projects/linux·blob master · refs

799 lines · 18.5 KiB

1// SPDX-License-Identifier: GPL-2.0
2#include <linux/async.h>
3#include <linux/delay.h>
4#include <linux/dirent.h>
5#include <linux/export.h>
6#include <linux/fcntl.h>
7#include <linux/file.h>
8#include <linux/fs.h>
9#include <linux/fs_struct.h>
10#include <linux/hex.h>
11#include <linux/init.h>
12#include <linux/init_syscalls.h>
13#include <linux/kstrtox.h>
14#include <linux/memblock.h>
15#include <linux/mm.h>
16#include <linux/namei.h>
17#include <linux/overflow.h>
18#include <linux/security.h>
19#include <linux/slab.h>
20#include <linux/string.h>
21#include <linux/syscalls.h>
22#include <linux/types.h>
23#include <linux/umh.h>
24#include <linux/utime.h>
25
26#include <asm/byteorder.h>
27
28#include "do_mounts.h"
29#include "initramfs_internal.h"
30
31static __initdata bool csum_present;
32static __initdata u32 io_csum;
33
34static ssize_t __init xwrite(struct file *file, const unsigned char *p,
35 size_t count, loff_t *pos)
36{
37 ssize_t out = 0;
38
39 /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
40 while (count) {
41 ssize_t rv = kernel_write(file, p, count, pos);
42
43 if (rv < 0) {
44 if (rv == -EINTR || rv == -EAGAIN)
45 continue;
46 return out ? out : rv;
47 } else if (rv == 0)
48 break;
49
50 if (csum_present) {
51 ssize_t i;
52
53 for (i = 0; i < rv; i++)
54 io_csum += p[i];
55 }
56
57 p += rv;
58 out += rv;
59 count -= rv;
60 }
61
62 return out;
63}
64
65static __initdata char *message;
66static void __init error(char *x)
67{
68 if (!message)
69 message = x;
70}
71
72#define panic_show_mem(fmt, ...) \
73 ({ show_mem(); panic(fmt, ##__VA_ARGS__); })
74
75/* link hash */
76
77#define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
78
79static __initdata struct hash {
80 int ino, minor, major;
81 umode_t mode;
82 struct hash *next;
83 char name[];
84} *head[32];
85static __initdata bool hardlink_seen;
86
87static inline int hash(int major, int minor, int ino)
88{
89 unsigned long tmp = ino + minor + (major << 3);
90 tmp += tmp >> 5;
91 return tmp & 31;
92}
93
94static char __init *find_link(int major, int minor, int ino,
95 umode_t mode, const char *name, size_t nlen)
96{
97 struct hash **p, *q;
98 for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
99 if ((*p)->ino != ino)
100 continue;
101 if ((*p)->minor != minor)
102 continue;
103 if ((*p)->major != major)
104 continue;
105 if (((*p)->mode ^ mode) & S_IFMT)
106 continue;
107 return (*p)->name;
108 }
109
110 q = kmalloc_flex(struct hash, name, nlen);
111 if (!q)
112 panic_show_mem("can't allocate link hash entry");
113 q->major = major;
114 q->minor = minor;
115 q->ino = ino;
116 q->mode = mode;
117 strscpy(q->name, name, nlen);
118 q->next = NULL;
119 *p = q;
120 hardlink_seen = true;
121 return NULL;
122}
123
124static void __init free_hash(void)
125{
126 struct hash **p, *q;
127 for (p = head; hardlink_seen && p < head + 32; p++) {
128 while (*p) {
129 q = *p;
130 *p = q->next;
131 kfree(q);
132 }
133 }
134 hardlink_seen = false;
135}
136
137#ifdef CONFIG_INITRAMFS_PRESERVE_MTIME
138static void __init do_utime(char *filename, time64_t mtime)
139{
140 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
141 init_utimes(filename, t);
142}
143
144static void __init do_utime_path(const struct path *path, time64_t mtime)
145{
146 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } };
147 vfs_utimes(path, t);
148}
149
150static __initdata LIST_HEAD(dir_list);
151struct dir_entry {
152 struct list_head list;
153 time64_t mtime;
154 char name[];
155};
156
157static void __init dir_add(const char *name, size_t nlen, time64_t mtime)
158{
159 struct dir_entry *de;
160
161 de = kmalloc_flex(*de, name, nlen);
162 if (!de)
163 panic_show_mem("can't allocate dir_entry buffer");
164 INIT_LIST_HEAD(&de->list);
165 strscpy(de->name, name, nlen);
166 de->mtime = mtime;
167 list_add(&de->list, &dir_list);
168}
169
170static void __init dir_utime(void)
171{
172 struct dir_entry *de, *tmp;
173 list_for_each_entry_safe(de, tmp, &dir_list, list) {
174 list_del(&de->list);
175 do_utime(de->name, de->mtime);
176 kfree(de);
177 }
178}
179#else
180static void __init do_utime(char *filename, time64_t mtime) {}
