initramfs.c

  1// SPDX-License-Identifier: GPL-2.0
  2#include <linux/init.h>
  3#include <linux/async.h>
  4#include <linux/export.h>
  5#include <linux/fs.h>
  6#include <linux/slab.h>
  7#include <linux/types.h>
  8#include <linux/fcntl.h>
  9#include <linux/delay.h>
 10#include <linux/string.h>
 11#include <linux/dirent.h>
 12#include <linux/syscalls.h>
 13#include <linux/utime.h>
 14#include <linux/file.h>
 15#include <linux/kstrtox.h>
 16#include <linux/memblock.h>
 17#include <linux/mm.h>
 18#include <linux/namei.h>
 19#include <linux/init_syscalls.h>
 20#include <linux/umh.h>
 21#include <linux/security.h>
 22#include <linux/overflow.h>
 23
 24#include "do_mounts.h"
 25#include "initramfs_internal.h"
 26
 27static __initdata bool csum_present;
 28static __initdata u32 io_csum;
 29
 30static ssize_t __init xwrite(struct file *file, const unsigned char *p,
 31		size_t count, loff_t *pos)
 32{
 33	ssize_t out = 0;
 34
 35	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
 36	while (count) {
 37		ssize_t rv = kernel_write(file, p, count, pos);
 38
 39		if (rv < 0) {
 40			if (rv == -EINTR || rv == -EAGAIN)
 41				continue;
 42			return out ? out : rv;
 43		} else if (rv == 0)
 44			break;
 45
 46		if (csum_present) {
 47			ssize_t i;
 48
 49			for (i = 0; i < rv; i++)
 50				io_csum += p[i];
 51		}
 52
 53		p += rv;
 54		out += rv;
 55		count -= rv;
 56	}
 57
 58	return out;
 59}
 60
 61static __initdata char *message;
 62static void __init error(char *x)
 63{
 64	if (!message)
 65		message = x;
 66}
 67
 68#define panic_show_mem(fmt, ...) \
 69	({ show_mem(); panic(fmt, ##__VA_ARGS__); })
 70
 71/* link hash */
 72
 73#define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
 74
 75static __initdata struct hash {
 76	int ino, minor, major;
 77	umode_t mode;
 78	struct hash *next;
 79	char name[N_ALIGN(PATH_MAX)];
 80} *head[32];
 81static __initdata bool hardlink_seen;
 82
 83static inline int hash(int major, int minor, int ino)
 84{
 85	unsigned long tmp = ino + minor + (major << 3);
 86	tmp += tmp >> 5;
 87	return tmp & 31;
 88}
 89
 90static char __init *find_link(int major, int minor, int ino,
 91			      umode_t mode, char *name)
 92{
 93	struct hash **p, *q;
 94	for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
 95		if ((*p)->ino != ino)
 96			continue;
 97		if ((*p)->minor != minor)
 98			continue;
 99		if ((*p)->major != major)
100			continue;
101		if (((*p)->mode ^ mode) & S_IFMT)
102			continue;
103		return (*p)->name;
104	}
105	q = kmalloc(sizeof(struct hash), GFP_KERNEL);
106	if (!q)
107		panic_show_mem("can't allocate link hash entry");
108	q->major = major;
109	q->minor = minor;
110	q->ino = ino;
111	q->mode = mode;
112	strscpy(q->name, name);
113	q->next = NULL;
114	*p = q;
115	hardlink_seen = true;
116	return NULL;
117}
118
119static void __init free_hash(void)
120{
121	struct hash **p, *q;
122	for (p = head; hardlink_seen && p < head + 32; p++) {
123		while (*p) {
124			q = *p;
125			*p = q->next;
126			kfree(q);
127		}
128	}
129	hardlink_seen = false;
130}
131
132#ifdef CONFIG_INITRAMFS_PRESERVE_MTIME
133static void __init do_utime(char *filename, time64_t mtime)
134{
135	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
136	init_utimes(filename, t);
137}
138
139static void __init do_utime_path(const struct path *path, time64_t mtime)
140{
141	struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
142	vfs_utimes(path, t);
143}
144
145static __initdata LIST_HEAD(dir_list);
146struct dir_entry {
147	struct list_head list;
148	time64_t mtime;
149	char name[];
150};
151
152static void __init dir_add(const char *name, size_t nlen, time64_t mtime)
153{
154	struct dir_entry *de;
155
156	de = kmalloc(struct_size(de, name, 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
165static 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
175static void __init do_utime(char *filename, time64_t mtime) {}
176static void __init do_utime_path(const struct path *path, time64_t mtime) {}
177static void __init dir_add(const char *name, size_t nlen, time64_t mtime) {}
178static void __init dir_utime(void) {}
