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) { 219 pr_warn("Parameter %s is obsolete, ignored\n", 220 p-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