xref: /linux/init/main.c (revision e406d57be7bd2a4e73ea512c1ae36a40a44e499e)
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/vmalloc.h>
40 #include <linux/kernel_stat.h>
41 #include <linux/start_kernel.h>
42 #include <linux/security.h>
43 #include <linux/smp.h>
44 #include <linux/profile.h>
45 #include <linux/kfence.h>
46 #include <linux/rcupdate.h>
47 #include <linux/srcu.h>
48 #include <linux/moduleparam.h>
49 #include <linux/kallsyms.h>
50 #include <linux/buildid.h>
51 #include <linux/writeback.h>
52 #include <linux/cpu.h>
53 #include <linux/cpuset.h>
54 #include <linux/memcontrol.h>
55 #include <linux/cgroup.h>
56 #include <linux/tick.h>
57 #include <linux/sched/isolation.h>
58 #include <linux/interrupt.h>
59 #include <linux/taskstats_kern.h>
60 #include <linux/delayacct.h>
61 #include <linux/unistd.h>
62 #include <linux/utsname.h>
63 #include <linux/rmap.h>
64 #include <linux/mempolicy.h>
65 #include <linux/key.h>
66 #include <linux/debug_locks.h>
67 #include <linux/debugobjects.h>
68 #include <linux/lockdep.h>
69 #include <linux/kmemleak.h>
70 #include <linux/padata.h>
71 #include <linux/pid_namespace.h>
72 #include <linux/device/driver.h>
73 #include <linux/kthread.h>
74 #include <linux/sched.h>
75 #include <linux/sched/init.h>
76 #include <linux/signal.h>
77 #include <linux/idr.h>
78 #include <linux/kgdb.h>
79 #include <linux/ftrace.h>
80 #include <linux/async.h>
81 #include <linux/shmem_fs.h>
82 #include <linux/slab.h>
83 #include <linux/perf_event.h>
84 #include <linux/ptrace.h>
85 #include <linux/pti.h>
86 #include <linux/blkdev.h>
87 #include <linux/sched/clock.h>
88 #include <linux/sched/task.h>
89 #include <linux/sched/task_stack.h>
90 #include <linux/context_tracking.h>
91 #include <linux/random.h>
92 #include <linux/moduleloader.h>
93 #include <linux/list.h>
94 #include <linux/integrity.h>
95 #include <linux/proc_ns.h>
96 #include <linux/io.h>
97 #include <linux/cache.h>
98 #include <linux/rodata_test.h>
99 #include <linux/jump_label.h>
100 #include <linux/kcsan.h>
101 #include <linux/init_syscalls.h>
102 #include <linux/stackdepot.h>
103 #include <linux/randomize_kstack.h>
104 #include <linux/pidfs.h>
105 #include <linux/ptdump.h>
106 #include <linux/time_namespace.h>
107 #include <net/net_namespace.h>
108 
109 #include <asm/io.h>
110 #include <asm/setup.h>
111 #include <asm/sections.h>
112 #include <asm/cacheflush.h>
113 
114 #define CREATE_TRACE_POINTS
115 #include <trace/events/initcall.h>
116 
117 #include <kunit/test.h>
118 
119 static int kernel_init(void *);
120 
121 /*
122  * Debug helper: via this flag we know that we are in 'early bootup code'
123  * where only the boot processor is running with IRQ disabled.  This means
124  * two things - IRQ must not be enabled before the flag is cleared and some
125  * operations which are not allowed with IRQ disabled are allowed while the
126  * flag is set.
127  */
128 bool early_boot_irqs_disabled __read_mostly;
129 
130 enum system_states system_state __read_mostly;
131 EXPORT_SYMBOL(system_state);
132 
133 /*
134  * Boot command-line arguments
135  */
136 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
137 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
138 
139 /* Default late time init is NULL. archs can override this later. */
140 void (*__initdata late_time_init)(void);
141 
142 /* Untouched command line saved by arch-specific code. */
143 char __initdata boot_command_line[COMMAND_LINE_SIZE];
144 /* Untouched saved command line (eg. for /proc) */
145 char *saved_command_line __ro_after_init;
146 unsigned int saved_command_line_len __ro_after_init;
147 /* Command line for parameter parsing */
148 static char *static_command_line;
149 /* Untouched extra command line */
150 static char *extra_command_line;
151 /* Extra init arguments */
152 static char *extra_init_args;
153 
154 #ifdef CONFIG_BOOT_CONFIG
155 /* Is bootconfig on command line? */
156 static bool bootconfig_found;
157 static size_t initargs_offs;
158 #else
159 # define bootconfig_found false
160 # define initargs_offs 0
161 #endif
162 
163 static char *execute_command;
164 static char *ramdisk_execute_command = "/init";
165 
166 /*
167  * Used to generate warnings if static_key manipulation functions are used
168  * before jump_label_init is called.
169  */
170 bool static_key_initialized __read_mostly;
171 EXPORT_SYMBOL_GPL(static_key_initialized);
172 
173 /*
174  * If set, this is an indication to the drivers that reset the underlying
175  * device before going ahead with the initialization otherwise driver might
176  * rely on the BIOS and skip the reset operation.
177  *
178  * This is useful if kernel is booting in an unreliable environment.
179  * For ex. kdump situation where previous kernel has crashed, BIOS has been
180  * skipped and devices will be in unknown state.
181  */
182 unsigned int reset_devices;
183 EXPORT_SYMBOL(reset_devices);
184 
set_reset_devices(char * str)185 static int __init set_reset_devices(char *str)
186 {
187 	reset_devices = 1;
188 	return 1;
189 }
190 
191 __setup("reset_devices", set_reset_devices);
192 
193 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
194 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
195 static const char *panic_later, *panic_param;
196 
obsolete_checksetup(char * line)197 static bool __init obsolete_checksetup(char *line)
198 {
199 	const struct obs_kernel_param *p;
200 	bool had_early_param = false;
201 
202 	p = __setup_start;
203 	do {
204 		int n = strlen(p->str);
205 		if (parameqn(line, p->str, n)) {
206 			if (p->early) {
207 				/* Already done in parse_early_param?
208 				 * (Needs exact match on param part).
