init/initramfs.c

   1 /*
   2  * Many of the syscalls used in this file expect some of the arguments
   3  * to be __user pointers not __kernel pointers.  To limit the sparse
   4  * noise, turn off sparse checking for this file.
   5  */
   6 #ifdef __CHECKER__
   7 #undef __CHECKER__
   8 #warning "Sparse checking disabled for this file"
   9 #endif
  10
  11 #include <linux/init.h>
  12 #include <linux/fs.h>
  13 #include <linux/slab.h>
  14 #include <linux/types.h>
  15 #include <linux/fcntl.h>
  16 #include <linux/delay.h>
  17 #include <linux/string.h>
  18 #include <linux/dirent.h>
  19 #include <linux/syscalls.h>
  20 #include <linux/utime.h>
  21 #include <linux/initramfs.h>
  22
  23 static ssize_t __init xwrite(int fd, const char *p, size_t count)
  24 {
  25         ssize_t out = 0;
  26
  27         /* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
  28         while (count) {
  29                 ssize_t rv = sys_write(fd, p, count);
  30
  31                 if (rv < 0) {
  32                         if (rv == -EINTR || rv == -EAGAIN)
  33                                 continue;
  34                         return out ? out : rv;
  35                 } else if (rv == 0)
  36                         break;
  37
  38                 p += rv;
  39                 out += rv;
  40                 count -= rv;
  41         }
  42
  43         return out;
  44 }
  45
  46 static __initdata char *message;
  47 static void __init error(char *x)
  48 {
  49         if (!message)
  50                 message = x;
  51 }
  52
  53 /* link hash */
  54
  55 #define N_ALIGN(len) ((((len) + 1) & ~3) + 2)
  56
  57 static __initdata struct hash {
  58         int ino, minor, major;
  59         umode_t mode;
  60         struct hash *next;
  61         char name[N_ALIGN(PATH_MAX)];
  62 } *head[32];
  63
  64 static inline int hash(int major, int minor, int ino)
  65 {
  66         unsigned long tmp = ino + minor + (major << 3);
  67         tmp += tmp >> 5;
  68         return tmp & 31;
  69 }
  70
  71 static char __init *find_link(int major, int minor, int ino,
  72                               umode_t mode, char *name)
  73 {
  74         struct hash **p, *q;
  75         for (p = head + hash(major, minor, ino); *p; p = &(*p)->next) {
  76                 if ((*p)->ino != ino)
  77                         continue;
  78                 if ((*p)->minor != minor)
  79                         continue;
  80                 if ((*p)->major != major)
  81                         continue;
  82                 if (((*p)->mode ^ mode) & S_IFMT)
  83                         continue;
  84                 return (*p)->name;
  85         }
  86         q = kmalloc(sizeof(struct hash), GFP_KERNEL);
  87         if (!q)
  88                 panic("can't allocate link hash entry");
  89         q->major = major;
  90         q->minor = minor;
  91         q->ino = ino;
  92         q->mode = mode;
  93         strcpy(q->name, name);
  94         q->next = NULL;
  95         *p = q;
  96         return NULL;
  97 }
  98
  99 static void __init free_hash(void)
 100 {
 101         struct hash **p, *q;
 102         for (p = head; p < head + 32; p++) {
 103                 while (*p) {
 104                         q = *p;
 105                         *p = q->next;
 106                         kfree(q);
 107                 }
 108         }
 109 }
 110
 111 static long __init do_utime(char *filename, time_t mtime)
 112 {
 113         struct timespec t[2];
 114
 115         t[0].tv_sec = mtime;
 116         t[0].tv_nsec = 0;
 117         t[1].tv_sec = mtime;
 118         t[1].tv_nsec = 0;
 119
 120         return do_utimes(AT_FDCWD, filename, t, AT_SYMLINK_NOFOLLOW);
 121 }
 122
 123 static __initdata LIST_HEAD(dir_list);
 124 struct dir_entry {
 125         struct list_head list;
 126         char *name;
 127         time_t mtime;
 128 };
 129
 130 static void __init dir_add(const char *name, time_t mtime)
 131 {
 132         struct dir_entry *de = kmalloc(sizeof(struct dir_entry), GFP_KERNEL);
 133         if (!de)
 134                 panic("can't allocate dir_entry buffer");
 135         INIT_LIST_HEAD(&de->list);
 136         de->name = kstrdup(name, GFP_KERNEL);
 137         de->mtime = mtime;
 138         list_add(&de->list, &dir_list);
 139 }
 140
 141 static void __init dir_utime(void)
 142 {
 143         struct dir_entry *de, *tmp;
 144         list_for_each_entry_safe(de, tmp, &dir_list, list) {
 145                 list_del(&de->list);
 146                 do_utime(de->name, de->mtime);
 147                 kfree(de->name);
