xref: /kernel/linux/linux-5.10/init/initramfs.c (revision 8c2ecf20) 
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
17static 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
41static __initdata char *message;
42static 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
52static __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
59static 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
66static 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
94static 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
106static 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
117static __initdata LIST_HEAD(dir_list);
118struct dir_entry {
119	struct list_head list;
120	char *name;
121	time64_t mtime;
122};
123
124static 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
135static 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
146static __initdata time64_t mtime;
147
148/* cpio header parsing */
149
150static __initdata unsigned long ino, major, minor, nlink;
151static __initdata umode_t mode;
152static __initdata unsigned long body_len, name_len;
153static __initdata uid_t uid;
154static __initdata gid_t gid;
155static __initdata unsigned rdev;
156
157static 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
183static __initdata enum state {
184	Start,
185	Collect,
186	GotHeader,
187	SkipIt,
188	GotName,
189	CopyFile,
190	GotSymlink,
191	Reset
192} state, next_state;
193
194static __initdata char *victim;
195static unsigned long byte_count __initdata;
196static __initdata loff_t this_header, next_header;
197
198static inline void __init eat(unsigned n)
199{
200	victim += n;
201	this_header += n;
202	byte_count -= n;
203}
204
205static __initdata char *collected;
206static long remains __initdata;
207static __initdata char *collect;
208
209static 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
223static __initdata char *header_buf, *symlink_buf, *name_buf;
224
225static int __init do_start(void)
226{
227	read_into(header_buf, 110, GotHeader);
228	return 0;
229}
230
231static 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
245static 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
275static 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
287static 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
296static 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
309static 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
321static __initdata struct file *wfile;
322static __initdata loff_t wfile_pos;
323
324static 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
367static 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
391static 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
403static __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
414static 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
424static 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
447static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
448
449#include <linux/decompress/generic.h>
450
451static 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
513static int __initdata do_retain_initrd;
514
515static 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
525static int __init keepinitrd_setup(char *__unused)
526{
527	do_retain_initrd = 1;
528	return 1;
529}
530__setup("keepinitrd", keepinitrd_setup);
531#endif
532
533extern char __initramfs_start[];
534extern unsigned long __initramfs_size;
535#include <linux/initrd.h>
536#include <linux/kexec.h>
537
538void __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;
577disable:
578	pr_cont(" - disabling initrd\n");
579	initrd_start = 0;
580	initrd_end = 0;
581}
582
583void __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
597static 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
620static inline bool kexec_free_initrd(void)
621{
622	return false;
623}
624#endif /* CONFIG_KEXEC_CORE */
625
626#ifdef CONFIG_BLK_DEV_RAM
627static 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
650static 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
674done:
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}
687rootfs_initcall(populate_rootfs);
688