xref: /linux/init/initramfs.c (revision b77e0ce62d63a761ffb7f7245a215a49f5921c2f)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/init.h>
3 #include <linux/fs.h>
4 #include <linux/slab.h>
5 #include <linux/types.h>
6 #include <linux/fcntl.h>
7 #include <linux/delay.h>
8 #include <linux/string.h>
9 #include <linux/dirent.h>
10 #include <linux/syscalls.h>
11 #include <linux/utime.h>
12 #include <linux/file.h>
13 #include <linux/memblock.h>
14 #include <linux/mm.h>
15 #include <linux/namei.h>
16 #include <linux/init_syscalls.h>
17 
18 static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
19 		loff_t *pos)
20 {
21 	ssize_t out = 0;
22 
23 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
24 	while (count) {
25 		ssize_t rv = kernel_write(file, p, count, pos);
26 
27 		if (rv < 0) {
28 			if (rv == -EINTR || rv == -EAGAIN)
29 				continue;
30 			return out ? out : rv;
31 		} else if (rv == 0)
32 			break;
33 
34 		p += rv;
35 		out += rv;
36 		count -= rv;
37 	}
38 
39 	return out;
40 }
41 
42 static __initdata char *message;
43 static void __init error(char *x)
44 {
45 	if (!message)
46 		message = x;
47 }
48 
49 static void panic_show_mem(const char *fmt, ...)
50 {
51 	va_list args;
52 
53 	show_mem(0, NULL);
54 	va_start(args, fmt);
55 	panic(fmt, args);
56 	va_end(args);
57 }
58 
59 /* link hash */
60 
61 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
62 
63 static __initdata struct hash {
64 	int ino, minor, major;
65 	umode_t mode;
66 	struct hash *next;
67 	char name[N_ALIGN(PATH_MAX)];
68 } *head[32];
69 
70 static inline int hash(int major, int minor, int ino)
71 {
72 	unsigned long tmp = ino + minor + (major << 3);
73 	tmp += tmp >> 5;
74 	return tmp & 31;
75 }
76 
77 static char __init *find_link(int major, int minor, int ino,
78 			      umode_t mode, char *name)
79 {
80 	struct hash **p, *q;
81 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
82 		if ((*p)->ino != ino)
83 			continue;
84 		if ((*p)->minor != minor)
85 			continue;
86 		if ((*p)->major != major)
87 			continue;
88 		if (((*p)->mode ^ mode) & S_IFMT)
89 			continue;
90 		return (*p)->name;
91 	}
92 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
93 	if (!q)
94 		panic_show_mem("can't allocate link hash entry");
95 	q->major = major;
96 	q->minor = minor;
97 	q->ino = ino;
98 	q->mode = mode;
99 	strcpy(q->name, name);
100 	q->next = NULL;
101 	*p = q;
102 	return NULL;
103 }
104 
105 static void __init free_hash(void)
106 {
107 	struct hash **p, *q;
108 	for (p = head; p < head + 32; p++) {
109 		while (*p) {
110 			q = *p;
111 			*p = q->next;
112 			kfree(q);
113 		}
114 	}
115 }
116 
117 static long __init do_utime(char *filename, time64_t mtime)
118 {
119 	struct timespec64 t[2];
120 
121 	t[0].tv_sec = mtime;
122 	t[0].tv_nsec = 0;
123 	t[1].tv_sec = mtime;
124 	t[1].tv_nsec = 0;
125 	return init_utimes(filename, t);
126 }
127 
128 static __initdata LIST_HEAD(dir_list);
129 struct dir_entry {
130 	struct list_head list;
131 	char *name;
132 	time64_t mtime;
133 };
134 
135 static void __init dir_add(const char *name, time64_t mtime)
136 {
137 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
138 	if (!de)
139 		panic_show_mem("can't allocate dir_entry buffer");
140 	INIT_LIST_HEAD(&de->list);
141 	de->name = kstrdup(name, GFP_KERNEL);
142 	de->mtime = mtime;
143 	list_add(&de->list, &dir_list);
144 }
145 
146 static void __init dir_utime(void)
147 {
148 	struct dir_entry *de, *tmp;
149 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