 148                 kfree(de);
 149         }
 150 }
 151
 152 static __initdata time_t mtime;
 153
 154 /* cpio header parsing */
 155
 156 static __initdata unsigned long ino, major, minor, nlink;
 157 static __initdata umode_t mode;
 158 static __initdata unsigned long body_len, name_len;
 159 static __initdata uid_t uid;
 160 static __initdata gid_t gid;
 161 static __initdata unsigned rdev;
 162
 163 static void __init parse_header(char *s)
 164 {
 165         unsigned long parsed[12];
 166         char buf[9];
 167         int i;
 168
 169         buf[8] = '\0';
 170         for (i = 0, s += 6; i < 12; i++, s += 8) {
 171                 memcpy(buf, s, 8);
 172                 parsed[i] = simple_strtoul(buf, NULL, 16);
 173         }
 174         ino = parsed[0];
 175         mode = parsed[1];
 176         uid = parsed[2];
 177         gid = parsed[3];
 178         nlink = parsed[4];
 179         mtime = parsed[5];
 180         body_len = parsed[6];
 181         major = parsed[7];
 182         minor = parsed[8];
 183         rdev = new_encode_dev(MKDEV(parsed[9], parsed[10]));
 184         name_len = parsed[11];
 185 }
 186
 187 /* FSM */
 188
 189 static __initdata enum state {
 190         Start,
 191         Collect,
 192         GotHeader,
 193         SkipIt,
 194         GotName,
 195         CopyFile,
 196         GotSymlink,
 197         Reset
 198 } state, next_state;
 199
 200 static __initdata char *victim;
 201 static unsigned long byte_count __initdata;
 202 static __initdata loff_t this_header, next_header;
 203
 204 static inline void __init eat(unsigned n)
 205 {
 206         victim += n;
 207         this_header += n;
 208         byte_count -= n;
 209 }
 210
 211 static __initdata char *vcollected;
 212 static __initdata char *collected;
 213 static long remains __initdata;
 214 static __initdata char *collect;
 215
 216 static void __init read_into(char *buf, unsigned size, enum state next)
 217 {
 218         if (byte_count >= size) {
 219                 collected = victim;
 220                 eat(size);
 221                 state = next;
 222         } else {
 223                 collect = collected = buf;
 224                 remains = size;
 225                 next_state = next;
 226                 state = Collect;
 227         }
 228 }
 229
 230 static __initdata char *header_buf, *symlink_buf, *name_buf;
 231
 232 static int __init do_start(void)
 233 {
 234         read_into(header_buf, 110, GotHeader);
 235         return 0;
 236 }
 237
 238 static int __init do_collect(void)
 239 {
 240         unsigned long n = remains;
 241         if (byte_count < n)
 242                 n = byte_count;
 243         memcpy(collect, victim, n);
 244         eat(n);
 245         collect += n;
 246         if ((remains -= n) != 0)
 247                 return 1;
 248         state = next_state;
 249         return 0;
 250 }
 251
 252 static int __init do_header(void)
 253 {
 254         if (memcmp(collected, "070707", 6)==0) {
 255                 error("incorrect cpio method used: use -H newc option");
 256                 return 1;
 257         }
 258         if (memcmp(collected, "070701", 6)) {
 259                 error("no cpio magic");
 260                 return 1;
 261         }
 262         parse_header(collected);
 263         next_header = this_header + N_ALIGN(name_len) + body_len;
 264         next_header = (next_header + 3) & ~3;
 265         state = SkipIt;
 266         if (name_len <= 0 || name_len > PATH_MAX)
 267                 return 0;
 268         if (S_ISLNK(mode)) {
 269                 if (body_len > PATH_MAX)
 270                         return 0;
 271                 collect = collected = symlink_buf;
 272                 remains = N_ALIGN(name_len) + body_len;
 273                 next_state = GotSymlink;
 274                 state = Collect;
 275                 return 0;
 276         }
 277         if (S_ISREG(mode) || !body_len)
 278                 read_into(name_buf, N_ALIGN(name_len), GotName);
 279         return 0;
 280 }
 281
 282 static int __init do_skip(void)
 283 {
 284         if (this_header + byte_count < next_header) {
 285                 eat(byte_count);
 286                 return 1;
 287         } else {
 288                 eat(next_header - this_header);
 289                 state = next_state;
 290                 return 0;
 291         }
 292 }
 293
 294 static int __init do_reset(void)
 295 {
 296         while (byte_count && *victim == '\0')
 297                 eat(1);
 298         if (byte_count && (this_header & 3))
 299                 error("broken padding");
