xref: /kernel/linux/linux-6.6/init/initramfs.c (revision 62306a36) 
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
22static __initdata bool csum_present;
23static __initdata u32 io_csum;
24
25static 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
56static __initdata char *message;
57static void __init error(char *x)
58{
59	if (!message)
60		message = x;
61}
62
63#define panic_show_mem(fmt, ...) \
64	({ show_mem(); panic(fmt, ##__VA_ARGS__); })
65
66/* link hash */
67
68#define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
69
70static __initdata struct hash {
71	int ino, minor, major;
72	umode_t mode;
73	struct hash *next;
74	char name[N_ALIGN(PATH_MAX)];
75} *head[32];
76
77static inline int hash(int major, int minor, int ino)
78{
79	unsigned long tmp = ino + minor + (major << 3);
80	tmp += tmp >> 5;
81	return tmp & 31;
82}
83
84static char __init *find_link(int major, int minor, int ino,
85			      umode_t mode, char *name)
86{
87	struct hash **p, *q;
88	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
89		if ((*p)->ino != ino)
90			continue;
91		if ((*p)->minor != minor)
92			continue;
93		if ((*p)->major != major)
94			continue;
95		if (((*p)->mode ^ mode) & S_IFMT)
96			continue;
97		return (*p)->name;
98	}
99	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
100	if (!q)
101		panic_show_mem("can't allocate link hash entry");
102	q->major = major;
103	q->minor = minor;
104	q->ino = ino;
105	q->mode = mode;
106	strcpy(q->name, name);
107	q->next = NULL;
108	*p = q;
109	return NULL;
110}
111
112static void __init free_hash(void)
113{
114	struct hash **p, *q;
115	for (p = head; p < head + 32; p++) {
116		while (*p) {
117			q = *p;
118			*p = q->next;
119			kfree(q);
120		}
121	}
122}
123
124#ifdef CONFIG_INITRAMFS_PRESERVE_MTIME
125static void __init do_utime(char *filename, time64_t mtime)
126{
127	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
128	init_utimes(filename, t);
129}
130
131static void __init do_utime_path(const struct path *path, time64_t mtime)
132{
133	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
134	vfs_utimes(path, t);
135}
136
137static __initdata LIST_HEAD(dir_list);
138struct dir_entry {
139	struct list_head list;
140	time64_t mtime;
141	char name[];
142};
143
144static void __init dir_add(const char *name, time64_t mtime)
145{
146	size_t nlen = strlen(name) + 1;
147	struct dir_entry *de;
148
149	de = kmalloc(sizeof(struct dir_entry) + nlen, GFP_KERNEL);
150	if (!de)
151		panic_show_mem("can't allocate dir_entry buffer");
152	INIT_LIST_HEAD(&de->list);
153	strscpy(de->name, name, nlen);
154	de->mtime = mtime;
155	list_add(&de->list, &dir_list);
156}
157
158static void __init dir_utime(void)
159{
160	struct dir_entry *de, *tmp;
161	list_for_each_entry_safe(de, tmp, &dir_list, list) {
162		list_del(&de->list);
163		do_utime(de->name, de->mtime);
164		kfree(de);
165	}
166}
167#else
168static void __init do_utime(char *filename, time64_t mtime) {}
169static void __init do_utime_path(const struct path *path, time64_t mtime) {}
170static void __init dir_add(const char *name, time64_t mtime) {}
171static void __init dir_utime(void) {}
172#endif
173
174static __initdata time64_t mtime;
175
176/* cpio header parsing */
177
178static __initdata unsigned long ino, major, minor, nlink;
179static __initdata umode_t mode;
180static __initdata unsigned long body_len, name_len;
181static __initdata uid_t uid;
182static __initdata gid_t gid;
183static __initdata unsigned rdev;
184static __initdata u32 hdr_csum;
185
186static void __init parse_header(char *s)
187{
188	unsigned long parsed[13];
189	char buf[9];
190	int i;
191
192	buf[8] = '\0';
193	for (i = 0, s += 6; i < 13; i++, s += 8) {
194		memcpy(buf, s, 8);
195		parsed[i] = simple_strtoul(buf, NULL, 16);
196	}
197	ino = parsed[0];
198	mode = parsed[1];
199	uid = parsed[2];
200	gid = parsed[3];
201	nlink = parsed[4];
202	mtime = parsed[5]; /* breaks in y2106 */
203	body_len = parsed[6];
204	major = parsed[7];
205	minor = parsed[8];
206	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
207	name_len = parsed[11];
208	hdr_csum = parsed[12];
209}
210
211/* FSM */
212
213static __initdata enum state {
214	Start,
215	Collect,
216	GotHeader,
217	SkipIt,
218	GotName,
219	CopyFile,
220	GotSymlink,
221	Reset
222} state, next_state;
223
224static __initdata char *victim;
225static unsigned long byte_count __initdata;
226static __initdata loff_t this_header, next_header;
227
228static inline void __init eat(unsigned n)
229{
230	victim += n;
231	this_header += n;
232	byte_count -= n;
