1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * linux/init/main.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * GK 2/5/95 - Changed to support mounting root fs via NFS 8 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96 9 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96 10 * Simplified starting of init: Michael A. Griffith <grif@acm.org> 11 */ 12 13 #define DEBUG /* Enable initcall_debug */ 14 15 #include <linux/types.h> 16 #include <linux/export.h> 17 #include <linux/extable.h> 18 #include <linux/module.h> 19 #include <linux/proc_fs.h> 20 #include <linux/binfmts.h> 21 #include <linux/kernel.h> 22 #include <linux/syscalls.h> 23 #include <linux/stackprotector.h> 24 #include <linux/string.h> 25 #include <linux/ctype.h> 26 #include <linux/delay.h> 27 #include <linux/ioport.h> 28 #include <linux/init.h> 29 #include <linux/initrd.h> 30 #include <linux/memblock.h> 31 #include <linux/acpi.h> 32 #include <linux/bootconfig.h> 33 #include <linux/console.h> 34 #include <linux/nmi.h> 35 #include <linux/percpu.h> 36 #include <linux/kmod.h> 37 #include <linux/kprobes.h> 38 #include <linux/kmsan.h> 39 #include <linux/ksysfs.h> 40 #include <linux/vmalloc.h> 41 #include <linux/kernel_stat.h> 42 #include <linux/start_kernel.h> 43 #include <linux/security.h> 44 #include <linux/smp.h> 45 #include <linux/profile.h> 46 #include <linux/kfence.h> 47 #include <linux/rcupdate.h> 48 #include <linux/srcu.h> 49 #include <linux/moduleparam.h> 50 #include <linux/kallsyms.h> 51 #include <linux/buildid.h> 52 #include <linux/writeback.h> 53 #include <linux/cpu.h> 54 #include <linux/cpuset.h> 55 #include <linux/memcontrol.h> 56 #include <linux/cgroup.h> 57 #include <linux/tick.h> 58 #include <linux/sched/isolation.h> 59 #include <linux/interrupt.h> 60 #include <linux/taskstats_kern.h> 61 #include <linux/delayacct.h> 62 #include <linux/unistd.h> 63 #include <linux/utsname.h> 64 #include <linux/rmap.h> 65 #include <linux/mempolicy.h> 66 #include <linux/key.h> 67 #include <linux/debug_locks.h> 68 #include <linux/debugobjects.h> 69 #include <linux/lockdep.h> 70 #include <linux/kmemleak.h> 71 #include <linux/padata.h> 72 #include <linux/pid_namespace.h> 73 #include <linux/device/driver.h> 74 #include <linux/kthread.h> 75 #include <linux/sched.h> 76 #include <linux/sched/init.h> 77 #include <linux/signal.h> 78 #include <linux/idr.h> 79 #include <linux/kgdb.h> 80 #include <linux/ftrace.h> 81 #include <linux/async.h> 82 #include <linux/shmem_fs.h> 83 #include <linux/slab.h> 84 #include <linux/perf_event.h> 85 #include <linux/ptrace.h> 86 #include <linux/pti.h> 87 #include <linux/blkdev.h> 88 #include <linux/sched/clock.h> 89 #include <linux/sched/task.h> 90 #include <linux/sched/task_stack.h> 91 #include <linux/context_tracking.h> 92 #include <linux/random.h> 93 #include <linux/moduleloader.h> 94 #include <linux/list.h> 95 #include <linux/integrity.h> 96 #include <linux/proc_ns.h> 97 #include <linux/io.h> 98 #include <linux/cache.h> 99 #include <linux/rodata_test.h> 100 #include <linux/jump_label.h> 101 #include <linux/kcsan.h> 102 #include <linux/init_syscalls.h> 103 #include <linux/stackdepot.h> 104 #include <linux/randomize_kstack.h> 105 #include <linux/pidfs.h> 106 #include <linux/fs_struct.h> 107 #include <linux/ptdump.h> 108 #include <linux/time_namespace.h> 109 #include <linux/unaligned.h> 110 #include <linux/vdso_datastore.h> 111 #include <net/net_namespace.h> 112 113 #include <asm/io.h> 114 #include <asm/setup.h> 115 #include <asm/sections.h> 116 #include <asm/cacheflush.h> 117 118 #define CREATE_TRACE_POINTS 119 #include <trace/events/initcall.h> 120 121 #include <kunit/test.h> 122 123 static int kernel_init(void *); 124 125 /* 126 * Debug helper: via this flag we know that we are in 'early bootup code' 127 * where only the boot processor is running with IRQ disabled. This means 128 * two things - IRQ must not be enabled before the flag is cleared and some 129 * operations which are not allowed with IRQ disabled are allowed while the 130 * flag is set. 131 */ 132 bool early_boot_irqs_disabled __read_mostly; 133 134 enum system_states system_state __read_mostly; 135 EXPORT_SYMBOL(system_state); 136 137 /* 138 * Boot command-line arguments 139 */ 140 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT 141 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT 142 143 /* Default late time init is NULL. archs can override this later. */ 144 void (*__initdata late_time_init)(void); 145 146 /* Untouched command line saved by arch-specific code. */ 147 char __initdata boot_command_line[COMMAND_LINE_SIZE]; 148 /* Untouched saved command line (eg. for /proc) */ 149 char *saved_command_line __ro_after_init; 150 unsigned int saved_command_line_len __ro_after_init; 151 /* Command line for parameter parsing */ 152 static char *static_command_line; 153 /* Untouched extra command line */ 154 static char *extra_command_line; 155 /* Extra init arguments */ 156 static char *extra_init_args; 157 158 #ifdef CONFIG_BOOT_CONFIG 159 /* Is bootconfig on command line? */ 160 static bool bootconfig_found; 161 static size_t initargs_offs; 162 #else 163 # define bootconfig_found false 164 # define initargs_offs 0 165 #endif 166 167 static char *execute_command; 168 static char *ramdisk_execute_command = "/init"; 169 static bool __initdata ramdisk_execute_command_set; 170 171 /* 172 * Used to generate warnings if static_key manipulation functions are used 173 * before jump_label_init is called. 174 */ 175 bool static_key_initialized __read_mostly; 176 EXPORT_SYMBOL_GPL(static_key_initialized); 177 178 /* 179 * If set, this is an indication to the drivers that reset the underlying 180 * device before going ahead with the initialization otherwise driver might 181 * rely on the BIOS and skip the reset operation. 182 * 183 * This is useful if kernel is booting in an unreliable environment. 184 * For ex. kdump situation where previous kernel has crashed, BIOS has been 185 * skipped and devices will be in unknown state. 