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