xref: /openbmc/linux/init/initramfs.c (revision 9dbbc3b9) 
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 
19 static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
20 		loff_t *pos)
21 {
22 	ssize_t out = 0;
23 
24 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
25 	while (count) {
26 		ssize_t rv = kernel_write(file, p, count, pos);
27 
28 		if (rv < 0) {
29 			if (rv == -EINTR || rv == -EAGAIN)
30 				continue;
31 			return out ? out : rv;
32 		} else if (rv == 0)
33 			break;
34 
35 		p += rv;
36 		out += rv;
37 		count -= rv;
38 	}
39 
40 	return out;
41 }
42 
43 static __initdata char *message;
44 static void __init error(char *x)
45 {
46 	if (!message)
47 		message = x;
48 }
49 
50 static void panic_show_mem(const char *fmt, ...)
51 {
52 	va_list args;
53 
54 	show_mem(0, NULL);
55 	va_start(args, fmt);
56 	panic(fmt, args);
57 	va_end(args);
58 }
59 
60 /* link hash */
61 
62 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
63 
64 static __initdata struct hash {
65 	int ino, minor, major;
66 	umode_t mode;
67 	struct hash *next;
68 	char name[N_ALIGN(PATH_MAX)];
69 } *head[32];
70 
71 static inline int hash(int major, int minor, int ino)
72 {
73 	unsigned long tmp = ino + minor + (major << 3);
74 	tmp += tmp >> 5;
75 	return tmp & 31;
76 }
77 
78 static char __init *find_link(int major, int minor, int ino,
79 			      umode_t mode, char *name)
80 {
81 	struct hash **p, *q;
82 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
83 		if ((*p)->ino != ino)
84 			continue;
85 		if ((*p)->minor != minor)
86 			continue;
87 		if ((*p)->major != major)
88 			continue;
89 		if (((*p)->mode ^ mode) & S_IFMT)
90 			continue;
91 		return (*p)->name;
92 	}
93 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
94 	if (!q)
95 		panic_show_mem("can't allocate link hash entry");
96 	q->major = major;
97 	q->minor = minor;
98 	q->ino = ino;
99 	q->mode = mode;
100 	strcpy(q->name, name);
101 	q->next = NULL;
102 	*p = q;
103 	return NULL;
104 }
105 
106 static void __init free_hash(void)
107 {
108 	struct hash **p, *q;
109 	for (p = head; p < head + 32; p++) {
110 		while (*p) {
111 			q = *p;
112 			*p = q->next;
113 			kfree(q);
114 		}
115 	}
116 }
117 
118 static long __init do_utime(char *filename, time64_t mtime)
119 {
120 	struct timespec64 t[2];
121 
122 	t[0].tv_sec = mtime;
123 	t[0].tv_nsec = 0;
124 	t[1].tv_sec = mtime;
125 	t[1].tv_nsec = 0;
126 	return init_utimes(filename, t);
127 }
128 
129 static __initdata LIST_HEAD(dir_list);
130 struct dir_entry {
131 	struct list_head list;
132 	char *name;
133 	time64_t mtime;
134 };
135 
136 static void __init dir_add(const char *name, time64_t mtime)
137 {
138 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
139 	if (!de)
140 		panic_show_mem("can't allocate dir_entry buffer");
141 	INIT_LIST_HEAD(&de->list);
142 	de->name = kstrdup(name, GFP_KERNEL);
143 	de->mtime = mtime;
144 	list_add(&de->list, &dir_list);
145 }
146 
147 static void __init dir_utime(void)
148 {
149 	struct dir_entry *de, *tmp;
150 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
151 		list_del(&de->list);
152 		do_utime(de->name, de->mtime);
153 		kfree(de->name);
154 		kfree(de);
155 	}
156 }
157 
158 static __initdata time64_t mtime;
159 
160 /* cpio header parsing */
161 
162 static __initdata unsigned long ino, major, minor, nlink;
163 static __initdata umode_t mode;
164 static __initdata unsigned long body_len, name_len;
165 static __initdata uid_t uid;
