xref: /openbmc/linux/init/initramfs.c (revision b7b3c35e) 
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/umh.h>
19 
20 static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
21 		loff_t *pos)
22 {
23 	ssize_t out = 0;
24 
25 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
26 	while (count) {
27 		ssize_t rv = kernel_write(file, p, count, pos);
28 
29 		if (rv < 0) {
30 			if (rv == -EINTR || rv == -EAGAIN)
31 				continue;
32 			return out ? out : rv;
33 		} else if (rv == 0)
34 			break;
35 
36 		p += rv;
37 		out += rv;
38 		count -= rv;
39 	}
40 
41 	return out;
42 }
43 
44 static __initdata char *message;
45 static void __init error(char *x)
46 {
47 	if (!message)
48 		message = x;
49 }
50 
51 static void panic_show_mem(const char *fmt, ...)
52 {
53 	va_list args;
54 
55 	show_mem(0, NULL);
56 	va_start(args, fmt);
57 	panic(fmt, args);
58 	va_end(args);
59 }
60 
61 /* link hash */
62 
63 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
64 
65 static __initdata struct hash {
66 	int ino, minor, major;
67 	umode_t mode;
68 	struct hash *next;
69 	char name[N_ALIGN(PATH_MAX)];
70 } *head[32];
71 
72 static inline int hash(int major, int minor, int ino)
73 {
74 	unsigned long tmp = ino + minor + (major << 3);
75 	tmp += tmp >> 5;
76 	return tmp & 31;
77 }
78 
79 static char __init *find_link(int major, int minor, int ino,
80 			      umode_t mode, char *name)
81 {
82 	struct hash **p, *q;
83 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
84 		if ((*p)->ino != ino)
85 			continue;
86 		if ((*p)->minor != minor)
87 			continue;
88 		if ((*p)->major != major)
89 			continue;
90 		if (((*p)->mode ^ mode) & S_IFMT)
91 			continue;
92 		return (*p)->name;
93 	}
94 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
95 	if (!q)
96 		panic_show_mem("can't allocate link hash entry");
97 	q->major = major;
98 	q->minor = minor;
99 	q->ino = ino;
100 	q->mode = mode;
101 	strcpy(q->name, name);
102 	q->next = NULL;
103 	*p = q;
104 	return NULL;
105 }
106 
107 static void __init free_hash(void)
108 {
109 	struct hash **p, *q;
110 	for (p = head; p < head + 32; p++) {
111 		while (*p) {
112 			q = *p;
113 			*p = q->next;
114 			kfree(q);
115 		}
116 	}
117 }
118 
119 static long __init do_utime(char *filename, time64_t mtime)
120 {
121 	struct timespec64 t[2];
122 
123 	t[0].tv_sec = mtime;
124 	t[0].tv_nsec = 0;
125 	t[1].tv_sec = mtime;
126 	t[1].tv_nsec = 0;
127 	return init_utimes(filename, t);
128 }
129 
130 static __initdata LIST_HEAD(dir_list);
131 struct dir_entry {
132 	struct list_head list;
133 	char *name;
134 	time64_t mtime;
135 };
136 
137 static void __init dir_add(const char *name, time64_t mtime)
138 {
139 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
140 	if (!de)
141 		panic_show_mem("can't allocate dir_entry buffer");
142 	INIT_LIST_HEAD(&de->list);
143 	de->name = kstrdup(name, GFP_KERNEL);
144 	de->mtime = mtime;
145 	list_add(&de->list, &dir_list);
146 }
147 
148 static void __init dir_utime(void)
149 {
150 	struct dir_entry *de, *tmp;
151 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
152 		list_del(&de->list);
153 		do_utime(de->name, de->mtime);
154 		kfree(de->name);
155 		kfree(de);
156 	}
157 }
158 
159 static __initdata time64_t mtime;
160 
161 /* cpio header parsing */
162 
163 static __initdata unsigned long ino, major, minor, nlink;
164 static __initdata umode_t mode;
165 static __initdata unsigned long body_len, name_len;
166 static __initdata uid_t uid;