181static void __init do_utime_path(const struct path *path, time64_t mtime) {}
182static void __init dir_add(const char *name, size_t nlen, time64_t mtime) {}
183static void __init dir_utime(void) {}
184#endif
185
186static __initdata time64_t mtime;
187
188/* cpio header parsing */
189
190static __initdata unsigned long ino, major, minor, nlink;
191static __initdata umode_t mode;
192static __initdata unsigned long body_len, name_len;
193static __initdata uid_t uid;
194static __initdata gid_t gid;
195static __initdata unsigned rdev;
196static __initdata u32 hdr_csum;
197
198static int __init parse_header(char *s)
199{
200 __be32 header[13];
201 int ret;
202
203 ret = hex2bin((u8 *)header, s + 6, sizeof(header));
204 if (ret) {
205 error("damaged header");
206 return ret;
207 }
208
209 ino = be32_to_cpu(header[0]);
210 mode = be32_to_cpu(header[1]);
211 uid = be32_to_cpu(header[2]);
212 gid = be32_to_cpu(header[3]);
213 nlink = be32_to_cpu(header[4]);
214 mtime = be32_to_cpu(header[5]); /* breaks in y2106 */
215 body_len = be32_to_cpu(header[6]);
216 major = be32_to_cpu(header[7]);
217 minor = be32_to_cpu(header[8]);
218 rdev = new_encode_dev(MKDEV(be32_to_cpu(header[9]), be32_to_cpu(header[10])));
219 name_len = be32_to_cpu(header[11]);
220 hdr_csum = be32_to_cpu(header[12]);
221 return 0;
222}
223
224/* Finite-state machine */
225
226static __initdata enum state {
227 Start,
228 Collect,
229 GotHeader,
230 SkipIt,
231 GotName,
232 CopyFile,
233 GotSymlink,
234 Reset
235} state, next_state;
236
237static __initdata char *victim;
238static unsigned long byte_count __initdata;
239static __initdata loff_t this_header, next_header;
240
241static inline void __init eat(unsigned n)
242{
243 victim += n;
244 this_header += n;
245 byte_count -= n;
246}
247
248static __initdata char *collected;
249static long remains __initdata;
250static __initdata char *collect;
251
252static void __init read_into(char *buf, unsigned size, enum state next)
253{
254 if (byte_count >= size) {
255 collected = victim;
256 eat(size);
257 state = next;
258 } else {
259 collect = collected = buf;
260 remains = size;
261 next_state = next;
262 state = Collect;
263 }
264}
265
266static __initdata char *header_buf, *symlink_buf, *name_buf;
267
268static int __init do_start(void)
269{
270 read_into(header_buf, CPIO_HDRLEN, GotHeader);
271 return 0;
272}
273
274static int __init do_collect(void)
275{
276 unsigned long n = remains;
277 if (byte_count < n)
278 n = byte_count;
279 memcpy(collect, victim, n);
280 eat(n);
281 collect += n;
282 if ((remains -= n) != 0)
283 return 1;
284 state = next_state;
285 return 0;
286}
287
288static int __init do_header(void)
289{
290 if (!memcmp(collected, "070701", 6)) {
291 csum_present = false;
292 } else if (!memcmp(collected, "070702", 6)) {
293 csum_present = true;
294 } else {
295 if (memcmp(collected, "070707", 6) == 0)
296 error("incorrect cpio method used: use -H newc option");
297 else
298 error("no cpio magic");
299 return 1;
300 }
301 if (parse_header(collected))
302 return 1;
303 next_header = this_header + N_ALIGN(name_len) + body_len;
304 next_header = (next_header + 3) & ~3;
305 state = SkipIt;
306 if (name_len <= 0 || name_len > PATH_MAX)
307 return 0;
308 if (S_ISLNK(mode)) {
309 if (body_len > PATH_MAX)
310 return 0;
311 collect = collected = symlink_buf;
312 remains = N_ALIGN(name_len) + body_len;
313 next_state = GotSymlink;
314 state = Collect;
315 return 0;
316 }
317 if (S_ISREG(mode) || !body_len)
318 read_into(name_buf, N_ALIGN(name_len), GotName);
319 return 0;
320}
321
322static int __init do_skip(void)
323{
324 if (this_header + byte_count < next_header) {
325 eat(byte_count);
326 return 1;
327 } else {
328 eat(next_header - this_header);
329 state = next_state;
330 return 0;
331 }
332}
333
334static int __init do_reset(void)
335{
336 while (byte_count && *victim == '\0')
337 eat(1);
338 if (byte_count && (this_header & 3))
339 error("broken padding");
340 return 1;
341}
342
343static void __init clean_path(char *path, umode_t fmode)
344{
345 struct kstat st;
346
347 if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) &&
348 (st.mode ^ fmode) & S_IFMT) {
349 if (S_ISDIR(st.mode))
350 init_rmdir(path);
351 else
352 init_unlink(path);
