init/do_mounts.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/module.h>
  12 #include <linux/sched.h>
  13 #include <linux/ctype.h>
  14 #include <linux/fd.h>
  15 #include <linux/tty.h>
  16 #include <linux/suspend.h>
  17 #include <linux/root_dev.h>
  18 #include <linux/security.h>
  19 #include <linux/delay.h>
  20 #include <linux/genhd.h>
  21 #include <linux/mount.h>
  22 #include <linux/device.h>
  23 #include <linux/init.h>
  24 #include <linux/fs.h>
  25 #include <linux/initrd.h>
  26 #include <linux/async.h>
  27 #include <linux/fs_struct.h>
  28 #include <linux/slab.h>
  29
  30 #include <linux/nfs_fs.h>
  31 #include <linux/nfs_fs_sb.h>
  32 #include <linux/nfs_mount.h>
  33
  34 #include "do_mounts.h"
  35
  36 int __initdata rd_doload;       /* 1 = load RAM disk, 0 = don't load */
  37
  38 int root_mountflags = MS_RDONLY | MS_SILENT;
  39 static char * __initdata root_device_name;
  40 static char __initdata saved_root_name[64];
  41 static int root_wait;
  42
  43 dev_t ROOT_DEV;
  44
  45 static int __init load_ramdisk(char *str)
  46 {
  47         rd_doload = simple_strtol(str,NULL,0) & 3;
  48         return 1;
  49 }
  50 __setup("load_ramdisk=", load_ramdisk);
  51
  52 static int __init readonly(char *str)
  53 {
  54         if (*str)
  55                 return 0;
  56         root_mountflags |= MS_RDONLY;
  57         return 1;
  58 }
  59
  60 static int __init readwrite(char *str)
  61 {
  62         if (*str)
  63                 return 0;
  64         root_mountflags &= ~MS_RDONLY;
  65         return 1;
  66 }
  67
  68 __setup("ro", readonly);
  69 __setup("rw", readwrite);
  70
  71 #ifdef CONFIG_BLOCK
  72 /**
  73  * match_dev_by_uuid - callback for finding a partition using its uuid
  74  * @dev:        device passed in by the caller
  75  * @data:       opaque pointer to a 36 byte char array with a UUID
  76  *
  77  * Returns 1 if the device matches, and 0 otherwise.
  78  */
  79 static int match_dev_by_uuid(struct device *dev, void *data)
  80 {
  81         u8 *uuid = data;
  82         struct hd_struct *part = dev_to_part(dev);
  83
  84         if (!part->info)
  85                 goto no_match;
  86
  87         if (memcmp(uuid, part->info->uuid, sizeof(part->info->uuid)))
  88                         goto no_match;
  89
  90         return 1;
  91 no_match:
  92         return 0;
  93 }
  94
  95
  96 /**
  97  * devt_from_partuuid - looks up the dev_t of a partition by its UUID
  98  * @uuid:       min 36 byte char array containing a hex ascii UUID
  99  *
 100  * The function will return the first partition which contains a matching
 101  * UUID value in its partition_meta_info struct.  This does not search
 102  * by filesystem UUIDs.
 103  *
 104  * If @uuid is followed by a "/PARTNROFF=%d", then the number will be
 105  * extracted and used as an offset from the partition identified by the UUID.
 106  *
 107  * Returns the matching dev_t on success or 0 on failure.