209 				 * Keep iterating, as we can have early
210 				 * params and __setups of same names 8( */
211 				if (line[n] == '\0' || line[n] == '=')
212 					had_early_param = true;
213 			} else if (!p->setup_func) {
214 				pr_warn("Parameter %s is obsolete, ignored\n",
215 					p->str);
216 				return true;
217 			} else if (p->setup_func(line + n))
218 				return true;
219 		}
220 		p++;
221 	} while (p < __setup_end);
222 
223 	return had_early_param;
224 }
225 
226 /*
227  * This should be approx 2 Bo*oMips to start (note initial shift), and will
228  * still work even if initially too large, it will just take slightly longer
229  */
230 unsigned long loops_per_jiffy = (1<<12);
231 EXPORT_SYMBOL(loops_per_jiffy);
232 
debug_kernel(char * str)233 static int __init debug_kernel(char *str)
234 {
235 	console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
236 	return 0;
237 }
238 
quiet_kernel(char * str)239 static int __init quiet_kernel(char *str)
240 {
241 	console_loglevel = CONSOLE_LOGLEVEL_QUIET;
242 	return 0;
243 }
244 
245 early_param("debug", debug_kernel);
246 early_param("quiet", quiet_kernel);
247 
loglevel(char * str)248 static int __init loglevel(char *str)
249 {
250 	int newlevel;
251 
252 	/*
253 	 * Only update loglevel value when a correct setting was passed,
254 	 * to prevent blind crashes (when loglevel being set to 0) that
255 	 * are quite hard to debug
256 	 */
257 	if (get_option(&str, &newlevel)) {
258 		console_loglevel = newlevel;
259 		return 0;
260 	}
261 
262 	return -EINVAL;
263 }
264 
265 early_param("loglevel", loglevel);
266 
267 #ifdef CONFIG_BLK_DEV_INITRD
get_boot_config_from_initrd(size_t * _size)268 static void * __init get_boot_config_from_initrd(size_t *_size)
269 {
270 	u32 size, csum;
271 	char *data;
272 	u32 *hdr;
273 	int i;
274 
275 	if (!initrd_end)
276 		return NULL;
277 
278 	data = (char *)initrd_end - BOOTCONFIG_MAGIC_LEN;
279 	/*
280 	 * Since Grub may align the size of initrd to 4, we must
281 	 * check the preceding 3 bytes as well.
282 	 */
283 	for (i = 0; i < 4; i++) {
284 		if (!memcmp(data, BOOTCONFIG_MAGIC, BOOTCONFIG_MAGIC_LEN))
285 			goto found;
286 		data--;
287 	}
288 	return NULL;
289 
290 found:
291 	hdr = (u32 *)(data - 8);
292 	size = le32_to_cpu(hdr[0]);
293 	csum = le32_to_cpu(hdr[1]);
294 
295 	data = ((void *)hdr) - size;
296 	if ((unsigned long)data < initrd_start) {
297 		pr_err("bootconfig size %d is greater than initrd size %ld\n",
298 			size, initrd_end - initrd_start);
299 		return NULL;
300 	}
301 
302 	if (xbc_calc_checksum(data, size) != csum) {
303 		pr_err("bootconfig checksum failed\n");
304 		return NULL;
305 	}
306 
307 	/* Remove bootconfig from initramfs/initrd */
308 	initrd_end = (unsigned long)data;
309 	if (_size)
310 		*_size = size;
311 
312 	return data;
313 }
314 #else
get_boot_config_from_initrd(size_t * _size)315 static void * __init get_boot_config_from_initrd(size_t *_size)
316 {
317 	return NULL;
318 }
319 #endif
320 
321 #ifdef CONFIG_BOOT_CONFIG
322 
323 static char xbc_namebuf[XBC_KEYLEN_MAX] __initdata;
324 
325 #define rest(dst, end) ((end) > (dst) ? (end) - (dst) : 0)
326 
xbc_snprint_cmdline(char * buf,size_t size,struct xbc_node * root)327 static int __init xbc_snprint_cmdline(char *buf, size_t size,
328 				      struct xbc_node *root)
329 {
330 	struct xbc_node *knode, *vnode;
331 	char *end = buf + size;
332 	const char *val, *q;
333 	int ret;
334 
335 	xbc_node_for_each_key_value(root, knode, val) {
336 		ret = xbc_node_compose_key_after(root, knode,
337 					xbc_namebuf, XBC_KEYLEN_MAX);
338 		if (ret < 0)
339 			return ret;
340 
341 		vnode = xbc_node_get_child(knode);
342 		if (!vnode) {
343 			ret = snprintf(buf, rest(buf, end), "%s ", xbc_namebuf);
344 			if (ret < 0)
345 				return ret;
346 			buf += ret;
347 			continue;
348 		}
349 		xbc_array_for_each_value(vnode, val) {
350 			/*
351 			 * For prettier and more readable /proc/cmdline, only
352 			 * quote the value when necessary, i.e. when it contains
353 			 * whitespace.
354 			 */
355 			q = strpbrk(val, " \t\r\n") ? "\"" : "";
356 			ret = snprintf(buf, rest(buf, end), "%s=%s%s%s ",
357 				       xbc_namebuf, q, val, q);
358 			if (ret < 0)
359 				return ret;
360 			buf += ret;
361 		}
362 	}
363 
364 	return buf - (end - size);
365 }
366 #undef rest
367 
368 /* Make an extra command line under given key word */
xbc_make_cmdline(const char * key)369 static char * __init xbc_make_cmdline(const char *key)
370 {
371 	struct xbc_node *root;
372 	char *new_cmdline;
373 	int ret, len = 0;
374 
375 	root = xbc_find_node(key);
376 	if (!root)
377 		return NULL;
378 
379 	/* Count required buffer size */
380 	len = xbc_snprint_cmdline(NULL, 0, root);
381 	if (len <= 0)
382 		return NULL;
383 
384 	new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES);
385 	if (!new_cmdline) {
386 		pr_err("Failed to allocate memory for extra kernel cmdline.\n");
387 		return NULL;
388 	}
389 
390 	ret = xbc_snprint_cmdline(new_cmdline, len + 1, root);
391 	if (ret < 0 || ret > len) {
392 		pr_err("Failed to print extra kernel cmdline.\n");
393 		memblock_free(new_cmdline, len + 1);
394 		return NULL;
395 	}
396 
397 	return new_cmdline;
398 }
399 
bootconfig_params(char * param,char * val,const char * unused,void * arg)400 static int __init bootconfig_params(char *param, char *val,
401 				    const char *unused, void *arg)
402 {
403 	if (strcmp(param, "bootconfig") == 0) {
404 		bootconfig_found = true;
405 	}
406 	return 0;
407 }
408 
warn_bootconfig(char * str)409 static int __init warn_bootconfig(char *str)
410 {
411 	/* The 'bootconfig' has been handled by bootconfig_params(). */
412 	return 0;
413 }
414 
setup_boot_config(void)415 static void __init setup_boot_config(void)
416 {
417 	static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
418 	const char *msg, *data;
419 	int pos, ret;
420 	size_t size;
421 	char *err;
422 
423 	/* Cut out the bootconfig data even if we have no bootconfig option */
424 	data = get_boot_config_from_initrd(&size);
425 	/* If there is no bootconfig in initrd, try embedded one. */
426 	if (!data)
427 		data = xbc_get_embedded_bootconfig(&size);
