]> git.ipfire.org Git - thirdparty/linux.git/blob - init/initramfs.c
btrfs: don't reserve space for checksums when writing to nocow files
[thirdparty/linux.git] / init / initramfs.c
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
22 static __initdata bool csum_present;
23 static __initdata u32 io_csum;
24
25 static 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
56 static __initdata char *message;
57 static 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
70 static __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
77 static 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
84 static 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
112 static 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
125 static 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
131 static 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
137 static __initdata LIST_HEAD(dir_list);
138 struct dir_entry {
139 struct list_head list;
140 time64_t mtime;
141 char name[];
142 };
143
144 static 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
158 static 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
168 static void __init do_utime(char *filename, time64_t mtime) {}
169 static void __init do_utime_path(const struct path *path, time64_t mtime) {}
170 static void __init dir_add(const char *name, time64_t mtime) {}
171 static void __init dir_utime(void) {}
172 #endif
173
174 static __initdata time64_t mtime;
175
176 /* cpio header parsing */
177
178 static __initdata unsigned long ino, major, minor, nlink;
179 static __initdata umode_t mode;
180 static __initdata unsigned long body_len, name_len;
181 static __initdata uid_t uid;
182 static __initdata gid_t gid;
183 static __initdata unsigned rdev;
184 static __initdata u32 hdr_csum;
185
186 static 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
213 static __initdata enum state {
214 Start,
215 Collect,
216 GotHeader,
217 SkipIt,
218 GotName,
219 CopyFile,
220 GotSymlink,
221 Reset
222 } state, next_state;
223
224 static __initdata char *victim;
225 static unsigned long byte_count __initdata;
226 static __initdata loff_t this_header, next_header;
227
228 static inline void __init eat(unsigned n)
229 {
230 victim += n;
231 this_header += n;
232 byte_count -= n;
233 }
234
235 static __initdata char *collected;
236 static long remains __initdata;
237 static __initdata char *collect;
238
239 static 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
253 static __initdata char *header_buf, *symlink_buf, *name_buf;
254
255 static int __init do_start(void)
256 {
257 read_into(header_buf, 110, GotHeader);
258 return 0;
259 }
260
261 static 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
275 static 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
308 static 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
320 static 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
329 static 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
342 static 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
354 static __initdata struct file *wfile;
355 static __initdata loff_t wfile_pos;
356
357 static 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
401 static 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
423 static 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
435 static __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
446 static 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
456 static 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
479 static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */
480
481 #include <linux/decompress/generic.h>
482
483 static 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
545 static int __initdata do_retain_initrd;
546
547 static 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
557 static int __init keepinitrd_setup(char *__unused)
558 {
559 do_retain_initrd = 1;
560 return 1;
561 }
562 __setup("keepinitrd", keepinitrd_setup);
563 #endif
564
565 static bool __initdata initramfs_async = true;
566 static int __init initramfs_async_setup(char *str)
567 {
568 return kstrtobool(str, &initramfs_async) == 0;
569 }
570 __setup("initramfs_async=", initramfs_async_setup);
571
572 extern char __initramfs_start[];
573 extern unsigned long __initramfs_size;
574 #include <linux/initrd.h>
575 #include <linux/kexec.h>
576
577 void __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;
616 disable:
617 pr_cont(" - disabling initrd\n");
618 initrd_start = 0;
619 initrd_end = 0;
620 }
621
622 void __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
636 static 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
659 static inline bool kexec_free_initrd(void)
660 {
661 return false;
662 }
663 #endif /* CONFIG_KEXEC_CORE */
664
665 #ifdef CONFIG_BLK_DEV_RAM
666 static 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
689 static 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
713 done:
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
727 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain);
728 static async_cookie_t initramfs_cookie;
729
730 void 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 }
744 EXPORT_SYMBOL_GPL(wait_for_initramfs);
745
746 static 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 }
755 rootfs_initcall(populate_rootfs);