int kprobe_fault_handler(struct pt_regs *regs, int trapnr) |
| 438 | { |
| 439 | struct kprobe *cur = kprobe_running(); |
| 440 | struct kprobe_ctlblk *kcb = get_kprobe_ctlblk(); |
| 441 | const struct exception_table_entry *entry; |
| 442 | |
| 443 | switch(kcb->kprobe_status) { |
| 444 | case KPROBE_HIT_SS: |
| 445 | case KPROBE_REENTER: |
| 446 | /* |
| 447 | * We are here because the instruction being single |
| 448 | * stepped caused a page fault. We reset the current |
| 449 | * kprobe and the nip points back to the probe address |
| 450 | * and allow the page fault handler to continue as a |
| 451 | * normal page fault. |
| 452 | */ |
| 453 | regs_set_return_ip(regs, (unsigned long)cur->addr); |
| 454 | /* Turn off 'trace' bits */ |
| 455 | regs_set_return_msr(regs, |
| 456 | (regs->msr & ~MSR_SINGLESTEP) | |
| 457 | kcb->kprobe_saved_msr); |
| 458 | if (kcb->kprobe_status == KPROBE_REENTER) |
| 459 | restore_previous_kprobe(kcb); |
| 460 | else |
| 461 | reset_current_kprobe(); |
| 462 | preempt_enable(); |
| 463 | break; |
| 464 | case KPROBE_HIT_ACTIVE: |
| 465 | case KPROBE_HIT_SSDONE: |
| 466 | /* |
| 467 | * In case the user-specified fault handler returned |
| 468 | * zero, try to fix up. |
| 469 | */ |
| 470 | if ((entry = search_exception_tables(add: regs->nip)) != NULL) { |
| 471 | regs_set_return_ip(regs, extable_fixup(entry)); |
| 472 | return 1; |
| 473 | } |
| 474 | |
| 475 | /* |
| 476 | * fixup_exception() could not handle it, |
| 477 | * Let do_page_fault() fix it. |
| 478 | */ |
| 479 | break; |
| 480 | default: |
| 481 | break; |
| 482 | } |
| 483 | return 0; |
| 484 | } |
| 485 | NOKPROBE_SYMBOL(kprobe_fault_handler); |
| 486 | |
| 487 | int arch_trampoline_kprobe(struct kprobe *p) |
| 488 | { |
| 489 | if (p->addr == (kprobe_opcode_t *)&arch_rethook_trampoline) |
| 490 | return 1; |
| 491 | |
| 492 | return 0; |
| 493 | } |
| 494 | NOKPROBE_SYMBOL(arch_trampoline_kprobe); |
| 495 | |