xref: /openbmc/linux/init/initramfs.c (revision ae213c44)
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 
14 static ssize_t __init xwrite(int fd, const char *p, size_t count)
15 {
16 	ssize_t out = 0;
17 
18 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
19 	while (count) {
20 		ssize_t rv = ksys_write(fd, p, count);
21 
22 		if (rv < 0) {
23 			if (rv == -EINTR || rv == -EAGAIN)
24 				continue;
25 			return out ? out : rv;
26 		} else if (rv == 0)
27 			break;
28 
29 		p += rv;
30 		out += rv;
31 		count -= rv;
32 	}
33 
34 	return out;
35 }
36 
37 static __initdata char *message;
38 static void __init error(char *x)
39 {
40 	if (!message)
41 		message = x;
42 }
43 
44 /* link hash */
45 
46 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
47 
48 static __initdata struct hash {
49 	int ino, minor, major;
50 	umode_t mode;
51 	struct hash *next;
52 	char name[N_ALIGN(PATH_MAX)];
53 } *head[32];
54 
55 static inline int hash(int major, int minor, int ino)
56 {
57 	unsigned long tmp = ino + minor + (major << 3);
58 	tmp += tmp >> 5;
59 	return tmp & 31;
60 }
61 
62 static char __init *find_link(int major, int minor, int ino,
63 			      umode_t mode, char *name)
64 {
65 	struct hash **p, *q;
66 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
67 		if ((*p)->ino != ino)
68 			continue;
69 		if ((*p)->minor != minor)
70 			continue;
71 		if ((*p)->major != major)
72 			continue;
73 		if (((*p)->mode ^ mode) & S_IFMT)
74 			continue;
75 		return (*p)->name;
76 	}
77 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
78 	if (!q)
79 		panic("can't allocate link hash entry");
80 	q->major = major;
81 	q->minor = minor;
82 	q->ino = ino;
83 	q->mode = mode;
84 	strcpy(q->name, name);
85 	q->next = NULL;
86 	*p = q;
87 	return NULL;
88 }
89 
90 static void __init free_hash(void)
91 {
92 	struct hash **p, *q;
93 	for (p = head; p < head + 32; p++) {
94 		while (*p) {
95 			q = *p;
96 			*p = q->next;
97 			kfree(q);
98 		}
99 	}
100 }
101 
102 static long __init do_utime(char *filename, time64_t mtime)
103 {
104 	struct timespec64 t[2];
105 
106 	t[0].tv_sec = mtime;
107 	t[0].tv_nsec = 0;
108 	t[1].tv_sec = mtime;
109 	t[1].tv_nsec = 0;
110 
111 	return do_utimes(AT_FDCWD, filename, t, AT_SYMLINK_NOFOLLOW);
112 }
113 
114 static __initdata LIST_HEAD(dir_list);
115 struct dir_entry {
116 	struct list_head list;
117 	char *name;
118 	time64_t mtime;
119 };
120 
121 static void __init dir_add(const char *name, time64_t mtime)
122 {
123 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
124 	if (!de)
125 		panic("can't allocate dir_entry buffer");
126 	INIT_LIST_HEAD(&de->list);
127 	de->name = kstrdup(name, GFP_KERNEL);
128 	de->mtime = mtime;
129 	list_add(&de->list, &dir_list);
130 }
131 
132 static void __init dir_utime(void)
133 {
134 	struct dir_entry *de, *tmp;
135 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
136 		list_del(&de->list);
137 		do_utime(de->name, de->mtime);
138 		kfree(de->name);
139 		kfree(de);
140 	}
141 }
142 
143 static __initdata time64_t mtime;
144 
145 /* cpio header parsing */
146 
147 static __initdata unsigned long ino, major, minor, nlink;
148 static __initdata umode_t mode;
149 static __initdata unsigned long body_len, name_len;
150 static __initdata uid_t uid;
151 static __initdata gid_t gid;
152 static __initdata unsigned rdev;
153 
154 static void __init parse_header(char *s)
155 {
156 	unsigned long parsed[12];
157 	char buf[9];
158 	int i;
159 
160 	buf[8] = '\0';
161 	for (i = 0, s += 6; i < 12; i++, s += 8) {
162 		memcpy(buf, s, 8);
163 		parsed[i] = simple_strtoul(buf, NULL, 16);
164 	}
165 	ino = parsed[0];
166 	mode = parsed[1];
167 	uid = parsed[2];
168 	gid = parsed[3];
169 	nlink = parsed[4];
170 	mtime = parsed[5]; /* breaks in y2106 */
171 	body_len = parsed[6];
172 	major = parsed[7];
173 	minor = parsed[8];
174 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
175 	name_len = parsed[11];
176 }
177 
178 /* FSM */
179 
180 static __initdata enum state {
181 	Start,
182 	Collect,
183 	GotHeader,
184 	SkipIt,
185 	GotName,
186 	CopyFile,
187 	GotSymlink,
188 	Reset
189 } state, next_state;
190 
191 static __initdata char *victim;
192 static unsigned long byte_count __initdata;
193 static __initdata loff_t this_header, next_header;
194 
195 static inline void __init eat(unsigned n)
196 {
197 	victim += n;
198 	this_header += n;
199 	byte_count -= n;
200 }
201 
