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 24 static __initdata bool csum_present; 25 static __initdata u32 io_csum; 26 27 static ssize_t __init xwrite(struct file *file, const unsigned char *p, 28 size_t count, loff_t *pos) 29 { 30 ssize_t out = 0; 31 32 /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */ 33 while (count) { 34 ssize_t rv = kernel_write(file, p, count, pos); 35 36 if (rv < 0) { 37 if (rv == -EINTR || rv == -EAGAIN) 38 continue; 39 return out ? out : rv; 40 } else if (rv == 0) 41 break; 42 43 if (csum_present) { 44 ssize_t i; 45 46 for (i = 0; i < rv; i++) 47 io_csum += p[i]; 48 } 49 50 p += rv; 51 out += rv; 52 count -= rv; 53 } 54 55 return out; 56 } 57 58 static __initdata char *message; 59 static void __init error(char *x) 60 { 61 if (!message) 62 message = x; 63 } 64 65 #define panic_show_mem(fmt, ...) \ 66 ({ show_mem(); panic(fmt, ##__VA_ARGS__); }) 67 68 /* link hash */ 69 70 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2) 71 72 static __initdata struct hash { 73 int ino, minor, major; 74 umode_t mode; 75 struct hash *next; 76 char name[N_ALIGN(PATH_MAX)]; 77 } *head[32]; 78 79 static inline int hash(int major, int minor, int ino) 80 { 81 unsigned long tmp = ino + minor + (major << 3); 82 tmp += tmp >> 5; 83 return tmp & 31; 84 } 85 86 static char __init *find_link(int major, int minor, int ino, 87 umode_t mode, char *name) 88 { 89 struct hash **p, *q; 90 for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) { 91 if ((*p)->ino != ino) 92 continue; 93 if ((*p)->minor != minor) 94 continue; 95 if ((*p)->major != major) 96 continue; 97 if (((*p)->mode ^ mode) & S_IFMT) 98 continue; 99 return (*p)->name; 100 } 101 q = kmalloc(sizeof(struct hash), GFP_KERNEL); 102 if (!q) 103 panic_show_mem("can't allocate link hash entry"); 104 q->major = major; 105 q->minor = minor; 106 q->ino = ino; 107 q->mode = mode; 108 strcpy(q->name, name); 109 q->next = NULL; 110 *p = q; 111 return NULL; 112 } 113 114 static void __init free_hash(void) 115 { 116 struct hash **p, *q; 117 for (p = head; p < head + 32; p++) { 118 while (*p) { 119 q = *p; 120 *p = q->next; 121 kfree(q); 122 } 123 } 124 } 125 126 #ifdef CONFIG_INITRAMFS_PRESERVE_MTIME 127 static void __init do_utime(char *filename, time64_t mtime) 128 { 129 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } }; 130 init_utimes(filename, t); 131 } 132 133 static void __init do_utime_path(const struct path *path, time64_t mtime) 134 { 135 struct timespec64 t[2] = { { .tv_sec = mtime }, { .tv_sec = mtime } }; 136 vfs_utimes(path, t); 137 } 138 139 static __initdata LIST_HEAD(dir_list); 140 struct dir_entry { 141 struct list_head list; 142 time64_t mtime; 143 char name[]; 144 }; 145 146 static void __init dir_add(const char *name, time64_t mtime) 147 { 148 size_t nlen = strlen(name) + 1; 149 struct dir_entry *de; 150 151 de = kmalloc(sizeof(struct dir_entry) + nlen, GFP_KERNEL); 152 if (!de) 153 panic_show_mem("can't allocate dir_entry buffer"); 154 INIT_LIST_HEAD(&de->list); 155 strscpy(de->name, name, nlen); 156 de->mtime = mtime; 157 list_add(&de->list, &dir_list); 158 } 159 160 static void __init dir_utime(void) 161 { 162 struct dir_entry *de, *tmp; 163 list_for_each_entry_safe(de, tmp, &dir_list, list) { 164 list_del(&de->list); 165 do_utime(de->name, de->mtime); 166 kfree(de); 167 } 168 } 169 #else 170 static void __init do_utime(char *filename, time64_t mtime) {} 171 static void __init do_utime_path(const struct path *path, time64_t mtime) {} 172 static void __init dir_add(const char *name, time64_t mtime) {} 173 static void __init dir_utime(void) {} 174 #endif 175 176 static __initdata time64_t mtime; 177 178 /* cpio header parsing */ 179 180 static __initdata unsigned long ino, major, minor, nlink; 181 static __initdata umode_t mode; 182 static __initdata unsigned long body_len, name_len; 183 static __initdata uid_t uid; 184 static __initdata gid_t gid; 185 static __initdata unsigned rdev; 186 static __initdata u32 hdr_csum; 187 188 static void __init parse_header(char *s) 189 { 190 unsigned long parsed[13]; 191 char buf[9]; 192 int i; 193 194 buf[8] = '\0'; 195 for (i = 0, s += 6; i < 13; i++, s += 8) { 196 memcpy(buf, s, 8); 197 parsed[i] = simple_strtoul(buf, NULL, 16); 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, 110, 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 if (strcmp(collected, "TRAILER!!!") == 0) { 364 free_hash(); 365 return 0; 366 } 367 clean_path(collected, mode); 368 if (S_ISREG(mode)) { 369 int ml = maybe_link(); 370 if (ml >= 0) { 371 int openflags = O_WRONLY|O_CREAT|O_LARGEFILE; 372 if (ml != 1) 373 openflags |= O_TRUNC; 374 wfile = filp_open(collected, openflags, mode); 375 if (IS_ERR(wfile)) 376 return 0; 377 wfile_pos = 0; 378 io_csum = 0; 379 380 vfs_fchown(wfile, uid, gid); 381 vfs_fchmod(wfile, mode); 382 if (body_len) 383 vfs_truncate(&wfile->f_path, body_len); 384 state = CopyFile; 385 } 386 } else if (S_ISDIR(mode)) { 387 init_mkdir(collected, mode); 388 init_chown(collected, uid, gid, 0); 389 init_chmod(collected, mode); 390 dir_add(collected, mtime); 391 } else if (S_ISBLK(mode) || S_ISCHR(mode) || 392 S_ISFIFO(mode) || S_ISSOCK(mode)) { 393 if (maybe_link() == 0) { 394 init_mknod(collected, mode, rdev); 395 init_chown(collected, uid, gid, 0); 396 init_chmod(collected, mode); 397 do_utime(collected, mtime); 398 } 399 } 400 return 0; 401 } 402 403 static int __init do_copy(void) 404 { 405 if (byte_count >= body_len) { 406 if (xwrite(wfile, victim, body_len, &wfile_pos) != body_len) 407 error("write error"); 408 409 do_utime_path(&wfile->f_path, mtime); 410 fput(wfile); 411 if (csum_present && io_csum != hdr_csum) 412 error("bad data checksum"); 413 eat(body_len); 414 state = SkipIt; 415 return 0; 416 } else { 417 if (xwrite(wfile, victim, byte_count, &wfile_pos) != byte_count) 418 error("write error"); 419 body_len -= byte_count; 420 eat(byte_count); 421 return 1; 422 } 423 } 424 425 static int __init do_symlink(void) 426 { 427 collected[N_ALIGN(name_len) + body_len] = '\0'; 428 clean_path(collected, 0); 429 init_symlink(collected + N_ALIGN(name_len), collected); 430 init_chown(collected, uid, gid, AT_SYMLINK_NOFOLLOW); 431 do_utime(collected, mtime); 432 state = SkipIt; 433 next_state = Reset; 434 return 0; 435 } 436 437 static __initdata int (*actions[])(void) = { 438 [Start] = do_start, 439 [Collect] = do_collect, 440 [GotHeader] = do_header, 441 [SkipIt] = do_skip, 442 [GotName] = do_name, 443 [CopyFile] = do_copy, 444 [GotSymlink] = do_symlink, 445 [Reset] = do_reset, 446 }; 447 448 static long __init write_buffer(char *buf, unsigned long len) 449 { 450 byte_count = len; 451 victim = buf; 452 453 while (!actions[state]()) 454 ; 455 return len - byte_count; 456 } 457 458 static long __init flush_buffer(void *bufv, unsigned long len) 459 { 460 char *buf = bufv; 461 long written; 462 long origLen = len; 463 if (message) 464 return -1; 465 while ((written = write_buffer(buf, len)) < len && !message) { 466 char c = buf[written]; 467 if (c == '0') { 468 buf += written; 469 len -= written; 470 state = Start; 471 } else if (c == 0) { 472 buf += written; 473 len -= written; 474 state = Reset; 475 } else 476 error("junk within compressed archive"); 477 } 478 return origLen; 479 } 480 481 static unsigned long my_inptr __initdata; /* index of next byte to be processed in inbuf */ 482 483 #include <linux/decompress/generic.h> 484 485 static char * __init unpack_to_rootfs(char *buf, unsigned long len) 486 { 487 long written; 488 decompress_fn decompress; 489 const char *compress_name; 490 static __initdata char msg_buf[64]; 491 492 header_buf = kmalloc(110, GFP_KERNEL); 493 symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL); 494 name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL); 495 496 if (!header_buf || !symlink_buf || !name_buf) 497 panic_show_mem("can't allocate buffers"); 498 499 state = Start; 500 this_header = 0; 501 message = NULL; 502 while (!message && len) { 503 loff_t saved_offset = this_header; 504 if (*buf == '0' && !