186 */ 187 unsigned int reset_devices; 188 EXPORT_SYMBOL(reset_devices); 189 190 static int __init set_reset_devices(char *str) 191 { 192 reset_devices = 1; 193 return 1; 194 } 195 196 __setup("reset_devices", set_reset_devices); 197 198 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, }; 199 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, }; 200 static const char *panic_later, *panic_param; 201 202 static bool __init obsolete_checksetup(char *line) 203 { 204 const struct obs_kernel_param *p; 205 bool had_early_param = false; 206 207 p = __setup_start; 208 do { 209 int n = strlen(p->str); 210 if (parameqn(line, p->str, n)) { 211 if (p->early) { 212 /* Already done in parse_early_param? 213 * (Needs exact match on param part). 214 * Keep iterating, as we can have early 215 * params and __setups of same names 8( */ 216 if (line[n] == '\0' || line[n] == '=') 217 had_early_param = true; 218 } else if (p->setup_func(line + n)) 219 return true; 220 } 221 p++; 222 } while (p < __setup_end); 223 224 return had_early_param; 225 } 226 227 /* 228 * This should be approx 2 Bo*oMips to start (note initial shift), and will 229 * still work even if initially too large, it will just take slightly longer 230 */ 231 unsigned long loops_per_jiffy = (1<<12); 232 EXPORT_SYMBOL(loops_per_jiffy); 233 234 static int __init debug_kernel(char *str) 235 { 236 console_loglevel = CONSOLE_LOGLEVEL_DEBUG; 237 return 0; 238 } 239 240 static int __init quiet_kernel(char *str) 241 { 242 console_loglevel = CONSOLE_LOGLEVEL_QUIET; 243 return 0; 244 } 245 246 early_param("debug", debug_kernel); 247 early_param("quiet", quiet_kernel); 248 249 static int __init loglevel(char *str) 250 { 251 int newlevel; 252 253 /* 254 * Only update loglevel value when a correct setting was passed, 255 * to prevent blind crashes (when loglevel being set to 0) that 256 * are quite hard to debug 257 */ 258 if (get_option(&str, &newlevel)) { 259 console_loglevel = newlevel; 260 return 0; 261 } 262 263 return -EINVAL; 264 } 265 266 early_param("loglevel", loglevel); 267 268 #ifdef CONFIG_BLK_DEV_INITRD 269 static void * __init get_boot_config_from_initrd(size_t *_size) 270 { 271 u32 size, csum; 272 char *data; 273 u8 *hdr; 274 int i; 275 276 if (!initrd_end || initrd_end < initrd_start || 277 initrd_end - initrd_start < BOOTCONFIG_FOOTER_SIZE) 278 return NULL; 279 280 data = (char *)initrd_end - BOOTCONFIG_MAGIC_LEN; 281 /* 282 * Since Grub may align the size of initrd to 4, we must 283 * check the preceding 3 bytes as well. 284 */ 285 for (i = 0; i < 4; i++) { 286 if (!memcmp(data, BOOTCONFIG_MAGIC, BOOTCONFIG_MAGIC_LEN)) 287 goto found; 288 data--; 289 } 290 return NULL; 291 292 found: 293 hdr = (u8 *)(data - 8); 294 if ((unsigned long)hdr < initrd_start) 295 return NULL; 296 297 size = get_unaligned_le32(hdr); 298 csum = get_unaligned_le32(hdr + 4); 299 300 if (size > XBC_DATA_MAX) { 301 pr_err("bootconfig size %u is greater than max size %d\n", 302 size, XBC_DATA_MAX); 303 return NULL; 304 } 305 306 if (size > ((unsigned long)hdr - initrd_start)) { 307 pr_err("bootconfig size %u is greater than initrd size %lu\n", 308 size, initrd_end - initrd_start); 309 return NULL; 310 } 311 312 data = ((void *)hdr) - size; 313 314 if (xbc_calc_checksum(data, size) != csum) { 315 pr_err("bootconfig checksum failed\n"); 316 return NULL; 317 } 318 319 /* Remove bootconfig from initramfs/initrd */ 320 initrd_end = (unsigned long)data; 321 if (_size) 322 *_size = size; 323 324 return data; 325 } 326 #else 327 static void * __init get_boot_config_from_initrd(size_t *_size) 328 { 329 return NULL; 330 } 331 #endif 332 333 #ifdef CONFIG_BOOT_CONFIG 334 335 /* Check if a parameter name matches an obs_kernel_param entry */ 336 static bool __init match_obs_param(const char *param, const char *pattern) 337 { 338 size_t n = strlen(pattern); 339 340 if (n > 0 && pattern[n - 1] == '=') 341 return parameqn(param, pattern, n - 1) && param[n - 1] == '\0'; 342 343 return parameq(param, pattern); 344 } 345 346 /* Return true if the given param is only defined by early_param(). */ 347 static bool __init is_early_only_param(const char *param) 348 { 349 const struct obs_kernel_param *p; 350 const struct kernel_param *kp; 351 bool has_early = false; 352 353 /* If handled by a regular module_param, it is not early-only */ 354 for (kp = __start___param; kp < __stop___param; kp++) { 355 if (parameq(param, kp->name)) 356 return false; 357 } 358 359 /* Check __setup / early_param table */ 360 for (p = __setup_start; p < __setup_end; p++) { 361 if (match_obs_param(param, p->str)) { 362 if (!p->early) 363 return false; 364 has_early = true; 365 } 366 } 367 return has_early; 368 } 369 370 struct xbc_filter_data { 371 bool warn; 372 }; 373 374 static bool __init xbc_kernel_param_filter(struct xbc_node *node, 375 const char *key, void *data) 376 { 377 struct xbc_filter_data *fdata = data; 378 379 if (is_early_only_param(key)) { 380 if (fdata && fdata->warn) 381 pr_warn("bootconfig: early_param 'kernel.%s' cannot be applied from bootconfig, skipping\n", 382 key); 383 return false; 384 } 385 return true; 386 } 387 388 /* Make an extra command line under given key word */ 389 static char * __init xbc_make_cmdline(const char *key, bool is_kernel) 390 { 391 struct xbc_filter_data fdata = { .warn = true }; 392 xbc_cmdline_filter_fn filter = is_kernel ? xbc_kernel_param_filter : NULL; 393 struct xbc_node *root; 394 char *new_cmdline; 395 int ret, len = 0; 396 397 root = xbc_find_node(key); 398 if (!root) 399 return NULL; 400 401 /* 402 * Pass 1: Count required buffer size. Emit warnings for skipped early 403 * params in this pass so they are logged even if all parameters under 404 * @key are skipped and len becomes 0. 405 */ 406 len = xbc_snprint_cmdline_filter(NULL, 0, root, filter, &fdata); 407 if (len <= 0) 408 return NULL; 409 410 new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES); 411 if (!new_cmdline) { 412 pr_err("Failed to allocate memory for extra kernel cmdline.\n"); 413 return NULL; 414 } 415 416 /* Pass 2: Render into buffer with warnings suppressed */ 417 fdata.warn = false; 418 ret = xbc_snprint_cmdline_filter(new_cmdline, len + 1, root, filter, &fdata); 419 if (ret < 0 || ret > len) { 420 pr_err("Failed to print extra kernel cmdline.