353 }
354}
355
356static int __init maybe_link(void)
357{
358 if (nlink >= 2) {
359 char *old = find_link(major, minor, ino, mode, collected, name_len);
360 if (old) {
361 clean_path(collected, 0);
362 return (init_link(old, collected) < 0) ? -1 : 1;
363 }
364 }
365 return 0;
366}
367
368static __initdata struct file *wfile;
369static __initdata loff_t wfile_pos;
370
371static int __init do_name(void)
372{
373 state = SkipIt;
374 next_state = Reset;
375
376 /* name_len > 0 && name_len <= PATH_MAX checked in do_header */
377 if (collected[name_len - 1] != '\0') {
378 pr_err("initramfs name without nulterm: %.*s\n",
379 (int)name_len, collected);
380 error("malformed archive");
381 return 1;
382 }
383
384 if (strcmp(collected, "TRAILER!!!") == 0) {
385 free_hash();
386 return 0;
387 }
388 clean_path(collected, mode);
389 if (S_ISREG(mode)) {
390 int ml = maybe_link();
391 if (ml >= 0) {
392 int openflags = O_WRONLY|O_CREAT|O_LARGEFILE;
393 if (ml != 1)
394 openflags |= O_TRUNC;
395 wfile = filp_open(collected, openflags, mode);
396 if (IS_ERR(wfile))
397 return 0;
398 wfile_pos = 0;
399 io_csum = 0;
400
401 vfs_fchown(wfile, uid, gid);
402 vfs_fchmod(wfile, mode);
403 if (body_len)
404 vfs_truncate(&wfile->f_path, body_len);
405 state = CopyFile;
406 }
407 } else if (S_ISDIR(mode)) {
408 init_mkdir(collected, mode);
409 init_chown(collected, uid, gid, 0);
410 init_chmod(collected, mode);
411 dir_add(collected, name_len, mtime);
412 } else if (S_ISBLK(mode) || S_ISCHR(mode) ||
413 S_ISFIFO(mode) || S_ISSOCK(mode)) {
414 if (maybe_link() == 0) {
415 init_mknod(collected, mode, rdev);
416 init_chown(collected, uid, gid, 0);
417 init_chmod(collected, mode);
418 do_utime(collected, mtime);
419 }
420 }
421 return 0;
422}
423
424static int __init do_copy(void)
425{
426 if (byte_count >= body_len) {
427 if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len)
428 error("write error");
429
430 do_utime_path(&wfile->f_path, mtime);
431 fput(wfile);
432 if (csum_present && io_csum != hdr_csum)
433 error("bad data checksum");
434 eat(body_len);
435 state = SkipIt;
436 return 0;
437 } else {
438 if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count)
439 error("write error");
440 body_len -= byte_count;
441 eat(byte_count);
442 return 1;
443 }
444}
445
446static int __init do_symlink(void)
447{
448 if (collected[name_len - 1] != '\0') {
449 pr_err("initramfs symlink without nulterm: %.*s\n",
450 (int)name_len, collected);
451 error("malformed archive");
452 return 1;
453 }
454 collected[N_ALIGN(name_len) + body_len] = '\0';
455 clean_path(collected, 0);
456 init_symlink(collected + N_ALIGN(name_len), collected);
457 init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW);
458 do_utime(collected, mtime);
459 state = SkipIt;
460 next_state = Reset;
461 return 0;
462}
463
464static __initdata int (*actions[])(void) = {
465 [Start] = do_start,
466 [Collect] = do_collect,
467 [GotHeader] = do_header,
468 [SkipIt] = do_skip,
469 [GotName] = do_name,
470 [CopyFile] = do_copy,
471 [GotSymlink] = do_symlink,
472 [Reset] = do_reset,
473};
474
475static long __init write_buffer(char *buf, unsigned long len)
476{
477 byte_count = len;
478 victim = buf;
479
480 while (!actions[state]())
481 ;
482 return len - byte_count;
483}
484
485static long __init flush_buffer(void *bufv, unsigned long len)
486{
487 char *buf = bufv;
488 long written;
489 long origLen = len;
490 if (message)
491 return -1;
492 while ((written = write_buffer(buf, len)) < len && !message) {
493 char c = buf[written];
494 if (c == '0') {
495 buf += written;
496 len -= written;
497 state = Start;
498 } else if (c == 0) {
499 buf += written;
500 len -= written;
501 state = Reset;
502 } else
503 error("junk within compressed archive");
504 }
505 return origLen;
506}
507
508static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
509
510#include <linux/decompress/generic.h>
511
512/**
513 * unpack_to_rootfs - decompress and extract an initramfs archive
514 * @buf: input initramfs archive to extract
515 * @len: length of initramfs data to process
516 *
517 * Returns: NULL for success or an error message string
518 *
519 * This symbol shouldn't be used externally. It's available for unit tests.
