xref: /linux/init/initramfs.c (revision 09e61efd884ca68a768717d60858f138685b161b)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/async.h>
4 #include <linux/fs.h>
5 #include <linux/slab.h>
6 #include <linux/types.h>
7 #include <linux/fcntl.h>
8 #include <linux/delay.h>
9 #include <linux/string.h>
10 #include <linux/dirent.h>
11 #include <linux/syscalls.h>
12 #include <linux/utime.h>
13 #include <linux/file.h>
14 #include <linux/memblock.h>
15 #include <linux/mm.h>
16 #include <linux/namei.h>
17 #include <linux/init_syscalls.h>
18 #include <linux/task_work.h>
19 #include <linux/umh.h>
20 
21 static __initdata bool csum_present;
22 static __initdata u32 io_csum;
23 
24 static ssize_t __init xwrite(struct file *file, const unsigned char *p,
25 		size_t count, loff_t *pos)
26 {
27 	ssize_t out = 0;
28 
29 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
30 	while (count) {
31 		ssize_t rv = kernel_write(file, p, count, pos);
32 
33 		if (rv < 0) {
34 			if (rv == -EINTR || rv == -EAGAIN)
35 				continue;
36 			return out ? out : rv;
37 		} else if (rv == 0)
38 			break;
39 
40 		if (csum_present) {
41 			ssize_t i;
42 
43 			for (i = 0; i < rv; i++)
44 				io_csum += p[i];
45 		}
46 
47 		p += rv;
48 		out += rv;
49 		count -= rv;
50 	}
51 
52 	return out;
53 }
54 
55 static __initdata char *message;
56 static void __init error(char *x)
57 {
58 	if (!message)
59 		message = x;
60 }
61 
62 static void panic_show_mem(const char *fmt, ...)
63 {
64 	va_list args;
65 
66 	show_mem(0, NULL);
67 	va_start(args, fmt);
68 	panic(fmt, args);
69 	va_end(args);
70 }
71 
72 /* link hash */
73 
74 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
75 
76 static __initdata struct hash {
77 	int ino, minor, major;
78 	umode_t mode;
79 	struct hash *next;
80 	char name[N_ALIGN(PATH_MAX)];
81 } *head[32];
82 
83 static inline int hash(int major, int minor, int ino)
84 {
85 	unsigned long tmp = ino + minor + (major << 3);
86 	tmp += tmp >> 5;
87 	return tmp & 31;
88 }
89 
90 static char __init *find_link(int major, int minor, int ino,
91 			      umode_t mode, char *name)
92 {
93 	struct hash **p, *q;
94 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
95 		if ((*p)->ino != ino)
96 			continue;
97 		if ((*p)->minor != minor)
98 			continue;
99 		if ((*p)->major != major)
100 			continue;
101 		if (((*p)->mode ^ mode) & S_IFMT)
102 			continue;
103 		return (*p)->name;
104 	}
105 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
106 	if (!q)
107 		panic_show_mem("can't allocate link hash entry");
108 	q->major = major;
109 	q->minor = minor;
110 	q->ino = ino;
111 	q->mode = mode;
112 	strcpy(q->name, name);
113 	q->next = NULL;
114 	*p = q;
115 	return NULL;
116 }
117 
118 static void __init free_hash(void)
119 {
120 	struct hash **p, *q;
121 	for (p = head; p < head + 32; p++) {
122 		while (*p) {
123 			q = *p;
124 			*p = q->next;
125 			kfree(q);
126 		}
127 	}
128 }
129 
130 #ifdef CONFIG_INITRAMFS_PRESERVE_MTIME
131 static void __init do_utime(char *filename, time64_t mtime)
132 {
133 	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
134 	init_utimes(filename, t);
135 }
136 
137 static void __init do_utime_path(const struct path *path, time64_t mtime)
138 {
139 	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
140 	vfs_utimes(path, t);
141 }
142 
143 static __initdata LIST_HEAD(dir_list);
144 struct dir_entry {
145 	struct list_head list;
146 	time64_t mtime;
147 	char name[];
148 };
149 
150 static void __init dir_add(const char *name, time64_t mtime)
151 {
152 	size_t nlen = strlen(name) + 1;
153 	struct dir_entry *de;
154 
155 	de = kmalloc(sizeof(struct dir_entry) + nlen, GFP_KERNEL);
156 	if (!de)
157 		panic_show_mem("can't allocate dir_entry buffer");
158 	INIT_LIST_HEAD(&de->list);
159 	strscpy(de->name, name, nlen);
160 	de->mtime = mtime;