\n"); 421 memblock_free(new_cmdline, len + 1); 422 return NULL; 423 } 424 425 return new_cmdline; 426 } 427 428 static int __init warn_bootconfig(char *str) 429 { 430 /* The 'bootconfig' option is handled by setup_boot_config(). */ 431 return 0; 432 } 433 434 static void __init setup_boot_config(void) 435 { 436 const char *msg, *data; 437 int pos, ret, offs; 438 size_t size; 439 bool from_embedded = false; 440 441 /* Cut out the bootconfig data even if we have no bootconfig option */ 442 data = get_boot_config_from_initrd(&size); 443 /* If there is no bootconfig in initrd, try embedded one. */ 444 if (!data) { 445 data = xbc_get_embedded_bootconfig(&size); 446 from_embedded = true; 447 } 448 449 bootconfig_found = bootconfig_cmdline_requested(boot_command_line, &offs); 450 if (!(bootconfig_found || IS_ENABLED(CONFIG_BOOT_CONFIG_FORCE))) 451 return; 452 453 /* Offset of the init arguments after a "--", located by the helper. */ 454 initargs_offs = offs; 455 456 if (!data) { 457 /* If user intended to use bootconfig, show an error level message */ 458 if (bootconfig_found) 459 pr_err("'bootconfig' found on command line, but no bootconfig found\n"); 460 else 461 pr_info("No bootconfig data provided, so skipping bootconfig"); 462 return; 463 } 464 465 ret = xbc_init(data, size, &msg, &pos); 466 if (ret < 0) { 467 if (pos < 0) 468 pr_err("Failed to init bootconfig: %s.\n", msg); 469 else 470 pr_err("Failed to parse bootconfig: %s at %d.\n", 471 msg, pos); 472 } else { 473 xbc_get_info(&ret, NULL); 474 pr_info("Load bootconfig: %ld bytes %d nodes\n", (long)size, ret); 475 /* 476 * keys starting with "kernel." are passed via cmdline. When 477 * this bootconfig came from the embedded source and 478 * setup_arch() already prepended the rendered "kernel" subtree 479 * to boot_command_line, rendering again here would duplicate 480 * the keys in saved_command_line and make accumulating handlers 481 * (console=, earlycon=, ...) re-register the same value. Skip 482 * only when the prepend really happened. 483 * 484 * On arches that do not select ARCH_SUPPORTS_CMDLINE_FROM_BOOTCONFIG, 485 * CONFIG_CMDLINE_FROM_BOOTCONFIG is unselectable and 486 * xbc_embedded_cmdline_applied() collapses to a stub returning 487 * false, so this path still runs and the embedded "kernel" 488 * keys reach the cmdline via the runtime parser exactly as 489 * before this series. 490 */ 491 if (!from_embedded || !xbc_embedded_cmdline_applied()) 492 extra_command_line = xbc_make_cmdline("kernel", true); 493 /* Also, "init." keys are init arguments */ 494 extra_init_args = xbc_make_cmdline("init", false); 495 } 496 return; 497 } 498 499 static void __init exit_boot_config(void) 500 { 501 xbc_exit(); 502 } 503 504 #else /* !CONFIG_BOOT_CONFIG */ 505 506 static void __init setup_boot_config(void) 507 { 508 /* Remove bootconfig data from initrd */ 509 get_boot_config_from_initrd(NULL); 510 } 511 512 static int __init warn_bootconfig(char *str) 513 { 514 pr_warn("WARNING: 'bootconfig' found on the kernel command line but CONFIG_BOOT_CONFIG is not set.\n"); 515 return 0; 516 } 517 518 #define exit_boot_config() do {} while (0) 519 520 #endif /* CONFIG_BOOT_CONFIG */ 521 522 early_param("bootconfig", warn_bootconfig); 523 524 bool __init cmdline_has_extra_options(void) 525 { 526 return extra_command_line || extra_init_args; 527 } 528 529 /* Change NUL term back to "=", to make "param" the whole string. */ 530 static void __init repair_env_string(char *param, char *val) 531 { 532 if (val) { 533 /* param=val or param="val"? */ 534 if (val == param+strlen(param)+1) 535 val[-1] = '='; 536 else if (val == param+strlen(param)+2) { 537 val[-2] = '='; 538 memmove(val-1, val, strlen(val)+1); 539 } else 540 BUG(); 541 } 542 } 543 544 /* Anything after -- gets handed straight to init. */ 545 static int __init set_init_arg(char *param, char *val, 546 const char *unused, void *arg) 547 { 548 unsigned int i; 549 550 if (panic_later) 551 return 0; 552 553 repair_env_string(param, val); 554 555 for (i = 0; argv_init[i]; i++) { 556 if (i == MAX_INIT_ARGS) { 557 panic_later = "init"; 558 panic_param = param; 559 return 0; 560 } 561 } 562 argv_init[i] = param; 563 return 0; 564 } 565 566 /* 567 * Unknown boot options get handed to init, unless they look like 568 * unused parameters (modprobe will find them in /proc/cmdline). 569 */ 570 static int __init unknown_bootoption(char *param, char *val, 571 const char *unused, void *arg) 572 { 573 size_t len = strlen(param); 574 /* 575 * Well-known bootloader identifiers: 576 * 1. LILO/Grub pass "BOOT_IMAGE=..."; 577 * 2. kexec/kdump (kexec-tools) pass "kexec". 578 */ 579 const char *bootloader[] = { "BOOT_IMAGE=", "kexec", NULL }; 580 581 /* Handle params aliased to sysctls */ 582 if (sysctl_is_alias(param)) 583 return 0; 584 585 repair_env_string(param, val); 586 587 /* Handle bootloader identifier */ 588 for (int i = 0; bootloader[i]; i++) { 589 if (strstarts(param, bootloader[i])) 590 return 0; 591 } 592 593 /* Handle obsolete-style parameters */ 594 if (obsolete_checksetup(param)) 595 return 0; 596 597 /* Unused module parameter. */ 598 if (strnchr(param, len, '.')) 599 return 0; 600 601 if (panic_later) 602 return 0; 603 604 if (val) { 605 /* Environment option */ 606 unsigned int i; 607 for (i = 0; envp_init[i]; i++) { 608 if (!strncmp(param, envp_init[i], len+1)) 609 break; 610 if (i == MAX_INIT_ENVS) { 611 panic_later = "env"; 612 panic_param = param; 613 } 614 } 615 envp_init[i] = param; 616 } else { 617 /* Command line option */ 618 unsigned int i; 619 for (i = 0; argv_init[i]; i++) { 620 if (i == MAX_INIT_ARGS) { 621 panic_later = "init"; 622 panic_param = param; 623 } 624 } 625 argv_init[i] = param; 626 } 627 return 0; 628 } 629 630 static int __init init_setup(char *str) 631 { 632 unsigned int i; 633 634 execute_command = str; 635 /* 636 * In case LILO is going to boot us with default command line, 637 * it prepends "auto" before the whole cmdline which makes 638 * the shell think it should execute a script with such name. 639 * So we ignore all arguments entered _before_ init=... [MJ] 640 */ 641 for (i = 1; i <= MAX_INIT_ARGS; i++) 642 argv_init[i] = NULL; 643 return 1; 644 } 645 __setup("init=", init_setup); 646 647 static int __init rdinit_setup(char *str) 648 { 649 unsigned int i; 650 651 ramdisk_execute_command = str; 652 ramdisk_execute_command_set = true; 653 /* See "auto" comment in init_setup */ 654 for (i = 1; i <= MAX_INIT_ARGS; i++) 655 argv_init[i] = NULL; 656 return 1; 657 } 658 __setup("rdinit=", rdinit_setup); 659 660 #ifndef CONFIG_SMP 661 static inline void setup_nr_cpu_ids(void) { } 662 static inline void smp_prepare_cpus(unsigned int maxcpus) { } 663 #endif 664 665 /* 666 * We need to store the untouched command line for future reference. 