(this_header & 3)) { 505 state = Start; 506 written = write_buffer(buf, len); 507 buf += written; 508 len -= written; 509 continue; 510 } 511 if (!*buf) { 512 buf++; 513 len--; 514 this_header++; 515 continue; 516 } 517 this_header = 0; 518 decompress = decompress_method(buf, len, &compress_name); 519 pr_debug("Detected %s compressed data\n", compress_name); 520 if (decompress) { 521 int res = decompress(buf, len, NULL, flush_buffer, NULL, 522 &my_inptr, error); 523 if (res) 524 error("decompressor failed"); 525 } else if (compress_name) { 526 if (!message) { 527 snprintf(msg_buf, sizeof msg_buf, 528 "compression method %s not configured", 529 compress_name); 530 message = msg_buf; 531 } 532 } else 533 error("invalid magic at start of compressed archive"); 534 if (state != Reset) 535 error("junk at the end of compressed archive"); 536 this_header = saved_offset + my_inptr; 537 buf += my_inptr; 538 len -= my_inptr; 539 } 540 dir_utime(); 541 kfree(name_buf); 542 kfree(symlink_buf); 543 kfree(header_buf); 544 return message; 545 } 546 547 static int __initdata do_retain_initrd; 548 549 static int __init retain_initrd_param(char *str) 550 { 551 if (*str) 552 return 0; 553 do_retain_initrd = 1; 554 return 1; 555 } 556 __setup("retain_initrd", retain_initrd_param); 557 558 #ifdef CONFIG_ARCH_HAS_KEEPINITRD 559 static int __init keepinitrd_setup(char *__unused) 560 { 561 do_retain_initrd = 1; 562 return 1; 563 } 564 __setup("keepinitrd", keepinitrd_setup); 565 #endif 566 567 static bool __initdata initramfs_async = true; 568 static int __init initramfs_async_setup(char *str) 569 { 570 return kstrtobool(str, &initramfs_async) == 0; 571 } 572 __setup("initramfs_async=", initramfs_async_setup); 573 574 extern char __initramfs_start[]; 575 extern unsigned long __initramfs_size; 576 #include <linux/initrd.h> 577 #include <linux/kexec.h> 578 579 static BIN_ATTR(initrd, 0440, sysfs_bin_attr_simple_read, NULL, 0); 580 581 void __init reserve_initrd_mem(void) 582 { 583 phys_addr_t start; 584 unsigned long size; 585 586 /* Ignore the virtul address computed during device tree parsing */ 587 initrd_start = initrd_end = 0; 588 589 if (!phys_initrd_size) 590 return; 591 /* 592 * Round the memory region to page boundaries as per free_initrd_mem() 593 * This allows us to detect whether the pages overlapping the initrd 594 * are in use, but more importantly, reserves the entire set of pages 595 * as we don't want these pages allocated for other purposes. 596 */ 597 start = round_down(phys_initrd_start, PAGE_SIZE); 598 size = phys_initrd_size + (phys_initrd_start - start); 599 size = round_up(size, PAGE_SIZE); 600 601 if (!memblock_is_region_memory(start, size)) { 602 pr_err("INITRD: 0x%08llx+0x%08lx is not a memory region", 603 (u64)start, size); 604 goto disable; 605 } 606 607 if (memblock_is_region_reserved(start, size)) { 608 pr_err("INITRD: 0x%08llx+0x%08lx overlaps in-use memory region\n", 609 (u64)start, size); 610 goto disable; 611 } 612 613 memblock_reserve(start, size); 614 /* Now convert initrd to virtual addresses */ 615 initrd_start = (unsigned long)__va(phys_initrd_start); 616 initrd_end = initrd_start + phys_initrd_size; 617 initrd_below_start_ok = 1; 618 619 return; 620 disable: 621 pr_cont(" - disabling initrd\n"); 622 initrd_start = 0; 623 initrd_end = 0; 624 } 625 626 void __weak __init free_initrd_mem(unsigned long start, unsigned long end) 627 { 628 #ifdef CONFIG_ARCH_KEEP_MEMBLOCK 629 unsigned long aligned_start = ALIGN_DOWN(start, PAGE_SIZE); 630 unsigned long aligned_end = ALIGN(end, PAGE_SIZE); 631 632 memblock_free((void *)aligned_start, aligned_end - aligned_start); 633 #endif 634 635 free_reserved_area((void *)start, (void *)end, POISON_FREE_INITMEM, 636 "initrd"); 637 } 638 639 #ifdef CONFIG_CRASH_RESERVE 640 static bool __init kexec_free_initrd(void) 641 { 642 unsigned long crashk_start = (unsigned long)__va(crashk_res.start); 643 unsigned long crashk_end = (unsigned long)__va(crashk_res.end); 644 645 /* 646 * If the initrd region is overlapped with crashkernel reserved region, 647 * free only memory that is not part of crashkernel region. 