xref: /openbmc/linux/init/initramfs.c (revision c72160fe) 
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 #include <linux/namei.h>
15 #include <linux/init_syscalls.h>
16 
17 static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
18 		loff_t *pos)
19 {
20 	ssize_t out = 0;
21 
22 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
23 	while (count) {
24 		ssize_t rv = kernel_write(file, p, count, pos);
25 
26 		if (rv < 0) {
27 			if (rv == -EINTR || rv == -EAGAIN)
28 				continue;
29 			return out ? out : rv;
30 		} else if (rv == 0)
31 			break;
32 
33 		p += rv;
34 		out += rv;
35 		count -= rv;
36 	}
37 
38 	return out;
39 }
40 
41 static __initdata char *message;
42 static void __init error(char *x)
43 {
44 	if (!message)
45 		message = x;
46 }
47 
48 /* link hash */
49 
50 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
51 
52 static __initdata struct hash {
53 	int ino, minor, major;
54 	umode_t mode;
55 	struct hash *next;
56 	char name[N_ALIGN(PATH_MAX)];
57 } *head[32];
58 
59 static inline int hash(int major, int minor, int ino)
60 {
61 	unsigned long tmp = ino + minor + (major << 3);
62 	tmp += tmp >> 5;
63 	return tmp & 31;
64 }
65 
66 static char __init *find_link(int major, int minor, int ino,
67 			      umode_t mode, char *name)
68 {
69 	struct hash **p, *q;
70 	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
71 		if ((*p)->ino != ino)
72 			continue;
73 		if ((*p)->minor != minor)
74 			continue;
75 		if ((*p)->major != major)
76 			continue;
77 		if (((*p)->mode ^ mode) & S_IFMT)
78 			continue;
79 		return (*p)->name;
80 	}
81 	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
82 	if (!q)
83 		panic("can't allocate link hash entry");
84 	q->major = major;
85 	q->minor = minor;
86 	q->ino = ino;
87 	q->mode = mode;
88 	strcpy(q->name, name);
89 	q->next = NULL;
90 	*p = q;
91 	return NULL;
92 }
93 
94 static void __init free_hash(void)
95 {
96 	struct hash **p, *q;
97 	for (p = head; p < head + 32; p++) {
98 		while (*p) {
99 			q = *p;
100 			*p = q->next;
101 			kfree(q);
102 		}
103 	}
104 }
105 
106 static long __init do_utime(char *filename, time64_t mtime)
107 {
108 	struct timespec64 t[2];
109 
110 	t[0].tv_sec = mtime;
111 	t[0].tv_nsec = 0;
112 	t[1].tv_sec = mtime;
113 	t[1].tv_nsec = 0;
114 	return init_utimes(filename, t);
115 }
116 
117 static __initdata LIST_HEAD(dir_list);
118 struct dir_entry {
119 	struct list_head list;
120 	char *name;
121 	time64_t mtime;
122 };
123 
124 static void __init dir_add(const char *name, time64_t mtime)
125 {
126 	struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
127 	if (!de)
128 		panic("can't allocate dir_entry buffer");
129 	INIT_LIST_HEAD(&de->list);
130 	de->name = kstrdup(name, GFP_KERNEL);
131 	de->mtime = mtime;
132 	list_add(&de->list, &dir_list);
133 }
134 
135 static void __init dir_utime(void)
136 {
137 	struct dir_entry *de, *tmp;
138 	list_for_each_entry_safe(de, tmp, &dir_list, list) {
139 		list_del(&de->list);
140 		do_utime(de->name, de->mtime);
141 		kfree(de->name);
142 		kfree(de);
143 	}
144 }
145 
146 static __initdata time64_t mtime;
147 
148 /* cpio header parsing */
149 
150 static __initdata unsigned long ino, major, minor, nlink;
151 static __initdata umode_t mode;
152 static __initdata unsigned long body_len, name_len;
153 static __initdata uid_t uid;
154 static __initdata gid_t gid;
155 static __initdata unsigned rdev;
156 
157 static void __init parse_header(char *s)
158 {
159 	unsigned long parsed[12];
160 	char buf[9];
161 	int i;
162 
163 	buf[8] = '\0';
164 	for (i = 0, s += 6; i < 12; i++, s += 8) {
165 		memcpy(buf, s, 8);
166 		parsed[i] = simple_strtoul(buf, NULL, 16);
167 	}
168 	ino = parsed[0];
169 	mode = parsed[1];
170 	uid = parsed[2];
171 	gid = parsed[3];
172 	nlink = parsed[4];
173 	mtime = parsed[5]; /* breaks in y2106 */
174 	body_len = parsed[6];
175 	major = parsed[7];
176 	minor = parsed[8];
177 	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
178 	name_len = parsed[11];
179 }
180 
181 /* FSM */
182 
183 static __initdata enum state {
184 	Start,
185 	Collect,
186 	GotHeader,
187 	SkipIt,
188 	GotName,
189 	CopyFile,
190 	GotSymlink,
191 	Reset
192 } state, next_state;
193 
194 static __initdata char *victim;
195 static unsigned long byte_count __initdata;
196 static __initdata loff_t this_header, next_header;
197 
198 static inline void __init eat(unsigned n)
199 {
200 	victim += n;
201 	this_header += n;
202 	byte_count -= n;
203 }
204 
205 static __initdata char *collected;
