xref: /linux/init/initramfs.c (revision cea0f76a483d1270ac6f6513964e3e75193dda48)
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 
15 static ssize_t __init xwrite(int fd, const char *p, size_t count)
16 {
17 	ssize_t out = 0;
18 
19 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
20 	while (count) {
21 		ssize_t rv = ksys_write(fd, p, count);
22 
23 		if (rv < 0) {
24 			if (rv == -EINTR || rv == -EAGAIN)
25 				continue;
26 			return out ? out : rv;
27 		} else if (rv == 0)
28 			break;
29 
30 		p += rv;
31 		out += rv;
32 		count -= rv;
33 	}
34 
35 	return out;
36 }
37 
38 static __initdata char *message;
39 static void __init error(char *x)
40 {
41 	if (!message)
42 		message = x;
43 }
44 
45 /* link hash */
46 
47 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
48 
49 static __initdata struct hash {
50 	int ino, minor, major;
51 	umode_t mode;
52 	struct hash *next;
53 	char name[N_ALIGN(PATH_MAX)];
54 } *head[32];
55 
56 static inline int hash(int major, int minor, int ino)
57 {
58 	unsigned long tmp = ino + minor + (major << 3);
59 	tmp += tmp >> 5;
60 	return tmp & 31;
61 }
62 
63 static char __init *find_link(int major, int minor, int ino,
64 			      umode_t mode, char *name)
65 {
66 	struct hash **p, *q;
67 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
68 		if ((*p)->ino != ino)
69 			continue;
70 		if ((*p)->minor != minor)
71 			continue;
72 		if ((*p)->major != major)
73 			continue;
74 		if (((*p)->mode ^ mode) & S_IFMT)
75 			continue;
76 		return (*p)->name;
77 	}
78 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
79 	if (!q)
80 		panic("can't allocate link hash entry");
81 	q->major = major;
82 	q->minor = minor;
83 	q->ino = ino;
84 	q->mode = mode;
85 	strcpy(q->name, name);
86 	q->next = NULL;
87 	*p = q;
88 	return NULL;
89 }
90 
91 static void __init free_hash(void)
92 {
93 	struct hash **p, *q;
94 	for (p = head; p < head + 32; p++) {
95 		while (*p) {
96 			q = *p;
97 			*p = q->next;
98 			kfree(q);
99 		}
100 	}
101 }
102 
103 static long __init do_utime(char *filename, time64_t mtime)
104 {
105 	struct timespec64 t[2];
106 
107 	t[0].tv_sec = mtime;
108 	t[0].tv_nsec = 0;
109 	t[1].tv_sec = mtime;
110 	t[1].tv_nsec = 0;
111 
112 	return do_utimes(AT_FDCWD, filename, t, AT_SYMLINK_NOFOLLOW);
113 }
114 
115 static __initdata LIST_HEAD(dir_list);
116 struct dir_entry {
117 	struct list_head list;
118 	char *name;
119 	time64_t mtime;
120 };
121 
122 static void __init dir_add(const char *name, time64_t mtime)
123 {
124 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
125 	if (!de)
126 		panic("can't allocate dir_entry buffer");
127 	INIT_LIST_HEAD(&de->list);
128 	de->name = kstrdup(name, GFP_KERNEL);
129 	de->mtime = mtime;
130 	list_add(&de->list, &dir_list);
131 }
132 
133 static void __init dir_utime(void)
134 {
135 	struct dir_entry *de, *tmp;
136 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
137 		list_del(&de->list);
138 		do_utime(de->name, de->mtime);
139 		kfree(de->name);
140 		kfree(de);
141 	}
142 }
143 
144 static __initdata time64_t mtime;
145 
146 /* cpio header parsing */
147 
148 static __initdata unsigned long ino, major, minor, nlink;
149 static __initdata umode_t mode;
150 static __initdata unsigned long body_len, name_len;
151 static __initdata uid_t uid;
152 static __initdata gid_t gid;
153 static __initdata unsigned rdev;
154 