667 * We also need to store the touched command line since the parameter 668 * parsing is performed in place, and we should allow a component to 669 * store reference of name/value for future reference. 670 */ 671 static void __init setup_command_line(char *command_line) 672 { 673 size_t len, xlen = 0, ilen = 0; 674 675 if (extra_command_line) 676 xlen = strlen(extra_command_line); 677 if (extra_init_args) { 678 extra_init_args = strim(extra_init_args); /* remove trailing space */ 679 ilen = strlen(extra_init_args) + 4; /* for " -- " */ 680 } 681 682 len = xlen + strlen(boot_command_line) + ilen + 1; 683 684 saved_command_line = memblock_alloc_or_panic(len, SMP_CACHE_BYTES); 685 686 len = xlen + strlen(command_line) + 1; 687 688 static_command_line = memblock_alloc_or_panic(len, SMP_CACHE_BYTES); 689 690 if (xlen) { 691 /* 692 * We have to put extra_command_line before boot command 693 * lines because there could be dashes (separator of init 694 * command line) in the command lines. 695 */ 696 strcpy(saved_command_line, extra_command_line); 697 strcpy(static_command_line, extra_command_line); 698 } 699 strcpy(saved_command_line + xlen, boot_command_line); 700 strcpy(static_command_line + xlen, command_line); 701 702 if (ilen) { 703 /* 704 * Append supplemental init boot args to saved_command_line 705 * so that user can check what command line options passed 706 * to init. 707 * The order should always be 708 * " -- "[bootconfig init-param][cmdline init-param] 709 */ 710 if (initargs_offs) { 711 len = xlen + initargs_offs; 712 strcpy(saved_command_line + len, extra_init_args); 713 len += ilen - 4; /* strlen(extra_init_args) */ 714 strcpy(saved_command_line + len, 715 boot_command_line + initargs_offs - 1); 716 } else { 717 len = strlen(saved_command_line); 718 strcpy(saved_command_line + len, " -- "); 719 len += 4; 720 strcpy(saved_command_line + len, extra_init_args); 721 } 722 } 723 724 saved_command_line_len = strlen(saved_command_line); 725 } 726 727 /* 728 * We need to finalize in a non-__init function or else race conditions 729 * between the root thread and the init thread may cause start_kernel to 730 * be reaped by free_initmem before the root thread has proceeded to 731 * cpu_idle. 732 * 733 * gcc-3.4 accidentally inlines this function, so use noinline. 734 */ 735 736 static __initdata DECLARE_COMPLETION(kthreadd_done); 737 738 static noinline void __ref __noreturn rest_init(void) 739 { 740 struct kernel_clone_args init_args = { 741 .flags = (CLONE_VM | CLONE_UNTRACED), 742 .fn = kernel_init, 743 .fn_arg = NULL, 744 }; 745 struct task_struct *tsk; 746 int pid; 747 748 rcu_scheduler_starting(); 749 /* 750 * We need to spawn init first so that it obtains pid 1, however 751 * the init task will end up wanting to create kthreads, which, if 752 * we schedule it before we create kthreadd, will OOPS. 753 */ 754 pid = kernel_clone(&init_args); 755 /* 756 * Pin init on the boot CPU. Task migration is not properly working 757 * until sched_init_smp() has been run. It will set the allowed 758 * CPUs for init to the non isolated CPUs. 759 */ 760 rcu_read_lock(); 761 tsk = find_task_by_pid_ns(pid, &init_pid_ns); 762 tsk->flags |= PF_NO_SETAFFINITY; 763 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id())); 764 rcu_read_unlock(); 765 766 numa_default_policy(); 767 pid = kernel_thread(kthreadd, NULL, NULL, CLONE_FS | CLONE_FILES); 768 rcu_read_lock(); 769 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns); 770 rcu_read_unlock(); 771 772 /* 773 * Enable might_sleep() and smp_processor_id() checks. 774 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y 775 * kernel_thread() would trigger might_sleep() splats. With 776 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled 777 * already, but it's stuck on the kthreadd_done completion. 778 */ 779 system_state = SYSTEM_SCHEDULING; 780 781 complete(&kthreadd_done); 782 783 /* 784 * The boot idle thread must execute schedule() 785 * at least once to get things moving: 786 */ 787 schedule_preempt_disabled(); 788 /* Call into cpu_idle with preempt disabled */ 789 cpu_startup_entry(CPUHP_ONLINE); 790 } 791 792 /* Check for early params. */ 793 static int __init do_early_param(char *param, char *val, 794 const char *unused, void *arg) 795 { 796 const struct obs_kernel_param *p; 797 798 for (p = __setup_start; p < __setup_end; p++) { 799 if (p->early && parameq(param, p->str)) { 800 if (p->setup_func(val) != 0) 801 pr_warn("Malformed early option '%s'\n", param); 802 } 803 } 804 /* We accept everything at this stage. */ 805 return 0; 806 } 807 808 void __init parse_early_options(char *cmdline) 809 { 810 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL, 811 do_early_param); 812 } 813 814 /* Arch code calls this early on, or if not, just before other parsing. */ 815 void __init parse_early_param(void) 816 { 817 static int done __initdata; 818 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata; 819 820 if (done) 821 return; 822 823 /* All fall through to do_early_param. */ 824 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE); 825 parse_early_options(tmp_cmdline); 826 done = 1; 827 } 828 829 void __init __weak arch_post_acpi_subsys_init(void) { } 830 831 void __init __weak smp_setup_processor_id(void) 832 { 833 } 834 835 void __init __weak smp_prepare_boot_cpu(void) 836 { 837 } 838 839 # if THREAD_SIZE >= PAGE_SIZE 840 void __init __weak thread_stack_cache_init(void) 841 { 842 } 843 #endif 844 845 void __init __weak poking_init(void) { } 846 847 void __init __weak pgtable_cache_init(void) { } 848 849 void __init __weak trap_init(void) { } 850 851 bool initcall_debug; 852 core_param(initcall_debug, initcall_debug, bool, 0644); 853 854 #ifdef TRACEPOINTS_ENABLED 855 static void __init initcall_debug_enable(void); 856 #else 857 static inline void initcall_debug_enable(void) 858 { 859 } 860 #endif 861 862 #ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET 863 DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT, 864 randomize_kstack_offset); 865 DEFINE_PER_CPU(struct rnd_state, kstack_rnd_state); 866 867 static int __init random_kstack_init(void) 868 { 869 prandom_seed_full_state(&kstack_rnd_state); 870 return 