206 static long remains __initdata;
207 static __initdata char *collect;
208 
209 static void __init read_into(char *buf, unsigned size, enum state next)
210 {
211 	if (byte_count >= size) {
212 		collected = victim;
213 		eat(size);
214 		state = next;
215 	} else {
216 		collect = collected = buf;
217 		remains = size;
218 		next_state = next;
219 		state = Collect;
220 	}
221 }
222 
223 static __initdata char *header_buf, *symlink_buf, *name_buf;
224 
225 static int __init do_start(void)
226 {
227 	read_into(header_buf, 110, GotHeader);
228 	return 0;
229 }
230 
231 static int __init do_collect(void)
232 {
233 	unsigned long n = remains;
234 	if (byte_count < n)
235 		n = byte_count;
236 	memcpy(collect, victim, n);
237 	eat(n);
238 	collect += n;
239 	if ((remains -= n) != 0)
240 		return 1;
241 	state = next_state;
242 	return 0;
243 }
244 
245 static int __init do_header(void)
246 {
247 	if (memcmp(collected, "070707", 6)==0) {
248 		error("incorrect cpio method used: use -H newc option");
249 		return 1;
250 	}
251 	if (memcmp(collected, "070701", 6)) {
252 		error("no cpio magic");
253 		return 1;
254 	}
255 	parse_header(collected);
256 	next_header = this_header + N_ALIGN(name_len) + body_len;
257 	next_header = (next_header + 3) & ~3;
258 	state = SkipIt;
259 	if (name_len <= 0 || name_len > PATH_MAX)
260 		return 0;
261 	if (S_ISLNK(mode)) {
262 		if (body_len > PATH_MAX)
263 			return 0;
264 		collect = collected = symlink_buf;
265 		remains = N_ALIGN(name_len) + body_len;
266 		next_state = GotSymlink;
267 		state = Collect;
268 		return 0;
269 	}
270 	if (S_ISREG(mode) || !body_len)
271 		read_into(name_buf, N_ALIGN(name_len), GotName);
272 	return 0;
273 }
274 
275 static int __init do_skip(void)
276 {
277 	if (this_header + byte_count < next_header) {
278 		eat(byte_count);
279 		return 1;
280 	} else {
281 		eat(next_header - this_header);
282 		state = next_state;
283 		return 0;
284 	}
285 }
286 
287 static int __init do_reset(void)
288 {
289 	while (byte_count && *victim == '\0')
290 		eat(1);
291 	if (byte_count && (this_header & 3))
292 		error("broken padding");
293 	return 1;
294 }
295 
296 static void __init clean_path(char *path, umode_t fmode)
297 {
298 	struct kstat st;
299 
300 	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
301 	    (st.mode ^ fmode) & S_IFMT) {
302 		if (S_ISDIR(st.mode))
303 			init_rmdir(path);
304 		else
305 			init_unlink(path);
306 	}
307 }
308 
309 static int __init maybe_link(void)
310 {
311 	if (nlink >= 2) {
312 		char *old = find_link(major, minor, ino, mode, collected);
313 		if (old) {
314 			clean_path(collected, 0);
315 			return (init_link(old, collected) < 0) ? -1 : 1;
316 		}
317 	}
318 	return 0;
319 }
320 
321 static __initdata struct file *wfile;
322 static __initdata loff_t wfile_pos;
323 
324 static int __init do_name(void)
325 {
326 	state = SkipIt;
327 	next_state = Reset;
328 	if (strcmp(collected, "TRAILER!!!") == 0) {
329 		free_hash();
330 		return 0;
331 	}
332 	clean_path(collected, mode);
333 	if (S_ISREG(mode)) {
334 		int ml = maybe_link();
335 		if (ml >= 0) {
336 			int openflags = O_WRONLY|O_CREAT;
337 			if (ml != 1)
338 				openflags |= O_TRUNC;
339 			wfile = filp_open(collected, openflags, mode);
340 			if (IS_ERR(wfile))
341 				return 0;
342 			wfile_pos = 0;
343 
344 			vfs_fchown(wfile, uid, gid);
345 			vfs_fchmod(wfile, mode);
346 			if (body_len)
347 				vfs_truncate(&wfile->f_path, body_len);
348 			state = CopyFile;
349 		}
350 	} else if (S_ISDIR(mode)) {
351 		init_mkdir(collected, mode);
352 		init_chown(collected, uid, gid, 0);
353 		init_chmod(collected, mode);
354 		dir_add(collected, mtime);
355 	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
356 		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
357 		if (maybe_link() == 0) {
358 			init_mknod(collected, mode, rdev);
359 			init_chown(collected, uid, gid, 0);
360 			init_chmod(collected, mode);
361 			do_utime(collected, mtime);
362 		}
363 	}
364 	return 0;
365 }
366 
367 static int __init do_copy(void)
368 {
369 	if (byte_count >= body_len) {
370 		struct timespec64 t[2] = { };
371 		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
372 			error("write error");
373 
374 		t[0].tv_sec = mtime;
375 		t[1].tv_sec = mtime;
376 		vfs_utimes(&wfile->f_path, t);
377 
378 		fput(wfile);
379 		eat(body_len);
380 		state = SkipIt;
381 		return 0;
382 	} else {
383 		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
384 			error("write error");