179#endif
180
181static __initdata time64_t mtime;
182
183/* cpio header parsing */
184
185static __initdata unsigned long ino, major, minor, nlink;
186static __initdata umode_t mode;
187static __initdata unsigned long body_len, name_len;
188static __initdata uid_t uid;
189static __initdata gid_t gid;
190static __initdata unsigned rdev;
191static __initdata u32 hdr_csum;
192
193static void __init parse_header(char *s)
194{
195	unsigned long parsed[13];
196	int i;
197
198	for (i = 0, s += 6; i < 13; i++, s += 8)
199		parsed[i] = simple_strntoul(s, NULL, 16, 8);
200
201	ino = parsed[0];
202	mode = parsed[1];
203	uid = parsed[2];
204	gid = parsed[3];
205	nlink = parsed[4];
206	mtime = parsed[5]; /* breaks in y2106 */
207	body_len = parsed[6];
208	major = parsed[7];
209	minor = parsed[8];
210	rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
211	name_len = parsed[11];
212	hdr_csum = parsed[12];
213}
214
215/* FSM */
216
217static __initdata enum state {
218	Start,
219	Collect,
220	GotHeader,
221	SkipIt,
222	GotName,
223	CopyFile,
224	GotSymlink,
225	Reset
226} state, next_state;
227
228static __initdata char *victim;
229static unsigned long byte_count __initdata;
230static __initdata loff_t this_header, next_header;
231
232static inline void __init eat(unsigned n)
233{
234	victim += n;
235	this_header += n;
236	byte_count -= n;
237}
238
239static __initdata char *collected;
240static long remains __initdata;
241static __initdata char *collect;
242
243static void __init read_into(char *buf, unsigned size, enum state next)
244{
245	if (byte_count >= size) {
246		collected = victim;
247		eat(size);
248		state = next;
249	} else {
250		collect = collected = buf;
251		remains = size;
252		next_state = next;
253		state = Collect;
254	}
255}
256
257static __initdata char *header_buf, *symlink_buf, *name_buf;
258
259static int __init do_start(void)
260{
261	read_into(header_buf, CPIO_HDRLEN, GotHeader);
262	return 0;
263}
264
265static int __init do_collect(void)
266{
267	unsigned long n = remains;
268	if (byte_count < n)
269		n = byte_count;
270	memcpy(collect, victim, n);
271	eat(n);
272	collect += n;
273	if ((remains -= n) != 0)
274		return 1;
275	state = next_state;
276	return 0;
277}
278
279static int __init do_header(void)
280{
281	if (!memcmp(collected, "070701", 6)) {
282		csum_present = false;
283	} else if (!memcmp(collected, "070702", 6)) {
284		csum_present = true;
285	} else {
286		if (memcmp(collected, "070707", 6) == 0)
287			error("incorrect cpio method used: use -H newc option");
288		else
289			error("no cpio magic");
290		return 1;
291	}
292	parse_header(collected);
293	next_header = this_header + N_ALIGN(name_len) + body_len;
294	next_header = (next_header + 3) & ~3;
295	state = SkipIt;
296	if (name_len <= 0 || name_len > PATH_MAX)
297		return 0;
298	if (S_ISLNK(mode)) {
299		if (body_len > PATH_MAX)
300			return 0;
301		collect = collected = symlink_buf;
302		remains = N_ALIGN(name_len) + body_len;
303		next_state = GotSymlink;
304		state = Collect;
305		return 0;
306	}
307	if (S_ISREG(mode) || !body_len)
308		read_into(name_buf, N_ALIGN(name_len), GotName);
309	return 0;
310}
311
312static int __init do_skip(void)
313{
314	if (this_header + byte_count < next_header) {
315		eat(byte_count);
316		return 1;
317	} else {
318		eat(next_header - this_header);
319		state = next_state;
320		return 0;
321	}
322}
323
324static int __init do_reset(void)
325{
326	while (byte_count && *victim == '\0')
327		eat(1);
328	if (byte_count && (this_header & 3))
329		error("broken padding");
330	return 1;
331}
332
333static void __init clean_path(char *path, umode_t fmode)
334{
335	struct kstat st;
336
337	if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
338	    (st.mode ^ fmode) & S_IFMT) {
339		if (S_ISDIR(st.mode))
340			init_rmdir(path);
341		else
342			init_unlink(path);
343	}
344}
345
346static int __init maybe_link(void)
347{
348	if (nlink >= 2) {
349		char *old = find_link(major, minor, ino, mode, collected);
350		if (old) {
351			clean_path(collected, 0);
352			return (init_link(old, collected) < 0) ? -1 : 1;