0; 871 } 872 late_initcall(random_kstack_init); 873 874 static int __init early_randomize_kstack_offset(char *buf) 875 { 876 int ret; 877 bool bool_result; 878 879 ret = kstrtobool(buf, &bool_result); 880 if (ret) 881 return ret; 882 883 if (bool_result) 884 static_branch_enable(&randomize_kstack_offset); 885 else 886 static_branch_disable(&randomize_kstack_offset); 887 return 0; 888 } 889 early_param("randomize_kstack_offset", early_randomize_kstack_offset); 890 #endif 891 892 static void __init print_unknown_bootoptions(void) 893 { 894 char *unknown_options; 895 char *end; 896 const char *const *p; 897 size_t len; 898 899 if (panic_later || (!argv_init[1] && !envp_init[2])) 900 return; 901 902 /* 903 * Determine how many options we have to print out, plus a space 904 * before each 905 */ 906 len = 1; /* null terminator */ 907 for (p = &argv_init[1]; *p; p++) { 908 len++; 909 len += strlen(*p); 910 } 911 for (p = &envp_init[2]; *p; p++) { 912 len++; 913 len += strlen(*p); 914 } 915 916 unknown_options = memblock_alloc(len, SMP_CACHE_BYTES); 917 if (!unknown_options) { 918 pr_err("%s: Failed to allocate %zu bytes\n", 919 __func__, len); 920 return; 921 } 922 end = unknown_options; 923 924 for (p = &argv_init[1]; *p; p++) 925 end += sprintf(end, " %s", *p); 926 for (p = &envp_init[2]; *p; p++) 927 end += sprintf(end, " %s", *p); 928 929 /* Start at unknown_options[1] to skip the initial space */ 930 pr_notice("Unknown kernel command line parameters \"%s\", will be passed to user space.\n", 931 &unknown_options[1]); 932 memblock_free(unknown_options, len); 933 } 934 935 static void __init early_numa_node_init(void) 936 { 937 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID 938 #ifndef cpu_to_node 939 int cpu; 940 941 /* The early_cpu_to_node() should be ready here. */ 942 for_each_possible_cpu(cpu) 943 set_cpu_numa_node(cpu, early_cpu_to_node(cpu)); 944 #endif 945 #endif 946 } 947 948 #define KERNEL_CMDLINE_PREFIX "Kernel command line: " 949 #define KERNEL_CMDLINE_PREFIX_LEN (sizeof(KERNEL_CMDLINE_PREFIX) - 1) 950 #define KERNEL_CMDLINE_CONTINUATION " \\" 951 #define KERNEL_CMDLINE_CONTINUATION_LEN (sizeof(KERNEL_CMDLINE_CONTINUATION) - 1) 952 953 #define MIN_CMDLINE_LOG_WRAP_IDEAL_LEN (KERNEL_CMDLINE_PREFIX_LEN + \ 954 KERNEL_CMDLINE_CONTINUATION_LEN) 955 #define CMDLINE_LOG_WRAP_IDEAL_LEN (CONFIG_CMDLINE_LOG_WRAP_IDEAL_LEN > \ 956 MIN_CMDLINE_LOG_WRAP_IDEAL_LEN ? \ 957 CONFIG_CMDLINE_LOG_WRAP_IDEAL_LEN : \ 958 MIN_CMDLINE_LOG_WRAP_IDEAL_LEN) 959 960 #define IDEAL_CMDLINE_LEN (CMDLINE_LOG_WRAP_IDEAL_LEN - KERNEL_CMDLINE_PREFIX_LEN) 961 #define IDEAL_CMDLINE_SPLIT_LEN (IDEAL_CMDLINE_LEN - KERNEL_CMDLINE_CONTINUATION_LEN) 962 963 /** 964 * print_kernel_cmdline() - Print the kernel cmdline with wrapping. 965 * @cmdline: The cmdline to print. 966 * 967 * Print the kernel command line, trying to wrap based on the Kconfig knob 968 * CONFIG_CMDLINE_LOG_WRAP_IDEAL_LEN. 969 * 970 * Wrapping is based on spaces, ignoring quotes. All lines are prefixed 971 * with "Kernel command line: " and lines that are not the last line have 972 * a " \" suffix added to them. The prefix and suffix count towards the 973 * line length for wrapping purposes. The ideal length will be exceeded 974 * if no appropriate place to wrap is found. 975 * 976 * Example output if CONFIG_CMDLINE_LOG_WRAP_IDEAL_LEN is 40: 977 * Kernel command line: loglevel=7 \ 978 * Kernel command line: init=/sbin/init \ 979 * Kernel command line: root=PARTUUID=8c3efc1a-768b-6642-8d0c-89eb782f19f0/PARTNROFF=1 \ 980 * Kernel command line: rootwait ro \ 981 * Kernel command line: my_quoted_arg="The \ 982 * Kernel command line: quick brown fox \ 983 * Kernel command line: jumps over the \ 984 * Kernel command line: lazy dog." 985 */ 986 static void __init print_kernel_cmdline(const char *cmdline) 987 { 988 size_t len; 989 990 /* Config option of 0 or anything longer than the max disables wrapping */ 991 if (CONFIG_CMDLINE_LOG_WRAP_IDEAL_LEN == 0 || 992 IDEAL_CMDLINE_LEN >= COMMAND_LINE_SIZE - 1) { 993 pr_notice("%s%s\n", KERNEL_CMDLINE_PREFIX, cmdline); 994 return; 995 } 996 997 len = strlen(cmdline); 998 while (len > IDEAL_CMDLINE_LEN) { 999 const char *first_space; 1000 const char *prev_cutoff; 1001 const char *cutoff; 1002 int to_print; 1003 size_t used; 1004 1005 /* Find the last ' ' that wouldn't make the line too long */ 1006 prev_cutoff = NULL; 1007 cutoff = cmdline; 1008 while (true) { 1009 cutoff = strchr(cutoff + 1, ' '); 1010 if (!cutoff || cutoff - cmdline > IDEAL_CMDLINE_SPLIT_LEN) 1011 break; 1012 prev_cutoff = cutoff; 1013 } 1014 if (prev_cutoff) 1015 cutoff = prev_cutoff; 1016 else if (!cutoff) 1017 break; 1018 1019 /* Find the beginning and end of the string of spaces */ 1020 first_space = cutoff; 1021 while (first_space > cmdline && first_space[-1] == ' ') 1022 first_space--; 1023 to_print = first_space - cmdline; 1024 while (*cutoff == ' ') 1025 cutoff++; 1026 used = cutoff - cmdline; 1027 1028 /* If the whole string is used, break and do the final printout */ 1029 if (len == used) 1030 break; 1031 1032 if (to_print) 1033 pr_notice("%s%.*s%s\n", KERNEL_CMDLINE_PREFIX, 1034 to_print, cmdline, KERNEL_CMDLINE_CONTINUATION); 1035 1036 len -= used; 1037 cmdline += used; 1038 } 1039 if (len) 1040 pr_notice("%s%s\n", KERNEL_CMDLINE_PREFIX, cmdline); 1041 } 1042 1043 asmlinkage __visible __init __no_sanitize_address __noreturn __no_stack_protector 1044 void start_kernel(void) 1045 { 1046 char *command_line; 1047 char *after_dashes; 1048 1049 set_task_stack_end_magic(&init_task); 1050 smp_setup_processor_id(); 1051 debug_objects_early_init(); 1052 init_vmlinux_build_id(); 1053 1054 cgroup_init_early(); 1055 1056 local_irq_disable(); 1057 early_boot_irqs_disabled = true; 1058 1059 /* 1060 * Interrupts are still disabled. Do necessary setups, then 1061 * enable them. 1062 */ 1063 boot_cpu_init(); 1064 page_address_init(); 1065 pr_notice("%s", linux_banner); 1066 setup_arch(&command_line); 1067 mm_core_init_early(); 1068 /* Static keys and static calls are needed by LSMs */ 1069 jump_label_init(); 1070 static_call_init(); 1071 early_security_init(); 1072 setup_boot_config(); 1073 setup_command_line(command_line); 1074 setup_nr_cpu_ids(); 1075 setup_per_cpu_areas(); 1076 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */ 1077 