385 		body_len -= byte_count;
386 		eat(byte_count);
387 		return 1;
388 	}
389 }
390 
391 static int __init do_symlink(void)
392 {
393 	collected[N_ALIGN(name_len) + body_len] = '\0';
394 	clean_path(collected, 0);
395 	init_symlink(collected + N_ALIGN(name_len), collected);
396 	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
397 	do_utime(collected, mtime);
398 	state = SkipIt;
399 	next_state = Reset;
400 	return 0;
401 }
402 
403 static __initdata int (*actions[])(void) = {
404 	[Start]		= do_start,
405 	[Collect]	= do_collect,
406 	[GotHeader]	= do_header,
407 	[SkipIt]	= do_skip,
408 	[GotName]	= do_name,
409 	[CopyFile]	= do_copy,
410 	[GotSymlink]	= do_symlink,
411 	[Reset]		= do_reset,
412 };
413 
414 static long __init write_buffer(char *buf, unsigned long len)
415 {
416 	byte_count = len;
417 	victim = buf;
418 
419 	while (!actions[state]())
420 		;
421 	return len - byte_count;
422 }
423 
424 static long __init flush_buffer(void *bufv, unsigned long len)
425 {
426 	char *buf = (char *) bufv;
427 	long written;
428 	long origLen = len;
429 	if (message)
430 		return -1;
431 	while ((written = write_buffer(buf, len)) < len && !message) {
432 		char c = buf[written];
433 		if (c == '0') {
434 			buf += written;
435 			len -= written;
436 			state = Start;
437 		} else if (c == 0) {
438 			buf += written;
439 			len -= written;
440 			state = Reset;
441 		} else
442 			error("junk within compressed archive");
443 	}
444 	return origLen;
445 }
446 
447 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
448 
449 #include <linux/decompress/generic.h>
450 
451 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
452 {
453 	long written;
454 	decompress_fn decompress;
455 	const char *compress_name;
456 	static __initdata char msg_buf[64];
457 
458 	header_buf = kmalloc(110, GFP_KERNEL);
459 	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
460 	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
461 
462 	if (!header_buf || !symlink_buf || !name_buf)
463 		panic("can't allocate buffers");
464 
465 	state = Start;
466 	this_header = 0;
467 	message = NULL;
468 	while (!message && len) {
469 		loff_t saved_offset = this_header;
470 		if (*buf == '0' && !(this_header & 3)) {
471 			state = Start;
472 			written = write_buffer(buf, len);
473 			buf += written;
474 			len -= written;
475 			continue;
476 		}
477 		if (!*buf) {
478 			buf++;
479 			len--;
480 			this_header++;
481 			continue;
482 		}
483 		this_header = 0;
484 		decompress = decompress_method(buf, len, &compress_name);
485 		pr_debug("Detected %s compressed data\n", compress_name);
486 		if (decompress) {
487 			int res = decompress(buf, len, NULL, flush_buffer, NULL,
488 				   &my_inptr, error);
489 			if (res)
490 				error("decompressor failed");
491 		} else if (compress_name) {
492 			if (!message) {
493 				snprintf(msg_buf, sizeof msg_buf,
494 					 "compression method %s not configured",
495 					 compress_name);
496 				message = msg_buf;
497 			}
498 		} else
499 			error("invalid magic at start of compressed archive");
500 		if (state != Reset)
501 			error("junk at the end of compressed archive");
502 		this_header = saved_offset + my_inptr;
503 		buf += my_inptr;
504 		len -= my_inptr;
505 	}
506 	dir_utime();
507 	kfree(name_buf);
508 	kfree(symlink_buf);
509 	kfree(header_buf);
510 	return message;
511 }
512 
513 static int __initdata do_retain_initrd;
514 
515 static int __init retain_initrd_param(char *str)
516 {
517 	if (*str)
518 		return 0;
519 	do_retain_initrd = 1;
520 	return 1;
521 }
522 __setup("retain_initrd", retain_initrd_param);
523 
524 #ifdef CONFIG_ARCH_HAS_KEEPINITRD
525 static int __init keepinitrd_setup(char *__unused)
526 {
527 	do_retain_initrd = 1;
528 	return 1;
529 }
530 __setup("keepinitrd", keepinitrd_setup);
531 #endif
532 
533 extern char __initramfs_start[];
534 extern unsigned long __initramfs_size;
535 #include <linux/initrd.h>
536 #include <linux/kexec.h>
537 
538 void __init reserve_initrd_mem(void)
539 {
540 	phys_addr_t start;
541 	unsigned long size;
542 
543 	/* Ignore the virtul address computed during device tree parsing */
544 	initrd_start = initrd_end = 0;
545 
546 	if (!phys_initrd_size)
547 		return;
548 	/*
549 	 * Round the memory region to page boundaries as per free_initrd_mem()
550 	 * This allows us to detect whether the pages overlapping the initrd
551 	 * are in use, but more importantly, reserves the entire set of pages
552 	 * as we don't want these pages allocated for other purposes.
