xref: /openbmc/linux/init/initramfs.c (revision c9dc580c) 
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/kstrtox.h>
15 #include <linux/memblock.h>
16 #include <linux/mm.h>
17 #include <linux/namei.h>
18 #include <linux/init_syscalls.h>
19 #include <linux/task_work.h>
20 #include <linux/umh.h>
21 
22 static __initdata bool csum_present;
23 static __initdata u32 io_csum;
24 
25 static ssize_t __init xwrite(struct file *file, const unsigned char *p,
26 		size_t count, loff_t *pos)
27 {
28 	ssize_t out = 0;
29 
30 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
31 	while (count) {
32 		ssize_t rv = kernel_write(file, p, count, pos);
33 
34 		if (rv < 0) {
35 			if (rv == -EINTR || rv == -EAGAIN)
36 				continue;
37 			return out ? out : rv;
38 		} else if (rv == 0)
39 			break;
40 
41 		if (csum_present) {
42 			ssize_t i;
43 
44 			for (i = 0; i < rv; i++)
45 				io_csum += p[i];
46 		}
47 
48 		p += rv;
49 		out += rv;
50 		count -= rv;
51 	}
52 
53 	return out;
54 }
55 
56 static __initdata char *message;
57 static void __init error(char *x)
58 {
59 	if (!message)
60 		message = x;
61 }
62 
63 static void panic_show_mem(const char *fmt, ...)
64 {
65 	va_list args;
66 
67 	show_mem(0, NULL);
68 	va_start(args, fmt);
69 	panic(fmt, args);
70 	va_end(args);
71 }
72 
73 /* link hash */
74 
75 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
76 
77 static __initdata struct hash {
78 	int ino, minor, major;
79 	umode_t mode;
80 	struct hash *next;
81 	char name[N_ALIGN(PATH_MAX)];
82 } *head[32];
83 
84 static inline int hash(int major, int minor, int ino)
85 {
86 	unsigned long tmp = ino + minor + (major << 3);
87 	tmp += tmp >> 5;
88 	return tmp & 31;
89 }
90 
91 static char __init *find_link(int major, int minor, int ino,
92 			      umode_t mode, char *name)
93 {
94 	struct hash **p, *q;
95 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
96 		if ((*p)->ino != ino)
97 			continue;
98 		if ((*p)->minor != minor)
99 			continue;
100 		if ((*p)->major != major)
101 			continue;
102 		if (((*p)->mode ^ mode) & S_IFMT)
103 			continue;
104 		return (*p)->name;
105 	}
106 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
107 	if (!q)
108 		panic_show_mem("can't allocate link hash entry");
109 	q->major = major;
110 	q->minor = minor;
111 	q->ino = ino;
112 	q->mode = mode;
113 	strcpy(q->name, name);
114 	q->next = NULL;
115 	*p = q;
116 	return NULL;
117 }
118 
119 static void __init free_hash(void)
120 {
121 	struct hash **p, *q;
122 	for (p = head; p < head + 32; p++) {
123 		while (*p) {
124 			q = *p;
125 			*p = q->next;
126 			kfree(q);
127 		}
128 	}
129 }
130 
131 #ifdef CONFIG_INITRAMFS_PRESERVE_MTIME
132 static void __init do_utime(char *filename, time64_t mtime)
133 {
134 	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
135 	init_utimes(filename, t);
136 }
137 
138 static void __init do_utime_path(const struct path *path, time64_t mtime)
139 {
140 	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
141 	vfs_utimes(path, t);
142 }
143 
144 static __initdata LIST_HEAD(dir_list);
145 struct dir_entry {
146 	struct list_head list;
147 	time64_t mtime;
148 	char name[];
149 };
150 
151 static void __init dir_add(const char *name, time64_t mtime)
152 {
153 	size_t nlen = strlen(name) + 1;
154 	struct dir_entry *de;
155 
156 	de = kmalloc(sizeof(struct dir_entry) + nlen, GFP_KERNEL);
157 	if (!de)
158 		panic_show_mem("can't allocate dir_entry buffer");
159 	INIT_LIST_HEAD(&de->list);
160 	strscpy(de->name, name, nlen);
161 	de->mtime = mtime;
162 	list_add(&de->list, &dir_list);
163 }
164 
165 static void __init dir_utime(void)
166 {
167 	struct dir_entry *de, *tmp;
168 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
169 		list_del(&de->list);