202 static __initdata char *vcollected;
203 static __initdata char *collected;
204 static long remains __initdata;
205 static __initdata char *collect;
206 
207 static void __init read_into(char *buf, unsigned size, enum state next)
208 {
209 	if (byte_count >= size) {
210 		collected = victim;
211 		eat(size);
212 		state = next;
213 	} else {
214 		collect = collected = buf;
215 		remains = size;
216 		next_state = next;
217 		state = Collect;
218 	}
219 }
220 
221 static __initdata char *header_buf, *symlink_buf, *name_buf;
222 
223 static int __init do_start(void)
224 {
225 	read_into(header_buf, 110, GotHeader);
226 	return 0;
227 }
228 
229 static int __init do_collect(void)
230 {
231 	unsigned long n = remains;
232 	if (byte_count < n)
233 		n = byte_count;
234 	memcpy(collect, victim, n);
235 	eat(n);
236 	collect += n;
237 	if ((remains -= n) != 0)
238 		return 1;
239 	state = next_state;
240 	return 0;
241 }
242 
243 static int __init do_header(void)
244 {
245 	if (memcmp(collected, "070707", 6)==0) {
246 		error("incorrect cpio method used: use -H newc option");
247 		return 1;
248 	}
249 	if (memcmp(collected, "070701", 6)) {
250 		error("no cpio magic");
251 		return 1;
252 	}
253 	parse_header(collected);
254 	next_header = this_header + N_ALIGN(name_len) + body_len;
255 	next_header = (next_header + 3) & ~3;
256 	state = SkipIt;
257 	if (name_len <= 0 || name_len > PATH_MAX)
258 		return 0;
259 	if (S_ISLNK(mode)) {
260 		if (body_len > PATH_MAX)
261 			return 0;
262 		collect = collected = symlink_buf;
263 		remains = N_ALIGN(name_len) + body_len;
264 		next_state = GotSymlink;
265 		state = Collect;
266 		return 0;
267 	}
268 	if (S_ISREG(mode) || !body_len)
269 		read_into(name_buf, N_ALIGN(name_len), GotName);
270 	return 0;
271 }
272 
273 static int __init do_skip(void)
274 {
275 	if (this_header + byte_count < next_header) {
276 		eat(byte_count);
277 		return 1;
278 	} else {
279 		eat(next_header - this_header);
280 		state = next_state;
281 		return 0;
282 	}
283 }
284 
285 static int __init do_reset(void)
286 {
287 	while (byte_count && *victim == '\0')
288 		eat(1);
289 	if (byte_count && (this_header & 3))
290 		error("broken padding");
291 	return 1;
292 }
293 
294 static void __init clean_path(char *path, umode_t fmode)
295 {
296 	struct kstat st;
297 
298 	if (!vfs_lstat(path, &st) && (st.mode ^ fmode) & S_IFMT) {
299 		if (S_ISDIR(st.mode))
300 			ksys_rmdir(path);
301 		else
302 			ksys_unlink(path);
303 	}
304 }
305 
306 static int __init maybe_link(void)
307 {
308 	if (nlink >= 2) {
309 		char *old = find_link(major, minor, ino, mode, collected);
310 		if (old) {
311 			clean_path(collected, 0);
312 			return (ksys_link(old, collected) < 0) ? -1 : 1;
313 		}
314 	}
315 	return 0;
316 }
317 
318 static __initdata int wfd;
319 
320 static int __init do_name(void)
321 {
322 	state = SkipIt;
323 	next_state = Reset;
324 	if (strcmp(collected, "TRAILER!!!") == 0) {
325 		free_hash();
326 		return 0;
327 	}
328 	clean_path(collected, mode);
329 	if (S_ISREG(mode)) {
330 		int ml = maybe_link();
331 		if (ml >= 0) {
332 			int openflags = O_WRONLY|O_CREAT;
333 			if (ml != 1)
334 				openflags |= O_TRUNC;
335 			wfd = ksys_open(collected, openflags, mode);
336 
337 			if (wfd >= 0) {
338 				ksys_fchown(wfd, uid, gid);
339 				ksys_fchmod(wfd, mode);
340 				if (body_len)
341 					ksys_ftruncate(wfd, body_len);
342 				vcollected = kstrdup(collected, GFP_KERNEL);
343 				state = CopyFile;
344 			}
345 		}
346 	} else if (S_ISDIR(mode)) {
347 		ksys_mkdir(collected, mode);
348 		ksys_chown(collected, uid, gid);
349 		ksys_chmod(collected, mode);
350 		dir_add(collected, mtime);
351 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
352 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
353 		if (maybe_link() == 0) {
354 			ksys_mknod(collected, mode, rdev);
355 			ksys_chown(collected, uid, gid);
356 			ksys_chmod(collected, mode);
357 			do_utime(collected, mtime);
358 		}
359 	}
360 	return 0;
361 }
362 
363 static int __init do_copy(void)
364 {
365 	if (byte_count >= body_len) {
366 		if (xwrite(wfd, victim, body_len) != body_len)
367 			error("write error");
368 		ksys_close(wfd);
369 		do_utime(vcollected, mtime);
370 		kfree(vcollected);
371 		eat(body_len);
372 		state = SkipIt;
373 		return 0;
374 	} else {
375 		if (xwrite(wfd, victim, byte_count) != byte_count)
376 			error("write error");
377 		body_len -= byte_count;