167 static __initdata gid_t gid;
168 static __initdata unsigned rdev;
169 
170 static void __init parse_header(char *s)
171 {
172 	unsigned long parsed[12];
173 	char buf[9];
174 	int i;
175 
176 	buf[8] = '\0';
177 	for (i = 0, s += 6; i < 12; i++, s += 8) {
178 		memcpy(buf, s, 8);
179 		parsed[i] = simple_strtoul(buf, NULL, 16);
180 	}
181 	ino = parsed[0];
182 	mode = parsed[1];
183 	uid = parsed[2];
184 	gid = parsed[3];
185 	nlink = parsed[4];
186 	mtime = parsed[5]; /* breaks in y2106 */
187 	body_len = parsed[6];
188 	major = parsed[7];
189 	minor = parsed[8];
190 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
191 	name_len = parsed[11];
192 }
193 
194 /* FSM */
195 
196 static __initdata enum state {
197 	Start,
198 	Collect,
199 	GotHeader,
200 	SkipIt,
201 	GotName,
202 	CopyFile,
203 	GotSymlink,
204 	Reset
205 } state, next_state;
206 
207 static __initdata char *victim;
208 static unsigned long byte_count __initdata;
209 static __initdata loff_t this_header, next_header;
210 
211 static inline void __init eat(unsigned n)
212 {
213 	victim += n;
214 	this_header += n;
215 	byte_count -= n;
216 }
217 
218 static __initdata char *collected;
219 static long remains __initdata;
220 static __initdata char *collect;
221 
222 static void __init read_into(char *buf, unsigned size, enum state next)
223 {
224 	if (byte_count >= size) {
225 		collected = victim;
226 		eat(size);
227 		state = next;
228 	} else {
229 		collect = collected = buf;
230 		remains = size;
231 		next_state = next;
232 		state = Collect;
233 	}
234 }
235 
236 static __initdata char *header_buf, *symlink_buf, *name_buf;
237 
238 static int __init do_start(void)
239 {
240 	read_into(header_buf, 110, GotHeader);
241 	return 0;
242 }
243 
244 static int __init do_collect(void)
245 {
246 	unsigned long n = remains;
247 	if (byte_count < n)
248 		n = byte_count;
249 	memcpy(collect, victim, n);
250 	eat(n);
251 	collect += n;
252 	if ((remains -= n) != 0)
253 		return 1;
254 	state = next_state;
255 	return 0;
256 }
257 
258 static int __init do_header(void)
259 {
260 	if (memcmp(collected, "070707", 6)==0) {
261 		error("incorrect cpio method used: use -H newc option");
262 		return 1;
263 	}
264 	if (memcmp(collected, "070701", 6)) {
265 		error("no cpio magic");
266 		return 1;
267 	}
268 	parse_header(collected);
269 	next_header = this_header + N_ALIGN(name_len) + body_len;
270 	next_header = (next_header + 3) & ~3;
271 	state = SkipIt;
272 	if (name_len <= 0 || name_len > PATH_MAX)
273 		return 0;
274 	if (S_ISLNK(mode)) {
275 		if (body_len > PATH_MAX)
276 			return 0;
277 		collect = collected = symlink_buf;
278 		remains = N_ALIGN(name_len) + body_len;
279 		next_state = GotSymlink;
280 		state = Collect;
281 		return 0;
282 	}
283 	if (S_ISREG(mode) || !body_len)
284 		read_into(name_buf, N_ALIGN(name_len), GotName);
285 	return 0;
286 }
287 
288 static int __init do_skip(void)
289 {
290 	if (this_header + byte_count < next_header) {
291 		eat(byte_count);
292 		return 1;
293 	} else {
294 		eat(next_header - this_header);
295 		state = next_state;
296 		return 0;
297 	}
298 }
299 
300 static int __init do_reset(void)
301 {
302 	while (byte_count && *victim == '\0')
303 		eat(1);
304 	if (byte_count && (this_header & 3))
305 		error("broken padding");
306 	return 1;
307 }
308 
309 static void __init clean_path(char *path, umode_t fmode)
310 {
311 	struct kstat st;
312 
313 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