 300         return 1;
 301 }
 302
 303 static int __init maybe_link(void)
 304 {
 305         if (nlink >= 2) {
 306                 char *old = find_link(major, minor, ino, mode, collected);
 307                 if (old)
 308                         return (sys_link(old, collected) < 0) ? -1 : 1;
 309         }
 310         return 0;
 311 }
 312
 313 static void __init clean_path(char *path, umode_t fmode)
 314 {
 315         struct stat st;
 316
 317         if (!sys_newlstat(path, &st) && (st.st_mode ^ fmode) & S_IFMT) {
 318                 if (S_ISDIR(st.st_mode))
 319                         sys_rmdir(path);
 320                 else
 321                         sys_unlink(path);
 322         }
 323 }
 324
 325 static __initdata int wfd;
 326
 327 static int __init do_name(void)
 328 {
 329         state = SkipIt;
 330         next_state = Reset;
 331         if (strcmp(collected, "TRAILER!!!") == 0) {
 332                 free_hash();
 333                 return 0;
 334         }
 335         clean_path(collected, mode);
 336         if (S_ISREG(mode)) {
 337                 int ml = maybe_link();
 338                 if (ml >= 0) {
 339                         int openflags = O_WRONLY|O_CREAT;
 340                         if (ml != 1)
 341                                 openflags |= O_TRUNC;
 342                         wfd = sys_open(collected, openflags, mode);
 343
 344                         if (wfd >= 0) {
 345                                 sys_fchown(wfd, uid, gid);
 346                                 sys_fchmod(wfd, mode);
 347                                 if (body_len)
 348                                         sys_ftruncate(wfd, body_len);
 349                                 vcollected = kstrdup(collected, GFP_KERNEL);
 350                                 state = CopyFile;
 351                         }
 352                 }
 353         } else if (S_ISDIR(mode)) {
 354                 sys_mkdir(collected, mode);
 355                 sys_chown(collected, uid, gid);
 356                 sys_chmod(collected, mode);
 357                 dir_add(collected, mtime);
 358         } else if (S_ISBLK(mode) || S_ISCHR(mode) ||
 359                    S_ISFIFO(mode) || S_ISSOCK(mode)) {
 360                 if (maybe_link() == 0) {
 361                         sys_mknod(collected, mode, rdev);
 362                         sys_chown(collected, uid, gid);
 363                         sys_chmod(collected, mode);
 364                         do_utime(collected, mtime);
 365                 }
 366         }
 367         return 0;
 368 }
 369
 370 static int __init do_copy(void)
 371 {
 372         if (byte_count >= body_len) {
 373                 if (xwrite(wfd, victim, body_len) != body_len)
 374                         error("write error");
 375                 sys_close(wfd);
 376                 do_utime(vcollected, mtime);
 377                 kfree(vcollected);
 378                 eat(body_len);
 379                 state = SkipIt;
 380                 return 0;
 381         } else {
 382                 if (xwrite(wfd, victim, byte_count) != byte_count)
 383                         error("write error");
 384                 body_len -= byte_count;
 385                 eat(byte_count);
 386                 return 1;
 387         }
 388 }
 389
 390 static int __init do_symlink(void)
 391 {
 392         collected[N_ALIGN(name_len) + body_len] = '\0';
 393         clean_path(collected, 0);
 394         sys_symlink(collected + N_ALIGN(name_len), collected);
 395         sys_lchown(collected, uid, gid);
 396         do_utime(collected, mtime);
 397         state = SkipIt;
 398         next_state = Reset;
 399         return 0;
 400 }
 401
 402 static __initdata int (*actions[])(void) = {
 403         [Start]         = do_start,
 404         [Collect]       = do_collect,
 405         [GotHeader]     = do_header,
 406         [SkipIt]        = do_skip,
 407         [GotName]       = do_name,
 408         [CopyFile]      = do_copy,
 409         [GotSymlink]    = do_symlink,
 410         [Reset]         = do_reset,
 411 };
 412
 413 static long __init write_buffer(char *buf, unsigned long len)
 414 {
 415         byte_count = len;
 416         victim = buf;
 417
 418         while (!actions[state]())
 419                 ;
 420         return len - byte_count;
 421 }
 422
 423 static long __init flush_buffer(void *bufv, unsigned long len)
 424 {
 425         char *buf = (char *) bufv;
 426         long written;
 427         long origLen = len;
 428         if (message)
 429                 return -1;