428 
429 	strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
430 	err = parse_args("bootconfig", tmp_cmdline, NULL, 0, 0, 0, NULL,
431 			 bootconfig_params);
432 
433 	if (IS_ERR(err) || !(bootconfig_found || IS_ENABLED(CONFIG_BOOT_CONFIG_FORCE)))
434 		return;
435 
436 	/* parse_args() stops at the next param of '--' and returns an address */
437 	if (err)
438 		initargs_offs = err - tmp_cmdline;
439 
440 	if (!data) {
441 		/* If user intended to use bootconfig, show an error level message */
442 		if (bootconfig_found)
443 			pr_err("'bootconfig' found on command line, but no bootconfig found\n");
444 		else
445 			pr_info("No bootconfig data provided, so skipping bootconfig");
446 		return;
447 	}
448 
449 	if (size >= XBC_DATA_MAX) {
450 		pr_err("bootconfig size %ld greater than max size %d\n",
451 			(long)size, XBC_DATA_MAX);
452 		return;
453 	}
454 
455 	ret = xbc_init(data, size, &msg, &pos);
456 	if (ret < 0) {
457 		if (pos < 0)
458 			pr_err("Failed to init bootconfig: %s.\n", msg);
459 		else
460 			pr_err("Failed to parse bootconfig: %s at %d.\n",
461 				msg, pos);
462 	} else {
463 		xbc_get_info(&ret, NULL);
464 		pr_info("Load bootconfig: %ld bytes %d nodes\n", (long)size, ret);
465 		/* keys starting with "kernel." are passed via cmdline */
466 		extra_command_line = xbc_make_cmdline("kernel");
467 		/* Also, "init." keys are init arguments */
468 		extra_init_args = xbc_make_cmdline("init");
469 	}
470 	return;
471 }
472 
exit_boot_config(void)473 static void __init exit_boot_config(void)
474 {
475 	xbc_exit();
476 }
477 
478 #else	/* !CONFIG_BOOT_CONFIG */
479 
setup_boot_config(void)480 static void __init setup_boot_config(void)
481 {
482 	/* Remove bootconfig data from initrd */
483 	get_boot_config_from_initrd(NULL);
484 }
485 
warn_bootconfig(char * str)486 static int __init warn_bootconfig(char *str)
487 {
488 	pr_warn("WARNING: 'bootconfig' found on the kernel command line but CONFIG_BOOT_CONFIG is not set.\n");
489 	return 0;
490 }
491 
492 #define exit_boot_config()	do {} while (0)
493 
494 #endif	/* CONFIG_BOOT_CONFIG */
495 
496 early_param("bootconfig", warn_bootconfig);
497 
cmdline_has_extra_options(void)498 bool __init cmdline_has_extra_options(void)
499 {
500 	return extra_command_line || extra_init_args;
501 }
502 
503 /* Change NUL term back to "=", to make "param" the whole string. */
repair_env_string(char * param,char * val)504 static void __init repair_env_string(char *param, char *val)
505 {
506 	if (val) {
507 		/* param=val or param="val"? */
508 		if (val == param+strlen(param)+1)
509 			val[-1] = '=';
510 		else if (val == param+strlen(param)+2) {
511 			val[-2] = '=';
512 			memmove(val-1, val, strlen(val)+1);
513 		} else
514 			BUG();
515 	}
516 }
517 
518 /* Anything after -- gets handed straight to init. */
set_init_arg(char * param,char * val,const char * unused,void * arg)519 static int __init set_init_arg(char *param, char *val,
520 			       const char *unused, void *arg)
521 {
522 	unsigned int i;
523 
524 	if (panic_later)
525 		return 0;
526 
527 	repair_env_string(param, val);
528 
529 	for (i = 0; argv_init[i]; i++) {
530 		if (i == MAX_INIT_ARGS) {
531 			panic_later = "init";
532 			panic_param = param;
533 			return 0;
534 		}
535 	}
536 	argv_init[i] = param;
537 	return 0;
538 }
539 
540 /*
541  * Unknown boot options get handed to init, unless they look like
542  * unused parameters (modprobe will find them in /proc/cmdline).
543  */
unknown_bootoption(char * param,char * val,const char * unused,void * arg)544 static int __init unknown_bootoption(char *param, char *val,
545 				     const char *unused, void *arg)
546 {
547 	size_t len = strlen(param);
548 	/*
549 	 * Well-known bootloader identifiers:
550 	 * 1. LILO/Grub pass "BOOT_IMAGE=...";
551 	 * 2. kexec/kdump (kexec-tools) pass "kexec".
552 	 */
553 	const char *bootloader[] = { "BOOT_IMAGE=", "kexec", NULL };
554 
555 	/* Handle params aliased to sysctls */
556 	if (sysctl_is_alias(param))
557 		return 0;
558 
559 	repair_env_string(param, val);
560 
561 	/* Handle bootloader identifier */
562 	for (int i = 0; bootloader[i]; i++) {
563 		if (strstarts(param, bootloader[i]))
564 			return 0;
565 	}
566 
567 	/* Handle obsolete-style parameters */
568 	if (obsolete_checksetup(param))
569 		return 0;
570 
571 	/* Unused module parameter. */
572 	if (strnchr(param, len, '.'))
573 		return 0;
574 
575 	if (panic_later)
576 		return 0;
577 
578 	if (val) {
579 		/* Environment option */
580 		unsigned int i;
581 		for (i = 0; envp_init[i]; i++) {
582 			if (i == MAX_INIT_ENVS) {
583 				panic_later = "env";
584 				panic_param = param;
585 			}
586 			if (!strncmp(param, envp_init[i], len+1))
587 				break;
588 		}
589 		envp_init[i] = param;
590 	} else {
591 		/* Command line option */
592 		unsigned int i;
593 		for (i = 0; argv_init[i]; i++) {
594 			if (i == MAX_INIT_ARGS) {
595 				panic_later = "init";
596 				panic_param = param;
597 			}
598 		}
599 		argv_init[i] = param;
600 	}
601 	return 0;
602 }
603 
init_setup(char * str)604 static int __init init_setup(char *str)
605 {
606 	unsigned int i;
607 
608 	execute_command = str;
609 	/*
610 	 * In case LILO is going to boot us with default command line,
611 	 * it prepends "auto" before the whole cmdline which makes
612 	 * the shell think it should execute a script with such name.
613 	 * So we ignore all arguments entered _before_ init=... [MJ]
614 	 */
615 	for (i = 1; i < MAX_INIT_ARGS; i++)
616 		argv_init[i] = NULL;
617 	return 1;
618 }
619 __setup("init=", init_setup);
620 
rdinit_setup(char * str)621 static int __init rdinit_setup(char *str)
622 {
623 	unsigned int i;
624 
625 	ramdisk_execute_command = str;
626 	/* See "auto" comment in init_setup */
627 	for (i = 1; i < MAX_INIT_ARGS; i++)
628 		argv_init[i] = NULL;
629 	return 1;
630 }
631 __setup("rdinit=", rdinit_setup);
632 
633 #ifndef CONFIG_SMP
setup_nr_cpu_ids(void)634 static inline void setup_nr_cpu_ids(void) { }
smp_prepare_cpus(unsigned int maxcpus)635 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
636 #endif
637 
638 /*
639  * We need to store the untouched command line for future reference.