166 static __initdata gid_t gid;
167 static __initdata unsigned rdev;
168 
169 static void __init parse_header(char *s)
170 {
171 	unsigned long parsed[12];
172 	char buf[9];
173 	int i;
174 
175 	buf[8] = '\0';
176 	for (i = 0, s += 6; i < 12; i++, s += 8) {
177 		memcpy(buf, s, 8);
178 		parsed[i] = simple_strtoul(buf, NULL, 16);
179 	}
180 	ino = parsed[0];
181 	mode = parsed[1];
182 	uid = parsed[2];
183 	gid = parsed[3];
184 	nlink = parsed[4];
185 	mtime = parsed[5]; /* breaks in y2106 */
186 	body_len = parsed[6];
187 	major = parsed[7];
188 	minor = parsed[8];
189 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
190 	name_len = parsed[11];
191 }
192 
193 /* FSM */
194 
195 static __initdata enum state {
196 	Start,
197 	Collect,
198 	GotHeader,
199 	SkipIt,
200 	GotName,
201 	CopyFile,
202 	GotSymlink,
203 	Reset
204 } state, next_state;
205 
206 static __initdata char *victim;
207 static unsigned long byte_count __initdata;
208 static __initdata loff_t this_header, next_header;
209 
210 static inline void __init eat(unsigned n)
211 {
212 	victim += n;
213 	this_header += n;
214 	byte_count -= n;
215 }
216 
217 static __initdata char *collected;
218 static long remains __initdata;
219 static __initdata char *collect;
220 
221 static void __init read_into(char *buf, unsigned size, enum state next)
222 {
223 	if (byte_count >= size) {
224 		collected = victim;
225 		eat(size);
226 		state = next;
227 	} else {
228 		collect = collected = buf;
229 		remains = size;
230 		next_state = next;
231 		state = Collect;
232 	}
233 }
234 
235 static __initdata char *header_buf, *symlink_buf, *name_buf;
236 
237 static int __init do_start(void)
238 {
239 	read_into(header_buf, 110, GotHeader);
240 	return 0;
241 }
242 
243 static int __init do_collect(void)
244 {
245 	unsigned long n = remains;
246 	if (byte_count < n)
247 		n = byte_count;
248 	memcpy(collect, victim, n);
249 	eat(n);
250 	collect += n;
251 	if ((remains -= n) != 0)
252 		return 1;
253 	state = next_state;
254 	return 0;
255 }
256 
257 static int __init do_header(void)
258 {
259 	if (memcmp(collected, "070707", 6)==0) {
260 		error("incorrect cpio method used: use -H newc option");
261 		return 1;
262 	}
263 	if (memcmp(collected, "070701", 6)) {
264 		error("no cpio magic");
265 		return 1;
266 	}
267 	parse_header(collected);
268 	next_header = this_header + N_ALIGN(name_len) + body_len;
269 	next_header = (next_header + 3) & ~3;
270 	state = SkipIt;
271 	if (name_len <= 0 || name_len > PATH_MAX)
272 		return 0;
273 	if (S_ISLNK(mode)) {
274 		if (body_len > PATH_MAX)
275 			return 0;
276 		collect = collected = symlink_buf;
277 		remains = N_ALIGN(name_len) + body_len;
278 		next_state = GotSymlink;
279 		state = Collect;
280 		return 0;
281 	}
282 	if (S_ISREG(mode) || !body_len)
283 		read_into(name_buf, N_ALIGN(name_len), GotName);
284 	return 0;
285 }
286 
287 static int __init do_skip(void)
288 {
289 	if (this_header + byte_count < next_header) {
290 		eat(byte_count);
291 		return 1;
292 	} else {
293 		eat(next_header - this_header);
294 		state = next_state;
295 		return 0;
296 	}
297 }
298 
299 static int __init do_reset(void)
300 {
301 	while (byte_count && *victim == '\0')
302 		eat(1);
303 	if (byte_count && (this_header & 3))
304 		error("broken padding");
305 	return 1;
306 }
307 
308 static void __init clean_path(char *path, umode_t fmode)
309 {
310 	struct kstat st;
311 
312 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