378 		eat(byte_count);
379 		return 1;
380 	}
381 }
382 
383 static int __init do_symlink(void)
384 {
385 	collected[N_ALIGN(name_len) + body_len] = '\0';
386 	clean_path(collected, 0);
387 	ksys_symlink(collected + N_ALIGN(name_len), collected);
388 	ksys_lchown(collected, uid, gid);
389 	do_utime(collected, mtime);
390 	state = SkipIt;
391 	next_state = Reset;
392 	return 0;
393 }
394 
395 static __initdata int (*actions[])(void) = {
396 	[Start]		= do_start,
397 	[Collect]	= do_collect,
398 	[GotHeader]	= do_header,
399 	[SkipIt]	= do_skip,
400 	[GotName]	= do_name,
401 	[CopyFile]	= do_copy,
402 	[GotSymlink]	= do_symlink,
403 	[Reset]		= do_reset,
404 };
405 
406 static long __init write_buffer(char *buf, unsigned long len)
407 {
408 	byte_count = len;
409 	victim = buf;
410 
411 	while (!actions[state]())
412 		;
413 	return len - byte_count;
414 }
415 
416 static long __init flush_buffer(void *bufv, unsigned long len)
417 {
418 	char *buf = (char *) bufv;
419 	long written;
420 	long origLen = len;
421 	if (message)
422 		return -1;
423 	while ((written = write_buffer(buf, len)) < len && !message) {
424 		char c = buf[written];
425 		if (c == '0') {
426 			buf += written;
427 			len -= written;
428 			state = Start;
429 		} else if (c == 0) {
430 			buf += written;
431 			len -= written;
432 			state = Reset;
433 		} else
434 			error("junk within compressed archive");
435 	}
436 	return origLen;
437 }
438 
439 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
440 
441 #include <linux/decompress/generic.h>
442 
443 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
444 {
445 	long written;
446 	decompress_fn decompress;
447 	const char *compress_name;
448 	static __initdata char msg_buf[64];
449 
450 	header_buf = kmalloc(110, GFP_KERNEL);
451 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
452 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
453 
454 	if (!header_buf || !symlink_buf || !name_buf)
455 		panic("can't allocate buffers");
456 
457 	state = Start;
458 	this_header = 0;
459 	message = NULL;
460 	while (!message && len) {
461 		loff_t saved_offset = this_header;
462 		if (*buf == '0' && !(this_header & 3)) {
463 			state = Start;
464 			written = write_buffer(buf, len);
465 			buf += written;
466 			len -= written;
467 			continue;
468 		}
469 		if (!*buf) {
470 			buf++;
471 			len--;
472 			this_header++;
473 			continue;
474 		}
475 		this_header = 0;
476 		decompress = decompress_method(buf, len, &compress_name);
477 		pr_debug("Detected %s compressed data\n", compress_name);
478 		if (decompress) {
479 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
480 				   &my_inptr, error);
481 			if (res)
482 				error("decompressor failed");
483 		} else if (compress_name) {
484 			if (!message) {
485 				snprintf(msg_buf, sizeof msg_buf,
486 					 "compression method %s not configured",
487 					 compress_name);
488 				message = msg_buf;
489 			}
490 		} else
491 			error("invalid magic at start of compressed archive");
492 		if (state != Reset)
493 			error("junk at the end of compressed archive");
494 		this_header = saved_offset + my_inptr;
495 		buf += my_inptr;
496 		len -= my_inptr;
497 	}
498 	dir_utime();
499 	kfree(name_buf);
500 	kfree(symlink_buf);
501 	kfree(header_buf);
502 	return message;
503 }
504 
505 static int __initdata do_retain_initrd;
506 
507 static int __init retain_initrd_param(char *str)
508 {
509 	if (*str)
510 		return 0;
511 	do_retain_initrd = 1;
512 	return 1;
513 }
514 __setup("retain_initrd", retain_initrd_param);
515 
516 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
517 static int __init keepinitrd_setup(char *__unused)
518 {
519 	do_retain_initrd = 1;
520 	return 1;
521 }
522 __setup("keepinitrd", keepinitrd_setup);
523 #endif
524 
525 extern char __initramfs_start[];
526 extern unsigned long __initramfs_size;
527 #include <linux/initrd.h>
528 #include <linux/kexec.h>
529 
530 void __weak free_initrd_mem(unsigned long start, unsigned long end)
531 {
532 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
533 			"initrd");
534 }
535 
536 #ifdef CONFIG_KEXEC_CORE
537 static bool kexec_free_initrd(void)
538 {
539 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
540 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
541 
542 	/*
543 	 * If the initrd region is overlapped with crashkernel reserved region,
544 	 * free only memory that is not part of crashkernel region.