640  * We also need to store the touched command line since the parameter
641  * parsing is performed in place, and we should allow a component to
642  * store reference of name/value for future reference.
643  */
setup_command_line(char * command_line)644 static void __init setup_command_line(char *command_line)
645 {
646 	size_t len, xlen = 0, ilen = 0;
647 
648 	if (extra_command_line)
649 		xlen = strlen(extra_command_line);
650 	if (extra_init_args) {
651 		extra_init_args = strim(extra_init_args); /* remove trailing space */
652 		ilen = strlen(extra_init_args) + 4; /* for " -- " */
653 	}
654 
655 	len = xlen + strlen(boot_command_line) + ilen + 1;
656 
657 	saved_command_line = memblock_alloc_or_panic(len, SMP_CACHE_BYTES);
658 
659 	len = xlen + strlen(command_line) + 1;
660 
661 	static_command_line = memblock_alloc_or_panic(len, SMP_CACHE_BYTES);
662 
663 	if (xlen) {
664 		/*
665 		 * We have to put extra_command_line before boot command
666 		 * lines because there could be dashes (separator of init
667 		 * command line) in the command lines.
668 		 */
669 		strcpy(saved_command_line, extra_command_line);
670 		strcpy(static_command_line, extra_command_line);
671 	}
672 	strcpy(saved_command_line + xlen, boot_command_line);
673 	strcpy(static_command_line + xlen, command_line);
674 
675 	if (ilen) {
676 		/*
677 		 * Append supplemental init boot args to saved_command_line
678 		 * so that user can check what command line options passed
679 		 * to init.
680 		 * The order should always be
681 		 * " -- "[bootconfig init-param][cmdline init-param]
682 		 */
683 		if (initargs_offs) {
684 			len = xlen + initargs_offs;
685 			strcpy(saved_command_line + len, extra_init_args);
686 			len += ilen - 4;	/* strlen(extra_init_args) */
687 			strcpy(saved_command_line + len,
688 				boot_command_line + initargs_offs - 1);
689 		} else {
690 			len = strlen(saved_command_line);
691 			strcpy(saved_command_line + len, " -- ");
692 			len += 4;
693 			strcpy(saved_command_line + len, extra_init_args);
694 		}
695 	}
696 
697 	saved_command_line_len = strlen(saved_command_line);
698 }
699 
700 /*
701  * We need to finalize in a non-__init function or else race conditions
702  * between the root thread and the init thread may cause start_kernel to
703  * be reaped by free_initmem before the root thread has proceeded to
704  * cpu_idle.
705  *
706  * gcc-3.4 accidentally inlines this function, so use noinline.
707  */
708 
709 static __initdata DECLARE_COMPLETION(kthreadd_done);
710 
rest_init(void)711 static noinline void __ref __noreturn rest_init(void)
712 {
713 	struct task_struct *tsk;
714 	int pid;
715 
716 	rcu_scheduler_starting();
717 	/*
718 	 * We need to spawn init first so that it obtains pid 1, however
719 	 * the init task will end up wanting to create kthreads, which, if
720 	 * we schedule it before we create kthreadd, will OOPS.
721 	 */
722 	pid = user_mode_thread(kernel_init, NULL, CLONE_FS);
723 	/*
724 	 * Pin init on the boot CPU. Task migration is not properly working
725 	 * until sched_init_smp() has been run. It will set the allowed
726 	 * CPUs for init to the non isolated CPUs.
727 	 */
728 	rcu_read_lock();
729 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
730 	tsk->flags |= PF_NO_SETAFFINITY;
731 	set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
732 	rcu_read_unlock();
733 
734 	numa_default_policy();
735 	pid = kernel_thread(kthreadd, NULL, NULL, CLONE_FS | CLONE_FILES);
736 	rcu_read_lock();
737 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
738 	rcu_read_unlock();
739 
740 	/*
741 	 * Enable might_sleep() and smp_processor_id() checks.
742 	 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
743 	 * kernel_thread() would trigger might_sleep() splats. With
744 	 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
745 	 * already, but it's stuck on the kthreadd_done completion.
746 	 */
747 	system_state = SYSTEM_SCHEDULING;
748 
749 	complete(&kthreadd_done);
750 
751 	/*
752 	 * The boot idle thread must execute schedule()
753 	 * at least once to get things moving:
754 	 */
755 	schedule_preempt_disabled();
756 	/* Call into cpu_idle with preempt disabled */
757 	cpu_startup_entry(CPUHP_ONLINE);
758 }
759 
760 /* Check for early params. */
do_early_param(char * param,char * val,const char * unused,void * arg)761 static int __init do_early_param(char *param, char *val,
762 				 const char *unused, void *arg)
763 {
764 	const struct obs_kernel_param *p;
765 
766 	for (p = __setup_start; p < __setup_end; p++) {
767 		if (p->early && parameq(param, p->str)) {
768 			if (p->setup_func(val) != 0)
769 				pr_warn("Malformed early option '%s'\n", param);
770 		}
771 	}
772 	/* We accept everything at this stage. */
773 	return 0;
774 }
775 
parse_early_options(char * cmdline)776 void __init parse_early_options(char *cmdline)
777 {
778 	parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
779 		   do_early_param);
780 }
781 
782 /* Arch code calls this early on, or if not, just before other parsing. */
parse_early_param(void)783 void __init parse_early_param(void)
784 {
785 	static int done __initdata;
786 	static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
787 
788 	if (done)
789 		return;
790 
791 	/* All fall through to do_early_param. */
792 	strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
793 	parse_early_options(tmp_cmdline);
794 	done = 1;
795 }
796 
arch_post_acpi_subsys_init(void)797 void __init __weak arch_post_acpi_subsys_init(void) { }
798 
smp_setup_processor_id(void)799 void __init __weak smp_setup_processor_id(void)
800 {
801 }
802 
smp_prepare_boot_cpu(void)803 void __init __weak smp_prepare_boot_cpu(void)
804 {
805 }
806 
807 # if THREAD_SIZE >= PAGE_SIZE
thread_stack_cache_init(void)808 void __init __weak thread_stack_cache_init(void)
809 {
810 }
811 #endif
812 
poking_init(void)813 void __init __weak poking_init(void) { }
814 
pgtable_cache_init(void)815 void __init __weak pgtable_cache_init(void) { }
816 
trap_init(void)817 void __init __weak trap_init(void) { }
818 
819 bool initcall_debug;
820 core_param(initcall_debug, initcall_debug, bool, 0644);
821 
822 #ifdef TRACEPOINTS_ENABLED
823 static void __init initcall_debug_enable(void);
824 #else
initcall_debug_enable(void)825 static inline void initcall_debug_enable(void)
826 {
827 }
828 #endif
829 