161 	list_add(&de->list, &dir_list);
162 }
163 
164 static void __init dir_utime(void)
165 {
166 	struct dir_entry *de, *tmp;
167 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
168 		list_del(&de->list);
169 		do_utime(de->name, de->mtime);
170 		kfree(de);
171 	}
172 }
173 #else
174 static void __init do_utime(char *filename, time64_t mtime) {}
175 static void __init do_utime_path(const struct path *path, time64_t mtime) {}
176 static void __init dir_add(const char *name, time64_t mtime) {}
177 static void __init dir_utime(void) {}
178 #endif
179 
180 static __initdata time64_t mtime;
181 
182 /* cpio header parsing */
183 
184 static __initdata unsigned long ino, major, minor, nlink;
185 static __initdata umode_t mode;
186 static __initdata unsigned long body_len, name_len;
187 static __initdata uid_t uid;
188 static __initdata gid_t gid;
189 static __initdata unsigned rdev;
190 static __initdata u32 hdr_csum;
191 
192 static void __init parse_header(char *s)
193 {
194 	unsigned long parsed[13];
195 	char buf[9];
196 	int i;
197 
198 	buf[8] = '\0';
199 	for (i = 0, s += 6; i < 13; i++, s += 8) {
200 		memcpy(buf, s, 8);
201 		parsed[i] = simple_strtoul(buf, NULL, 16);
202 	}
203 	ino = parsed[0];
204 	mode = parsed[1];
205 	uid = parsed[2];
206 	gid = parsed[3];
207 	nlink = parsed[4];
208 	mtime = parsed[5]; /* breaks in y2106 */
209 	body_len = parsed[6];
210 	major = parsed[7];
211 	minor = parsed[8];
212 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
213 	name_len = parsed[11];
214 	hdr_csum = parsed[12];
215 }
216 
217 /* FSM */
218 
219 static __initdata enum state {
220 	Start,
221 	Collect,
222 	GotHeader,
223 	SkipIt,
224 	GotName,
225 	CopyFile,
226 	GotSymlink,
227 	Reset
228 } state, next_state;
229 
230 static __initdata char *victim;
231 static unsigned long byte_count __initdata;
232 static __initdata loff_t this_header, next_header;
233 
234 static inline void __init eat(unsigned n)
235 {
236 	victim += n;
237 	this_header += n;
238 	byte_count -= n;
239 }
240 
241 static __initdata char *collected;
242 static long remains __initdata;
243 static __initdata char *collect;
244 
245 static void __init read_into(char *buf, unsigned size, enum state next)
246 {
247 	if (byte_count >= size) {
248 		collected = victim;
249 		eat(size);
250 		state = next;
251 	} else {
252 		collect = collected = buf;
253 		remains = size;
254 		next_state = next;
255 		state = Collect;
256 	}
257 }
258 
259 static __initdata char *header_buf, *symlink_buf, *name_buf;
260 
261 static int __init do_start(void)
262 {
263 	read_into(header_buf, 110, GotHeader);
264 	return 0;
265 }
266 
267 static int __init do_collect(void)
268 {
269 	unsigned long n = remains;
270 	if (byte_count < n)
271 		n = byte_count;
272 	memcpy(collect, victim, n);
273 	eat(n);
274 	collect += n;
275 	if ((remains -= n) != 0)
276 		return 1;
277 	state = next_state;
278 	return 0;
279 }
280 
281 static int __init do_header(void)
282 {
283 	if (!memcmp(collected, "070701", 6)) {
284 		csum_present = false;
285 	} else if (!memcmp(collected, "070702", 6)) {
286 		csum_present = true;
287 	} else {
288 		if (memcmp(collected, "070707", 6) == 0)
289 			error("incorrect cpio method used: use -H newc option");
290 		else
291 			error("no cpio magic");
292 		return 1;
293 	}
294 	parse_header(collected);
295 	next_header = this_header + N_ALIGN(name_len) + body_len;
296 	next_header = (next_header + 3) & ~3;
297 	state = SkipIt;
298 	if (name_len <= 0 || name_len > PATH_MAX)
299 		return 0;
300 	if (S_ISLNK(mode)) {
301 		if (body_len > PATH_MAX)
302 			return 0;
303 		collect = collected = symlink_buf;
304 		remains = N_ALIGN(name_len) + body_len;
305 		next_state = GotSymlink;
306 		state = Collect;
307 		return 0;
308 	}
309 	if (S_ISREG(mode) || !body_len)