early_numa_node_init(); 1078 boot_cpu_hotplug_init(); 1079 1080 print_kernel_cmdline(saved_command_line); 1081 /* parameters may set static keys */ 1082 parse_early_param(); 1083 after_dashes = parse_args("Booting kernel", 1084 static_command_line, __start___param, 1085 __stop___param - __start___param, 1086 -1, -1, NULL, &unknown_bootoption); 1087 print_unknown_bootoptions(); 1088 if (!IS_ERR_OR_NULL(after_dashes)) 1089 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1, 1090 NULL, set_init_arg); 1091 if (extra_init_args) 1092 parse_args("Setting extra init args", extra_init_args, 1093 NULL, 0, -1, -1, NULL, set_init_arg); 1094 1095 /* Architectural and non-timekeeping rng init, before allocator init */ 1096 random_init_early(command_line); 1097 1098 /* 1099 * These use large bootmem allocations and must precede 1100 * initalization of page allocator 1101 */ 1102 setup_log_buf(0); 1103 vfs_caches_init_early(); 1104 sort_main_extable(); 1105 trap_init(); 1106 mm_core_init(); 1107 maple_tree_init(); 1108 poking_init(); 1109 ftrace_init(); 1110 1111 /* trace_printk can be enabled here */ 1112 early_trace_init(); 1113 1114 /* 1115 * Set up the scheduler prior starting any interrupts (such as the 1116 * timer interrupt). Full topology setup happens at smp_init() 1117 * time - but meanwhile we still have a functioning scheduler. 1118 */ 1119 sched_init(); 1120 1121 if (WARN(!irqs_disabled(), 1122 "Interrupts were enabled *very* early, fixing it\n")) 1123 local_irq_disable(); 1124 radix_tree_init(); 1125 1126 /* 1127 * Set up housekeeping before setting up workqueues to allow the unbound 1128 * workqueue to take non-housekeeping into account. 1129 */ 1130 housekeeping_init(); 1131 1132 /* 1133 * Allow workqueue creation and work item queueing/cancelling 1134 * early. Work item execution depends on kthreads and starts after 1135 * workqueue_init(). 1136 */ 1137 workqueue_init_early(); 1138 1139 rcu_init(); 1140 kvfree_rcu_init(); 1141 1142 /* Trace events are available after this */ 1143 trace_init(); 1144 1145 if (initcall_debug) 1146 initcall_debug_enable(); 1147 1148 context_tracking_init(); 1149 /* init some links before init_ISA_irqs() */ 1150 early_irq_init(); 1151 init_IRQ(); 1152 tick_init(); 1153 rcu_init_nohz(); 1154 timers_init(); 1155 srcu_init(); 1156 hrtimers_init(); 1157 softirq_init(); 1158 vdso_setup_data_pages(); 1159 timekeeping_init(); 1160 time_init(); 1161 1162 /* This must be after timekeeping is initialized */ 1163 random_init(); 1164 1165 /* These make use of the fully initialized rng */ 1166 kfence_init(); 1167 boot_init_stack_canary(); 1168 1169 perf_event_init(); 1170 profile_init(); 1171 call_function_init(); 1172 WARN(!irqs_disabled(), "Interrupts were enabled early\n"); 1173 1174 early_boot_irqs_disabled = false; 1175 local_irq_enable(); 1176 1177 kmem_cache_init_late(); 1178 1179 /* 1180 * HACK ALERT! This is early. We're enabling the console before 1181 * we've done PCI setups etc, and console_init() must be aware of 1182 * this. But we do want output early, in case something goes wrong. 1183 */ 1184 console_init(); 1185 if (panic_later) 1186 panic("Too many boot %s vars at `%s'", panic_later, 1187 panic_param); 1188 1189 lockdep_init(); 1190 1191 /* 1192 * Need to run this when irqs are enabled, because it wants 1193 * to self-test [hard/soft]-irqs on/off lock inversion bugs 1194 * too: 1195 */ 1196 locking_selftest(); 1197 1198 #ifdef CONFIG_BLK_DEV_INITRD 1199 if (initrd_start && !initrd_below_start_ok && 1200 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) { 1201 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n", 1202 page_to_pfn(virt_to_page((void *)initrd_start)), 1203 min_low_pfn); 1204 initrd_start = 0; 1205 } 1206 #endif 1207 setup_per_cpu_pageset(); 1208 numa_policy_init(); 1209 acpi_early_init(); 1210 if (late_time_init) 1211 late_time_init(); 1212 sched_clock_init(); 1213 calibrate_delay(); 1214 1215 arch_cpu_finalize_init(); 1216 1217 pid_idr_init(); 1218 anon_vma_init(); 1219 thread_stack_cache_init(); 1220 cred_init(); 1221 fork_init(); 1222 proc_caches_init(); 1223 uts_ns_init(); 1224 time_ns_init(); 1225 key_init(); 1226 security_init(); 1227 dbg_late_init(); 1228 net_ns_init(); 1229 vfs_caches_init(); 1230 pagecache_init(); 1231 signals_init(); 1232 seq_file_init(); 1233 proc_root_init(); 1234 nsfs_init(); 1235 pidfs_init(); 1236 cpuset_init(); 1237 mem_cgroup_init(); 1238 cgroup_init(); 1239 taskstats_init_early(); 1240 delayacct_init(); 1241 1242 acpi_subsystem_init(); 1243 arch_post_acpi_subsys_init(); 1244 kcsan_init(); 1245 1246 /* Do the rest non-__init'ed, we're now alive */ 1247 rest_init(); 1248 1249 /* 1250 * Avoid stack canaries in callers of boot_init_stack_canary for gcc-10 1251 * and older. 1252 */ 1253 #if !__has_attribute(__no_stack_protector__) 1254 prevent_tail_call_optimization(); 1255 #endif 1256 } 1257 1258 /* Call all constructor functions linked into the kernel. */ 1259 static void __init do_ctors(void) 1260 { 1261 /* 1262 * For UML, the constructors have already been called by the 1263 * normal setup code as it's just a normal ELF binary, so we 1264 * cannot do it again - but we do need CONFIG_CONSTRUCTORS 1265 * even on UML for modules. 1266 */ 1267 #if defined(CONFIG_CONSTRUCTORS) && !defined(CONFIG_UML) 1268 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start; 1269 1270 for (; fn < (ctor_fn_t *) __ctors_end; fn++) 1271 (*fn)(); 1272 #endif 1273 } 1274 1275 #ifdef CONFIG_KALLSYMS 1276 struct blacklist_entry { 1277 struct list_head next; 1278 char *buf; 1279 }; 1280 1281 static __initdata_or_module LIST_HEAD(blacklisted_initcalls); 1282 1283 static int __init initcall_blacklist(char *str) 1284 { 1285 char *str_entry; 1286 struct blacklist_entry *entry; 1287 1288 /* str argument is a comma-separated list of functions */ 1289 do { 1290 str_entry = strsep(&str, ","); 1291 if (str_entry) { 1292 pr_debug("blacklisting initcall %s\n", str_entry); 1293 entry = memblock_alloc_or_panic(sizeof(*entry), 1294 SMP_CACHE_BYTES); 1295 entry->buf = memblock_alloc_or_panic(strlen(str_entry) + 1, 1296 SMP_CACHE_BYTES); 1297 strcpy(entry->buf, str_entry); 1298 list_add(&entry->next, &blacklisted_initcalls); 1299 } 1300 } while (str_entry); 1301 1302 return 1; 1303 } 1304 1305 static bool __init_or_module initcall_blacklisted(initcall_t fn) 1306 { 1307 struct blacklist_entry *entry; 1308 char fn_name[KSYM_SYMBOL_LEN]; 1309 unsigned long addr; 1310 1311 if (list_empty(&blacklisted_initcalls)) 1312 return false; 1313 1314 addr = (unsigned long) dereference_function_descriptor(fn); 1315 sprint_symbol_no_offset(fn_name, addr); 1316 1317 /* 1318 * fn will be "function_name [module_name]" where [module_name] is not 1319 * displayed for built-in init functions. Strip off the [module_name]. 