520 */
521char * __init unpack_to_rootfs(char *buf, unsigned long len)
522{
523 long written;
524 decompress_fn decompress;
525 const char *compress_name;
526 struct {
527 char header[CPIO_HDRLEN];
528 char symlink[PATH_MAX + N_ALIGN(PATH_MAX) + 1];
529 char name[N_ALIGN(PATH_MAX)];
530 } *bufs = kmalloc_obj(*bufs);
531
532 if (!bufs)
533 panic_show_mem("can't allocate buffers");
534
535 header_buf = bufs->header;
536 symlink_buf = bufs->symlink;
537 name_buf = bufs->name;
538
539 state = Start;
540 this_header = 0;
541 message = NULL;
542 while (!message && len) {
543 loff_t saved_offset = this_header;
544 if (*buf == '0' && !(this_header & 3)) {
545 state = Start;
546 written = write_buffer(buf, len);
547 buf += written;
548 len -= written;
549 continue;
550 }
551 if (!*buf) {
552 buf++;
553 len--;
554 this_header++;
555 continue;
556 }
557 this_header = 0;
558 decompress = decompress_method(buf, len, &compress_name);
559 pr_debug("Detected %s compressed data\n", compress_name);
560 if (decompress) {
561 int res = decompress(buf, len, NULL, flush_buffer, NULL,
562 &my_inptr, error);
563 if (res)
564 error("decompressor failed");
565 } else if (compress_name) {
566 pr_err("compression method %s not configured\n",
567 compress_name);
568 error("decompressor failed");
569 } else
570 error("invalid magic at start of compressed archive");
571 if (state != Reset)
572 error("junk at the end of compressed archive");
573 this_header = saved_offset + my_inptr;
574 buf += my_inptr;
575 len -= my_inptr;
576 }
577 dir_utime();
578 /* free any hardlink state collected without optional TRAILER!!! */
579 free_hash();
580 kfree(bufs);
581 return message;
582}
583
584static int __initdata do_retain_initrd;
585
586static int __init retain_initrd_param(char *str)
587{
588 if (*str)
589 return 0;
590 do_retain_initrd = 1;
591 return 1;
592}
593__setup("retain_initrd", retain_initrd_param);
594
595#ifdef CONFIG_ARCH_HAS_KEEPINITRD
596static int __init keepinitrd_setup(char *__unused)
597{
598 do_retain_initrd = 1;
599 return 1;
600}
601__setup("keepinitrd", keepinitrd_setup);
602#endif
603
604static bool __initdata initramfs_async = true;
605static int __init initramfs_async_setup(char *str)
606{
607 return kstrtobool(str, &initramfs_async) == 0;
608}
609__setup("initramfs_async=", initramfs_async_setup);
610
611extern char __initramfs_start[];
612extern unsigned long __initramfs_size;
613#include <linux/initrd.h>
614#include <linux/kexec.h>
615
616static BIN_ATTR(initrd, 0440, sysfs_bin_attr_simple_read, NULL, 0);
617
618void __init reserve_initrd_mem(void)
619{
620 phys_addr_t start;
621 unsigned long size;
622
623 /* Ignore the virtual address computed during device tree parsing */
624 initrd_start = initrd_end = 0;
625
626 if (!phys_initrd_size)
627 return;
628 /*
629 * Round the memory region to page boundaries as per free_initrd_mem()
630 * This allows us to detect whether the pages overlapping the initrd
631 * are in use, but more importantly, reserves the entire set of pages
632 * as we don't want these pages allocated for other purposes.