353		}
354	}
355	return 0;
356}
357
358static __initdata struct file *wfile;
359static __initdata loff_t wfile_pos;
360
361static int __init do_name(void)
362{
363	state = SkipIt;
364	next_state = Reset;
365
366	/* name_len > 0 && name_len <= PATH_MAX checked in do_header */
367	if (collected[name_len - 1] != '\0') {
368		pr_err("initramfs name without nulterm: %.*s\n",
369		       (int)name_len, collected);
370		error("malformed archive");
371		return 1;
372	}
373
374	if (strcmp(collected, "TRAILER!!!") == 0) {
375		free_hash();
376		return 0;
377	}
378	clean_path(collected, mode);
379	if (S_ISREG(mode)) {
380		int ml = maybe_link();
381		if (ml >= 0) {
382			int openflags = O_WRONLY|O_CREAT|O_LARGEFILE;
383			if (ml != 1)
384				openflags |= O_TRUNC;
385			wfile = filp_open(collected, openflags, mode);
386			if (IS_ERR(wfile))
387				return 0;
388			wfile_pos = 0;
389			io_csum = 0;
390
391			vfs_fchown(wfile, uid, gid);
392			vfs_fchmod(wfile, mode);
393			if (body_len)
394				vfs_truncate(&wfile->f_path, body_len);
395			state = CopyFile;
396		}
397	} else if (S_ISDIR(mode)) {
398		init_mkdir(collected, mode);
399		init_chown(collected, uid, gid, 0);
400		init_chmod(collected, mode);
401		dir_add(collected, name_len, mtime);
402	} else if (S_ISBLK(mode) || S_ISCHR(mode) ||
403		   S_ISFIFO(mode) || S_ISSOCK(mode)) {
404		if (maybe_link() == 0) {
405			init_mknod(collected, mode, rdev);
406			init_chown(collected, uid, gid, 0);
407			init_chmod(collected, mode);
408			do_utime(collected, mtime);
409		}
410	}
411	return 0;
412}
413
414static int __init do_copy(void)
415{
416	if (byte_count >= body_len) {
417		if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
418			error("write error");
419
420		do_utime_path(&wfile->f_path, mtime);
421		fput(wfile);
422		if (csum_present && io_csum != hdr_csum)
423			error("bad data checksum");
424		eat(body_len);
425		state = SkipIt;
426		return 0;
427	} else {
428		if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
429			error("write error");
430		body_len -= byte_count;
431		eat(byte_count);
432		return 1;
433	}
434}
435
436static int __init do_symlink(void)
437{
438	if (collected[name_len - 1] != '\0') {
439		pr_err("initramfs symlink without nulterm: %.*s\n",
440		       (int)name_len, collected);
441		error("malformed archive");
442		return 1;
443	}
444	collected[N_ALIGN(name_len) + body_len] = '\0';
445	clean_path(collected, 0);
446	init_symlink(collected + N_ALIGN(name_len), collected);
447	init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
448	do_utime(collected, mtime);
449	state = SkipIt;
450	next_state = Reset;
451	return 0;
452}
453
454static __initdata int (*actions[])(void) = {
455	[Start]		= do_start,
456	[Collect]	= do_collect,
457	[GotHeader]	= do_header,
458	[SkipIt]	= do_skip,
459	[GotName]	= do_name,
460	[CopyFile]	= do_copy,
461	[GotSymlink]	= do_symlink,
462	[Reset]		= do_reset,
463};
464
465static long __init write_buffer(char *buf, unsigned long len)
466{
467	byte_count = len;
468	victim = buf;
469
470	while (!actions[state]())
471		;
472	return len - byte_count;
473}
474
475static long __init flush_buffer(void *bufv, unsigned long len)
476{
477	char *buf = bufv;
478	long written;
479	long origLen = len;
480	if (message)
481		return -1;
482	while ((written = write_buffer(buf, len)) < len && !message) {
483		char c = buf[written];
484		if (c == '0') {
485			buf += written;
486			len -= written;
487			state = Start;
488		} else if (c == 0) {
489			buf += written;
490			len -= written;
491			state = Reset;
492		} else
493			error("junk within compressed archive");
494	}
495	return origLen;
496}
497
498static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
499
500#include <linux/decompress/generic.h>
501
502/**
503 * unpack_to_rootfs - decompress and extract an initramfs archive
504 * @buf: input initramfs archive to extract
505 * @len: length of initramfs data to process
506 *
507 * Returns: NULL for success or an error message string
508 *
509 * This symbol shouldn't be used externally. It's available for unit tests.