1320 */ 1321 strreplace(fn_name, ' ', '\0'); 1322 1323 list_for_each_entry(entry, &blacklisted_initcalls, next) { 1324 if (!strcmp(fn_name, entry->buf)) { 1325 pr_debug("initcall %s blacklisted\n", fn_name); 1326 return true; 1327 } 1328 } 1329 1330 return false; 1331 } 1332 #else 1333 static int __init initcall_blacklist(char *str) 1334 { 1335 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n"); 1336 return 0; 1337 } 1338 1339 static bool __init_or_module initcall_blacklisted(initcall_t fn) 1340 { 1341 return false; 1342 } 1343 #endif 1344 __setup("initcall_blacklist=", initcall_blacklist); 1345 1346 static __init_or_module void 1347 trace_initcall_start_cb(void *data, initcall_t fn) 1348 { 1349 ktime_t *calltime = data; 1350 1351 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current)); 1352 *calltime = ktime_get(); 1353 } 1354 1355 static __init_or_module void 1356 trace_initcall_finish_cb(void *data, initcall_t fn, int ret) 1357 { 1358 ktime_t rettime, *calltime = data; 1359 1360 rettime = ktime_get(); 1361 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n", 1362 fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime)); 1363 } 1364 1365 static __init_or_module void 1366 trace_initcall_level_cb(void *data, const char *level) 1367 { 1368 printk(KERN_DEBUG "entering initcall level: %s\n", level); 1369 } 1370 1371 static ktime_t initcall_calltime; 1372 1373 #ifdef TRACEPOINTS_ENABLED 1374 static void __init initcall_debug_enable(void) 1375 { 1376 int ret; 1377 1378 ret = register_trace_initcall_start(trace_initcall_start_cb, 1379 &initcall_calltime); 1380 ret |= register_trace_initcall_finish(trace_initcall_finish_cb, 1381 &initcall_calltime); 1382 ret |= register_trace_initcall_level(trace_initcall_level_cb, NULL); 1383 WARN(ret, "Failed to register initcall tracepoints\n"); 1384 } 1385 # define do_trace_initcall_start trace_initcall_start 1386 # define do_trace_initcall_finish trace_initcall_finish 1387 # define do_trace_initcall_level trace_initcall_level 1388 #else 1389 static inline void do_trace_initcall_start(initcall_t fn) 1390 { 1391 if (!initcall_debug) 1392 return; 1393 trace_initcall_start_cb(&initcall_calltime, fn); 1394 } 1395 static inline void do_trace_initcall_finish(initcall_t fn, int ret) 1396 { 1397 if (!initcall_debug) 1398 return; 1399 trace_initcall_finish_cb(&initcall_calltime, fn, ret); 1400 } 1401 static inline void do_trace_initcall_level(const char *level) 1402 { 1403 if (!initcall_debug) 1404 return; 1405 trace_initcall_level_cb(NULL, level); 1406 } 1407 #endif /* !TRACEPOINTS_ENABLED */ 1408 1409 int __init_or_module do_one_initcall(initcall_t fn) 1410 { 1411 int count = preempt_count(); 1412 char msgbuf[64]; 1413 int ret; 1414 1415 if (initcall_blacklisted(fn)) 1416 return -EPERM; 1417 1418 do_trace_initcall_start(fn); 1419 ret = fn(); 1420 do_trace_initcall_finish(fn, ret); 1421 1422 msgbuf[0] = 0; 1423 1424 if (preempt_count() != count) { 1425 sprintf(msgbuf, "preemption imbalance "); 1426 preempt_count_set(count); 1427 } 1428 if (irqs_disabled()) { 1429 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf)); 1430 local_irq_enable(); 1431 } 1432 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf); 1433 1434 add_latent_entropy(); 1435 return ret; 1436 } 1437 1438 1439 static initcall_entry_t *initcall_levels[] __initdata = { 1440 __initcall0_start, 1441 __initcall1_start, 1442 __initcall2_start, 1443 __initcall3_start, 1444 __initcall4_start, 1445 __initcall5_start, 1446 __initcall6_start, 1447 __initcall7_start, 1448 __initcall_end, 1449 }; 1450 1451 /* Keep these in sync with initcalls in include/linux/init.h */ 1452 static const char *initcall_level_names[] __initdata = { 1453 "pure", 1454 "core", 1455 "postcore", 1456 "arch", 1457 "subsys", 1458 "fs", 1459 "device", 1460 "late", 1461 }; 1462 1463 static int __init ignore_unknown_bootoption(char *param, char *val, 1464 const char *unused, void *arg) 1465 { 1466 return 0; 1467 } 1468 1469 static void __init do_initcall_level(int level, char *command_line) 1470 { 1471 initcall_entry_t *fn; 1472 1473 parse_args(initcall_level_names[level], 1474 command_line, __start___param, 1475 __stop___param - __start___param, 1476 level, level, 1477 NULL, ignore_unknown_bootoption); 1478 1479 do_trace_initcall_level(initcall_level_names[level]); 1480 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++) 1481 do_one_initcall(initcall_from_entry(fn)); 1482 } 1483 1484 static void __init do_initcalls(void) 1485 { 1486 int level; 1487 size_t len = saved_command_line_len + 1; 1488 char *command_line; 1489 1490 command_line = kzalloc(len, GFP_KERNEL); 1491 if (!command_line) 1492 panic("%s: Failed to allocate %zu bytes\n", __func__, len); 1493 1494 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) { 1495 /* Parser modifies command_line, restore it each time */ 1496 strcpy(command_line, saved_command_line); 1497 do_initcall_level(level, command_line); 1498 } 1499 1500 kfree(command_line); 1501 } 1502 1503 /* 1504 * Ok, the machine is now initialized. None of the devices 1505 * have been touched yet, but the CPU subsystem is up and 1506 * running, and memory and process management works. 