 108  */
 109 static dev_t devt_from_partuuid(char *uuid_str)
 110 {
 111         dev_t res = 0;
 112         struct device *dev = NULL;
 113         u8 uuid[16];
 114         struct gendisk *disk;
 115         struct hd_struct *part;
 116         int offset = 0;
 117
 118         if (strlen(uuid_str) < 36)
 119                 goto done;
 120
 121         /* Check for optional partition number offset attributes. */
 122         if (uuid_str[36]) {
 123                 char c = 0;
 124                 /* Explicitly fail on poor PARTUUID syntax. */
 125                 if (sscanf(&uuid_str[36],
 126                            "/PARTNROFF=%d%c", &offset, &c) != 1) {
 127                         printk(KERN_ERR "VFS: PARTUUID= is invalid.\n"
 128                          "Expected PARTUUID=<valid-uuid-id>[/PARTNROFF=%%d]\n");
 129                         if (root_wait)
 130                                 printk(KERN_ERR
 131                                      "Disabling rootwait; root= is invalid.\n");
 132                         root_wait = 0;
 133                         goto done;
 134                 }
 135         }
 136
 137         /* Pack the requested UUID in the expected format. */
 138         part_pack_uuid(uuid_str, uuid);
 139
 140         dev = class_find_device(&block_class, NULL, uuid, &match_dev_by_uuid);
 141         if (!dev)
 142                 goto done;
 143
 144         res = dev->devt;
 145
 146         /* Attempt to find the partition by offset. */
 147         if (!offset)
 148                 goto no_offset;
 149
 150         res = 0;
 151         disk = part_to_disk(dev_to_part(dev));
 152         part = disk_get_part(disk, dev_to_part(dev)->partno + offset);
 153         if (part) {
 154                 res = part_devt(part);
 155                 put_device(part_to_dev(part));
 156         }
 157
 158 no_offset:
 159         put_device(dev);
 160 done:
 161         return res;
 162 }
 163 #endif
 164
 165 /*
 166  *      Convert a name into device number.  We accept the following variants:
 167  *
 168  *      1) device number in hexadecimal represents itself
 169  *      2) /dev/nfs represents Root_NFS (0xff)
 170  *      3) /dev/<disk_name> represents the device number of disk
 171  *      4) /dev/<disk_name><decimal> represents the device number
 172  *         of partition - device number of disk plus the partition number
 173  *      5) /dev/<disk_name>p<decimal> - same as the above, that form is
 174  *         used when disk name of partitioned disk ends on a digit.
 175  *      6) PARTUUID=00112233-4455-6677-8899-AABBCCDDEEFF representing the
 176  *         unique id of a partition if the partition table provides it.
 177  *      7) PARTUUID=<UUID>/PARTNROFF=<int> to select a partition in relation to
 178  *         a partition with a known unique id.
 179  *
 180  *      If name doesn't have fall into the categories above, we return (0,0).
 181  *      block_class is used to check if something is a disk name. If the disk
 182  *      name contains slashes, the device name has them replaced with
 183  *      bangs.
 184  */
 185
 186 dev_t name_to_dev_t(char *name)
 187 {
 188         char s[32];
 189         char *p;
 190         dev_t res = 0;
 191         int part;
 192
 193 #ifdef CONFIG_BLOCK
 194         if (strncmp(name, "PARTUUID=", 9) == 0) {
 195                 name += 9;
 196                 res = devt_from_partuuid(name);
 197                 if (!res)
 198                         goto fail;
 199                 goto done;
 200         }
 201 #endif
 202
 203         if (strncmp(name, "/dev/", 5) != 0) {
 204                 unsigned maj, min;
 205
 206                 if (sscanf(name, "%u:%u", &maj, &min) == 2) {
 207                         res = MKDEV(maj, min);
 208                         if (maj != MAJOR(res) || min != MINOR(res))
 209                                 goto fail;
 210                 } else {
 211                         res = new_decode_dev(simple_strtoul(name, &p, 16));
 212                         if (*p)
 213                                 goto fail;
 214                 }
 215                 goto done;
 216         }
 217
 218         name += 5;
 219         res = Root_NFS;
 220         if (strcmp(name, "nfs") == 0)
 221                 goto done;
 222         res = Root_RAM0;
 223         if (strcmp(name, "ram") == 0)
 224                 goto done;
 225
 226         if (strlen(name) > 31)
 227                 goto fail;
 228         strcpy(s, name);
 229         for (p = s; *p; p++)
 230                 if (*p == '/')
 231                         *p = '!';
 232         res = blk_lookup_devt(s, 0);
 233         if (res)