314 	    (st.mode ^ fmode) & S_IFMT) {
315 		if (S_ISDIR(st.mode))
316 			init_rmdir(path);
317 		else
318 			init_unlink(path);
319 	}
320 }
321 
322 static int __init maybe_link(void)
323 {
324 	if (nlink >= 2) {
325 		char *old = find_link(major, minor, ino, mode, collected);
326 		if (old) {
327 			clean_path(collected, 0);
328 			return (init_link(old, collected) < 0) ? -1 : 1;
329 		}
330 	}
331 	return 0;
332 }
333 
334 static __initdata struct file *wfile;
335 static __initdata loff_t wfile_pos;
336 
337 static int __init do_name(void)
338 {
339 	state = SkipIt;
340 	next_state = Reset;
341 	if (strcmp(collected, "TRAILER!!!") == 0) {
342 		free_hash();
343 		return 0;
344 	}
345 	clean_path(collected, mode);
346 	if (S_ISREG(mode)) {
347 		int ml = maybe_link();
348 		if (ml >= 0) {
349 			int openflags = O_WRONLY|O_CREAT;
350 			if (ml != 1)
351 				openflags |= O_TRUNC;
352 			wfile = filp_open(collected, openflags, mode);
353 			if (IS_ERR(wfile))
354 				return 0;
355 			wfile_pos = 0;
356 
357 			vfs_fchown(wfile, uid, gid);
358 			vfs_fchmod(wfile, mode);
359 			if (body_len)
360 				vfs_truncate(&wfile->f_path, body_len);
361 			state = CopyFile;
362 		}
363 	} else if (S_ISDIR(mode)) {
364 		init_mkdir(collected, mode);
365 		init_chown(collected, uid, gid, 0);
366 		init_chmod(collected, mode);
367 		dir_add(collected, mtime);
368 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
369 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
370 		if (maybe_link() == 0) {
371 			init_mknod(collected, mode, rdev);
372 			init_chown(collected, uid, gid, 0);
373 			init_chmod(collected, mode);
374 			do_utime(collected, mtime);
375 		}
376 	}
377 	return 0;
378 }
379 
380 static int __init do_copy(void)
381 {
382 	if (byte_count >= body_len) {
383 		struct timespec64 t[2] = { };
384 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
385 			error("write error");
386 
387 		t[0].tv_sec = mtime;
388 		t[1].tv_sec = mtime;
389 		vfs_utimes(&wfile->f_path, t);
390 
391 		fput(wfile);
392 		eat(body_len);
393 		state = SkipIt;
394 		return 0;
395 	} else {
396 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
397 			error("write error");
398 		body_len -= byte_count;
399 		eat(byte_count);
400 		return 1;
401 	}
402 }
403 
404 static int __init do_symlink(void)
405 {
406 	collected[N_ALIGN(name_len) + body_len] = '\0';
407 	clean_path(collected, 0);
408 	init_symlink(collected + N_ALIGN(name_len), collected);
409 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
410 	do_utime(collected, mtime);
411 	state = SkipIt;
412 	next_state = Reset;
413 	return 0;
414 }
415 
416 static __initdata int (*actions[])(void) = {
417 	[Start]		= do_start,
418 	[Collect]	= do_collect,
419 	[GotHeader]	= do_header,
420 	[SkipIt]	= do_skip,
421 	[GotName]	= do_name,
422 	[CopyFile]	= do_copy,
423 	[GotSymlink]	= do_symlink,
424 	[Reset]		= do_reset,
425 };
426 
427 static long __init write_buffer(char *buf, unsigned long len)
428 {
429 	byte_count = len;
430 	victim = buf;
431 
432 	while (!actions[state]())
433 		;
434 	return len - byte_count;
435 }
436 
437 static long __init flush_buffer(void *bufv, unsigned long len)
438 {
439 	char *buf = (char *) bufv;
440 	long written;
441 	long origLen = len;
442 	if (message)
443 		return -1;
444 	while ((written = write_buffer(buf, len)) < len && !message) {
445 		char c = buf[written];
446 		if (c == '0') {
447 			buf += written;
448 			len -= written;
449 			state = Start;
