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 #include <linux/security.h> 21 22 #include "do_mounts.h" 23 #include "initramfs_internal.h" 24 25 static __initdata bool csum_present; 26 static __initdata u32 io_csum; 27 28 static ssize_t __init xwrite(struct file *file, const unsigned char *p, 29 size_t count, loff_t *pos) 30 { 31 ssize_t out = 0; 32 33 /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */ 34 while (count) { 35 ssize_t rv = kernel_write(file, p, count, pos); 36 37 if (rv < 0) { 38 if (rv == -EINTR || rv == -EAGAIN) 39 continue; 40 return out ? out : rv; 41 } else if (rv == 0) 42 break; 43 44 if (csum_present) { 45 ssize_t i; 46 47 for (i = 0; i < rv; i++) 48 io_csum += p[i]; 49 } 50 51 p += rv; 52 out += rv; 53 count -= rv; 54 } 55 56 return out; 57 } 58 59 static __initdata char *message; 60 static void __init error(char *x) 61 { 62 if (!message) 63 message = x; 64 } 65 66 #define panic_show_mem(fmt, ...) \ 67 ({ show_mem(); panic(fmt, ##__VA_ARGS__); }) 68 69 /* link hash */ 70 71 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2) 72 73 static __initdata struct hash { 74 int ino, minor, major; 75 umode_t mode; 76 struct hash *next; 77 char name[N_ALIGN(PATH_MAX)]; 78 } *head[32]; 79 static __initdata bool hardlink_seen; 80 81 static inline int hash(int major, int minor, int ino) 82 { 83 unsigned long tmp = ino + minor + (major << 3); 84 tmp += tmp >> 5; 85 return tmp & 31; 86 } 87 88 static char __init *find_link(int major, int minor, int ino, 89 umode_t mode, char *name) 90 { 91 struct hash **p, *q; 92 for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) { 93 if ((*p)->ino != ino) 94 continue; 95 if ((*p)->minor != minor) 96 continue; 97 if ((*p)->major != major) 98 continue; 99 if (((*p)->mode ^ mode) & S_IFMT) 100 continue; 101 return (*p)->name; 102 } 103 q = kmalloc(sizeof(struct hash), GFP_KERNEL); 104 if (!q) 105 panic_show_mem("can't allocate link hash entry"); 106 q->major = major; 107 q->minor = minor; 108 q->ino = ino; 109 q->mode = mode; 110 strcpy(q->name, name); 111 q->next = NULL; 112 *p = q; 113 hardlink_seen = true; 114 return NULL; 115 } 116 117 static void __init free_hash(void) 118 { 119 struct hash **p, *q; 120 for (p = head; hardlink_seen && p < head + 32; p++) { 121 while (*p) { 122 q = *p; 123 *p = q->next; 124 kfree(q); 125 } 126 } 127 hardlink_seen = false; 128 } 129 130 #ifdef CONFIG_INITRAMFS_PRESERVE_MTIME 131 static void __init do_utime(char *filename, time64_t mtime) 132 { 133 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } }; 134 init_utimes(filename, t); 135 } 136 137 static void __init do_utime_path(const struct path *path, time64_t mtime) 138 { 139 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } }; 140 vfs_utimes(path, t); 141 } 142 143 static __initdata LIST_HEAD(dir_list); 144 struct dir_entry { 145 struct list_head list; 146 time64_t mtime; 147 char name[]; 148 }; 149 150 static void __init dir_add(const char *name, size_t nlen, time64_t mtime) 151 { 152 struct dir_entry *de; 153 154 de = kmalloc(sizeof(struct dir_entry) + nlen, GFP_KERNEL); 155 if (!de) 156 panic_show_mem("can't allocate dir_entry buffer"); 157 INIT_LIST_HEAD(&de->list); 158 strscpy(de->name, name, nlen); 159 de->mtime = mtime; 160 list_add(&de->list, &dir_list); 161 } 162 163 static void __init dir_utime(void) 164 { 165 struct dir_entry *de, *tmp; 166 list_for_each_entry_safe(de, tmp, &dir_list, list) { 167 list_del(&de->list); 168 do_utime(de->name, de->mtime); 169 kfree(de); 170 } 171 } 172 #else 173 static void __init do_utime(char *filename, time64_t mtime) {} 174 static void __init do_utime_path(const struct path *path, time64_t mtime) {} 175 static void __init dir_add(const char *name, size_t nlen, time64_t mtime) {} 176 static void __init dir_utime(void) {} 177 #endif 178 179 static __initdata time64_t mtime; 180 181 /* cpio header parsing */ 182 183 static __initdata unsigned long ino, major, minor, nlink; 184 static __initdata umode_t mode; 185 static __initdata unsigned long body_len, name_len; 