150 		list_del(&de->list);
151 		do_utime(de->name, de->mtime);
152 		kfree(de->name);
153 		kfree(de);
154 	}
155 }
156 
157 static __initdata time64_t mtime;
158 
159 /* cpio header parsing */
160 
161 static __initdata unsigned long ino, major, minor, nlink;
162 static __initdata umode_t mode;
163 static __initdata unsigned long body_len, name_len;
164 static __initdata uid_t uid;
165 static __initdata gid_t gid;
166 static __initdata unsigned rdev;
167 
168 static void __init parse_header(char *s)
169 {
170 	unsigned long parsed[12];
171 	char buf[9];
172 	int i;
173 
174 	buf[8] = '\0';
175 	for (i = 0, s += 6; i < 12; i++, s += 8) {
176 		memcpy(buf, s, 8);
177 		parsed[i] = simple_strtoul(buf, NULL, 16);
178 	}
179 	ino = parsed[0];
180 	mode = parsed[1];
181 	uid = parsed[2];
182 	gid = parsed[3];
183 	nlink = parsed[4];
184 	mtime = parsed[5]; /* breaks in y2106 */
185 	body_len = parsed[6];
186 	major = parsed[7];
187 	minor = parsed[8];
188 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
189 	name_len = parsed[11];
190 }
191 
192 /* FSM */
193 
194 static __initdata enum state {
195 	Start,
196 	Collect,
197 	GotHeader,
198 	SkipIt,
199 	GotName,
200 	CopyFile,
201 	GotSymlink,
202 	Reset
203 } state, next_state;
204 
205 static __initdata char *victim;
206 static unsigned long byte_count __initdata;
207 static __initdata loff_t this_header, next_header;
208 
209 static inline void __init eat(unsigned n)
210 {
211 	victim += n;
212 	this_header += n;
213 	byte_count -= n;
214 }
215 
216 static __initdata char *collected;
217 static long remains __initdata;
218 static __initdata char *collect;
219 
220 static void __init read_into(char *buf, unsigned size, enum state next)
221 {
222 	if (byte_count >= size) {
223 		collected = victim;
224 		eat(size);
225 		state = next;
226 	} else {
227 		collect = collected = buf;
228 		remains = size;
229 		next_state = next;
230 		state = Collect;
231 	}
232 }
233 
234 static __initdata char *header_buf, *symlink_buf, *name_buf;
235 
236 static int __init do_start(void)
237 {
238 	read_into(header_buf, 110, GotHeader);
239 	return 0;
240 }
241 
242 static int __init do_collect(void)
243 {
244 	unsigned long n = remains;
245 	if (byte_count < n)
246 		n = byte_count;
247 	memcpy(collect, victim, n);
248 	eat(n);
249 	collect += n;
250 	if ((remains -= n) != 0)
251 		return 1;
252 	state = next_state;
253 	return 0;
254 }
255 
256 static int __init do_header(void)
257 {
258 	if (memcmp(collected, "070707", 6)==0) {
259 		error("incorrect cpio method used: use -H newc option");
260 		return 1;
261 	}
262 	if (memcmp(collected, "070701", 6)) {
263 		error("no cpio magic");
264 		return 1;
265 	}
266 	parse_header(collected);
267 	next_header = this_header + N_ALIGN(name_len) + body_len;
268 	next_header = (next_header + 3) & ~3;
269 	state = SkipIt;
270 	if (name_len <= 0 || name_len > PATH_MAX)
271 		return 0;
272 	if (S_ISLNK(mode)) {
273 		if (body_len > PATH_MAX)
274 			return 0;
275 		collect = collected = symlink_buf;
276 		remains = N_ALIGN(name_len) + body_len;
277 		next_state = GotSymlink;
278 		state = Collect;
279 		return 0;
280 	}
281 	if (S_ISREG(mode) || !body_len)
282 		read_into(name_buf, N_ALIGN(name_len), GotName);
283 	return 0;
284 }
285 
286 static int __init do_skip(void)
287 {
288 	if (this_header + byte_count < next_header) {
289 		eat(byte_count);
290 		return 1;
291 	} else {
292 		eat(next_header - this_header);
293 		state = next_state;
294 		return 0;
295 	}
296 }
297 
298 static int __init do_reset(void)