233}
234
235static __initdata char *collected;
236static long remains __initdata;
237static __initdata char *collect;
238
239static void __init read_into(char *buf, unsigned size, enum state next)
240{
241	if (byte_count >= size) {
242		collected = victim;
243		eat(size);
244		state = next;
245	} else {
246		collect = collected = buf;
247		remains = size;
248		next_state = next;
249		state = Collect;
250	}
251}
252
253static __initdata char *header_buf, *symlink_buf, *name_buf;
254
255static int __init do_start(void)
256{
257	read_into(header_buf, 110, GotHeader);
258	return 0;
259}
260
261static int __init do_collect(void)
262{
263	unsigned long n = remains;
264	if (byte_count < n)
265		n = byte_count;
266	memcpy(collect, victim, n);
267	eat(n);
268	collect += n;
269	if ((remains -= n) != 0)
270		return 1;
271	state = next_state;
272	return 0;
273}
274
275static int __init do_header(void)
276{
277	if (!memcmp(collected, "070701", 6)) {
278		csum_present = false;
279	} else if (!memcmp(collected, "070702", 6)) {
280		csum_present = true;
281	} else {
282		if (memcmp(collected, "070707", 6) == 0)
283			error("incorrect cpio method used: use -H newc option");
284		else
285			error("no cpio magic");
286		return 1;
287	}
288	parse_header(collected);
289	next_header = this_header + N_ALIGN(name_len) + body_len;
290	next_header = (next_header + 3) & ~3;
291	state = SkipIt;
292	if (name_len <= 0 || name_len > PATH_MAX)
293		return 0;
294	if (S_ISLNK(mode)) {
295		if (body_len > PATH_MAX)
296			return 0;
297		collect = collected = symlink_buf;
298		remains = N_ALIGN(name_len) + body_len;
299		next_state = GotSymlink;
300		state = Collect;
301		return 0;
302	}
303	if (S_ISREG(mode) || !body_len)
304		read_into(name_buf, N_ALIGN(name_len), GotName);
305	return 0;
306}
307
308static int __init do_skip(void)
309{
310	if (this_header + byte_count < next_header) {
311		eat(byte_count);
312		return 1;
313	} else {
314		eat(next_header - this_header);
315		state = next_state;
316		return 0;
317	}
318}
319
320static int __init do_reset(void)
321{
322	while (byte_count && *victim == '\0')
323		eat(1);
324	if (byte_count && (this_header & 3))
325		error("broken padding");
326	return 1;
327}
328
329static void __init clean_path(char *path, umode_t fmode)
330{
331	struct kstat st;
332
333	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
334	    (st.mode ^ fmode) & S_IFMT) {
335		if (S_ISDIR(st.mode))
336			init_rmdir(path);
337		else
338			init_unlink(path);
339	}
340}
341
342static int __init maybe_link(void)
343{
344	if (nlink >= 2) {
345		char *old = find_link(major, minor, ino, mode, collected);
346		if (old) {
347			clean_path(collected, 0);
348			return (init_link(old, collected) < 0) ? -1 : 1;
349		}
350	}
351	return 0;
352}
353
354static __initdata struct file *wfile;
355static __initdata loff_t wfile_pos;
356
357static int __init do_name(void)
358{
359	state = SkipIt;
360	next_state = Reset;
361	if (strcmp(collected, "TRAILER!!!") == 0) {
362		free_hash();
363		return 0;
364	}
365	clean_path(collected, mode);
366	if (S_ISREG(mode)) {
367		int ml = maybe_link();
368		if (ml >= 0) {
369			int openflags = O_WRONLY|O_CREAT;
370			if (ml != 1)
371				openflags |= O_TRUNC;
372			wfile = filp_open(collected, openflags, mode);
373			if (IS_ERR(wfile))
374				return 0;
375			wfile_pos = 0;
376			io_csum = 0;
377
378			vfs_fchown(wfile, uid, gid);
379			vfs_fchmod(wfile, mode);
380			if (body_len)
381				vfs_truncate(&wfile->f_path, body_len);
382			state = CopyFile;
383		}
384	} else if (S_ISDIR(mode)) {
385		init_mkdir(collected, mode);
386		init_chown(collected, uid, gid, 0);
387		init_chmod(collected, mode);
388		dir_add(collected, mtime);
389	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
390		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
391		if (maybe_link() == 0) {
392			init_mknod(collected, mode, rdev);
393			init_chown(collected, uid, gid, 0);
394			init_chmod(collected, mode);
395			do_utime(collected, mtime);
396		}
397	}
398	return 0;
399}
400
401static int __init do_copy(void)
402{
403	if (byte_count >= body_len) {
404		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
405			error("write error");
406
407		do_utime_path(&wfile->f_path, mtime);
408		fput(wfile);
409		if (csum_present && io_csum != hdr_csum)
410			error("bad data checksum");
411		eat(body_len);
412		state = SkipIt;
413		return 0;
414	} else {
415		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
416			error("write error");
417		body_len -= byte_count;
418		eat(byte_count);