313 	    (st.mode ^ fmode) & S_IFMT) {
314 		if (S_ISDIR(st.mode))
315 			init_rmdir(path);
316 		else
317 			init_unlink(path);
318 	}
319 }
320 
321 static int __init maybe_link(void)
322 {
323 	if (nlink >= 2) {
324 		char *old = find_link(major, minor, ino, mode, collected);
325 		if (old) {
326 			clean_path(collected, 0);
327 			return (init_link(old, collected) < 0) ? -1 : 1;
328 		}
329 	}
330 	return 0;
331 }
332 
333 static __initdata struct file *wfile;
334 static __initdata loff_t wfile_pos;
335 
336 static int __init do_name(void)
337 {
338 	state = SkipIt;
339 	next_state = Reset;
340 	if (strcmp(collected, "TRAILER!!!") == 0) {
341 		free_hash();
342 		return 0;
343 	}
344 	clean_path(collected, mode);
345 	if (S_ISREG(mode)) {
346 		int ml = maybe_link();
347 		if (ml >= 0) {
348 			int openflags = O_WRONLY|O_CREAT;
349 			if (ml != 1)
350 				openflags |= O_TRUNC;
351 			wfile = filp_open(collected, openflags, mode);
352 			if (IS_ERR(wfile))
353 				return 0;
354 			wfile_pos = 0;
355 
356 			vfs_fchown(wfile, uid, gid);
357 			vfs_fchmod(wfile, mode);
358 			if (body_len)
359 				vfs_truncate(&wfile->f_path, body_len);
360 			state = CopyFile;
361 		}
362 	} else if (S_ISDIR(mode)) {
363 		init_mkdir(collected, mode);
364 		init_chown(collected, uid, gid, 0);
365 		init_chmod(collected, mode);
366 		dir_add(collected, mtime);
367 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
368 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
369 		if (maybe_link() == 0) {
370 			init_mknod(collected, mode, rdev);
371 			init_chown(collected, uid, gid, 0);
372 			init_chmod(collected, mode);
373 			do_utime(collected, mtime);
374 		}
375 	}
376 	return 0;
377 }
378 
379 static int __init do_copy(void)
380 {
381 	if (byte_count >= body_len) {
382 		struct timespec64 t[2] = { };
383 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
384 			error("write error");
385 
386 		t[0].tv_sec = mtime;
387 		t[1].tv_sec = mtime;
388 		vfs_utimes(&wfile->f_path, t);
389 
390 		fput(wfile);
391 		eat(body_len);
392 		state = SkipIt;
393 		return 0;
394 	} else {
395 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
396 			error("write error");
397 		body_len -= byte_count;
398 		eat(byte_count);
399 		return 1;
400 	}
401 }
402 
403 static int __init do_symlink(void)
404 {
405 	collected[N_ALIGN(name_len) + body_len] = '\0';
406 	clean_path(collected, 0);
407 	init_symlink(collected + N_ALIGN(name_len), collected);
408 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
409 	do_utime(collected, mtime);
410 	state = SkipIt;
411 	next_state = Reset;
412 	return 0;
413 }
414 
415 static __initdata int (*actions[])(void) = {
416 	[Start]		= do_start,
417 	[Collect]	= do_collect,
418 	[GotHeader]	= do_header,
419 	[SkipIt]	= do_skip,
420 	[GotName]	= do_name,
421 	[CopyFile]	= do_copy,
422 	[GotSymlink]	= do_symlink,
423 	[Reset]		= do_reset,
424 };
425 
426 static long __init write_buffer(char *buf, unsigned long len)
427 {
428 	byte_count = len;
429 	victim = buf;
430 
431 	while (!actions[state]())
432 		;
433 	return len - byte_count;
434 }
435 
436 static long __init flush_buffer(void *bufv, unsigned long len)
437 {
438 	char *buf = (char *) bufv;
439 	long written;
440 	long origLen = len;
441 	if (message)
442 		return -1;
443 	while ((written = write_buffer(buf, len)) < len && !message) {
444 		char c = buf[written];
445 		if (c == '0') {
446 			buf += written;
447 			len -= written;
448 			state = Start;