310 		read_into(name_buf, N_ALIGN(name_len), GotName);
311 	return 0;
312 }
313 
314 static int __init do_skip(void)
315 {
316 	if (this_header + byte_count < next_header) {
317 		eat(byte_count);
318 		return 1;
319 	} else {
320 		eat(next_header - this_header);
321 		state = next_state;
322 		return 0;
323 	}
324 }
325 
326 static int __init do_reset(void)
327 {
328 	while (byte_count && *victim == '\0')
329 		eat(1);
330 	if (byte_count && (this_header & 3))
331 		error("broken padding");
332 	return 1;
333 }
334 
335 static void __init clean_path(char *path, umode_t fmode)
336 {
337 	struct kstat st;
338 
339 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
340 	    (st.mode ^ fmode) & S_IFMT) {
341 		if (S_ISDIR(st.mode))
342 			init_rmdir(path);
343 		else
344 			init_unlink(path);
345 	}
346 }
347 
348 static int __init maybe_link(void)
349 {
350 	if (nlink >= 2) {
351 		char *old = find_link(major, minor, ino, mode, collected);
352 		if (old) {
353 			clean_path(collected, 0);
354 			return (init_link(old, collected) < 0) ? -1 : 1;
355 		}
356 	}
357 	return 0;
358 }
359 
360 static __initdata struct file *wfile;
361 static __initdata loff_t wfile_pos;
362 
363 static int __init do_name(void)
364 {
365 	state = SkipIt;
366 	next_state = Reset;
367 	if (strcmp(collected, "TRAILER!!!") == 0) {
368 		free_hash();
369 		return 0;
370 	}
371 	clean_path(collected, mode);
372 	if (S_ISREG(mode)) {
373 		int ml = maybe_link();
374 		if (ml >= 0) {
375 			int openflags = O_WRONLY|O_CREAT;
376 			if (ml != 1)
377 				openflags |= O_TRUNC;
378 			wfile = filp_open(collected, openflags, mode);
379 			if (IS_ERR(wfile))
380 				return 0;
381 			wfile_pos = 0;
382 			io_csum = 0;
383 
384 			vfs_fchown(wfile, uid, gid);
385 			vfs_fchmod(wfile, mode);
386 			if (body_len)
387 				vfs_truncate(&wfile->f_path, body_len);
388 			state = CopyFile;
389 		}
390 	} else if (S_ISDIR(mode)) {
391 		init_mkdir(collected, mode);
392 		init_chown(collected, uid, gid, 0);
393 		init_chmod(collected, mode);
394 		dir_add(collected, mtime);
395 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
396 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
397 		if (maybe_link() == 0) {
398 			init_mknod(collected, mode, rdev);
399 			init_chown(collected, uid, gid, 0);
400 			init_chmod(collected, mode);
401 			do_utime(collected, mtime);
402 		}
403 	}
404 	return 0;
405 }
406 
407 static int __init do_copy(void)
408 {
409 	if (byte_count >= body_len) {
410 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
411 			error("write error");
412 
413 		do_utime_path(&wfile->f_path, mtime);
414 		fput(wfile);
415 		if (csum_present && io_csum != hdr_csum)
416 			error("bad data checksum");
417 		eat(body_len);
418 		state = SkipIt;
419 		return 0;
420 	} else {
421 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
422 			error("write error");
423 		body_len -= byte_count;
424 		eat(byte_count);
425 		return 1;
426 	}
427 }
428 
429 static int __init do_symlink(void)
430 {
431 	collected[N_ALIGN(name_len) + body_len] = '\0';
432 	clean_path(collected, 0);
433 	init_symlink(collected + N_ALIGN(name_len), collected);
434 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
435 	do_utime(collected, mtime);
436 	state = SkipIt;
437 	next_state = Reset;
438 	return 0;
439 }
440 
441 static __initdata int (*actions[])(void) = {
442 	[Start]		= do_start,
443 	[Collect]	= do_collect,
444 	[GotHeader]	= do_header,
445 	[SkipIt]	= do_skip,
446 	[GotName]	= do_name,
447 	[CopyFile]	= do_copy,
448 	[GotSymlink]	= do_symlink,
449 	[Reset]		= do_reset,
450 };
451 
452 static long __init write_buffer(char *buf, unsigned long len)
453 {
454 	byte_count = len;
455 	victim = buf;
456 
457 	while (!actions[state]())
458 		;
459 	return len - byte_count;
460 }
461 