299 {
300 	while (byte_count && *victim == '\0')
301 		eat(1);
302 	if (byte_count && (this_header & 3))
303 		error("broken padding");
304 	return 1;
305 }
306 
307 static void __init clean_path(char *path, umode_t fmode)
308 {
309 	struct kstat st;
310 
311 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
312 	    (st.mode ^ fmode) & S_IFMT) {
313 		if (S_ISDIR(st.mode))
314 			init_rmdir(path);
315 		else
316 			init_unlink(path);
317 	}
318 }
319 
320 static int __init maybe_link(void)
321 {
322 	if (nlink >= 2) {
323 		char *old = find_link(major, minor, ino, mode, collected);
324 		if (old) {
325 			clean_path(collected, 0);
326 			return (init_link(old, collected) < 0) ? -1 : 1;
327 		}
328 	}
329 	return 0;
330 }
331 
332 static __initdata struct file *wfile;
333 static __initdata loff_t wfile_pos;
334 
335 static int __init do_name(void)
336 {
337 	state = SkipIt;
338 	next_state = Reset;
339 	if (strcmp(collected, "TRAILER!!!") == 0) {
340 		free_hash();
341 		return 0;
342 	}
343 	clean_path(collected, mode);
344 	if (S_ISREG(mode)) {
345 		int ml = maybe_link();
346 		if (ml >= 0) {
347 			int openflags = O_WRONLY|O_CREAT;
348 			if (ml != 1)
349 				openflags |= O_TRUNC;
350 			wfile = filp_open(collected, openflags, mode);
351 			if (IS_ERR(wfile))
352 				return 0;
353 			wfile_pos = 0;
354 
355 			vfs_fchown(wfile, uid, gid);
356 			vfs_fchmod(wfile, mode);
357 			if (body_len)
358 				vfs_truncate(&wfile->f_path, body_len);
359 			state = CopyFile;
360 		}
361 	} else if (S_ISDIR(mode)) {
362 		init_mkdir(collected, mode);
363 		init_chown(collected, uid, gid, 0);
364 		init_chmod(collected, mode);
365 		dir_add(collected, mtime);
366 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
367 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
368 		if (maybe_link() == 0) {
369 			init_mknod(collected, mode, rdev);
370 			init_chown(collected, uid, gid, 0);
371 			init_chmod(collected, mode);
372 			do_utime(collected, mtime);
373 		}
374 	}
375 	return 0;
376 }
377 
378 static int __init do_copy(void)
379 {
380 	if (byte_count >= body_len) {
381 		struct timespec64 t[2] = { };
382 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
383 			error("write error");
384 
385 		t[0].tv_sec = mtime;
386 		t[1].tv_sec = mtime;
387 		vfs_utimes(&wfile->f_path, t);
388 
389 		fput(wfile);
390 		eat(body_len);
391 		state = SkipIt;
392 		return 0;
393 	} else {
394 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
395 			error("write error");
396 		body_len -= byte_count;
397 		eat(byte_count);
398 		return 1;
399 	}
400 }
401 
402 static int __init do_symlink(void)
403 {
404 	collected[N_ALIGN(name_len) + body_len] = '\0';
405 	clean_path(collected, 0);
406 	init_symlink(collected + N_ALIGN(name_len), collected);
407 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
408 	do_utime(collected, mtime);
409 	state = SkipIt;
410 	next_state = Reset;
411 	return 0;
412 }
413 
414 static __initdata int (*actions[])(void) = {
415 	[Start]		= do_start,
416 	[Collect]	= do_collect,
417 	[GotHeader]	= do_header,
418 	[SkipIt]	= do_skip,
419 	[GotName]	= do_name,
420 	[CopyFile]	= do_copy,
421 	[GotSymlink]	= do_symlink,
422 	[Reset]		= do_reset,
423 };
424 
425 static long __init write_buffer(char *buf, unsigned long len)
426 {
427 	byte_count = len;
428 	victim = buf;
429 
430 	while (!actions[state]())
431 		;
432 	return len - byte_count;
433 }
434 
435 static long __init flush_buffer(void *bufv, unsigned long len)
436 {
437 	char *buf = (char *) bufv;