449 		} else if (c == 0) {
450 			buf += written;
451 			len -= written;
452 			state = Reset;
453 		} else
454 			error("junk within compressed archive");
455 	}
456 	return origLen;
457 }
458 
459 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
460 
461 #include <linux/decompress/generic.h>
462 
463 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
464 {
465 	long written;
466 	decompress_fn decompress;
467 	const char *compress_name;
468 	static __initdata char msg_buf[64];
469 
470 	header_buf = kmalloc(110, GFP_KERNEL);
471 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
472 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
473 
474 	if (!header_buf || !symlink_buf || !name_buf)
475 		panic_show_mem("can't allocate buffers");
476 
477 	state = Start;
478 	this_header = 0;
479 	message = NULL;
480 	while (!message && len) {
481 		loff_t saved_offset = this_header;
482 		if (*buf == '0' && !(this_header & 3)) {
483 			state = Start;
484 			written = write_buffer(buf, len);
485 			buf += written;
486 			len -= written;
487 			continue;
488 		}
489 		if (!*buf) {
490 			buf++;
491 			len--;
492 			this_header++;
493 			continue;
494 		}
495 		this_header = 0;
496 		decompress = decompress_method(buf, len, &compress_name);
497 		pr_debug("Detected %s compressed data\n", compress_name);
498 		if (decompress) {
499 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
500 				   &my_inptr, error);
501 			if (res)
502 				error("decompressor failed");
503 		} else if (compress_name) {
504 			if (!message) {
505 				snprintf(msg_buf, sizeof msg_buf,
506 					 "compression method %s not configured",
507 					 compress_name);
508 				message = msg_buf;
509 			}
510 		} else
511 			error("invalid magic at start of compressed archive");
512 		if (state != Reset)
513 			error("junk at the end of compressed archive");
514 		this_header = saved_offset + my_inptr;
515 		buf += my_inptr;
516 		len -= my_inptr;
517 	}
518 	dir_utime();
519 	kfree(name_buf);
520 	kfree(symlink_buf);
521 	kfree(header_buf);
522 	return message;
523 }
524 
525 static int __initdata do_retain_initrd;
526 
527 static int __init retain_initrd_param(char *str)
528 {
529 	if (*str)
530 		return 0;
531 	do_retain_initrd = 1;
532 	return 1;
533 }
534 __setup("retain_initrd", retain_initrd_param);
535 
536 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
537 static int __init keepinitrd_setup(char *__unused)
538 {
539 	do_retain_initrd = 1;
540 	return 1;
541 }
542 __setup("keepinitrd", keepinitrd_setup);
543 #endif
544 
545 static bool __initdata initramfs_async = true;
546 static int __init initramfs_async_setup(char *str)
547 {
548 	strtobool(str, &initramfs_async);
549 	return 1;
550 }
551 __setup("initramfs_async=", initramfs_async_setup);
552 
553 extern char __initramfs_start[];
554 extern unsigned long __initramfs_size;
555 #include <linux/initrd.h>
556 #include <linux/kexec.h>
557 
558 void __init reserve_initrd_mem(void)
559 {
560 	phys_addr_t start;
561 	unsigned long size;
562 
563 	/* Ignore the virtul address computed during device tree parsing */
564 	initrd_start = initrd_end = 0;
565 
566 	if (!phys_initrd_size)
567 		return;
568 	/*
569 	 * Round the memory region to page boundaries as per free_initrd_mem()
570 	 * This allows us to detect whether the pages overlapping the initrd
571 	 * are in use, but more importantly, reserves the entire set of pages
572 	 * as we don't want these pages allocated for other purposes.