186 static __initdata uid_t uid; 187 static __initdata gid_t gid; 188 static __initdata unsigned rdev; 189 static __initdata u32 hdr_csum; 190 191 static void __init parse_header(char *s) 192 { 193 unsigned long parsed[13]; 194 int i; 195 196 for (i = 0, s += 6; i < 13; i++, s += 8) 197 parsed[i] = simple_strntoul(s, NULL, 16, 8); 198 199 ino = parsed[0]; 200 mode = parsed[1]; 201 uid = parsed[2]; 202 gid = parsed[3]; 203 nlink = parsed[4]; 204 mtime = parsed[5]; /* breaks in y2106 */ 205 body_len = parsed[6]; 206 major = parsed[7]; 207 minor = parsed[8]; 208 rdev = new_encode_dev(MKDEV(parsed[9], parsed[10])); 209 name_len = parsed[11]; 210 hdr_csum = parsed[12]; 211 } 212 213 /* FSM */ 214 215 static __initdata enum state { 216 Start, 217 Collect, 218 GotHeader, 219 SkipIt, 220 GotName, 221 CopyFile, 222 GotSymlink, 223 Reset 224 } state, next_state; 225 226 static __initdata char *victim; 227 static unsigned long byte_count __initdata; 228 static __initdata loff_t this_header, next_header; 229 230 static inline void __init eat(unsigned n) 231 { 232 victim += n; 233 this_header += n; 234 byte_count -= n; 235 } 236 237 static __initdata char *collected; 238 static long remains __initdata; 239 static __initdata char *collect; 240 241 static void __init read_into(char *buf, unsigned size, enum state next) 242 { 243 if (byte_count >= size) { 244 collected = victim; 245 eat(size); 246 state = next; 247 } else { 248 collect = collected = buf; 249 remains = size; 250 next_state = next; 251 state = Collect; 252 } 253 } 254 255 static __initdata char *header_buf, *symlink_buf, *name_buf; 256 257 static int __init do_start(void) 258 { 259 read_into(header_buf, CPIO_HDRLEN, GotHeader); 260 return 0; 261 } 262 263 static int __init do_collect(void) 264 { 265 unsigned long n = remains; 266 if (byte_count < n) 267 n = byte_count; 268 memcpy(collect, victim, n); 269 eat(n); 270 collect += n; 271 if ((remains -= n) != 0) 272 return 1; 273 state = next_state; 274 return 0; 275 } 276 277 static int __init do_header(void) 278 { 279 if (!memcmp(collected, "070701", 6)) { 280 csum_present = false; 281 } else if (!memcmp(collected, "070702", 6)) { 282 csum_present = true; 283 } else { 284 if (memcmp(collected, "070707", 6) == 0) 285 error("incorrect cpio method used: use -H newc option"); 286 else 287 error("no cpio magic"); 288 return 1; 289 } 290 parse_header(collected); 291 next_header = this_header + N_ALIGN(name_len) + body_len; 292 next_header = (next_header + 3) & ~3; 293 state = SkipIt; 294 if (name_len <= 0 || name_len > PATH_MAX) 295 return 0; 296 if (S_ISLNK(mode)) { 297 if (body_len > PATH_MAX) 298 return 0; 299 collect = collected = symlink_buf; 300 remains = N_ALIGN(name_len) + body_len; 301 next_state = GotSymlink; 302 state = Collect; 303 return 0; 304 } 305 if (S_ISREG(mode) || !body_len) 306 read_into(name_buf, N_ALIGN(name_len), GotName); 307 return 0; 308 } 309 310 static int __init do_skip(void) 311 { 312 if (this_header + byte_count < next_header) { 313 eat(byte_count); 314 return 1; 315 } else { 316 eat(next_header - this_header); 317 state = next_state; 318 return 0; 319 } 320 } 321 322 static int __init do_reset(void) 323 { 324 while (byte_count && *victim == '\0') 325 eat(1); 326 if (byte_count && (this_header & 3)) 327 error("broken padding"); 328 return 1; 329 } 330 331 static void __init clean_path(char *path, umode_t fmode) 332 { 333 struct kstat st; 334 335 if (!init_stat(path, &st, AT_SYMLINK_NOFOLLOW) && 336 (st.mode ^ fmode) & S_IFMT) { 337 if (S_ISDIR(st.mode)) 338 init_rmdir(path); 339 else 340 init_unlink(path); 341 } 342 } 343 344 static int __init maybe_link(void) 345 { 346 if (nlink >= 2) { 347 char *old = find_link(major, minor, ino, mode, collected); 348 if (old) { 349 clean_path(collected, 0); 350 return (init_link(old, collected) < 0) ? -1 : 1; 351 } 352 } 353 return 0; 354 } 355 356 static __initdata struct file *wfile; 357 static __initdata loff_t wfile_pos; 358 359 static int __init do_name(void) 360 { 361 state = SkipIt; 362 next_state = Reset; 363 364 /* name_len > 0 && name_len <= PATH_MAX checked in do_header */ 365 if (collected[name_len - 1] != '\0') { 366 pr_err("initramfs name without nulterm: %.