633 */
634 start = round_down(phys_initrd_start, PAGE_SIZE);
635 size = phys_initrd_size + (phys_initrd_start - start);
636 size = round_up(size, PAGE_SIZE);
637
638 if (!memblock_is_region_memory(start, size)) {
639 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region",
640 (u64)start, size);
641 goto disable;
642 }
643
644 if (memblock_is_region_reserved(start, size)) {
645 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n",
646 (u64)start, size);
647 goto disable;
648 }
649
650 memblock_reserve(start, size);
651 /* Now convert initrd to virtual addresses */
652 initrd_start = (unsigned long)__va(phys_initrd_start);
653 initrd_end = initrd_start + phys_initrd_size;
654 initrd_below_start_ok = 1;
655
656 return;
657disable:
658 pr_cont(" - disabling initrd\n");
659 initrd_start = 0;
660 initrd_end = 0;
661}
662
663void __weak __init free_initrd_mem(unsigned long start, unsigned long end)
664{
665 free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM,
666 "initrd");
667}
668
669#ifdef CONFIG_CRASH_RESERVE
670static bool __init kexec_free_initrd(void)
671{
672 unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
673 unsigned long crashk_end = (unsigned long)__va(crashk_res.end);
674
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 (initrd_start >= crashk_end || initrd_end <= crashk_start)
680 return false;
681
682 /*
683 * Initialize initrd memory region since the kexec boot does not do.
684 */
685 memset((void *)initrd_start, 0, initrd_end - initrd_start);
686 if (initrd_start < crashk_start)
687 free_initrd_mem(initrd_start, crashk_start);
688 if (initrd_end > crashk_end)
689 free_initrd_mem(crashk_end, initrd_end);
690 return true;
691}
692#else
693static inline bool kexec_free_initrd(void)
694{
695 return false;
696}
697#endif /* CONFIG_KEXEC_CORE */
698
699#ifdef CONFIG_BLK_DEV_RAM
700static void __init populate_initrd_image(char *err)
701{
702 ssize_t written;
703 struct file *file;
704 loff_t pos = 0;
705
706 printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n",
707 err);
708 file = filp_open("/initrd.image", O_WRONLY|O_CREAT|O_LARGEFILE, 0700);
709 if (IS_ERR(file))
710 return;
711
712 written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start,
713 &pos);
714 if (written != initrd_end - initrd_start)
715 pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
716 written, initrd_end - initrd_start);
717 fput(file);
718}
719#endif /* CONFIG_BLK_DEV_RAM */
720
721static void __init unpack_initramfs(async_cookie_t cookie)
722{
723 /* Load the built in initramfs */
724 char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
725 if (err)
726 panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */
727
728 if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE))
729 return;
730
731 if (IS_ENABLED(CONFIG_BLK_DEV_RAM))
732 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
733 else
734 printk(KERN_INFO "Unpacking initramfs...\n");
735
736 err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start);
737 if (err) {
738#ifdef CONFIG_BLK_DEV_RAM
739 populate_initrd_image(err);
740#else
741 printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
742#endif
743 }
744}
745
746static void __init do_populate_rootfs(void *unused, async_cookie_t cookie)
747{
748 scoped_with_init_fs() {
749 unpack_initramfs(cookie);
750 security_initramfs_populated();
751 }
752
753 /*
754 * If the initrd region is overlapped with crashkernel reserved region,
755 * free only memory that is not part of crashkernel region.
756 */
757 if (!do_retain_initrd && initrd_start && !kexec_free_initrd()) {
758 free_initrd_mem(initrd_start, initrd_end);
759 } else if (do_retain_initrd && initrd_start) {
760 bin_attr_initrd.size = initrd_end - initrd_start;
761 bin_attr_initrd.private = (void *)initrd_start;
762 if (sysfs_create_bin_file(firmware_kobj, &bin_attr_initrd))
763 pr_err("Failed to create initrd sysfs file");
764 }
765 initrd_start = 0;
766 initrd_end = 0;
767
768 init_flush_fput();
769}
770
771static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
772static async_cookie_t initramfs_cookie;
773
774void wait_for_initramfs(void)
775{
776 if (!initramfs_cookie) {
777 /*
778 * Something before rootfs_initcall wants to access
779 * the filesystem/initramfs. Probably a bug. Make a
780 * note, avoid deadlocking the machine, and let the
781 * caller's access fail as it used to.
782 */
783 pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n");
784 return;
785 }
786 async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain);
787}
788EXPORT_SYMBOL_GPL(wait_for_initramfs);
789
790static int __init populate_rootfs(void)
791{
792 initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
793 &initramfs_domain);
794 usermodehelper_enable();
795 if (!initramfs_async)
796 wait_for_initramfs();
797 return 0;
798}
799rootfs_initcall(populate_rootfs);