462 static long __init flush_buffer(void *bufv, unsigned long len)
463 {
464 	char *buf = bufv;
465 	long written;
466 	long origLen = len;
467 	if (message)
468 		return -1;
469 	while ((written = write_buffer(buf, len)) < len && !message) {
470 		char c = buf[written];
471 		if (c == '0') {
472 			buf += written;
473 			len -= written;
474 			state = Start;
475 		} else if (c == 0) {
476 			buf += written;
477 			len -= written;
478 			state = Reset;
479 		} else
480 			error("junk within compressed archive");
481 	}
482 	return origLen;
483 }
484 
485 static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
486 
487 #include <linux/decompress/generic.h>
488 
489 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
490 {
491 	long written;
492 	decompress_fn decompress;
493 	const char *compress_name;
494 	static __initdata char msg_buf[64];
495 
496 	header_buf = kmalloc(110, GFP_KERNEL);
497 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
498 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
499 
500 	if (!header_buf || !symlink_buf || !name_buf)
501 		panic_show_mem("can't allocate buffers");
502 
503 	state = Start;
504 	this_header = 0;
505 	message = NULL;
506 	while (!message && len) {
507 		loff_t saved_offset = this_header;
508 		if (*buf == '0' && !(this_header & 3)) {
509 			state = Start;
510 			written = write_buffer(buf, len);
511 			buf += written;
512 			len -= written;
513 			continue;
514 		}
515 		if (!*buf) {
516 			buf++;
517 			len--;
518 			this_header++;
519 			continue;
520 		}
521 		this_header = 0;
522 		decompress = decompress_method(buf, len, &compress_name);
523 		pr_debug("Detected %s compressed data\n", compress_name);
524 		if (decompress) {
525 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
526 				   &my_inptr, error);
527 			if (res)
528 				error("decompressor failed");
529 		} else if (compress_name) {
530 			if (!message) {
531 				snprintf(msg_buf, sizeof msg_buf,
532 					 "compression method %s not configured",
533 					 compress_name);
534 				message = msg_buf;
535 			}
536 		} else
537 			error("invalid magic at start of compressed archive");
538 		if (state != Reset)
539 			error("junk at the end of compressed archive");
540 		this_header = saved_offset + my_inptr;
541 		buf += my_inptr;
542 		len -= my_inptr;
543 	}
544 	dir_utime();
545 	kfree(name_buf);
546 	kfree(symlink_buf);
547 	kfree(header_buf);
548 	return message;
549 }
550 
551 static int __initdata do_retain_initrd;
552 
553 static int __init retain_initrd_param(char *str)
554 {
555 	if (*str)
556 		return 0;
557 	do_retain_initrd = 1;
558 	return 1;
559 }
560 __setup("retain_initrd", retain_initrd_param);
561 
562 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
563 static int __init keepinitrd_setup(char *__unused)
564 {
565 	do_retain_initrd = 1;
566 	return 1;
567 }
568 __setup("keepinitrd", keepinitrd_setup);
569 #endif
570 
571 static bool __initdata initramfs_async = true;
572 static int __init initramfs_async_setup(char *str)
573 {
574 	strtobool(str, &initramfs_async);
575 	return 1;
576 }
577 __setup("initramfs_async=", initramfs_async_setup);
578 
579 extern char __initramfs_start[];
580 extern unsigned long __initramfs_size;
581 #include <linux/initrd.h>
582 #include <linux/kexec.h>
583 
584 void __init reserve_initrd_mem(void)
585 {
586 	phys_addr_t start;
587 	unsigned long size;
588 
589 	/* Ignore the virtul address computed during device tree parsing */
590 	initrd_start = initrd_end = 0;
591 
592 	if (!phys_initrd_size)
593 		return;
594 	/*
595 	 * Round the memory region to page boundaries as per free_initrd_mem()
596 	 * This allows us to detect whether the pages overlapping the initrd
597 	 * are in use, but more importantly, reserves the entire set of pages
598 	 * as we don't want these pages allocated for other purposes.