*s\n", 367 (int)name_len, collected); 368 error("malformed archive"); 369 return 1; 370 } 371 372 if (strcmp(collected, "TRAILER!!!") == 0) { 373 free_hash(); 374 return 0; 375 } 376 clean_path(collected, mode); 377 if (S_ISREG(mode)) { 378 int ml = maybe_link(); 379 if (ml >= 0) { 380 int openflags = O_WRONLY|O_CREAT|O_LARGEFILE; 381 if (ml != 1) 382 openflags |= O_TRUNC; 383 wfile = filp_open(collected, openflags, mode); 384 if (IS_ERR(wfile)) 385 return 0; 386 wfile_pos = 0; 387 io_csum = 0; 388 389 vfs_fchown(wfile, uid, gid); 390 vfs_fchmod(wfile, mode); 391 if (body_len) 392 vfs_truncate(&wfile->f_path, body_len); 393 state = CopyFile; 394 } 395 } else if (S_ISDIR(mode)) { 396 init_mkdir(collected, mode); 397 init_chown(collected, uid, gid, 0); 398 init_chmod(collected, mode); 399 dir_add(collected, name_len, mtime); 400 } else if (S_ISBLK(mode) || S_ISCHR(mode) || 401 S_ISFIFO(mode) || S_ISSOCK(mode)) { 402 if (maybe_link() == 0) { 403 init_mknod(collected, mode, rdev); 404 init_chown(collected, uid, gid, 0); 405 init_chmod(collected, mode); 406 do_utime(collected, mtime); 407 } 408 } 409 return 0; 410 } 411 412 static int __init do_copy(void) 413 { 414 if (byte_count >= body_len) { 415 if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len) 416 error("write error"); 417 418 do_utime_path(&wfile->f_path, mtime); 419 fput(wfile); 420 if (csum_present && io_csum != hdr_csum) 421 error("bad data checksum"); 422 eat(body_len); 423 state = SkipIt; 424 return 0; 425 } else { 426 if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count) 427 error("write error"); 428 body_len -= byte_count; 429 eat(byte_count); 430 return 1; 431 } 432 } 433 434 static int __init do_symlink(void) 435 { 436 if (collected[name_len - 1] != '\0') { 437 pr_err("initramfs symlink without nulterm: %.*s\n", 438 (int)name_len, collected); 439 error("malformed archive"); 440 return 1; 441 } 442 collected[N_ALIGN(name_len) + body_len] = '\0'; 443 clean_path(collected, 0); 444 init_symlink(collected + N_ALIGN(name_len), collected); 445 init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW); 446 do_utime(collected, mtime); 447 state = SkipIt; 448 next_state = Reset; 449 return 0; 450 } 451 452 static __initdata int (*actions[])(void) = { 453 [Start] = do_start, 454 [Collect] = do_collect, 455 [GotHeader] = do_header, 456 [SkipIt] = do_skip, 457 [GotName] = do_name, 458 [CopyFile] = do_copy, 459 [GotSymlink] = do_symlink, 460 [Reset] = do_reset, 461 }; 462 463 static long __init write_buffer(char *buf, unsigned long len) 464 { 465 byte_count = len; 466 victim = buf; 467 468 while (!actions[state]()) 469 ; 470 return len - byte_count; 471 } 472 473 static long __init flush_buffer(void *bufv, unsigned long len) 474 { 475 char *buf = bufv; 476 long written; 477 long origLen = len; 478 if (message) 479 return -1; 480 while ((written = write_buffer(buf, len)) < len && !message) { 481 char c = buf[written]; 482 if (c == '0') { 483 buf += written; 484 len -= written; 485 state = Start; 486 } else if (c == 0) { 487 buf += written; 488 len -= written; 489 state = Reset; 490 } else 491 error("junk within compressed archive"); 492 } 493 return origLen; 494 } 495 496 static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */ 497 498 #include <linux/decompress/generic.h> 499 500 /** 501 * unpack_to_rootfs - decompress and extract an initramfs archive 502 * @buf: input initramfs archive to extract 503 * @len: length of initramfs data to process 504 * 505 * Returns: NULL for success or an error message string 506 * 507 * This symbol shouldn't be used externally. It's available for unit tests. 