170 		do_utime(de->name, de->mtime);
171 		kfree(de);
172 	}
173 }
174 #else
175 static void __init do_utime(char *filename, time64_t mtime) {}
176 static void __init do_utime_path(const struct path *path, time64_t mtime) {}
177 static void __init dir_add(const char *name, time64_t mtime) {}
178 static void __init dir_utime(void) {}
179 #endif
180 
181 static __initdata time64_t mtime;
182 
183 /* cpio header parsing */
184 
185 static __initdata unsigned long ino, major, minor, nlink;
186 static __initdata umode_t mode;
187 static __initdata unsigned long body_len, name_len;
188 static __initdata uid_t uid;
189 static __initdata gid_t gid;
190 static __initdata unsigned rdev;
191 static __initdata u32 hdr_csum;
192 
193 static void __init parse_header(char *s)
194 {
195 	unsigned long parsed[13];
196 	char buf[9];
197 	int i;
198 
199 	buf[8] = '\0';
200 	for (i = 0, s += 6; i < 13; i++, s += 8) {
201 		memcpy(buf, s, 8);
202 		parsed[i] = simple_strtoul(buf, NULL, 16);
203 	}
204 	ino = parsed[0];
205 	mode = parsed[1];
206 	uid = parsed[2];
207 	gid = parsed[3];
208 	nlink = parsed[4];
209 	mtime = parsed[5]; /* breaks in y2106 */
210 	body_len = parsed[6];
211 	major = parsed[7];
212 	minor = parsed[8];
213 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
214 	name_len = parsed[11];
215 	hdr_csum = parsed[12];
216 }
217 
218 /* FSM */
219 
220 static __initdata enum state {
221 	Start,
222 	Collect,
223 	GotHeader,
224 	SkipIt,
225 	GotName,
226 	CopyFile,
227 	GotSymlink,
228 	Reset
229 } state, next_state;
230 
231 static __initdata char *victim;
232 static unsigned long byte_count __initdata;
233 static __initdata loff_t this_header, next_header;
234 
235 static inline void __init eat(unsigned n)
236 {
237 	victim += n;
238 	this_header += n;
239 	byte_count -= n;
240 }
241 
242 static __initdata char *collected;
243 static long remains __initdata;
244 static __initdata char *collect;
245 
246 static void __init read_into(char *buf, unsigned size, enum state next)
247 {
248 	if (byte_count >= size) {
249 		collected = victim;
250 		eat(size);
251 		state = next;
252 	} else {
253 		collect = collected = buf;
254 		remains = size;
255 		next_state = next;
256 		state = Collect;
257 	}
258 }
259 
260 static __initdata char *header_buf, *symlink_buf, *name_buf;
261 
262 static int __init do_start(void)
263 {
264 	read_into(header_buf, 110, GotHeader);
265 	return 0;
266 }
267 
268 static int __init do_collect(void)
269 {
270 	unsigned long n = remains;
271 	if (byte_count < n)
272 		n = byte_count;
273 	memcpy(collect, victim, n);
274 	eat(n);
275 	collect += n;
276 	if ((remains -= n) != 0)
277 		return 1;
278 	state = next_state;
279 	return 0;
280 }
281 
282 static int __init do_header(void)
283 {
284 	if (!memcmp(collected, "070701", 6)) {
285 		csum_present = false;
286 	} else if (!memcmp(collected, "070702", 6)) {
287 		csum_present = true;
288 	} else {
289 		if (memcmp(collected, "070707", 6) == 0)
290 			error("incorrect cpio method used: use -H newc option");
291 		else
292 			error("no cpio magic");
293 		return 1;
294 	}
295 	parse_header(collected);
296 	next_header = this_header + N_ALIGN(name_len) + body_len;
297 	next_header = (next_header + 3) & ~3;
298 	state = SkipIt;
299 	if (name_len <= 0 || name_len > PATH_MAX)
300 		return 0;
301 	if (S_ISLNK(mode)) {
302 		if (body_len > PATH_MAX)
303 			return 0;
304 		collect = collected = symlink_buf;
305 		remains = N_ALIGN(name_len) + body_len;
306 		next_state = GotSymlink;
307 		state = Collect;
308 		return 0;
309 	}
310 	if (S_ISREG(mode) || !body_len)
311 		read_into(name_buf, N_ALIGN(name_len), GotName);
312 	return 0;
313 }
314 
315 static int __init do_skip(void)
316 {
317 	if (this_header + byte_count < next_header) {