155 static void __init parse_header(char *s)
156 {
157 	unsigned long parsed[12];
158 	char buf[9];
159 	int i;
160 
161 	buf[8] = '\0';
162 	for (i = 0, s += 6; i < 12; i++, s += 8) {
163 		memcpy(buf, s, 8);
164 		parsed[i] = simple_strtoul(buf, NULL, 16);
165 	}
166 	ino = parsed[0];
167 	mode = parsed[1];
168 	uid = parsed[2];
169 	gid = parsed[3];
170 	nlink = parsed[4];
171 	mtime = parsed[5]; /* breaks in y2106 */
172 	body_len = parsed[6];
173 	major = parsed[7];
174 	minor = parsed[8];
175 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
176 	name_len = parsed[11];
177 }
178 
179 /* FSM */
180 
181 static __initdata enum state {
182 	Start,
183 	Collect,
184 	GotHeader,
185 	SkipIt,
186 	GotName,
187 	CopyFile,
188 	GotSymlink,
189 	Reset
190 } state, next_state;
191 
192 static __initdata char *victim;
193 static unsigned long byte_count __initdata;
194 static __initdata loff_t this_header, next_header;
195 
196 static inline void __init eat(unsigned n)
197 {
198 	victim += n;
199 	this_header += n;
200 	byte_count -= n;
201 }
202 
203 static __initdata char *vcollected;
204 static __initdata char *collected;
205 static long remains __initdata;
206 static __initdata char *collect;
207 
208 static void __init read_into(char *buf, unsigned size, enum state next)
209 {
210 	if (byte_count >= size) {
211 		collected = victim;
212 		eat(size);
213 		state = next;
214 	} else {
215 		collect = collected = buf;
216 		remains = size;
217 		next_state = next;
218 		state = Collect;
219 	}
220 }
221 
222 static __initdata char *header_buf, *symlink_buf, *name_buf;
223 
224 static int __init do_start(void)
225 {
226 	read_into(header_buf, 110, GotHeader);
227 	return 0;
228 }
229 
230 static int __init do_collect(void)
231 {
232 	unsigned long n = remains;
233 	if (byte_count < n)
234 		n = byte_count;
235 	memcpy(collect, victim, n);
236 	eat(n);
237 	collect += n;
238 	if ((remains -= n) != 0)
239 		return 1;
240 	state = next_state;
241 	return 0;
242 }
243 
244 static int __init do_header(void)
245 {
246 	if (memcmp(collected, "070707", 6)==0) {
247 		error("incorrect cpio method used: use -H newc option");
248 		return 1;
249 	}
250 	if (memcmp(collected, "070701", 6)) {
251 		error("no cpio magic");
252 		return 1;
253 	}
254 	parse_header(collected);
255 	next_header = this_header + N_ALIGN(name_len) + body_len;
256 	next_header = (next_header + 3) & ~3;
257 	state = SkipIt;
258 	if (name_len <= 0 || name_len > PATH_MAX)
259 		return 0;
260 	if (S_ISLNK(mode)) {
261 		if (body_len > PATH_MAX)
262 			return 0;
263 		collect = collected = symlink_buf;
264 		remains = N_ALIGN(name_len) + body_len;
265 		next_state = GotSymlink;
266 		state = Collect;
267 		return 0;
268 	}
269 	if (S_ISREG(mode) || !body_len)
270 		read_into(name_buf, N_ALIGN(name_len), GotName);
271 	return 0;
272 }
273 
274 static int __init do_skip(void)
275 {
276 	if (this_header + byte_count < next_header) {
277 		eat(byte_count);
278 		return 1;
279 	} else {
280 		eat(next_header - this_header);
281 		state = next_state;
282 		return 0;
283 	}
284 }
285 
286 static int __init do_reset(void)
287 {
288 	while (byte_count && *victim == '\0')
289 		eat(1);
290 	if (byte_count && (this_header & 3))
291 		error("broken padding");
292 	return 1;
293 }
294 
295 static void __init clean_path(char *path, umode_t fmode)
296 {
297 	struct kstat st;
298 
299 	if (!vfs_lstat(path, &st) && (st.mode ^ fmode) & S_IFMT) {