508 */ 509 char * __init unpack_to_rootfs(char *buf, unsigned long len) 510 { 511 long written; 512 decompress_fn decompress; 513 const char *compress_name; 514 struct { 515 char header[CPIO_HDRLEN]; 516 char symlink[PATH_MAX + N_ALIGN(PATH_MAX) + 1]; 517 char name[N_ALIGN(PATH_MAX)]; 518 } *bufs = kmalloc(sizeof(*bufs), GFP_KERNEL); 519 520 if (!bufs) 521 panic_show_mem("can't allocate buffers"); 522 523 header_buf = bufs->header; 524 symlink_buf = bufs->symlink; 525 name_buf = bufs->name; 526 527 state = Start; 528 this_header = 0; 529 message = NULL; 530 while (!message && len) { 531 loff_t saved_offset = this_header; 532 if (*buf == '0' && !(this_header & 3)) { 533 state = Start; 534 written = write_buffer(buf, len); 535 buf += written; 536 len -= written; 537 continue; 538 } 539 if (!*buf) { 540 buf++; 541 len--; 542 this_header++; 543 continue; 544 } 545 this_header = 0; 546 decompress = decompress_method(buf, len, &compress_name); 547 pr_debug("Detected %s compressed data\n", compress_name); 548 if (decompress) { 549 int res = decompress(buf, len, NULL, flush_buffer, NULL, 550 &my_inptr, error); 551 if (res) 552 error("decompressor failed"); 553 } else if (compress_name) { 554 pr_err("compression method %s not configured\n", 555 compress_name); 556 error("decompressor failed"); 557 } else 558 error("invalid magic at start of compressed archive"); 559 if (state != Reset) 560 error("junk at the end of compressed archive"); 561 this_header = saved_offset + my_inptr; 562 buf += my_inptr; 563 len -= my_inptr; 564 } 565 dir_utime(); 566 /* free any hardlink state collected without optional TRAILER!!! */ 567 free_hash(); 568 kfree(bufs); 569 return message; 570 } 571 572 static int __initdata do_retain_initrd; 573 574 static int __init retain_initrd_param(char *str) 575 { 576 if (*str) 577 return 0; 578 do_retain_initrd = 1; 579 return 1; 580 } 581 __setup("retain_initrd", retain_initrd_param); 582 583 #ifdef CONFIG_ARCH_HAS_KEEPINITRD 584 static int __init keepinitrd_setup(char *__unused) 585 { 586 do_retain_initrd = 1; 587 return 1; 588 } 589 __setup("keepinitrd", keepinitrd_setup); 590 #endif 591 592 static bool __initdata initramfs_async = true; 593 static int __init initramfs_async_setup(char *str) 594 { 595 return kstrtobool(str, &initramfs_async) == 0; 596 } 597 __setup("initramfs_async=", initramfs_async_setup); 598 599 extern char __initramfs_start[]; 600 extern unsigned long __initramfs_size; 601 #include <linux/initrd.h> 602 #include <linux/kexec.h> 603 604 static BIN_ATTR(initrd, 0440, sysfs_bin_attr_simple_read, NULL, 0); 605 606 void __init reserve_initrd_mem(void) 607 { 608 phys_addr_t start; 609 unsigned long size; 610 611 /* Ignore the virtul address computed during device tree parsing */ 612 initrd_start = initrd_end = 0; 613 614 if (!phys_initrd_size) 615 return; 616 /* 617 * Round the memory region to page boundaries as per free_initrd_mem() 618 * This allows us to detect whether the pages overlapping the initrd 619 * are in use, but more importantly, reserves the entire set of pages 620 * as we don't want these pages allocated for other purposes. 621 */ 622 start = round_down(phys_initrd_start, PAGE_SIZE); 623 size = phys_initrd_size + (phys_initrd_start - start); 624 size = round_up(size, PAGE_SIZE); 625 626 if (!memblock_is_region_memory(start, size)) { 627 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region", 628 (u64)start, size); 629 goto disable; 630 } 631 632 if (memblock_is_region_reserved(start, size)) { 633 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n", 634 (u64)start, size); 635 goto disable; 636 } 637 638 memblock_reserve(start, size); 639 /* Now convert initrd to virtual addresses */ 640 initrd_start = (unsigned long)__va(phys_initrd_start); 641 initrd_end = initrd_start + phys_initrd_size; 642 initrd_below_start_ok = 1; 643 644 return; 645 disable: 646 pr_cont(" - disabling initrd\n"); 647 initrd_start = 0; 648 initrd_end = 0; 649 } 650 651 void __weak __init free_initrd_mem(unsigned long start, unsigned long end) 652 { 653 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 654 unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE); 655 unsigned long aligned_end = ALIGN(end, PAGE_SIZE); 656 657 memblock_free((void *)aligned_start, aligned_end - aligned_start); 658 #endif 659 660 free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM, 661 "initrd"); 662 } 663 664 #ifdef CONFIG_CRASH_RESERVE 665 static bool __init kexec_free_initrd(void) 666 { 667 unsigned long crashk_start = (unsigned long)__va(crashk_res.start); 668 unsigned long crashk_end = (unsigned long)__va(crashk_res.end); 669 670 /* 671 * If the initrd region is overlapped with crashkernel reserved region, 672 * free only memory that is not part of crashkernel region. 