318 		eat(byte_count);
319 		return 1;
320 	} else {
321 		eat(next_header - this_header);
322 		state = next_state;
323 		return 0;
324 	}
325 }
326 
327 static int __init do_reset(void)
328 {
329 	while (byte_count && *victim == '\0')
330 		eat(1);
331 	if (byte_count && (this_header & 3))
332 		error("broken padding");
333 	return 1;
334 }
335 
336 static void __init clean_path(char *path, umode_t fmode)
337 {
338 	struct kstat st;
339 
340 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
341 	    (st.mode ^ fmode) & S_IFMT) {
342 		if (S_ISDIR(st.mode))
343 			init_rmdir(path);
344 		else
345 			init_unlink(path);
346 	}
347 }
348 
349 static int __init maybe_link(void)
350 {
351 	if (nlink >= 2) {
352 		char *old = find_link(major, minor, ino, mode, collected);
353 		if (old) {
354 			clean_path(collected, 0);
355 			return (init_link(old, collected) < 0) ? -1 : 1;
356 		}
357 	}
358 	return 0;
359 }
360 
361 static __initdata struct file *wfile;
362 static __initdata loff_t wfile_pos;
363 
364 static int __init do_name(void)
365 {
366 	state = SkipIt;
367 	next_state = Reset;
368 	if (strcmp(collected, "TRAILER!!!") == 0) {
369 		free_hash();
370 		return 0;
371 	}
372 	clean_path(collected, mode);
373 	if (S_ISREG(mode)) {
374 		int ml = maybe_link();
375 		if (ml >= 0) {
376 			int openflags = O_WRONLY|O_CREAT;
377 			if (ml != 1)
378 				openflags |= O_TRUNC;
379 			wfile = filp_open(collected, openflags, mode);
380 			if (IS_ERR(wfile))
381 				return 0;
382 			wfile_pos = 0;
383 			io_csum = 0;
384 
385 			vfs_fchown(wfile, uid, gid);
386 			vfs_fchmod(wfile, mode);
387 			if (body_len)
388 				vfs_truncate(&wfile->f_path, body_len);
389 			state = CopyFile;
390 		}
391 	} else if (S_ISDIR(mode)) {
392 		init_mkdir(collected, mode);
393 		init_chown(collected, uid, gid, 0);
394 		init_chmod(collected, mode);
395 		dir_add(collected, mtime);
396 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
397 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
398 		if (maybe_link() == 0) {
399 			init_mknod(collected, mode, rdev);
400 			init_chown(collected, uid, gid, 0);
401 			init_chmod(collected, mode);
402 			do_utime(collected, mtime);
403 		}
404 	}
405 	return 0;
406 }
407 
408 static int __init do_copy(void)
409 {
410 	if (byte_count >= body_len) {
411 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
412 			error("write error");
413 
414 		do_utime_path(&wfile->f_path, mtime);
415 		fput(wfile);
416 		if (csum_present && io_csum != hdr_csum)
417 			error("bad data checksum");
418 		eat(body_len);
419 		state = SkipIt;
420 		return 0;
421 	} else {
422 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
423 			error("write error");
424 		body_len -= byte_count;
425 		eat(byte_count);
426 		return 1;
427 	}
428 }
429 
430 static int __init do_symlink(void)
431 {
432 	collected[N_ALIGN(name_len) + body_len] = '\0';
433 	clean_path(collected, 0);
434 	init_symlink(collected + N_ALIGN(name_len), collected);
435 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
436 	do_utime(collected, mtime);
437 	state = SkipIt;
438 	next_state = Reset;
439 	return 0;
440 }
441 
442 static __initdata int (*actions[])(void) = {
443 	[Start]		= do_start,
444 	[Collect]	= do_collect,
445 	[GotHeader]	= do_header,
446 	[SkipIt]	= do_skip,
447 	[GotName]	= do_name,
448 	[CopyFile]	= do_copy,
449 	[GotSymlink]	= do_symlink,
450 	[Reset]		= do_reset,
451 };
452 
453 static long __init write_buffer(char *buf, unsigned long len)
454 {
455 	byte_count = len;
456 	victim = buf;
457 
458 	while (!actions[state]())
459 		;
460 	return len - byte_count;
461 }
462 
463 static long __init flush_buffer(void *bufv, unsigned long len)
464 {
465 	char *buf = bufv;