553 	 */
554 	start = round_down(phys_initrd_start, PAGE_SIZE);
555 	size = phys_initrd_size + (phys_initrd_start - start);
556 	size = round_up(size, PAGE_SIZE);
557 
558 	if (!memblock_is_region_memory(start, size)) {
559 		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
560 		       (u64)start, size);
561 		goto disable;
562 	}
563 
564 	if (memblock_is_region_reserved(start, size)) {
565 		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
566 		       (u64)start, size);
567 		goto disable;
568 	}
569 
570 	memblock_reserve(start, size);
571 	/* Now convert initrd to virtual addresses */
572 	initrd_start = (unsigned long)__va(phys_initrd_start);
573 	initrd_end = initrd_start + phys_initrd_size;
574 	initrd_below_start_ok = 1;
575 
576 	return;
577 disable:
578 	pr_cont(" - disabling initrd\n");
579 	initrd_start = 0;
580 	initrd_end = 0;
581 }
582 
583 void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
584 {
585 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK
586 	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
587 	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
588 
589 	memblock_free(__pa(aligned_start), aligned_end - aligned_start);
590 #endif
591 
592 	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
593 			"initrd");
594 }
595 
596 #ifdef CONFIG_KEXEC_CORE
597 static bool __init kexec_free_initrd(void)
598 {
599 	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
600 	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
601 
602 	/*
603 	 * If the initrd region is overlapped with crashkernel reserved region,
604 	 * free only memory that is not part of crashkernel region.
605 	 */
606 	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
607 		return false;
608 
609 	/*
610 	 * Initialize initrd memory region since the kexec boot does not do.
611 	 */
612 	memset((void *)initrd_start, 0, initrd_end - initrd_start);
613 	if (initrd_start < crashk_start)
614 		free_initrd_mem(initrd_start, crashk_start);
615 	if (initrd_end > crashk_end)
616 		free_initrd_mem(crashk_end, initrd_end);
617 	return true;
618 }
619 #else
620 static inline bool kexec_free_initrd(void)
621 {
622 	return false;
623 }
624 #endif /* CONFIG_KEXEC_CORE */
625 
626 #ifdef CONFIG_BLK_DEV_RAM
627 static void __init populate_initrd_image(char *err)
628 {
629 	ssize_t written;
630 	struct file *file;
631 	loff_t pos = 0;
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 	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
638 	if (IS_ERR(file))
639 		return;
640 
641 	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
642 			&pos);
643 	if (written != initrd_end - initrd_start)
644 		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
645 		       written, initrd_end - initrd_start);
646 	fput(file);
647 }
648 #endif /* CONFIG_BLK_DEV_RAM */
649 
650 static int __init populate_rootfs(void)
651 {
652 	/* Load the built in initramfs */
653 	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
654 	if (err)
655 		panic("%s", err); /* Failed to decompress INTERNAL initramfs */
656 
657 	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
658 		goto done;
659 
660 	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
661 		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
662 	else
663 		printk(KERN_INFO "Unpacking initramfs...\n");
664 
665 	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
666 	if (err) {
667 #ifdef CONFIG_BLK_DEV_RAM
668 		populate_initrd_image(err);
669 #else
670 		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
671 #endif
672 	}
673 
674 done:
675 	/*
676 	 * If the initrd region is overlapped with crashkernel reserved region,
677 	 * free only memory that is not part of crashkernel region.
678 	 */
679 	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
680 		free_initrd_mem(initrd_start, initrd_end);
681 	initrd_start = 0;
682 	initrd_end = 0;
683 
684 	flush_delayed_fput();
685 	return 0;
686 }
687 rootfs_initcall(populate_rootfs);
688