 234                 goto done;
 235
 236         /*
 237          * try non-existent, but valid partition, which may only exist
 238          * after revalidating the disk, like partitioned md devices
 239          */
 240         while (p > s && isdigit(p[-1]))
 241                 p--;
 242         if (p == s || !*p || *p == '0')
 243                 goto fail;
 244
 245         /* try disk name without <part number> */
 246         part = simple_strtoul(p, NULL, 10);
 247         *p = '\0';
 248         res = blk_lookup_devt(s, part);
 249         if (res)
 250                 goto done;
 251
 252         /* try disk name without p<part number> */
 253         if (p < s + 2 || !isdigit(p[-2]) || p[-1] != 'p')
 254                 goto fail;
 255         p[-1] = '\0';
 256         res = blk_lookup_devt(s, part);
 257         if (res)
 258                 goto done;
 259
 260 fail:
 261         return 0;
 262 done:
 263         return res;
 264 }
 265
 266 static int __init root_dev_setup(char *line)
 267 {
 268         strlcpy(saved_root_name, line, sizeof(saved_root_name));
 269         return 1;
 270 }
 271
 272 __setup("root=", root_dev_setup);
 273
 274 static int __init rootwait_setup(char *str)
 275 {
 276         if (*str)
 277                 return 0;
 278         root_wait = 1;
 279         return 1;
 280 }
 281
 282 __setup("rootwait", rootwait_setup);
 283
 284 static char * __initdata root_mount_data;
 285 static int __init root_data_setup(char *str)
 286 {
 287         root_mount_data = str;
 288         return 1;
 289 }
 290
 291 static char * __initdata root_fs_names;
 292 static int __init fs_names_setup(char *str)
 293 {
 294         root_fs_names = str;
 295         return 1;
 296 }
 297
 298 static unsigned int __initdata root_delay;
 299 static int __init root_delay_setup(char *str)
 300 {
 301         root_delay = simple_strtoul(str, NULL, 0);
 302         return 1;
 303 }
 304
 305 __setup("rootflags=", root_data_setup);
 306 __setup("rootfstype=", fs_names_setup);
 307 __setup("rootdelay=", root_delay_setup);
 308
 309 static void __init get_fs_names(char *page)
 310 {
 311         char *s = page;
 312
 313         if (root_fs_names) {
 314                 strcpy(page, root_fs_names);
 315                 while (*s++) {
 316                         if (s[-1] == ',')
 317                                 s[-1] = '\0';
 318                 }
 319         } else {
 320                 int len = get_filesystem_list(page);
 321                 char *p, *next;
 322
 323                 page[len] = '\0';
 324                 for (p = page-1; p; p = next) {
 325                         next = strchr(++p, '\n');
 326                         if (*p++ != '\t')
 327                                 continue;
 328                         while ((*s++ = *p++) != '\n')
 329                                 ;
 330                         s[-1] = '\0';
 331                 }
 332         }
 333         *s = '\0';
 334 }
 335
 336 static int __init do_mount_root(char *name, char *fs, int flags, void *data)
 337 {
 338         struct super_block *s;
 339         int err = sys_mount(name, "/root", fs, flags, data);
 340         if (err)
 341                 return err;
 342
 343         sys_chdir("/root");
 344         s = current->fs->pwd.dentry->d_sb;
 345         ROOT_DEV = s->s_dev;
 346         printk(KERN_INFO
 347                "VFS: Mounted root (%s filesystem)%s on device %u:%u.\n",
 348                s->s_type->name,
 349                s->s_flags & MS_RDONLY ?  " readonly" : "",
 350                MAJOR(ROOT_DEV), MINOR(ROOT_DEV));
 351         return 0;
 352 }
 353
 354 void __init mount_block_root(char *name, int flags)
 355 {
 356         char *fs_names = __getname_gfp(GFP_KERNEL
 357                 | __GFP_NOTRACK_FALSE_POSITIVE);
 358         char *p;
 359 #ifdef CONFIG_BLOCK
 360         char b[BDEVNAME_SIZE];
 361 #else
 362         const char *b = name;
 363 #endif
 364
 365         get_fs_names(fs_names);
 366 retry:
 367         for (p = fs_names; *p; p += strlen(p)+1) {
 368                 int err = do_mount_root(name, p, flags, root_mount_data);
 369                 switch (err) {
 370                         case 0:
 371                                 goto out;
 372                         case -EACCES:
 373                                 flags |= MS_RDONLY;
 374                                 goto retry;
 375                         case -EINVAL:
 376                                 continue;
 377                 }
 378                 /*
 379                  * Allow the user to distinguish between failed sys_open
 380                  * and bad superblock on root device.