419		return 1;
420	}
421}
422
423static int __init do_symlink(void)
424{
425	collected[N_ALIGN(name_len) + body_len] = '\0';
426	clean_path(collected, 0);
427	init_symlink(collected + N_ALIGN(name_len), collected);
428	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
429	do_utime(collected, mtime);
430	state = SkipIt;
431	next_state = Reset;
432	return 0;
433}
434
435static __initdata int (*actions[])(void) = {
436	[Start]		= do_start,
437	[Collect]	= do_collect,
438	[GotHeader]	= do_header,
439	[SkipIt]	= do_skip,
440	[GotName]	= do_name,
441	[CopyFile]	= do_copy,
442	[GotSymlink]	= do_symlink,
443	[Reset]		= do_reset,
444};
445
446static long __init write_buffer(char *buf, unsigned long len)
447{
448	byte_count = len;
449	victim = buf;
450
451	while (!actions[state]())
452		;
453	return len - byte_count;
454}
455
456static long __init flush_buffer(void *bufv, unsigned long len)
457{
458	char *buf = bufv;
459	long written;
460	long origLen = len;
461	if (message)
462		return -1;
463	while ((written = write_buffer(buf, len)) < len && !message) {
464		char c = buf[written];
465		if (c == '0') {
466			buf += written;
467			len -= written;
468			state = Start;
469		} else if (c == 0) {
470			buf += written;
471			len -= written;
472			state = Reset;
473		} else
474			error("junk within compressed archive");
475	}
476	return origLen;
477}
478
479static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
480
481#include <linux/decompress/generic.h>
482
483static char * __init unpack_to_rootfs(char *buf, unsigned long len)
484{
485	long written;
486	decompress_fn decompress;
487	const char *compress_name;
488	static __initdata char msg_buf[64];
489
490	header_buf = kmalloc(110, GFP_KERNEL);
491	symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
492	name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
493
494	if (!header_buf || !symlink_buf || !name_buf)
495		panic_show_mem("can't allocate buffers");
496
497	state = Start;
498	this_header = 0;
499	message = NULL;
500	while (!message && len) {
501		loff_t saved_offset = this_header;
502		if (*buf == '0' && !(this_header & 3)) {
503			state = Start;
504			written = write_buffer(buf, len);
505			buf += written;
506			len -= written;
507			continue;
508		}
509		if (!*buf) {
510			buf++;
511			len--;
512			this_header++;
513			continue;
514		}
515		this_header = 0;
516		decompress = decompress_method(buf, len, &compress_name);
517		pr_debug("Detected %s compressed data\n", compress_name);
518		if (decompress) {
519			int res = decompress(buf, len, NULL, flush_buffer, NULL,
520				   &my_inptr, error);
521			if (res)
522				error("decompressor failed");
523		} else if (compress_name) {
524			if (!message) {
525				snprintf(msg_buf, sizeof msg_buf,
526					 "compression method %s not configured",
527					 compress_name);
528				message = msg_buf;
529			}
530		} else
531			error("invalid magic at start of compressed archive");
532		if (state != Reset)
533			error("junk at the end of compressed archive");
534		this_header = saved_offset + my_inptr;
535		buf += my_inptr;
536		len -= my_inptr;
537	}
538	dir_utime();
539	kfree(name_buf);
540	kfree(symlink_buf);
541	kfree(header_buf);
542	return message;
543}
544
545static int __initdata do_retain_initrd;
546
547static int __init retain_initrd_param(char *str)
548{
549	if (*str)
550		return 0;
551	do_retain_initrd = 1;
552	return 1;
553}
554__setup("retain_initrd", retain_initrd_param);
555
556#ifdef CONFIG_ARCH_HAS_KEEPINITRD
557static int __init keepinitrd_setup(char *__unused)
558{
559	do_retain_initrd = 1;
560	return 1;
561}
562__setup("keepinitrd", keepinitrd_setup);
563#endif
564
565static bool __initdata initramfs_async = true;
566static int __init initramfs_async_setup(char *str)
567{
568	return kstrtobool(str, &initramfs_async) == 0;
569}
570__setup("initramfs_async=", initramfs_async_setup);
571
572extern char __initramfs_start[];
573extern unsigned long __initramfs_size;
574#include <linux/initrd.h>
575#include <linux/kexec.h>
576
577void __init reserve_initrd_mem(void)
578{
579	phys_addr_t start;
580	unsigned long size;
581
582	/* Ignore the virtul address computed during device tree parsing */
583	initrd_start = initrd_end = 0;
584
585	if (!phys_initrd_size)
586		return;
587	/*
588	 * Round the memory region to page boundaries as per free_initrd_mem()
589	 * This allows us to detect whether the pages overlapping the initrd
590	 * are in use, but more importantly, reserves the entire set of pages
591	 * as we don't want these pages allocated for other purposes.