830 #ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
831 DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
832 			   randomize_kstack_offset);
833 DEFINE_PER_CPU(u32, kstack_offset);
834 
early_randomize_kstack_offset(char * buf)835 static int __init early_randomize_kstack_offset(char *buf)
836 {
837 	int ret;
838 	bool bool_result;
839 
840 	ret = kstrtobool(buf, &bool_result);
841 	if (ret)
842 		return ret;
843 
844 	if (bool_result)
845 		static_branch_enable(&randomize_kstack_offset);
846 	else
847 		static_branch_disable(&randomize_kstack_offset);
848 	return 0;
849 }
850 early_param("randomize_kstack_offset", early_randomize_kstack_offset);
851 #endif
852 
print_unknown_bootoptions(void)853 static void __init print_unknown_bootoptions(void)
854 {
855 	char *unknown_options;
856 	char *end;
857 	const char *const *p;
858 	size_t len;
859 
860 	if (panic_later || (!argv_init[1] && !envp_init[2]))
861 		return;
862 
863 	/*
864 	 * Determine how many options we have to print out, plus a space
865 	 * before each
866 	 */
867 	len = 1; /* null terminator */
868 	for (p = &argv_init[1]; *p; p++) {
869 		len++;
870 		len += strlen(*p);
871 	}
872 	for (p = &envp_init[2]; *p; p++) {
873 		len++;
874 		len += strlen(*p);
875 	}
876 
877 	unknown_options = memblock_alloc(len, SMP_CACHE_BYTES);
878 	if (!unknown_options) {
879 		pr_err("%s: Failed to allocate %zu bytes\n",
880 			__func__, len);
881 		return;
882 	}
883 	end = unknown_options;
884 
885 	for (p = &argv_init[1]; *p; p++)
886 		end += sprintf(end, " %s", *p);
887 	for (p = &envp_init[2]; *p; p++)
888 		end += sprintf(end, " %s", *p);
889 
890 	/* Start at unknown_options[1] to skip the initial space */
891 	pr_notice("Unknown kernel command line parameters \"%s\", will be passed to user space.\n",
892 		&unknown_options[1]);
893 	memblock_free(unknown_options, len);
894 }
895 
early_numa_node_init(void)896 static void __init early_numa_node_init(void)
897 {
898 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
899 #ifndef cpu_to_node
900 	int cpu;
901 
902 	/* The early_cpu_to_node() should be ready here. */
903 	for_each_possible_cpu(cpu)
904 		set_cpu_numa_node(cpu, early_cpu_to_node(cpu));
905 #endif
906 #endif
907 }
908 
909 asmlinkage __visible __init __no_sanitize_address __noreturn __no_stack_protector
start_kernel(void)910 void start_kernel(void)
911 {
912 	char *command_line;
913 	char *after_dashes;
914 
915 	set_task_stack_end_magic(&init_task);
916 	smp_setup_processor_id();
917 	debug_objects_early_init();
918 	init_vmlinux_build_id();
919 
920 	cgroup_init_early();
921 
922 	local_irq_disable();
923 	early_boot_irqs_disabled = true;
924 
925 	/*
926 	 * Interrupts are still disabled. Do necessary setups, then
927 	 * enable them.
928 	 */
929 	boot_cpu_init();
930 	page_address_init();
931 	pr_notice("%s", linux_banner);
932 	setup_arch(&command_line);
933 	/* Static keys and static calls are needed by LSMs */
934 	jump_label_init();
935 	static_call_init();
936 	early_security_init();
937 	setup_boot_config();
938 	setup_command_line(command_line);
939 	setup_nr_cpu_ids();
940 	setup_per_cpu_areas();
941 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
942 	early_numa_node_init();
943 	boot_cpu_hotplug_init();
944 
945 	pr_notice("Kernel command line: %s\n", saved_command_line);
946 	/* parameters may set static keys */
947 	parse_early_param();
948 	after_dashes = parse_args("Booting kernel",
949 				  static_command_line, __start___param,
950 				  __stop___param - __start___param,
951 				  -1, -1, NULL, &unknown_bootoption);
952 	print_unknown_bootoptions();
953 	if (!IS_ERR_OR_NULL(after_dashes))
954 		parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
955 			   NULL, set_init_arg);
956 	if (extra_init_args)
957 		parse_args("Setting extra init args", extra_init_args,
958 			   NULL, 0, -1, -1, NULL, set_init_arg);
959 
960 	/* Architectural and non-timekeeping rng init, before allocator init */
961 	random_init_early(command_line);
962 
963 	/*
964 	 * These use large bootmem allocations and must precede
965 	 * initalization of page allocator
966 	 */
967 	setup_log_buf(0);
968 	vfs_caches_init_early();
969 	sort_main_extable();
970 	trap_init();
971 	mm_core_init();
972 	maple_tree_init();
973 	poking_init();
974 	ftrace_init();
975 
976 	/* trace_printk can be enabled here */
977 	early_trace_init();
978 
979 	/*
980 	 * Set up the scheduler prior starting any interrupts (such as the
981 	 * timer interrupt). Full topology setup happens at smp_init()
982 	 * time - but meanwhile we still have a functioning scheduler.
983 	 */
984 	sched_init();
985 
986 	if (WARN(!irqs_disabled(),
987 		 "Interrupts were enabled *very* early, fixing it\n"))
988 		local_irq_disable();
989 	radix_tree_init();
990 
991 	/*
992 	 * Set up housekeeping before setting up workqueues to allow the unbound
993 	 * workqueue to take non-housekeeping into account.
994 	 */
995 	housekeeping_init();
996 
997 	/*
998 	 * Allow workqueue creation and work item queueing/cancelling
999 	 * early.  Work item execution depends on kthreads and starts after
1000 	 * workqueue_init().
1001 	 */
1002 	workqueue_init_early();
1003 
1004 	rcu_init();
1005 	kvfree_rcu_init();
1006 
1007 	/* Trace events are available after this */
1008 	trace_init();
1009 
1010 	if (initcall_debug)
1011 		initcall_debug_enable();
1012 
1013 	context_tracking_init();
1014 	/* init some links before init_ISA_irqs() */
1015 	early_irq_init();
1016 	init_IRQ();
1017 	tick_init();
1018 	rcu_init_nohz();
1019 	timers_init();
1020 	srcu_init();
1021 	hrtimers_init();
1022 	softirq_init();
1023 	timekeeping_init();
1024 	time_init();
1025 
1026 	/* This must be after timekeeping is initialized */
1027 	random_init();
1028 
1029 	/* These make use of the fully initialized rng */
1030 	kfence_init();
1031 	boot_init_stack_canary();
1032 
1033 	perf_event_init();
1034 	profile_init();
1035 	call_function_init();
1036 	WARN(!irqs_disabled(), "Interrupts were enabled early\n");
1037 
1038 	early_boot_irqs_disabled = false;
1039 	local_irq_enable();
1040 
1041 	kmem_cache_init_late();
1042 
1043 	/*
1044 	 * HACK ALERT! This is early. We're enabling the console before
1045 	 * we've done PCI setups etc, and console_init() must be aware of
1046 	 * this. But we do want output early, in case something goes wrong.