599 	 */
600 	start = round_down(phys_initrd_start, PAGE_SIZE);
601 	size = phys_initrd_size + (phys_initrd_start - start);
602 	size = round_up(size, PAGE_SIZE);
603 
604 	if (!memblock_is_region_memory(start, size)) {
605 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
606 		       (u64)start, size);
607 		goto disable;
608 	}
609 
610 	if (memblock_is_region_reserved(start, size)) {
611 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
612 		       (u64)start, size);
613 		goto disable;
614 	}
615 
616 	memblock_reserve(start, size);
617 	/* Now convert initrd to virtual addresses */
618 	initrd_start = (unsigned long)__va(phys_initrd_start);
619 	initrd_end = initrd_start + phys_initrd_size;
620 	initrd_below_start_ok = 1;
621 
622 	return;
623 disable:
624 	pr_cont(" - disabling initrd\n");
625 	initrd_start = 0;
626 	initrd_end = 0;
627 }
628 
629 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
630 {
631 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
632 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
633 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
634 
635 	memblock_free((void *)aligned_start, aligned_end - aligned_start);
636 #endif
637 
638 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
639 			"initrd");
640 }
641 
642 #ifdef CONFIG_KEXEC_CORE
643 static bool __init kexec_free_initrd(void)
644 {
645 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
646 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
647 
648 	/*
649 	 * If the initrd region is overlapped with crashkernel reserved region,
650 	 * free only memory that is not part of crashkernel region.
651 	 */
652 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
653 		return false;
654 
655 	/*
656 	 * Initialize initrd memory region since the kexec boot does not do.
657 	 */
658 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
659 	if (initrd_start < crashk_start)
660 		free_initrd_mem(initrd_start, crashk_start);
661 	if (initrd_end > crashk_end)
662 		free_initrd_mem(crashk_end, initrd_end);
663 	return true;
664 }
665 #else
666 static inline bool kexec_free_initrd(void)
667 {
668 	return false;
669 }
670 #endif /* CONFIG_KEXEC_CORE */
671 
672 #ifdef CONFIG_BLK_DEV_RAM
673 static void __init populate_initrd_image(char *err)
674 {
675 	ssize_t written;
676 	struct file *file;
677 	loff_t pos = 0;
678 
679 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
680 
681 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
682 			err);
683 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
684 	if (IS_ERR(file))
685 		return;
686 
687 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
688 			&pos);
689 	if (written != initrd_end - initrd_start)
690 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
691 		       written, initrd_end - initrd_start);
692 	fput(file);
693 }
694 #endif /* CONFIG_BLK_DEV_RAM */
695 
696 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
697 {
698 	/* Load the built in initramfs */
699 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
700 	if (err)
701 		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
702 
703 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
704 		goto done;
705 
706 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
707 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
708 	else
709 		printk(KERN_INFO "Unpacking initramfs...\n");
710 
711 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
712 	if (err) {
713 #ifdef CONFIG_BLK_DEV_RAM
714 		populate_initrd_image(err);
715 #else
716 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
717 #endif
718 	}
719 
720 done:
721 	/*
722 	 * If the initrd region is overlapped with crashkernel reserved region,
723 	 * free only memory that is not part of crashkernel region.
724 	 */
725 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
726 		free_initrd_mem(initrd_start, initrd_end);
727 	initrd_start = 0;
728 	initrd_end = 0;
729 
730 	flush_delayed_fput();
731 	task_work_run();
732 }
733 
734 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
735 static async_cookie_t initramfs_cookie;
736 
737 void wait_for_initramfs(void)
738 {
739 	if (!initramfs_cookie) {
740 		/*
741 		 * Something before rootfs_initcall wants to access
742 		 * the filesystem/initramfs. Probably a bug. Make a
743 		 * note, avoid deadlocking the machine, and let the
744 		 * caller's access fail as it used to.
745 		 */
746 		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
747 		return;
748 	}
749 	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
750 }
751 EXPORT_SYMBOL_GPL(wait_for_initramfs);
752 
753 static int __init populate_rootfs(void)
754 {
755 	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
756 						 &initramfs_domain);
757 	usermodehelper_enable();
758 	if (!initramfs_async)
759 		wait_for_initramfs();
760 	return 0;
761 }
762 rootfs_initcall(populate_rootfs);
763