466 	long written;
467 	long origLen = len;
468 	if (message)
469 		return -1;
470 	while ((written = write_buffer(buf, len)) < len && !message) {
471 		char c = buf[written];
472 		if (c == '0') {
473 			buf += written;
474 			len -= written;
475 			state = Start;
476 		} else if (c == 0) {
477 			buf += written;
478 			len -= written;
479 			state = Reset;
480 		} else
481 			error("junk within compressed archive");
482 	}
483 	return origLen;
484 }
485 
486 static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
487 
488 #include <linux/decompress/generic.h>
489 
490 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
491 {
492 	long written;
493 	decompress_fn decompress;
494 	const char *compress_name;
495 	static __initdata char msg_buf[64];
496 
497 	header_buf = kmalloc(110, GFP_KERNEL);
498 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
499 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
500 
501 	if (!header_buf || !symlink_buf || !name_buf)
502 		panic_show_mem("can't allocate buffers");
503 
504 	state = Start;
505 	this_header = 0;
506 	message = NULL;
507 	while (!message && len) {
508 		loff_t saved_offset = this_header;
509 		if (*buf == '0' && !(this_header & 3)) {
510 			state = Start;
511 			written = write_buffer(buf, len);
512 			buf += written;
513 			len -= written;
514 			continue;
515 		}
516 		if (!*buf) {
517 			buf++;
518 			len--;
519 			this_header++;
520 			continue;
521 		}
522 		this_header = 0;
523 		decompress = decompress_method(buf, len, &compress_name);
524 		pr_debug("Detected %s compressed data\n", compress_name);
525 		if (decompress) {
526 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
527 				   &my_inptr, error);
528 			if (res)
529 				error("decompressor failed");
530 		} else if (compress_name) {
531 			if (!message) {
532 				snprintf(msg_buf, sizeof msg_buf,
533 					 "compression method %s not configured",
534 					 compress_name);
535 				message = msg_buf;
536 			}
537 		} else
538 			error("invalid magic at start of compressed archive");
539 		if (state != Reset)
540 			error("junk at the end of compressed archive");
541 		this_header = saved_offset + my_inptr;
542 		buf += my_inptr;
543 		len -= my_inptr;
544 	}
545 	dir_utime();
546 	kfree(name_buf);
547 	kfree(symlink_buf);
548 	kfree(header_buf);
549 	return message;
550 }
551 
552 static int __initdata do_retain_initrd;
553 
554 static int __init retain_initrd_param(char *str)
555 {
556 	if (*str)
557 		return 0;
558 	do_retain_initrd = 1;
559 	return 1;
560 }
561 __setup("retain_initrd", retain_initrd_param);
562 
563 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
564 static int __init keepinitrd_setup(char *__unused)
565 {
566 	do_retain_initrd = 1;
567 	return 1;
568 }
569 __setup("keepinitrd", keepinitrd_setup);
570 #endif
571 
572 static bool __initdata initramfs_async = true;
573 static int __init initramfs_async_setup(char *str)
574 {
575 	return kstrtobool(str, &initramfs_async) == 0;
576 }
577 __setup("initramfs_async=", initramfs_async_setup);
578 
579 extern char __initramfs_start[];
580 extern unsigned long __initramfs_size;
581 #include <linux/initrd.h>
582 #include <linux/kexec.h>
583 
584 void __init reserve_initrd_mem(void)
585 {
586 	phys_addr_t start;
587 	unsigned long size;
588 
589 	/* Ignore the virtul address computed during device tree parsing */
590 	initrd_start = initrd_end = 0;
591 
592 	if (!phys_initrd_size)
593 		return;
594 	/*
595 	 * Round the memory region to page boundaries as per free_initrd_mem()
596 	 * This allows us to detect whether the pages overlapping the initrd
597 	 * are in use, but more importantly, reserves the entire set of pages
598 	 * as we don't want these pages allocated for other purposes.