1047 	 */
1048 	console_init();
1049 	if (panic_later)
1050 		panic("Too many boot %s vars at `%s'", panic_later,
1051 		      panic_param);
1052 
1053 	lockdep_init();
1054 
1055 	/*
1056 	 * Need to run this when irqs are enabled, because it wants
1057 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
1058 	 * too:
1059 	 */
1060 	locking_selftest();
1061 
1062 #ifdef CONFIG_BLK_DEV_INITRD
1063 	if (initrd_start && !initrd_below_start_ok &&
1064 	    page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
1065 		pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
1066 		    page_to_pfn(virt_to_page((void *)initrd_start)),
1067 		    min_low_pfn);
1068 		initrd_start = 0;
1069 	}
1070 #endif
1071 	setup_per_cpu_pageset();
1072 	numa_policy_init();
1073 	acpi_early_init();
1074 	if (late_time_init)
1075 		late_time_init();
1076 	sched_clock_init();
1077 	calibrate_delay();
1078 
1079 	arch_cpu_finalize_init();
1080 
1081 	pid_idr_init();
1082 	anon_vma_init();
1083 	thread_stack_cache_init();
1084 	cred_init();
1085 	fork_init();
1086 	proc_caches_init();
1087 	uts_ns_init();
1088 	time_ns_init();
1089 	key_init();
1090 	security_init();
1091 	dbg_late_init();
1092 	net_ns_init();
1093 	vfs_caches_init();
1094 	pagecache_init();
1095 	signals_init();
1096 	seq_file_init();
1097 	proc_root_init();
1098 	nsfs_init();
1099 	pidfs_init();
1100 	cpuset_init();
1101 	mem_cgroup_init();
1102 	cgroup_init();
1103 	taskstats_init_early();
1104 	delayacct_init();
1105 
1106 	acpi_subsystem_init();
1107 	arch_post_acpi_subsys_init();
1108 	kcsan_init();
1109 
1110 	/* Do the rest non-__init'ed, we're now alive */
1111 	rest_init();
1112 
1113 	/*
1114 	 * Avoid stack canaries in callers of boot_init_stack_canary for gcc-10
1115 	 * and older.
1116 	 */
1117 #if !__has_attribute(__no_stack_protector__)
1118 	prevent_tail_call_optimization();
1119 #endif
1120 }
1121 
1122 /* Call all constructor functions linked into the kernel. */
do_ctors(void)1123 static void __init do_ctors(void)
1124 {
1125 /*
1126  * For UML, the constructors have already been called by the
1127  * normal setup code as it's just a normal ELF binary, so we
1128  * cannot do it again - but we do need CONFIG_CONSTRUCTORS
1129  * even on UML for modules.
1130  */
1131 #if defined(CONFIG_CONSTRUCTORS) && !defined(CONFIG_UML)
1132 	ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
1133 
1134 	for (; fn < (ctor_fn_t *) __ctors_end; fn++)
1135 		(*fn)();
1136 #endif
1137 }
1138 
1139 #ifdef CONFIG_KALLSYMS
1140 struct blacklist_entry {
1141 	struct list_head next;
1142 	char *buf;
1143 };
1144 
1145 static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
1146 
initcall_blacklist(char * str)1147 static int __init initcall_blacklist(char *str)
1148 {
1149 	char *str_entry;
1150 	struct blacklist_entry *entry;
1151 
1152 	/* str argument is a comma-separated list of functions */
1153 	do {
1154 		str_entry = strsep(&str, ",");
1155 		if (str_entry) {
1156 			pr_debug("blacklisting initcall %s\n", str_entry);
1157 			entry = memblock_alloc_or_panic(sizeof(*entry),
1158 					       SMP_CACHE_BYTES);
1159 			entry->buf = memblock_alloc_or_panic(strlen(str_entry) + 1,
1160 						    SMP_CACHE_BYTES);
1161 			strcpy(entry->buf, str_entry);
1162 			list_add(&entry->next, &blacklisted_initcalls);
1163 		}
1164 	} while (str_entry);
1165 
1166 	return 1;
1167 }
1168 
initcall_blacklisted(initcall_t fn)1169 static bool __init_or_module initcall_blacklisted(initcall_t fn)
1170 {
1171 	struct blacklist_entry *entry;
1172 	char fn_name[KSYM_SYMBOL_LEN];
1173 	unsigned long addr;
1174 
1175 	if (list_empty(&blacklisted_initcalls))
1176 		return false;
1177 
1178 	addr = (unsigned long) dereference_function_descriptor(fn);
1179 	sprint_symbol_no_offset(fn_name, addr);
1180 
1181 	/*
1182 	 * fn will be "function_name [module_name]" where [module_name] is not
1183 	 * displayed for built-in init functions.  Strip off the [module_name].