648 */ 649 if (initrd_start >= crashk_end || initrd_end <= crashk_start) 650 return false; 651 652 /* 653 * Initialize initrd memory region since the kexec boot does not do. 654 */ 655 memset((void *)initrd_start, 0, initrd_end - initrd_start); 656 if (initrd_start < crashk_start) 657 free_initrd_mem(initrd_start, crashk_start); 658 if (initrd_end > crashk_end) 659 free_initrd_mem(crashk_end, initrd_end); 660 return true; 661 } 662 #else 663 static inline bool kexec_free_initrd(void) 664 { 665 return false; 666 } 667 #endif /* CONFIG_KEXEC_CORE */ 668 669 #ifdef CONFIG_BLK_DEV_RAM 670 static void __init populate_initrd_image(char *err) 671 { 672 ssize_t written; 673 struct file *file; 674 loff_t pos = 0; 675 676 printk(KERN_INFO "rootfs image is not initramfs (%s); looks like an initrd\n", 677 err); 678 file = filp_open("/initrd.image", O_WRONLY|O_CREAT|O_LARGEFILE, 0700); 679 if (IS_ERR(file)) 680 return; 681 682 written = xwrite(file, (char *)initrd_start, initrd_end - initrd_start, 683 &pos); 684 if (written != initrd_end - initrd_start) 685 pr_err("/initrd.image: incomplete write (%zd != %ld)\n", 686 written, initrd_end - initrd_start); 687 fput(file); 688 } 689 #endif /* CONFIG_BLK_DEV_RAM */ 690 691 static void __init do_populate_rootfs(void *unused, async_cookie_t cookie) 692 { 693 /* Load the built in initramfs */ 694 char *err = unpack_to_rootfs(__initramfs_start, __initramfs_size); 695 if (err) 696 panic_show_mem("%s", err); /* Failed to decompress INTERNAL initramfs */ 697 698 if (!initrd_start || IS_ENABLED(CONFIG_INITRAMFS_FORCE)) 699 goto done; 700 701 if (IS_ENABLED(CONFIG_BLK_DEV_RAM)) 702 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n"); 703 else 704 printk(KERN_INFO "Unpacking initramfs...\n"); 705 706 err = unpack_to_rootfs((char *)initrd_start, initrd_end - initrd_start); 707 if (err) { 708 #ifdef CONFIG_BLK_DEV_RAM 709 populate_initrd_image(err); 710 #else 711 printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err); 712 #endif 713 } 714 715 done: 716 security_initramfs_populated(); 717 718 /* 719 * If the initrd region is overlapped with crashkernel reserved region, 720 * free only memory that is not part of crashkernel region. 721 */ 722 if (!do_retain_initrd && initrd_start && !kexec_free_initrd()) { 723 free_initrd_mem(initrd_start, initrd_end); 724 } else if (do_retain_initrd && initrd_start) { 725 bin_attr_initrd.size = initrd_end - initrd_start; 726 bin_attr_initrd.private = (void *)initrd_start; 727 if (sysfs_create_bin_file(firmware_kobj, &bin_attr_initrd)) 728 pr_err("Failed to create initrd sysfs file"); 729 } 730 initrd_start = 0; 731 initrd_end = 0; 732 733 init_flush_fput(); 734 } 735 736 static ASYNC_DOMAIN_EXCLUSIVE(initramfs_domain); 737 static async_cookie_t initramfs_cookie; 738 739 void wait_for_initramfs(void) 740 { 741 if (!initramfs_cookie) { 742 /* 743 * Something before rootfs_initcall wants to access 744 * the filesystem/initramfs. Probably a bug. Make a 745 * note, avoid deadlocking the machine, and let the 746 * caller's access fail as it used to. 747 */ 748 pr_warn_once("wait_for_initramfs() called before rootfs_initcalls\n"); 749 return; 750 } 751 async_synchronize_cookie_domain(initramfs_cookie + 1, &initramfs_domain); 752 } 753 EXPORT_SYMBOL_GPL(wait_for_initramfs); 754 755 static int __init populate_rootfs(void) 756 { 757 initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL, 758 &initramfs_domain); 759 usermodehelper_enable(); 760 if (!initramfs_async) 761 wait_for_initramfs(); 762 return 0; 763 } 764 rootfs_initcall(populate_rootfs); 765