 430         while ((written = write_buffer(buf, len)) < len && !message) {
 431                 char c = buf[written];
 432                 if (c == '0') {
 433                         buf += written;
 434                         len -= written;
 435                         state = Start;
 436                 } else if (c == 0) {
 437                         buf += written;
 438                         len -= written;
 439                         state = Reset;
 440                 } else
 441                         error("junk in compressed archive");
 442         }
 443         return origLen;
 444 }
 445
 446 static unsigned long my_inptr; /* index of next byte to be processed in inbuf */
 447
 448 #include <linux/decompress/generic.h>
 449
 450 static char * __init unpack_to_rootfs(char *buf, unsigned long len)
 451 {
 452         long written;
 453         decompress_fn decompress;
 454         const char *compress_name;
 455         static __initdata char msg_buf[64];
 456
 457         header_buf = kmalloc(110, GFP_KERNEL);
 458         symlink_buf = kmalloc(PATH_MAX + N_ALIGN(PATH_MAX) + 1, GFP_KERNEL);
 459         name_buf = kmalloc(N_ALIGN(PATH_MAX), GFP_KERNEL);
 460
 461         if (!header_buf || !symlink_buf || !name_buf)
 462                 panic("can't allocate buffers");
 463
 464         state = Start;
 465         this_header = 0;
 466         message = NULL;
 467         while (!message && len) {
 468                 loff_t saved_offset = this_header;
 469                 if (*buf == '0' && !(this_header & 3)) {
 470                         state = Start;
 471                         written = write_buffer(buf, len);
 472                         buf += written;
 473                         len -= written;
 474                         continue;
 475                 }
 476                 if (!*buf) {
 477                         buf++;
 478                         len--;
 479                         this_header++;
 480                         continue;
 481                 }
 482                 this_header = 0;
 483                 decompress = decompress_method(buf, len, &compress_name);
 484                 pr_debug("Detected %s compressed data\n", compress_name);
 485                 if (decompress) {
 486                         int res = decompress(buf, len, NULL, flush_buffer, NULL,
 487                                    &my_inptr, error);
 488                         if (res)
 489                                 error("decompressor failed");
 490                 } else if (compress_name) {
 491                         if (!message) {
 492                                 snprintf(msg_buf, sizeof msg_buf,
 493                                          "compression method %s not configured",
 494                                          compress_name);
 495                                 message = msg_buf;
 496                         }
 497                 } else
 498                         error("junk in compressed archive");
 499                 if (state != Reset)
 500                         error("junk in compressed archive");
 501                 this_header = saved_offset + my_inptr;
 502                 buf += my_inptr;
 503                 len -= my_inptr;
 504         }
 505         dir_utime();
 506         kfree(name_buf);
 507         kfree(symlink_buf);
 508         kfree(header_buf);
 509         return message;
 510 }
 511
 512 static int __initdata do_retain_initrd;
 513
 514 static int __init retain_initrd_param(char *str)
 515 {
 516         if (*str)
 517                 return 0;
 518         do_retain_initrd = 1;
 519         return 1;
 520 }
 521 __setup("retain_initrd", retain_initrd_param);
 522
 523 extern char __initramfs_start[];
 524 extern unsigned long __initramfs_size;
 525 #include <linux/initrd.h>
 526 #include <linux/kexec.h>
 527
 528 static void __init free_initrd(void)
 529 {
 530 #ifdef CONFIG_KEXEC_CORE
 531         unsigned long crashk_start = (unsigned long)__va(crashk_res.start);
 532         unsigned long crashk_end   = (unsigned long)__va(crashk_res.end);
 533 #endif
 534         if (do_retain_initrd)
 535                 goto skip;
 536
 537 #ifdef CONFIG_KEXEC_CORE
 538         /*
 539          * If the initrd region is overlapped with crashkernel reserved region,
 540          * free only memory that is not part of crashkernel region.
 541          */
 542         if (initrd_start < crashk_end && initrd_end > crashk_start) {
 543                 /*
 544                  * Initialize initrd memory region since the kexec boot does
 545                  * not do.