1184 	 */
1185 	strreplace(fn_name, ' ', '\0');
1186 
1187 	list_for_each_entry(entry, &blacklisted_initcalls, next) {
1188 		if (!strcmp(fn_name, entry->buf)) {
1189 			pr_debug("initcall %s blacklisted\n", fn_name);
1190 			return true;
1191 		}
1192 	}
1193 
1194 	return false;
1195 }
1196 #else
initcall_blacklist(char * str)1197 static int __init initcall_blacklist(char *str)
1198 {
1199 	pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
1200 	return 0;
1201 }
1202 
initcall_blacklisted(initcall_t fn)1203 static bool __init_or_module initcall_blacklisted(initcall_t fn)
1204 {
1205 	return false;
1206 }
1207 #endif
1208 __setup("initcall_blacklist=", initcall_blacklist);
1209 
1210 static __init_or_module void
trace_initcall_start_cb(void * data,initcall_t fn)1211 trace_initcall_start_cb(void *data, initcall_t fn)
1212 {
1213 	ktime_t *calltime = data;
1214 
1215 	printk(KERN_DEBUG "calling  %pS @ %i\n", fn, task_pid_nr(current));
1216 	*calltime = ktime_get();
1217 }
1218 
1219 static __init_or_module void
trace_initcall_finish_cb(void * data,initcall_t fn,int ret)1220 trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
1221 {
1222 	ktime_t rettime, *calltime = data;
1223 
1224 	rettime = ktime_get();
1225 	printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
1226 		 fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime));
1227 }
1228 
1229 static __init_or_module void
trace_initcall_level_cb(void * data,const char * level)1230 trace_initcall_level_cb(void *data, const char *level)
1231 {
1232 	printk(KERN_DEBUG "entering initcall level: %s\n", level);
1233 }
1234 
1235 static ktime_t initcall_calltime;
1236 
1237 #ifdef TRACEPOINTS_ENABLED
initcall_debug_enable(void)1238 static void __init initcall_debug_enable(void)
1239 {
1240 	int ret;
1241 
1242 	ret = register_trace_initcall_start(trace_initcall_start_cb,
1243 					    &initcall_calltime);
1244 	ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
1245 					      &initcall_calltime);
1246 	ret |= register_trace_initcall_level(trace_initcall_level_cb, NULL);
1247 	WARN(ret, "Failed to register initcall tracepoints\n");
1248 }
1249 # define do_trace_initcall_start	trace_initcall_start
1250 # define do_trace_initcall_finish	trace_initcall_finish
1251 # define do_trace_initcall_level	trace_initcall_level
1252 #else
do_trace_initcall_start(initcall_t fn)1253 static inline void do_trace_initcall_start(initcall_t fn)
1254 {
1255 	if (!initcall_debug)
1256 		return;
1257 	trace_initcall_start_cb(&initcall_calltime, fn);
1258 }
do_trace_initcall_finish(initcall_t fn,int ret)1259 static inline void do_trace_initcall_finish(initcall_t fn, int ret)
1260 {
1261 	if (!initcall_debug)
1262 		return;
1263 	trace_initcall_finish_cb(&initcall_calltime, fn, ret);
1264 }
do_trace_initcall_level(const char * level)1265 static inline void do_trace_initcall_level(const char *level)
1266 {
1267 	if (!initcall_debug)
1268 		return;
1269 	trace_initcall_level_cb(NULL, level);
1270 }
1271 #endif /* !TRACEPOINTS_ENABLED */
1272 
do_one_initcall(initcall_t fn)1273 int __init_or_module do_one_initcall(initcall_t fn)
1274 {
1275 	int count = preempt_count();
1276 	char msgbuf[64];
1277 	int ret;
1278 
1279 	if (initcall_blacklisted(fn))
1280 		return -EPERM;
1281 
1282 	do_trace_initcall_start(fn);
1283 	ret = fn();
1284 	do_trace_initcall_finish(fn, ret);
1285 
1286 	msgbuf[0] = 0;
1287 
1288 	if (preempt_count() != count) {
1289 		sprintf(msgbuf, "preemption imbalance ");
1290 		preempt_count_set(count);
1291 	}
1292 	if (irqs_disabled()) {
1293 		strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
1294 		local_irq_enable();
1295 	}
1296 	WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
1297 
1298 	add_latent_entropy();
1299 	return ret;
1300 }
1301 
1302 
1303 static initcall_entry_t *initcall_levels[] __initdata = {
1304 	__initcall0_start,
1305 	__initcall1_start,
1306 	__initcall2_start,
1307 	__initcall3_start,
1308 	__initcall4_start,
1309 	__initcall5_start,
1310 	__initcall6_start,
1311 	__initcall7_start,
1312 	__initcall_end,
1313 };
1314 
1315 /* Keep these in sync with initcalls in include/linux/init.h */
1316 static const char *initcall_level_names[] __initdata = {
1317 	"pure",
1318 	"core",
1319 	"postcore",
1320 	"arch",
1321 	"subsys",
1322 	"fs",
1323 	"device",
1324 	"late",
1325 };
1326 
ignore_unknown_bootoption(char * param,char * val,const char * unused,void * arg)1327 static int __init ignore_unknown_bootoption(char *param, char *val,
1328 			       const char *unused, void *arg)
1329 {
1330 	return 0;
1331 }
1332 
do_initcall_level(int level,char * command_line)1333 static void __init do_initcall_level(int level, char *command_line)
1334 {
1335 	initcall_entry_t *fn;
1336 
1337 	parse_args(initcall_level_names[level],
1338 		   command_line, __start___param,
1339 		   __stop___param - __start___param,
1340 		   level, level,
1341 		   NULL, ignore_unknown_bootoption);
1342 
1343 	do_trace_initcall_level(initcall_level_names[level]);
1344 	for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
1345 		do_one_initcall(initcall_from_entry(fn));
1346 }
1347 
do_initcalls(void)1348 static void __init do_initcalls(void)
1349 {
1350 	int level;
1351 	size_t len = saved_command_line_len + 1;
1352 	char *command_line;
1353 
1354 	command_line = kzalloc(len, GFP_KERNEL);
1355 	if (!command_line)
1356 		panic("%s: Failed to allocate %zu bytes\n", __func__, len);
1357 
1358 	for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
1359 		/* Parser modifies command_line, restore it each time */
1360 		strcpy(command_line, saved_command_line);
1361 		do_initcall_level(level, command_line);
1362 	}
1363 
1364 	kfree(command_line);
1365 }
1366 
1367 /*
1368  * Ok, the machine is now initialized. None of the devices
1369  * have been touched yet, but the CPU subsystem is up and
1370  * running, and memory and process management works.
1371  *
1372  * Now we can finally start doing some real work..
1373  */
do_basic_setup(void)1374 static void __init do_basic_setup(void)
1375 {
1376 	cpuset_init_smp();
1377 	driver_init();
1378 	init_irq_proc();
1379 	do_ctors();
1380 	do_initcalls();
1381 }
1382 
do_pre_smp_initcalls(void)1383 static void __init do_pre_smp_initcalls(void)
1384 {
1385 	initcall_entry_t *fn;
1386 
1387 	do_trace_initcall_level("early");
1388 	for (fn = __initcall_start; fn < __initcall0_start; fn++)
1389 		do_one_initcall(initcall_from_entry(fn));
1390 }
1391 
run_init_process(const char * init_filename)1392 static int run_init_process(const char *init_filename)
1393 {
1394 	const char *const *p;
1395 
1396 	argv_init[0] = init_filename;
1397 	pr_info("Run %s as init process\n", init_filename);
1398 	pr_debug("  with arguments:\n");
1399 	for (p = argv_init; *p; p++)
1400 		pr_debug("    %s\n", *p);
1401 	pr_debug("  with environment:\n");
1402 	for (p = envp_init; *p; p++)
1403 		pr_debug("    %s\n", *p);
1404 	return kernel_execve(init_filename, argv_init, envp_init);
1405 }
1406 
try_to_run_init_process(const char * init_filename)1407 static int try_to_run_init_process(const char *init_filename)
1408 {
1409 	int ret;
1410 
1411 	ret = run_init_process(init_filename);
1412 
1413 	if (ret && ret != -ENOENT) {
1414 		pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1415 		       init_filename, ret);
1416 	}
1417 
1418 	return ret;
1419 }
1420 
1421 static noinline void __init kernel_init_freeable(void);
1422 
1423 #if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1424 bool rodata_enabled __ro_after_init = true;
1425 
1426 #ifndef arch_parse_debug_rodata
arch_parse_debug_rodata(char * str)1427 static inline bool arch_parse_debug_rodata(char *str) { return false; }
1428 #endif
1429 
set_debug_rodata(char * str)1430 static int __init set_debug_rodata(char *str)
1431 {
1432 	if (arch_parse_debug_rodata(str))
1433 		return 0;
1434 
1435 	if (str && !strcmp(str, "on"))
1436 		rodata_enabled = true;
1437 	else if (str && !strcmp(str, "off"))
1438 		rodata_enabled = false;
1439 	else
1440 		pr_warn("Invalid option string for rodata: '%s'\n", str);
1441 	return 0;
1442 }
1443 early_param("rodata", set_debug_rodata);
1444 #endif
1445 
mark_readonly(void)1446 static void mark_readonly(void)
1447 {
1448 	if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX) && rodata_enabled) {
1449 		/*
1450 		 * load_module() results in W+X mappings, which are cleaned
1451 		 * up with init_free_wq. Let's make sure that queued work is
1452 		 * flushed so that we don't hit false positives looking for
1453 		 * insecure pages which are W+X.