592	 */
593	start = round_down(phys_initrd_start, PAGE_SIZE);
594	size = phys_initrd_size + (phys_initrd_start - start);
595	size = round_up(size, PAGE_SIZE);
596
597	if (!memblock_is_region_memory(start, size)) {
598		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
599		       (u64)start, size);
600		goto disable;
601	}
602
603	if (memblock_is_region_reserved(start, size)) {
604		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
605		       (u64)start, size);
606		goto disable;
607	}
608
609	memblock_reserve(start, size);
610	/* Now convert initrd to virtual addresses */
611	initrd_start = (unsigned long)__va(phys_initrd_start);
612	initrd_end = initrd_start + phys_initrd_size;
613	initrd_below_start_ok = 1;
614
615	return;
616disable:
617	pr_cont(" - disabling initrd\n");
618	initrd_start = 0;
619	initrd_end = 0;
620}
621
622void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
623{
624#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
625	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
626	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
627
628	memblock_free((void *)aligned_start, aligned_end - aligned_start);
629#endif
630
631	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
632			"initrd");
633}
634
635#ifdef CONFIG_KEXEC_CORE
636static bool __init kexec_free_initrd(void)
637{
638	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
639	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
640
641	/*
642	 * If the initrd region is overlapped with crashkernel reserved region,
643	 * free only memory that is not part of crashkernel region.
644	 */
645	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
646		return false;
647
648	/*
649	 * Initialize initrd memory region since the kexec boot does not do.
650	 */
651	memset((void *)initrd_start, 0, initrd_end - initrd_start);
652	if (initrd_start < crashk_start)
653		free_initrd_mem(initrd_start, crashk_start);
654	if (initrd_end > crashk_end)
655		free_initrd_mem(crashk_end, initrd_end);
656	return true;
657}
658#else
659static inline bool kexec_free_initrd(void)
660{
661	return false;
662}
663#endif /* CONFIG_KEXEC_CORE */
664
665#ifdef CONFIG_BLK_DEV_RAM
666static void __init populate_initrd_image(char *err)
667{
668	ssize_t written;
669	struct file *file;
670	loff_t pos = 0;
671
672	unpack_to_rootfs(__initramfs_start, __initramfs_size);
673
674	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
675			err);
676	file = filp_open("/initrd.image", O_WRONLY | O_CREAT, 0700);
677	if (IS_ERR(file))
678		return;
679
680	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
681			&pos);
682	if (written != initrd_end - initrd_start)
683		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
684		       written, initrd_end - initrd_start);
685	fput(file);
686}
687#endif /* CONFIG_BLK_DEV_RAM */
688
689static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
690{
691	/* Load the built in initramfs */
692	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
693	if (err)
694		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
695
696	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
697		goto done;
698
699	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
700		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
701	else
702		printk(KERN_INFO "Unpacking initramfs...\n");
703
704	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
705	if (err) {
706#ifdef CONFIG_BLK_DEV_RAM
707		populate_initrd_image(err);
708#else
709		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
710#endif
711	}
712
713done:
714	/*
715	 * If the initrd region is overlapped with crashkernel reserved region,
716	 * free only memory that is not part of crashkernel region.
717	 */
718	if (!do_retain_initrd && initrd_start && !kexec_free_initrd())
719		free_initrd_mem(initrd_start, initrd_end);
720	initrd_start = 0;
721	initrd_end = 0;
722
723	flush_delayed_fput();
724	task_work_run();
725}
726
727static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
728static async_cookie_t initramfs_cookie;
729
730void wait_for_initramfs(void)
731{
732	if (!initramfs_cookie) {
733		/*
734		 * Something before rootfs_initcall wants to access
735		 * the filesystem/initramfs. Probably a bug. Make a
736		 * note, avoid deadlocking the machine, and let the
737		 * caller's access fail as it used to.
738		 */
739		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
740		return;
741	}
742	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
743}
744EXPORT_SYMBOL_GPL(wait_for_initramfs);
745
746static int __init populate_rootfs(void)
747{
748	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
749						 &initramfs_domain);
750	usermodehelper_enable();
751	if (!initramfs_async)
752		wait_for_initramfs();
753	return 0;
754}
755rootfs_initcall(populate_rootfs);
756