510 */
511char * __init unpack_to_rootfs(char *buf, unsigned long len)
512{
513	long written;
514	decompress_fn decompress;
515	const char *compress_name;
516	struct {
517		char header[CPIO_HDRLEN];
518		char symlink[PATH_MAX + N_ALIGN(PATH_MAX) + 1];
519		char name[N_ALIGN(PATH_MAX)];
520	} *bufs = kmalloc(sizeof(*bufs), GFP_KERNEL);
521
522	if (!bufs)
523		panic_show_mem("can't allocate buffers");
524
525	header_buf = bufs->header;
526	symlink_buf = bufs->symlink;
527	name_buf = bufs->name;
528
529	state = Start;
530	this_header = 0;
531	message = NULL;
532	while (!message && len) {
533		loff_t saved_offset = this_header;
534		if (*buf == '0' && !(this_header & 3)) {
535			state = Start;
536			written = write_buffer(buf, len);
537			buf += written;
538			len -= written;
539			continue;
540		}
541		if (!*buf) {
542			buf++;
543			len--;
544			this_header++;
545			continue;
546		}
547		this_header = 0;
548		decompress = decompress_method(buf, len, &compress_name);
549		pr_debug("Detected %s compressed data\n", compress_name);
550		if (decompress) {
551			int res = decompress(buf, len, NULL, flush_buffer, NULL,
552				   &my_inptr, error);
553			if (res)
554				error("decompressor failed");
555		} else if (compress_name) {
556			pr_err("compression method %s not configured\n",
557			       compress_name);
558			error("decompressor failed");
559		} else
560			error("invalid magic at start of compressed archive");
561		if (state != Reset)
562			error("junk at the end of compressed archive");
563		this_header = saved_offset + my_inptr;
564		buf += my_inptr;
565		len -= my_inptr;
566	}
567	dir_utime();
568	/* free any hardlink state collected without optional TRAILER!!! */
569	free_hash();
570	kfree(bufs);
571	return message;
572}
573
574static int __initdata do_retain_initrd;
575
576static int __init retain_initrd_param(char *str)
577{
578	if (*str)
579		return 0;
580	do_retain_initrd = 1;
581	return 1;
582}
583__setup("retain_initrd", retain_initrd_param);
584
585#ifdef CONFIG_ARCH_HAS_KEEPINITRD
586static int __init keepinitrd_setup(char *__unused)
587{
588	do_retain_initrd = 1;
589	return 1;
590}
591__setup("keepinitrd", keepinitrd_setup);
592#endif
593
594static bool __initdata initramfs_async = true;
595static int __init initramfs_async_setup(char *str)
596{
597	return kstrtobool(str, &initramfs_async) == 0;
598}
599__setup("initramfs_async=", initramfs_async_setup);
600
601extern char __initramfs_start[];
602extern unsigned long __initramfs_size;
603#include <linux/initrd.h>
604#include <linux/kexec.h>
605
606static BIN_ATTR(initrd, 0440, sysfs_bin_attr_simple_read, NULL, 0);
607
608void __init reserve_initrd_mem(void)
609{
610	phys_addr_t start;
611	unsigned long size;
612
613	/* Ignore the virtul address computed during device tree parsing */
614	initrd_start = initrd_end = 0;
615
616	if (!phys_initrd_size)
617		return;
618	/*
619	 * Round the memory region to page boundaries as per free_initrd_mem()
620	 * This allows us to detect whether the pages overlapping the initrd
621	 * are in use, but more importantly, reserves the entire set of pages
622	 * as we don't want these pages allocated for other purposes.