438 	long written;
439 	long origLen = len;
440 	if (message)
441 		return -1;
442 	while ((written = write_buffer(buf, len)) < len && !message) {
443 		char c = buf[written];
444 		if (c == '0') {
445 			buf += written;
446 			len -= written;
447 			state = Start;
448 		} else if (c == 0) {
449 			buf += written;
450 			len -= written;
451 			state = Reset;
452 		} else
453 			error("junk within compressed archive");
454 	}
455 	return origLen;
456 }
457 
458 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
459 
460 #include <linux/decompress/generic.h>
461 
462 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
463 {
464 	long written;
465 	decompress_fn decompress;
466 	const char *compress_name;
467 	static __initdata char msg_buf[64];
468 
469 	header_buf = kmalloc(110, GFP_KERNEL);
470 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
471 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
472 
473 	if (!header_buf || !symlink_buf || !name_buf)
474 		panic_show_mem("can't allocate buffers");
475 
476 	state = Start;
477 	this_header = 0;
478 	message = NULL;
479 	while (!message && len) {
480 		loff_t saved_offset = this_header;
481 		if (*buf == '0' && !(this_header & 3)) {
482 			state = Start;
483 			written = write_buffer(buf, len);
484 			buf += written;
485 			len -= written;
486 			continue;
487 		}
488 		if (!*buf) {
489 			buf++;
490 			len--;
491 			this_header++;
492 			continue;
493 		}
494 		this_header = 0;
495 		decompress = decompress_method(buf, len, &compress_name);
496 		pr_debug("Detected %s compressed data\n", compress_name);
497 		if (decompress) {
498 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
499 				   &my_inptr, error);
500 			if (res)
501 				error("decompressor failed");
502 		} else if (compress_name) {
503 			if (!message) {
504 				snprintf(msg_buf, sizeof msg_buf,
505 					 "compression method %s not configured",
506 					 compress_name);
507 				message = msg_buf;
508 			}
509 		} else
510 			error("invalid magic at start of compressed archive");
511 		if (state != Reset)
512 			error("junk at the end of compressed archive");
513 		this_header = saved_offset + my_inptr;
514 		buf += my_inptr;
515 		len -= my_inptr;
516 	}
517 	dir_utime();
518 	kfree(name_buf);
519 	kfree(symlink_buf);
520 	kfree(header_buf);
521 	return message;
522 }
523 
524 static int __initdata do_retain_initrd;
525 
526 static int __init retain_initrd_param(char *str)
527 {
528 	if (*str)
529 		return 0;
530 	do_retain_initrd = 1;
531 	return 1;
532 }
533 __setup("retain_initrd", retain_initrd_param);
534 
535 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
536 static int __init keepinitrd_setup(char *__unused)
537 {
538 	do_retain_initrd = 1;
539 	return 1;
540 }
541 __setup("keepinitrd", keepinitrd_setup);
542 #endif
543 
544 extern char __initramfs_start[];
545 extern unsigned long __initramfs_size;
546 #include <linux/initrd.h>
547 #include <linux/kexec.h>
548 
549 void __init reserve_initrd_mem(void)
550 {
551 	phys_addr_t start;
552 	unsigned long size;
553 
554 	/* Ignore the virtul address computed during device tree parsing */
555 	initrd_start = initrd_end = 0;
556 
557 	if (!phys_initrd_size)
558 		return;
559 	/*
560 	 * Round the memory region to page boundaries as per free_initrd_mem()
561 	 * This allows us to detect whether the pages overlapping the initrd
562 	 * are in use, but more importantly, reserves the entire set of pages
563 	 * as we don't want these pages allocated for other purposes.