300 		if (S_ISDIR(st.mode))
301 			ksys_rmdir(path);
302 		else
303 			ksys_unlink(path);
304 	}
305 }
306 
307 static int __init maybe_link(void)
308 {
309 	if (nlink >= 2) {
310 		char *old = find_link(major, minor, ino, mode, collected);
311 		if (old) {
312 			clean_path(collected, 0);
313 			return (ksys_link(old, collected) < 0) ? -1 : 1;
314 		}
315 	}
316 	return 0;
317 }
318 
319 static __initdata int wfd;
320 
321 static int __init do_name(void)
322 {
323 	state = SkipIt;
324 	next_state = Reset;
325 	if (strcmp(collected, "TRAILER!!!") == 0) {
326 		free_hash();
327 		return 0;
328 	}
329 	clean_path(collected, mode);
330 	if (S_ISREG(mode)) {
331 		int ml = maybe_link();
332 		if (ml >= 0) {
333 			int openflags = O_WRONLY|O_CREAT;
334 			if (ml != 1)
335 				openflags |= O_TRUNC;
336 			wfd = ksys_open(collected, openflags, mode);
337 
338 			if (wfd >= 0) {
339 				ksys_fchown(wfd, uid, gid);
340 				ksys_fchmod(wfd, mode);
341 				if (body_len)
342 					ksys_ftruncate(wfd, body_len);
343 				vcollected = kstrdup(collected, GFP_KERNEL);
344 				state = CopyFile;
345 			}
346 		}
347 	} else if (S_ISDIR(mode)) {
348 		ksys_mkdir(collected, mode);
349 		ksys_chown(collected, uid, gid);
350 		ksys_chmod(collected, mode);
351 		dir_add(collected, mtime);
352 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
353 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
354 		if (maybe_link() == 0) {
355 			ksys_mknod(collected, mode, rdev);
356 			ksys_chown(collected, uid, gid);
357 			ksys_chmod(collected, mode);
358 			do_utime(collected, mtime);
359 		}
360 	}
361 	return 0;
362 }
363 
364 static int __init do_copy(void)
365 {
366 	if (byte_count >= body_len) {
367 		if (xwrite(wfd, victim, body_len) != body_len)
368 			error("write error");
369 		ksys_close(wfd);
370 		do_utime(vcollected, mtime);
371 		kfree(vcollected);
372 		eat(body_len);
373 		state = SkipIt;
374 		return 0;
375 	} else {
376 		if (xwrite(wfd, victim, byte_count) != byte_count)
377 			error("write error");
378 		body_len -= byte_count;
379 		eat(byte_count);
380 		return 1;
381 	}
382 }
383 
384 static int __init do_symlink(void)
385 {
386 	collected[N_ALIGN(name_len) + body_len] = '\0';
387 	clean_path(collected, 0);
388 	ksys_symlink(collected + N_ALIGN(name_len), collected);
389 	ksys_lchown(collected, uid, gid);
390 	do_utime(collected, mtime);
391 	state = SkipIt;
392 	next_state = Reset;
393 	return 0;
394 }
395 
396 static __initdata int (*actions[])(void) = {
397 	[Start]		= do_start,
398 	[Collect]	= do_collect,
399 	[GotHeader]	= do_header,
400 	[SkipIt]	= do_skip,
401 	[GotName]	= do_name,
402 	[CopyFile]	= do_copy,
403 	[GotSymlink]	= do_symlink,
404 	[Reset]		= do_reset,
405 };
406 
407 static long __init write_buffer(char *buf, unsigned long len)
408 {
409 	byte_count = len;
410 	victim = buf;
411 
412 	while (!actions[state]())
413 		;
414 	return len - byte_count;
415 }
416 
417 static long __init flush_buffer(void *bufv, unsigned long len)
418 {
419 	char *buf = (char *) bufv;
420 	long written;
421 	long origLen = len;
422 	if (message)
423 		return -1;
424 	while ((written = write_buffer(buf, len)) < len && !message) {
425 		char c = buf[written];
426 		if (c == '0') {
427 			buf += written;
428 			len -= written;
429 			state = Start;
430 		} else if (c == 0) {
431 			buf += written;
432 			len -= written;