599 	 */
600 	start = round_down(phys_initrd_start, PAGE_SIZE);
601 	size = phys_initrd_size + (phys_initrd_start - start);
602 	size = round_up(size, PAGE_SIZE);
603 
604 	if (!memblock_is_region_memory(start, size)) {
605 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
606 		       (u64)start, size);
607 		goto disable;
608 	}
609 
610 	if (memblock_is_region_reserved(start, size)) {
611 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
612 		       (u64)start, size);
613 		goto disable;
614 	}
615 
616 	memblock_reserve(start, size);
617 	/* Now convert initrd to virtual addresses */
618 	initrd_start = (unsigned long)__va(phys_initrd_start);
619 	initrd_end = initrd_start + phys_initrd_size;
620 	initrd_below_start_ok = 1;
621 
622 	return;
623 disable:
624 	pr_cont(" - disabling initrd\n");
625 	initrd_start = 0;
626 	initrd_end = 0;
627 }
628 
629 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
630 {
631 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
632 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
633 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
634 
635 	memblock_free((void *)aligned_start, aligned_end - aligned_start);
636 #endif
637 
638 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
639 			"initrd");
640 }
641 
642 #ifdef CONFIG_KEXEC_CORE
643 static bool __init kexec_free_initrd(void)
644 {
645 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
646 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
647 
648 	/*
649 	 * If the initrd region is overlapped with crashkernel reserved region,
650 	 * free only memory that is not part of crashkernel region.
651 	 */
652 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
653 		return false;
654 
655 	/*
656 	 * Initialize initrd memory region since the kexec boot does not do.
657 	 */
658 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
659 	if (initrd_start < crashk_start)
660 		free_initrd_mem(initrd_start, crashk_start);
661 	if (initrd_end > crashk_end)
662 		free_initrd_mem(crashk_end, initrd_end);
663 	return true;
664 }
665 #else
666 static inline bool kexec_free_initrd(void)
667 {
668 	return false;
669 }
670 #endif /* CONFIG_KEXEC_CORE */
671 
672 #ifdef CONFIG_BLK_DEV_RAM
673 static void __init populate_initrd_image(char *err)
674 {
675 	ssize_t written;
676 	struct file *file;
677 	loff_t pos = 0;
678 
679 	unpack_to_rootfs(__initramfs_start, __initramfs_size);
680 
681 	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
682 			err);
683 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
684 	if (IS_ERR(file))
685 		return;
686 
687 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
688 			&pos);
689 	if (written != initrd_end - initrd_start)
690 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
691 		       written, initrd_end - initrd_start);
692 	fput(file);
693 }
694 #endif /* CONFIG_BLK_DEV_RAM */
695 
696 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
697 {
698 	/* Load the built in initramfs */
699 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
700 	if (err)
701 		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
702 
703 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
704 		goto done;
705 
706 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
707 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
708 	else
709 		printk(KERN_INFO "Unpacking initramfs...\n");
710 
711 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
712 	if (err) {
713 #ifdef CONFIG_BLK_DEV_RAM
714 		populate_initrd_image(err);
715 #else
716 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
717 #endif
718 	}
719 
720 done:
721 	/*
722 	 * If the initrd region is overlapped with crashkernel reserved region,
723 	 * free only memory that is not part of crashkernel region.
724 	 */
725 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
726 		free_initrd_mem(initrd_start, initrd_end);
727 	initrd_start = 0;
728 	initrd_end = 0;
729 
730 	flush_delayed_fput();
731 	task_work_run();
732 }
733 
734 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
735 static async_cookie_t initramfs_cookie;
736 
737 void wait_for_initramfs(void)
738 {
739 	if (!initramfs_cookie) {
740 		/*
741 		 * Something before rootfs_initcall wants to access
742 		 * the filesystem/initramfs. Probably a bug. Make a
743 		 * note, avoid deadlocking the machine, and let the
744 		 * caller's access fail as it used to.
745 		 */
746 		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
747 		return;
748 	}
749 	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
750 }
751 EXPORT_SYMBOL_GPL(wait_for_initramfs);
752 
753 static int __init populate_rootfs(void)
754 {
755 	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
756 						 &initramfs_domain);
757 	usermodehelper_enable();
758 	if (!initramfs_async)
759 		wait_for_initramfs();
760 	return 0;
761 }
762 rootfs_initcall(populate_rootfs);
763