450 		} else if (c == 0) {
451 			buf += written;
452 			len -= written;
453 			state = Reset;
454 		} else
455 			error("junk within compressed archive");
456 	}
457 	return origLen;
458 }
459 
460 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
461 
462 #include <linux/decompress/generic.h>
463 
464 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
465 {
466 	long written;
467 	decompress_fn decompress;
468 	const char *compress_name;
469 	static __initdata char msg_buf[64];
470 
471 	header_buf = kmalloc(110, GFP_KERNEL);
472 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
473 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
474 
475 	if (!header_buf || !symlink_buf || !name_buf)
476 		panic_show_mem("can't allocate buffers");
477 
478 	state = Start;
479 	this_header = 0;
480 	message = NULL;
481 	while (!message && len) {
482 		loff_t saved_offset = this_header;
483 		if (*buf == '0' && !(this_header & 3)) {
484 			state = Start;
485 			written = write_buffer(buf, len);
486 			buf += written;
487 			len -= written;
488 			continue;
489 		}
490 		if (!*buf) {
491 			buf++;
492 			len--;
493 			this_header++;
494 			continue;
495 		}
496 		this_header = 0;
497 		decompress = decompress_method(buf, len, &compress_name);
498 		pr_debug("Detected %s compressed data\n", compress_name);
499 		if (decompress) {
500 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
501 				   &my_inptr, error);
502 			if (res)
503 				error("decompressor failed");
504 		} else if (compress_name) {
505 			if (!message) {
506 				snprintf(msg_buf, sizeof msg_buf,
507 					 "compression method %s not configured",
508 					 compress_name);
509 				message = msg_buf;
510 			}
511 		} else
512 			error("invalid magic at start of compressed archive");
513 		if (state != Reset)
514 			error("junk at the end of compressed archive");
515 		this_header = saved_offset + my_inptr;
516 		buf += my_inptr;
517 		len -= my_inptr;
518 	}
519 	dir_utime();
520 	kfree(name_buf);
521 	kfree(symlink_buf);
522 	kfree(header_buf);
523 	return message;
524 }
525 
526 static int __initdata do_retain_initrd;
527 
528 static int __init retain_initrd_param(char *str)
529 {
530 	if (*str)
531 		return 0;
532 	do_retain_initrd = 1;
533 	return 1;
534 }
535 __setup("retain_initrd", retain_initrd_param);
536 
537 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
538 static int __init keepinitrd_setup(char *__unused)
539 {
540 	do_retain_initrd = 1;
541 	return 1;
542 }
543 __setup("keepinitrd", keepinitrd_setup);
544 #endif
545 
546 static bool __initdata initramfs_async = true;
547 static int __init initramfs_async_setup(char *str)
548 {
549 	strtobool(str, &initramfs_async);
550 	return 1;
551 }
552 __setup("initramfs_async=", initramfs_async_setup);
553 
554 extern char __initramfs_start[];
555 extern unsigned long __initramfs_size;
556 #include <linux/initrd.h>
557 #include <linux/kexec.h>
558 
559 void __init reserve_initrd_mem(void)
560 {
561 	phys_addr_t start;
562 	unsigned long size;
563 
564 	/* Ignore the virtul address computed during device tree parsing */
565 	initrd_start = initrd_end = 0;
566 
567 	if (!phys_initrd_size)
568 		return;
569 	/*
570 	 * Round the memory region to page boundaries as per free_initrd_mem()
571 	 * This allows us to detect whether the pages overlapping the initrd
572 	 * are in use, but more importantly, reserves the entire set of pages
573 	 * as we don't want these pages allocated for other purposes.