673 */ 674 if (initrd_start >= crashk_end || initrd_end <= crashk_start) 675 return false; 676 677 /* 678 * Initialize initrd memory region since the kexec boot does not do. 679 */ 680 memset((void *)initrd_start, 0, initrd_end - initrd_start); 681 if (initrd_start < crashk_start) 682 free_initrd_mem(initrd_start, crashk_start); 683 if (initrd_end > crashk_end) 684 free_initrd_mem(crashk_end, initrd_end); 685 return true; 686 } 687 #else 688 static inline bool kexec_free_initrd(void) 689 { 690 return false; 691 } 692 #endif /* CONFIG_KEXEC_CORE */ 693 694 #ifdef CONFIG_BLK_DEV_RAM 695 static void __init populate_initrd_image(char *err) 696 { 697 ssize_t written; 698 struct file *file; 699 loff_t pos = 0; 700 701 printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n", 702 err); 703 file = filp_open("/initrd.image", O_WRONLY|O_CREAT|O_LARGEFILE, 0700); 704 if (IS_ERR(file)) 705 return; 706 707 written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start, 708 &pos); 709 if (written != initrd_end - initrd_start) 710 pr_err("/initrd.image: incomplete write (%zd != %ld)\n", 711 written, initrd_end - initrd_start); 712 fput(file); 713 } 714 #endif /* CONFIG_BLK_DEV_RAM */ 715 716 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie) 717 { 718 /* Load the built in initramfs */ 719 char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size); 720 if (err) 721 panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */ 722 723 if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE)) 724 goto done; 725 726 if (IS_ENABLED(CONFIG_BLK_DEV_RAM)) 727 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n"); 728 else 729 printk(KERN_INFO "Unpacking initramfs...\n"); 730 731 err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start); 732 if (err) { 733 #ifdef CONFIG_BLK_DEV_RAM 734 populate_initrd_image(err); 735 #else 736 printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err); 737 #endif 738 } 739 740 done: 741 security_initramfs_populated(); 742 743 /* 744 * If the initrd region is overlapped with crashkernel reserved region, 745 * free only memory that is not part of crashkernel region. 746 */ 747 if (!do_retain_initrd && initrd_start && !kexec_free_initrd()) { 748 free_initrd_mem(initrd_start, initrd_end); 749 } else if (do_retain_initrd && initrd_start) { 750 bin_attr_initrd.size = initrd_end - initrd_start; 751 bin_attr_initrd.private = (void *)initrd_start; 752 if (sysfs_create_bin_file(firmware_kobj, &bin_attr_initrd)) 753 pr_err("Failed to create initrd sysfs file"); 754 } 755 initrd_start = 0; 756 initrd_end = 0; 757 758 init_flush_fput(); 759 } 760 761 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain); 762 static async_cookie_t initramfs_cookie; 763 764 void wait_for_initramfs(void) 765 { 766 if (!initramfs_cookie) { 767 /* 768 * Something before rootfs_initcall wants to access 769 * the filesystem/initramfs. Probably a bug. Make a 770 * note, avoid deadlocking the machine, and let the 771 * caller's access fail as it used to. 772 */ 773 pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n"); 774 return; 775 } 776 async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain); 777 } 778 EXPORT_SYMBOL_GPL(wait_for_initramfs); 779 780 static int __init populate_rootfs(void) 781 { 782 initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL, 783 &initramfs_domain); 784 usermodehelper_enable(); 785 if (!initramfs_async) 786 wait_for_initramfs(); 787 return 0; 788 } 789 rootfs_initcall(populate_rootfs); 790