433 			state = Reset;
434 		} else
435 			error("junk within compressed archive");
436 	}
437 	return origLen;
438 }
439 
440 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
441 
442 #include <linux/decompress/generic.h>
443 
444 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
445 {
446 	long written;
447 	decompress_fn decompress;
448 	const char *compress_name;
449 	static __initdata char msg_buf[64];
450 
451 	header_buf = kmalloc(110, GFP_KERNEL);
452 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
453 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
454 
455 	if (!header_buf || !symlink_buf || !name_buf)
456 		panic("can't allocate buffers");
457 
458 	state = Start;
459 	this_header = 0;
460 	message = NULL;
461 	while (!message && len) {
462 		loff_t saved_offset = this_header;
463 		if (*buf == '0' && !(this_header & 3)) {
464 			state = Start;
465 			written = write_buffer(buf, len);
466 			buf += written;
467 			len -= written;
468 			continue;
469 		}
470 		if (!*buf) {
471 			buf++;
472 			len--;
473 			this_header++;
474 			continue;
475 		}
476 		this_header = 0;
477 		decompress = decompress_method(buf, len, &compress_name);
478 		pr_debug("Detected %s compressed data\n", compress_name);
479 		if (decompress) {
480 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
481 				   &my_inptr, error);
482 			if (res)
483 				error("decompressor failed");
484 		} else if (compress_name) {
485 			if (!message) {
486 				snprintf(msg_buf, sizeof msg_buf,
487 					 "compression method %s not configured",
488 					 compress_name);
489 				message = msg_buf;
490 			}
491 		} else
492 			error("invalid magic at start of compressed archive");
493 		if (state != Reset)
494 			error("junk at the end of compressed archive");
495 		this_header = saved_offset + my_inptr;
496 		buf += my_inptr;
497 		len -= my_inptr;
498 	}
499 	dir_utime();
500 	kfree(name_buf);
501 	kfree(symlink_buf);
502 	kfree(header_buf);
503 	return message;
504 }
505 
506 static int __initdata do_retain_initrd;
507 
508 static int __init retain_initrd_param(char *str)
509 {
510 	if (*str)
511 		return 0;
512 	do_retain_initrd = 1;
513 	return 1;
514 }
515 __setup("retain_initrd", retain_initrd_param);
516 
517 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
518 static int __init keepinitrd_setup(char *__unused)
519 {
520 	do_retain_initrd = 1;
521 	return 1;
522 }
523 __setup("keepinitrd", keepinitrd_setup);
524 #endif
525 
526 extern char __initramfs_start[];
527 extern unsigned long __initramfs_size;
528 #include <linux/initrd.h>
529 #include <linux/kexec.h>
530 
531 void __weak free_initrd_mem(unsigned long start, unsigned long end)
532 {
533 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
534 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
535 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
536 
537 	memblock_free(__pa(aligned_start), aligned_end - aligned_start);
538 #endif
539 
540 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
541 			"initrd");
542 }
543 
544 #ifdef CONFIG_KEXEC_CORE
545 static bool __init kexec_free_initrd(void)
546 {
547 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
548 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
549 
550 	/*
551 	 * If the initrd region is overlapped with crashkernel reserved region,
552 	 * free only memory that is not part of crashkernel region.