573 	 */
574 	start = round_down(phys_initrd_start, PAGE_SIZE);
575 	size = phys_initrd_size + (phys_initrd_start - start);
576 	size = round_up(size, PAGE_SIZE);
577 
578 	if (!memblock_is_region_memory(start, size)) {
579 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
580 		       (u64)start, size);
581 		goto disable;
582 	}
583 
584 	if (memblock_is_region_reserved(start, size)) {
585 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
586 		       (u64)start, size);
587 		goto disable;
588 	}
589 
590 	memblock_reserve(start, size);
591 	/* Now convert initrd to virtual addresses */
592 	initrd_start = (unsigned long)__va(phys_initrd_start);
593 	initrd_end = initrd_start + phys_initrd_size;
594 	initrd_below_start_ok = 1;
595 
596 	return;
597 disable:
598 	pr_cont(" - disabling initrd\n");
599 	initrd_start = 0;
600 	initrd_end = 0;
601 }
602 
603 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
604 {
605 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
606 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
607 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
608 
609 	memblock_free(__pa(aligned_start), aligned_end - aligned_start);
610 #endif
611 
612 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
613 			"initrd");
614 }
615 
616 #ifdef CONFIG_KEXEC_CORE
617 static bool __init kexec_free_initrd(void)
618 {
619 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
620 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
621 
622 	/*
623 	 * If the initrd region is overlapped with crashkernel reserved region,
624 	 * free only memory that is not part of crashkernel region.
625 	 */
626 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
627 		return false;
628 
629 	/*
630 	 * Initialize initrd memory region since the kexec boot does not do.
631 	 */
632 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
633 	if (initrd_start < crashk_start)
634 		free_initrd_mem(initrd_start, crashk_start);
635 	if (initrd_end > crashk_end)
636 		free_initrd_mem(crashk_end, initrd_end);
637 	return true;
638 }
639 #else
640 static inline bool kexec_free_initrd(void)
641 {
642 	return false;
643 }
644 #endif /* CONFIG_KEXEC_CORE */
645 
646 #ifdef CONFIG_BLK_DEV_RAM
647 static void __init populate_initrd_image(char *err)
648 {
649 	ssize_t written;
650 	struct file *file;
651 	loff_t pos = 0;
652 
653 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
654 
655 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
656 			err);
657 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
658 	if (IS_ERR(file))
659 		return;
660 
661 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
662 			&pos);
663 	if (written != initrd_end - initrd_start)
664 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
665 		       written, initrd_end - initrd_start);
666 	fput(file);
667 }
668 #endif /* CONFIG_BLK_DEV_RAM */
669 
670 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
671 {
672 	/* Load the built in initramfs */
673 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
674 	if (err)
675 		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
676 
677 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
678 		goto done;
679 
680 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
681 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
682 	else
683 		printk(KERN_INFO "Unpacking initramfs...\n");
684 
685 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
686 	if (err) {
687 #ifdef CONFIG_BLK_DEV_RAM
688 		populate_initrd_image(err);
689 #else
690 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
691 #endif
692 	}
693 
694 done:
695 	/*
696 	 * If the initrd region is overlapped with crashkernel reserved region,
697 	 * free only memory that is not part of crashkernel region.
698 	 */
699 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
700 		free_initrd_mem(initrd_start, initrd_end);
701 	initrd_start = 0;
702 	initrd_end = 0;
703 
704 	flush_delayed_fput();
705 }
706 
707 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
708 static async_cookie_t initramfs_cookie;
709 
710 void wait_for_initramfs(void)
711 {
712 	if (!initramfs_cookie) {
713 		/*
714 		 * Something before rootfs_initcall wants to access
715 		 * the filesystem/initramfs. Probably a bug. Make a
716 		 * note, avoid deadlocking the machine, and let the
717 		 * caller's access fail as it used to.
718 		 */
719 		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
720 		return;
721 	}
722 	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
723 }
724 EXPORT_SYMBOL_GPL(wait_for_initramfs);
725 
726 static int __init populate_rootfs(void)
727 {
728 	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
729 						 &initramfs_domain);
730 	if (!initramfs_async)
731 		wait_for_initramfs();
732 	return 0;
733 }
734 rootfs_initcall(populate_rootfs);
735