1454 		 */
1455 		flush_module_init_free_work();
1456 		jump_label_init_ro();
1457 		mark_rodata_ro();
1458 		debug_checkwx();
1459 		rodata_test();
1460 	} else if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) {
1461 		pr_info("Kernel memory protection disabled.\n");
1462 	} else if (IS_ENABLED(CONFIG_ARCH_HAS_STRICT_KERNEL_RWX)) {
1463 		pr_warn("Kernel memory protection not selected by kernel config.\n");
1464 	} else {
1465 		pr_warn("This architecture does not have kernel memory protection.\n");
1466 	}
1467 }
1468 
free_initmem(void)1469 void __weak free_initmem(void)
1470 {
1471 	free_initmem_default(POISON_FREE_INITMEM);
1472 }
1473 
kernel_init(void * unused)1474 static int __ref kernel_init(void *unused)
1475 {
1476 	int ret;
1477 
1478 	/*
1479 	 * Wait until kthreadd is all set-up.
1480 	 */
1481 	wait_for_completion(&kthreadd_done);
1482 
1483 	kernel_init_freeable();
1484 	/* need to finish all async __init code before freeing the memory */
1485 	async_synchronize_full();
1486 
1487 	system_state = SYSTEM_FREEING_INITMEM;
1488 	kprobe_free_init_mem();
1489 	ftrace_free_init_mem();
1490 	kgdb_free_init_mem();
1491 	exit_boot_config();
1492 	free_initmem();
1493 	mark_readonly();
1494 
1495 	/*
1496 	 * Kernel mappings are now finalized - update the userspace page-table
1497 	 * to finalize PTI.
1498 	 */
1499 	pti_finalize();
1500 
1501 	system_state = SYSTEM_RUNNING;
1502 	numa_default_policy();
1503 
1504 	rcu_end_inkernel_boot();
1505 
1506 	do_sysctl_args();
1507 
1508 	if (ramdisk_execute_command) {
1509 		ret = run_init_process(ramdisk_execute_command);
1510 		if (!ret)
1511 			return 0;
1512 		pr_err("Failed to execute %s (error %d)\n",
1513 		       ramdisk_execute_command, ret);
1514 	}
1515 
1516 	/*
1517 	 * We try each of these until one succeeds.
1518 	 *
1519 	 * The Bourne shell can be used instead of init if we are
1520 	 * trying to recover a really broken machine.
1521 	 */
1522 	if (execute_command) {
1523 		ret = run_init_process(execute_command);
1524 		if (!ret)
1525 			return 0;
1526 		panic("Requested init %s failed (error %d).",
1527 		      execute_command, ret);
1528 	}
1529 
1530 	if (CONFIG_DEFAULT_INIT[0] != '\0') {
1531 		ret = run_init_process(CONFIG_DEFAULT_INIT);
1532 		if (ret)
1533 			pr_err("Default init %s failed (error %d)\n",
1534 			       CONFIG_DEFAULT_INIT, ret);
1535 		else
1536 			return 0;
1537 	}
1538 
1539 	if (!try_to_run_init_process("/sbin/init") ||
1540 	    !try_to_run_init_process("/etc/init") ||
1541 	    !try_to_run_init_process("/bin/init") ||
1542 	    !try_to_run_init_process("/bin/sh"))
1543 		return 0;
1544 
1545 	panic("No working init found.  Try passing init= option to kernel. "
1546 	      "See Linux Documentation/admin-guide/init.rst for guidance.");
1547 }
1548 
1549 /* Open /dev/console, for stdin/stdout/stderr, this should never fail */
console_on_rootfs(void)1550 void __init console_on_rootfs(void)
1551 {
1552 	struct file *file = filp_open("/dev/console", O_RDWR, 0);
1553 
1554 	if (IS_ERR(file)) {
1555 		pr_err("Warning: unable to open an initial console.\n");
1556 		return;
1557 	}
1558 	init_dup(file);
1559 	init_dup(file);
1560 	init_dup(file);
1561 	fput(file);
1562 }
1563 
kernel_init_freeable(void)1564 static noinline void __init kernel_init_freeable(void)
1565 {
1566 	/* Now the scheduler is fully set up and can do blocking allocations */
1567 	gfp_allowed_mask = __GFP_BITS_MASK;
1568 
1569 	/*
1570 	 * init can allocate pages on any node
1571 	 */
1572 	set_mems_allowed(node_states[N_MEMORY]);
1573 
1574 	cad_pid = get_pid(task_pid(current));
1575 
1576 	smp_prepare_cpus(setup_max_cpus);
1577 
1578 	workqueue_init();
1579 
1580 	init_mm_internals();
1581 
1582 	do_pre_smp_initcalls();
1583 	lockup_detector_init();
1584 
1585 	smp_init();
1586 	sched_init_smp();
1587 
1588 	workqueue_init_topology();
1589 	async_init();
1590 	padata_init();
1591 	page_alloc_init_late();
1592 
1593 	do_basic_setup();
1594 
1595 	kunit_run_all_tests();
1596 
1597 	wait_for_initramfs();
1598 	console_on_rootfs();
1599 
1600 	/*
1601 	 * check if there is an early userspace init.  If yes, let it do all
1602 	 * the work
1603 	 */
1604 	int ramdisk_command_access;
1605 	ramdisk_command_access = init_eaccess(ramdisk_execute_command);
1606 	if (ramdisk_command_access != 0) {
1607 		pr_warn("check access for rdinit=%s failed: %i, ignoring\n",
1608 			ramdisk_execute_command, ramdisk_command_access);
1609 		ramdisk_execute_command = NULL;
1610 		prepare_namespace();
1611 	}
1612 
1613 	/*
1614 	 * Ok, we have completed the initial bootup, and
1615 	 * we're essentially up and running. Get rid of the
1616 	 * initmem segments and start the user-mode stuff..
1617 	 *
1618 	 * rootfs is available now, try loading the public keys
1619 	 * and default modules
1620 	 */
1621 
1622 	integrity_load_keys();
1623 }
1624