574 	 */
575 	start = round_down(phys_initrd_start, PAGE_SIZE);
576 	size = phys_initrd_size + (phys_initrd_start - start);
577 	size = round_up(size, PAGE_SIZE);
578 
579 	if (!memblock_is_region_memory(start, size)) {
580 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
581 		       (u64)start, size);
582 		goto disable;
583 	}
584 
585 	if (memblock_is_region_reserved(start, size)) {
586 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
587 		       (u64)start, size);
588 		goto disable;
589 	}
590 
591 	memblock_reserve(start, size);
592 	/* Now convert initrd to virtual addresses */
593 	initrd_start = (unsigned long)__va(phys_initrd_start);
594 	initrd_end = initrd_start + phys_initrd_size;
595 	initrd_below_start_ok = 1;
596 
597 	return;
598 disable:
599 	pr_cont(" - disabling initrd\n");
600 	initrd_start = 0;
601 	initrd_end = 0;
602 }
603 
604 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
605 {
606 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
607 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
608 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
609 
610 	memblock_free((void *)aligned_start, aligned_end - aligned_start);
611 #endif
612 
613 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
614 			"initrd");
615 }
616 
617 #ifdef CONFIG_KEXEC_CORE
618 static bool __init kexec_free_initrd(void)
619 {
620 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
621 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
622 
623 	/*
624 	 * If the initrd region is overlapped with crashkernel reserved region,
625 	 * free only memory that is not part of crashkernel region.
626 	 */
627 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
628 		return false;
629 
630 	/*
631 	 * Initialize initrd memory region since the kexec boot does not do.
632 	 */
633 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
634 	if (initrd_start < crashk_start)
635 		free_initrd_mem(initrd_start, crashk_start);
636 	if (initrd_end > crashk_end)
637 		free_initrd_mem(crashk_end, initrd_end);
638 	return true;
639 }
640 #else
641 static inline bool kexec_free_initrd(void)
642 {
643 	return false;
644 }
645 #endif /* CONFIG_KEXEC_CORE */
646 
647 #ifdef CONFIG_BLK_DEV_RAM
648 static void __init populate_initrd_image(char *err)
649 {
650 	ssize_t written;
651 	struct file *file;
652 	loff_t pos = 0;
653 
654 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
655 
656 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
657 			err);
658 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
659 	if (IS_ERR(file))
660 		return;
661 
662 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
663 			&pos);
664 	if (written != initrd_end - initrd_start)
665 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
666 		       written, initrd_end - initrd_start);
667 	fput(file);
668 }
669 #endif /* CONFIG_BLK_DEV_RAM */
670 
671 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
672 {
673 	/* Load the built in initramfs */
674 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
675 	if (err)
676 		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
677 
678 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
679 		goto done;
680 
681 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
682 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
683 	else
684 		printk(KERN_INFO "Unpacking initramfs...\n");
685 
686 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
687 	if (err) {
688 #ifdef CONFIG_BLK_DEV_RAM
689 		populate_initrd_image(err);
690 #else
691 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
692 #endif
693 	}
694 
695 done:
696 	/*
697 	 * If the initrd region is overlapped with crashkernel reserved region,
698 	 * free only memory that is not part of crashkernel region.
699 	 */
700 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
701 		free_initrd_mem(initrd_start, initrd_end);
702 	initrd_start = 0;
703 	initrd_end = 0;
704 
705 	flush_delayed_fput();
706 }
707 
708 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
709 static async_cookie_t initramfs_cookie;
710 
711 void wait_for_initramfs(void)
712 {
713 	if (!initramfs_cookie) {
714 		/*
715 		 * Something before rootfs_initcall wants to access
716 		 * the filesystem/initramfs. Probably a bug. Make a
717 		 * note, avoid deadlocking the machine, and let the
718 		 * caller's access fail as it used to.
719 		 */
720 		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
721 		return;
722 	}
723 	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
724 }
725 EXPORT_SYMBOL_GPL(wait_for_initramfs);
726 
727 static int __init populate_rootfs(void)
728 {
729 	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
730 						 &initramfs_domain);
731 	usermodehelper_enable();
732 	if (!initramfs_async)
733 		wait_for_initramfs();
734 	return 0;
735 }
736 rootfs_initcall(populate_rootfs);
737