 546                  */
 547                 memset((void *)initrd_start, 0, initrd_end - initrd_start);
 548                 if (initrd_start < crashk_start)
 549                         free_initrd_mem(initrd_start, crashk_start);
 550                 if (initrd_end > crashk_end)
 551                         free_initrd_mem(crashk_end, initrd_end);
 552         } else
 553 #endif
 554                 free_initrd_mem(initrd_start, initrd_end);
 555 skip:
 556         initrd_start = 0;
 557         initrd_end = 0;
 558 }
 559
 560 #ifdef CONFIG_BLK_DEV_RAM
 561 #define BUF_SIZE 1024
 562 static void __init clean_rootfs(void)
 563 {
 564         int fd;
 565         void *buf;
 566         struct linux_dirent64 *dirp;
 567         int num;
 568
 569         fd = sys_open("/", O_RDONLY, 0);
 570         WARN_ON(fd < 0);
 571         if (fd < 0)
 572                 return;
 573         buf = kzalloc(BUF_SIZE, GFP_KERNEL);
 574         WARN_ON(!buf);
 575         if (!buf) {
 576                 sys_close(fd);
 577                 return;
 578         }
 579
 580         dirp = buf;
 581         num = sys_getdents64(fd, dirp, BUF_SIZE);
 582         while (num > 0) {
 583                 while (num > 0) {
 584                         struct stat st;
 585                         int ret;
 586
 587                         ret = sys_newlstat(dirp->d_name, &st);
 588                         WARN_ON_ONCE(ret);
 589                         if (!ret) {
 590                                 if (S_ISDIR(st.st_mode))
 591                                         sys_rmdir(dirp->d_name);
 592                                 else
 593                                         sys_unlink(dirp->d_name);
 594                         }
 595
 596                         num -= dirp->d_reclen;
 597                         dirp = (void *)dirp + dirp->d_reclen;
 598                 }
 599                 dirp = buf;
 600                 memset(buf, 0, BUF_SIZE);
 601                 num = sys_getdents64(fd, dirp, BUF_SIZE);
 602         }
 603
 604         sys_close(fd);
 605         kfree(buf);
 606 }
 607 #endif
 608
 609 static int __initdata do_skip_initramfs;
 610
 611 static int __init skip_initramfs_param(char *str)
 612 {
 613         if (*str)
 614                 return 0;
 615         do_skip_initramfs = 1;
 616         return 1;
 617 }
 618 __setup("skip_initramfs", skip_initramfs_param);
 619
 620 static int __init populate_rootfs(void)
 621 {
 622         char *err;
 623
 624         if (do_skip_initramfs) {
 625                 if (initrd_start)
 626                         free_initrd();
 627                 return default_rootfs();
 628         }
 629
 630         err = unpack_to_rootfs(__initramfs_start, __initramfs_size);
 631         if (err)
 632                 panic("%s", err); /* Failed to decompress INTERNAL initramfs */
 633         if (initrd_start) {
 634 #ifdef CONFIG_BLK_DEV_RAM
 635                 int fd;
 636                 printk(KERN_INFO "Trying to unpack rootfs image as initramfs...\n");
 637                 err = unpack_to_rootfs((char *)initrd_start,
 638                         initrd_end - initrd_start);
 639                 if (!err) {
 640                         free_initrd();
 641                         goto done;
 642                 } else {
 643                         clean_rootfs();
 644                         unpack_to_rootfs(__initramfs_start, __initramfs_size);
 645                 }
 646                 printk(KERN_INFO "rootfs image is not initramfs (%s)"
 647                                 "; looks like an initrd\n", err);
 648                 fd = sys_open("/initrd.image",
 649                               O_WRONLY|O_CREAT, 0700);
 650                 if (fd >= 0) {
 651                         ssize_t written = xwrite(fd, (char *)initrd_start,
 652                                                 initrd_end - initrd_start);
 653
 654                         if (written != initrd_end - initrd_start)
 655                                 pr_err("/initrd.image: incomplete write (%zd != %ld)\n",
 656                                        written, initrd_end - initrd_start);
 657
 658                         sys_close(fd);
 659                         free_initrd();
 660                 }
 661         done:
 662 #else
 663                 printk(KERN_INFO "Unpacking initramfs...\n");
 664                 err = unpack_to_rootfs((char *)initrd_start,
 665                         initrd_end - initrd_start);
 666                 if (err)
 667                         printk(KERN_EMERG "Initramfs unpacking failed: %s\n", err);
 668                 free_initrd();
 669 #endif
 670                 /*
 671                  * Try loading default modules from initramfs.  This gives
 672                  * us a chance to load before device_initcalls.
 673                  */
 674                 load_default_modules();
 675         }
 676         return 0;
 677 }
 678 rootfs_initcall(populate_rootfs);