545 	 */
546 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
547 		return false;
548 
549 	/*
550 	 * Initialize initrd memory region since the kexec boot does not do.
551 	 */
552 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
553 	if (initrd_start < crashk_start)
554 		free_initrd_mem(initrd_start, crashk_start);
555 	if (initrd_end > crashk_end)
556 		free_initrd_mem(crashk_end, initrd_end);
557 	return true;
558 }
559 #else
560 static inline bool kexec_free_initrd(void)
561 {
562 	return false;
563 }
564 #endif /* CONFIG_KEXEC_CORE */
565 
566 #ifdef CONFIG_BLK_DEV_RAM
567 #define BUF_SIZE 1024
568 static void __init clean_rootfs(void)
569 {
570 	int fd;
571 	void *buf;
572 	struct linux_dirent64 *dirp;
573 	int num;
574 
575 	fd = ksys_open("/", O_RDONLY, 0);
576 	WARN_ON(fd < 0);
577 	if (fd < 0)
578 		return;
579 	buf = kzalloc(BUF_SIZE, GFP_KERNEL);
580 	WARN_ON(!buf);
581 	if (!buf) {
582 		ksys_close(fd);
583 		return;
584 	}
585 
586 	dirp = buf;
587 	num = ksys_getdents64(fd, dirp, BUF_SIZE);
588 	while (num > 0) {
589 		while (num > 0) {
590 			struct kstat st;
591 			int ret;
592 
593 			ret = vfs_lstat(dirp->d_name, &st);
594 			WARN_ON_ONCE(ret);
595 			if (!ret) {
596 				if (S_ISDIR(st.mode))
597 					ksys_rmdir(dirp->d_name);
598 				else
599 					ksys_unlink(dirp->d_name);
600 			}
601 
602 			num -= dirp->d_reclen;
603 			dirp = (void *)dirp + dirp->d_reclen;
604 		}
605 		dirp = buf;
606 		memset(buf, 0, BUF_SIZE);
607 		num = ksys_getdents64(fd, dirp, BUF_SIZE);
608 	}
609 
610 	ksys_close(fd);
611 	kfree(buf);
612 }
613 #else
614 static inline void clean_rootfs(void)
615 {
616 }
617 #endif /* CONFIG_BLK_DEV_RAM */
618 
619 #ifdef CONFIG_BLK_DEV_RAM
620 static void populate_initrd_image(char *err)
621 {
622 	ssize_t written;
623 	int fd;
624 
625 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
626 
627 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
628 			err);
629 	fd = ksys_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
630 	if (fd < 0)
631 		return;
632 
633 	written = xwrite(fd, (char *)initrd_start, initrd_end - initrd_start);
634 	if (written != initrd_end - initrd_start)
635 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
636 		       written, initrd_end - initrd_start);
637 	ksys_close(fd);
638 }
639 #else
640 static void populate_initrd_image(char *err)
641 {
642 	printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
643 }
644 #endif /* CONFIG_BLK_DEV_RAM */
645 
646 static int __init populate_rootfs(void)
647 {
648 	/* Load the built in initramfs */
649 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
650 	if (err)
651 		panic("%s", err); /* Failed to decompress INTERNAL initramfs */
652 
653 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
654 		goto done;
655 
656 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
657 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
658 	else
659 		printk(KERN_INFO "Unpacking initramfs...\n");
660 
661 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
662 	if (err) {
663 		clean_rootfs();
664 		populate_initrd_image(err);
665 	}
666 
667 done:
668 	/*
669 	 * If the initrd region is overlapped with crashkernel reserved region,
670 	 * free only memory that is not part of crashkernel region.
671 	 */
672 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
673 		free_initrd_mem(initrd_start, initrd_end);
674 	initrd_start = 0;
675 	initrd_end = 0;
676 
677 	flush_delayed_fput();
678 	return 0;
679 }
680 rootfs_initcall(populate_rootfs);
681