553 	 */
554 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
555 		return false;
556 
557 	/*
558 	 * Initialize initrd memory region since the kexec boot does not do.
559 	 */
560 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
561 	if (initrd_start < crashk_start)
562 		free_initrd_mem(initrd_start, crashk_start);
563 	if (initrd_end > crashk_end)
564 		free_initrd_mem(crashk_end, initrd_end);
565 	return true;
566 }
567 #else
568 static inline bool kexec_free_initrd(void)
569 {
570 	return false;
571 }
572 #endif /* CONFIG_KEXEC_CORE */
573 
574 #ifdef CONFIG_BLK_DEV_RAM
575 #define BUF_SIZE 1024
576 static void __init clean_rootfs(void)
577 {
578 	int fd;
579 	void *buf;
580 	struct linux_dirent64 *dirp;
581 	int num;
582 
583 	fd = ksys_open("/", O_RDONLY, 0);
584 	WARN_ON(fd < 0);
585 	if (fd < 0)
586 		return;
587 	buf = kzalloc(BUF_SIZE, GFP_KERNEL);
588 	WARN_ON(!buf);
589 	if (!buf) {
590 		ksys_close(fd);
591 		return;
592 	}
593 
594 	dirp = buf;
595 	num = ksys_getdents64(fd, dirp, BUF_SIZE);
596 	while (num > 0) {
597 		while (num > 0) {
598 			struct kstat st;
599 			int ret;
600 
601 			ret = vfs_lstat(dirp->d_name, &st);
602 			WARN_ON_ONCE(ret);
603 			if (!ret) {
604 				if (S_ISDIR(st.mode))
605 					ksys_rmdir(dirp->d_name);
606 				else
607 					ksys_unlink(dirp->d_name);
608 			}
609 
610 			num -= dirp->d_reclen;
611 			dirp = (void *)dirp + dirp->d_reclen;
612 		}
613 		dirp = buf;
614 		memset(buf, 0, BUF_SIZE);
615 		num = ksys_getdents64(fd, dirp, BUF_SIZE);
616 	}
617 
618 	ksys_close(fd);
619 	kfree(buf);
620 }
621 #else
622 static inline void clean_rootfs(void)
623 {
624 }
625 #endif /* CONFIG_BLK_DEV_RAM */
626 
627 #ifdef CONFIG_BLK_DEV_RAM
628 static void __init populate_initrd_image(char *err)
629 {
630 	ssize_t written;
631 	int fd;
632 
633 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
634 
635 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
636 			err);
637 	fd = ksys_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
638 	if (fd < 0)
639 		return;
640 
641 	written = xwrite(fd, (char *)initrd_start, initrd_end - initrd_start);
642 	if (written != initrd_end - initrd_start)
643 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
644 		       written, initrd_end - initrd_start);
645 	ksys_close(fd);
646 }
647 #else
648 static void __init populate_initrd_image(char *err)
649 {
650 	printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
651 }
652 #endif /* CONFIG_BLK_DEV_RAM */
653 
654 static int __init populate_rootfs(void)
655 {
656 	/* Load the built in initramfs */
657 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
658 	if (err)
659 		panic("%s", err); /* Failed to decompress INTERNAL initramfs */
660 
661 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
662 		goto done;
663 
664 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
665 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
666 	else
667 		printk(KERN_INFO "Unpacking initramfs...\n");
668 
669 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
670 	if (err) {
671 		clean_rootfs();
672 		populate_initrd_image(err);
673 	}
674 
675 done:
676 	/*
677 	 * If the initrd region is overlapped with crashkernel reserved region,
678 	 * free only memory that is not part of crashkernel region.
679 	 */
680 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
681 		free_initrd_mem(initrd_start, initrd_end);
682 	initrd_start = 0;
683 	initrd_end = 0;
684 
685 	flush_delayed_fput();
686 	return 0;
687 }
688 rootfs_initcall(populate_rootfs);
689