1507 * 1508 * Now we can finally start doing some real work.. 1509 */ 1510 static void __init do_basic_setup(void) 1511 { 1512 cpuset_init_smp(); 1513 ksysfs_init(); 1514 driver_init(); 1515 init_irq_proc(); 1516 do_ctors(); 1517 do_initcalls(); 1518 } 1519 1520 static void __init do_pre_smp_initcalls(void) 1521 { 1522 initcall_entry_t *fn; 1523 1524 do_trace_initcall_level("early"); 1525 for (fn = __initcall_start; fn < __initcall0_start; fn++) 1526 do_one_initcall(initcall_from_entry(fn)); 1527 } 1528 1529 static int run_init_process(const char *init_filename) 1530 { 1531 const char *const *p; 1532 1533 argv_init[0] = init_filename; 1534 pr_info("Run %s as init process\n", init_filename); 1535 pr_debug(" with arguments:\n"); 1536 for (p = argv_init; *p; p++) 1537 pr_debug(" %s\n", *p); 1538 pr_debug(" with environment:\n"); 1539 for (p = envp_init; *p; p++) 1540 pr_debug(" %s\n", *p); 1541 return kernel_execve(init_filename, argv_init, envp_init); 1542 } 1543 1544 static int try_to_run_init_process(const char *init_filename) 1545 { 1546 int ret; 1547 1548 ret = run_init_process(init_filename); 1549 1550 if (ret && ret != -ENOENT) { 1551 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n", 1552 init_filename, ret); 1553 } 1554 1555 return ret; 1556 } 1557 1558 static noinline void __init kernel_init_freeable(void); 1559 1560 #if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX) 1561 bool rodata_enabled __ro_after_init = true; 1562 1563 #ifndef arch_parse_debug_rodata 1564 static inline bool arch_parse_debug_rodata(char *str) { return false; } 1565 #endif 1566 1567 static int __init set_debug_rodata(char *str) 1568 { 1569 if (arch_parse_debug_rodata(str)) 1570 return 0; 1571 1572 if (str && !strcmp(str, "on")) 1573 rodata_enabled = true; 1574 else if (str && !strcmp(str, "off")) 1575 rodata_enabled = false; 1576 else 1577 pr_warn("Invalid option string for rodata: '%s'\n", str); 1578 return 0; 1579 } 1580 early_param("rodata", set_debug_rodata); 1581 #endif 1582 1583 static void mark_readonly(void) 1584 { 1585 if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX) && rodata_enabled) { 1586 /* 1587 * load_module() results in W+X mappings, which are cleaned 1588 * up with init_free_wq. Let's make sure that queued work is 1589 * flushed so that we don't hit false positives looking for 1590 * insecure pages which are W+X. 1591 */ 1592 flush_module_init_free_work(); 1593 jump_label_init_ro(); 1594 mark_rodata_ro(); 1595 pgtable_checkwx(); 1596 rodata_test(); 1597 } else if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) { 1598 pr_info("Kernel memory protection disabled.\n"); 1599 } else if (IS_ENABLED(CONFIG_ARCH_HAS_STRICT_KERNEL_RWX)) { 1600 pr_warn("Kernel memory protection not selected by kernel config.\n"); 1601 } else { 1602 pr_warn("This architecture does not have kernel memory protection.\n"); 1603 } 1604 } 1605 1606 void __weak free_initmem(void) 1607 { 1608 free_initmem_default(POISON_FREE_INITMEM); 1609 } 1610 1611 static int __ref kernel_init(void *unused) 1612 { 1613 int ret; 1614 1615 init_userspace_fs(); 1616 1617 /* 1618 * Wait until kthreadd is all set-up. 1619 */ 1620 wait_for_completion(&kthreadd_done); 1621 1622 kernel_init_freeable(); 1623 /* need to finish all async __init code before freeing the memory */ 1624 async_synchronize_full(); 1625 1626 system_state = SYSTEM_FREEING_INITMEM; 1627 kprobe_free_init_mem(); 1628 ftrace_free_init_mem(); 1629 kgdb_free_init_mem(); 1630 exit_boot_config(); 1631 free_initmem(); 1632 mark_readonly(); 1633 1634 /* 1635 * Kernel mappings are now finalized - update the userspace page-table 1636 * to finalize PTI. 1637 */ 1638 pti_finalize(); 1639 1640 system_state = SYSTEM_RUNNING; 1641 numa_default_policy(); 1642 1643 rcu_end_inkernel_boot(); 1644 1645 do_sysctl_args(); 1646 1647 if (ramdisk_execute_command) { 1648 ret = run_init_process(ramdisk_execute_command); 1649 if (!ret) 1650 return 0; 1651 pr_err("Failed to execute %s (error %d)\n", 1652 ramdisk_execute_command, ret); 1653 } 1654 1655 /* 1656 * We try each of these until one succeeds. 1657 * 1658 * The Bourne shell can be used instead of init if we are 1659 * trying to recover a really broken machine. 1660 */ 1661 if (execute_command) { 1662 ret = run_init_process(execute_command); 1663 if (!ret) 1664 return 0; 1665 panic("Requested init %s failed (error %d).", 1666 execute_command, ret); 1667 } 1668 1669 if (CONFIG_DEFAULT_INIT[0] != '\0') { 1670 ret = run_init_process(CONFIG_DEFAULT_INIT); 1671 if (ret) 1672 pr_err("Default init %s failed (error %d)\n", 1673 CONFIG_DEFAULT_INIT, ret); 1674 else 1675 return 0; 1676 } 1677 1678 if (!try_to_run_init_process("/sbin/init") || 1679 !try_to_run_init_process("/etc/init") || 1680 !try_to_run_init_process("/bin/init") || 1681 !try_to_run_init_process("/bin/sh")) 1682 return 0; 1683 1684 panic("No working init found. Try passing init= option to kernel. " 1685 "See Linux Documentation/admin-guide/init.rst for guidance."); 1686 } 1687 1688 /* Open /dev/console, for stdin/stdout/stderr, this should never fail */ 1689 void __init console_on_rootfs(void) 1690 { 1691 struct file *file = filp_open("/dev/console", O_RDWR, 0); 1692 1693 if (IS_ERR(file)) { 1694 pr_err("Warning: unable to open an initial console.\n"); 1695 return; 1696 } 1697 init_dup(file); 1698 init_dup(file); 1699 init_dup(file); 1700 fput(file); 1701 } 1702 1703 static noinline void __init kernel_init_freeable(void) 1704 { 1705 /* Now the scheduler is fully set up and can do blocking allocations */ 1706 gfp_allowed_mask = __GFP_BITS_MASK; 1707 1708 /* 1709 * init can allocate pages on any node 1710 */ 1711 set_mems_allowed(node_states[N_MEMORY]); 1712 1713 rcu_assign_pointer(cad_pid, get_pid(task_pid(current))); 1714 1715 smp_prepare_cpus(setup_max_cpus); 1716 1717 workqueue_init(); 1718 1719 init_mm_internals(); 1720 1721 do_pre_smp_initcalls(); 1722 lockup_detector_init(); 1723 1724 smp_init(); 1725 sched_init_smp(); 1726 1727 workqueue_init_topology(); 1728 async_init(); 1729 padata_init(); 1730 page_alloc_init_late(); 1731 1732 do_basic_setup(); 1733 1734 kunit_run_all_tests(); 1735 1736 wait_for_initramfs(); 1737 console_on_rootfs(); 1738 1739 /* 1740 * check if there is an early userspace init. If yes, let it do all 1741 * the work 1742 */ 1743 int ramdisk_command_access; 1744 ramdisk_command_access = init_eaccess(ramdisk_execute_command); 1745 if (ramdisk_command_access != 0) { 1746 if (ramdisk_execute_command_set) 1747 pr_warn("check access for rdinit=%s failed: %i, ignoring\n", 1748 ramdisk_execute_command, ramdisk_command_access); 1749 ramdisk_execute_command = NULL; 1750 prepare_namespace(); 1751 } 1752 1753 /* 1754 * Ok, we have completed the initial bootup, and 1755 * we're essentially up and running. Get rid of the 1756 * initmem segments and start the user-mode stuff.. 1757 * 1758 * rootfs is available now, try loading the public keys 1759 * and default modules 1760 */ 1761 1762 integrity_load_keys(); 1763 } 1764