564 	 */
565 	start = round_down(phys_initrd_start, PAGE_SIZE);
566 	size = phys_initrd_size + (phys_initrd_start - start);
567 	size = round_up(size, PAGE_SIZE);
568 
569 	if (!memblock_is_region_memory(start, size)) {
570 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
571 		       (u64)start, size);
572 		goto disable;
573 	}
574 
575 	if (memblock_is_region_reserved(start, size)) {
576 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
577 		       (u64)start, size);
578 		goto disable;
579 	}
580 
581 	memblock_reserve(start, size);
582 	/* Now convert initrd to virtual addresses */
583 	initrd_start = (unsigned long)__va(phys_initrd_start);
584 	initrd_end = initrd_start + phys_initrd_size;
585 	initrd_below_start_ok = 1;
586 
587 	return;
588 disable:
589 	pr_cont(" - disabling initrd\n");
590 	initrd_start = 0;
591 	initrd_end = 0;
592 }
593 
594 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
595 {
596 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
597 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
598 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
599 
600 	memblock_free(__pa(aligned_start), aligned_end - aligned_start);
601 #endif
602 
603 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
604 			"initrd");
605 }
606 
607 #ifdef CONFIG_KEXEC_CORE
608 static bool __init kexec_free_initrd(void)
609 {
610 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
611 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
612 
613 	/*
614 	 * If the initrd region is overlapped with crashkernel reserved region,
615 	 * free only memory that is not part of crashkernel region.
616 	 */
617 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
618 		return false;
619 
620 	/*
621 	 * Initialize initrd memory region since the kexec boot does not do.
622 	 */
623 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
624 	if (initrd_start < crashk_start)
625 		free_initrd_mem(initrd_start, crashk_start);
626 	if (initrd_end > crashk_end)
627 		free_initrd_mem(crashk_end, initrd_end);
628 	return true;
629 }
630 #else
631 static inline bool kexec_free_initrd(void)
632 {
633 	return false;
634 }
635 #endif /* CONFIG_KEXEC_CORE */
636 
637 #ifdef CONFIG_BLK_DEV_RAM
638 static void __init populate_initrd_image(char *err)
639 {
640 	ssize_t written;
641 	struct file *file;
642 	loff_t pos = 0;
643 
644 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
645 
646 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
647 			err);
648 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
649 	if (IS_ERR(file))
650 		return;
651 
652 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
653 			&pos);
654 	if (written != initrd_end - initrd_start)
655 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
656 		       written, initrd_end - initrd_start);
657 	fput(file);
658 }
659 #endif /* CONFIG_BLK_DEV_RAM */
660 
661 static int __init populate_rootfs(void)
662 {
663 	/* Load the built in initramfs */
664 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
665 	if (err)
666 		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
667 
668 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
669 		goto done;
670 
671 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
672 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
673 	else
674 		printk(KERN_INFO "Unpacking initramfs...\n");
675 
676 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
677 	if (err) {
678 #ifdef CONFIG_BLK_DEV_RAM
679 		populate_initrd_image(err);
680 #else
681 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
682 #endif
683 	}
684 
685 done:
686 	/*
687 	 * If the initrd region is overlapped with crashkernel reserved region,
688 	 * free only memory that is not part of crashkernel region.
689 	 */
690 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
691 		free_initrd_mem(initrd_start, initrd_end);
692 	initrd_start = 0;
693 	initrd_end = 0;
694 
695 	flush_delayed_fput();
696 	return 0;
697 }
698 rootfs_initcall(populate_rootfs);
699