623	 */
624	start = round_down(phys_initrd_start, PAGE_SIZE);
625	size = phys_initrd_size + (phys_initrd_start - start);
626	size = round_up(size, PAGE_SIZE);
627
628	if (!memblock_is_region_memory(start, size)) {
629		pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
630		       (u64)start, size);
631		goto disable;
632	}
633
634	if (memblock_is_region_reserved(start, size)) {
635		pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
636		       (u64)start, size);
637		goto disable;
638	}
639
640	memblock_reserve(start, size);
641	/* Now convert initrd to virtual addresses */
642	initrd_start = (unsigned long)__va(phys_initrd_start);
643	initrd_end = initrd_start + phys_initrd_size;
644	initrd_below_start_ok = 1;
645
646	return;
647disable:
648	pr_cont(" - disabling initrd\n");
649	initrd_start = 0;
650	initrd_end = 0;
651}
652
653void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
654{
655#ifdef CONFIG_ARCH_KEEP_MEMBLOCK
656	unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE);
657	unsigned long aligned_end = ALIGN(end, PAGE_SIZE);
658
659	memblock_free((void *)aligned_start, aligned_end - aligned_start);
660#endif
661
662	free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
663			"initrd");
664}
665
666#ifdef CONFIG_CRASH_RESERVE
667static bool __init kexec_free_initrd(void)
668{
669	unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
670	unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
671
672	/*
673	 * If the initrd region is overlapped with crashkernel reserved region,
674	 * free only memory that is not part of crashkernel region.
675	 */
676	if (initrd_start >= crashk_end || initrd_end <= crashk_start)
677		return false;
678
679	/*
680	 * Initialize initrd memory region since the kexec boot does not do.
681	 */
682	memset((void *)initrd_start, 0, initrd_end - initrd_start);
683	if (initrd_start < crashk_start)
684		free_initrd_mem(initrd_start, crashk_start);
685	if (initrd_end > crashk_end)
686		free_initrd_mem(crashk_end, initrd_end);
687	return true;
688}
689#else
690static inline bool kexec_free_initrd(void)
691{
692	return false;
693}
694#endif /* CONFIG_KEXEC_CORE */
695
696#ifdef CONFIG_BLK_DEV_RAM
697static void __init populate_initrd_image(char *err)
698{
699	ssize_t written;
700	struct file *file;
701	loff_t pos = 0;
702
703	printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
704			err);
705	file = filp_open("/initrd.image", O_WRONLY|O_CREAT|O_LARGEFILE, 0700);
706	if (IS_ERR(file))
707		return;
708
709	written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
710			&pos);
711	if (written != initrd_end - initrd_start)
712		pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
713		       written, initrd_end - initrd_start);
714	fput(file);
715}
716#endif /* CONFIG_BLK_DEV_RAM */
717
718static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
719{
720	/* Load the built in initramfs */
721	char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
722	if (err)
723		panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
724
725	if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
726		goto done;
727
728	if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
729		printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
730	else
731		printk(KERN_INFO "Unpacking initramfs...\n");
732
733	err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
734	if (err) {
735#ifdef CONFIG_BLK_DEV_RAM
736		populate_initrd_image(err);
737#else
738		printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
739#endif
740	}
741
742done:
743	security_initramfs_populated();
744
745	/*
746	 * If the initrd region is overlapped with crashkernel reserved region,
747	 * free only memory that is not part of crashkernel region.
748	 */
749	if (!do_retain_initrd && initrd_start && !kexec_free_initrd()) {
750		free_initrd_mem(initrd_start, initrd_end);
751	} else if (do_retain_initrd && initrd_start) {
752		bin_attr_initrd.size = initrd_end - initrd_start;
753		bin_attr_initrd.private = (void *)initrd_start;
754		if (sysfs_create_bin_file(firmware_kobj, &bin_attr_initrd))
755			pr_err("Failed to create initrd sysfs file");
756	}
757	initrd_start = 0;
758	initrd_end = 0;
759
760	init_flush_fput();
761}
762
763static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
764static async_cookie_t initramfs_cookie;
765
766void wait_for_initramfs(void)
767{
768	if (!initramfs_cookie) {
769		/*
770		 * Something before rootfs_initcall wants to access
771		 * the filesystem/initramfs. Probably a bug. Make a
772		 * note, avoid deadlocking the machine, and let the
773		 * caller's access fail as it used to.
774		 */
775		pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
776		return;
777	}
778	async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
779}
780EXPORT_SYMBOL_GPL(wait_for_initramfs);
781
782static int __init populate_rootfs(void)
783{
784	initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
785						 &initramfs_domain);
786	usermodehelper_enable();
787	if (!initramfs_async)
788		wait_for_initramfs();
789	return 0;
790}
791rootfs_initcall(populate_rootfs);