 381                  * and give them a list of the available devices
 382                  */
 383 #ifdef CONFIG_BLOCK
 384                 __bdevname(ROOT_DEV, b);
 385 #endif
 386                 printk("VFS: Cannot open root device \"%s\" or %s: error %d\n",
 387                                 root_device_name, b, err);
 388                 printk("Please append a correct \"root=\" boot option; here are the available partitions:\n");
 389
 390                 printk_all_partitions();
 391 #ifdef CONFIG_DEBUG_BLOCK_EXT_DEVT
 392                 printk("DEBUG_BLOCK_EXT_DEVT is enabled, you need to specify "
 393                        "explicit textual name for \"root=\" boot option.\n");
 394 #endif
 395                 panic("VFS: Unable to mount root fs on %s", b);
 396         }
 397
 398         printk("List of all partitions:\n");
 399         printk_all_partitions();
 400         printk("No filesystem could mount root, tried: ");
 401         for (p = fs_names; *p; p += strlen(p)+1)
 402                 printk(" %s", p);
 403         printk("\n");
 404 #ifdef CONFIG_BLOCK
 405         __bdevname(ROOT_DEV, b);
 406 #endif
 407         panic("VFS: Unable to mount root fs on %s", b);
 408 out:
 409         putname(fs_names);
 410 }
 411
 412 #ifdef CONFIG_ROOT_NFS
 413
 414 #define NFSROOT_TIMEOUT_MIN     5
 415 #define NFSROOT_TIMEOUT_MAX     30
 416 #define NFSROOT_RETRY_MAX       5
 417
 418 static int __init mount_nfs_root(void)
 419 {
 420         char *root_dev, *root_data;
 421         unsigned int timeout;
 422         int try, err;
 423
 424         err = nfs_root_data(&root_dev, &root_data);
 425         if (err != 0)
 426                 return 0;
 427
 428         /*
 429          * The server or network may not be ready, so try several
 430          * times.  Stop after a few tries in case the client wants
 431          * to fall back to other boot methods.
 432          */
 433         timeout = NFSROOT_TIMEOUT_MIN;
 434         for (try = 1; ; try++) {
 435                 err = do_mount_root(root_dev, "nfs",
 436                                         root_mountflags, root_data);
 437                 if (err == 0)
 438                         return 1;
 439                 if (try > NFSROOT_RETRY_MAX)
 440                         break;
 441
 442                 /* Wait, in case the server refused us immediately */
 443                 ssleep(timeout);
 444                 timeout <<= 1;
 445                 if (timeout > NFSROOT_TIMEOUT_MAX)
 446                         timeout = NFSROOT_TIMEOUT_MAX;
 447         }
 448         return 0;
 449 }
 450 #endif
 451
 452 #if defined(CONFIG_BLK_DEV_RAM) || defined(CONFIG_BLK_DEV_FD)
 453 void __init change_floppy(char *fmt, ...)
 454 {
 455         struct termios termios;
 456         char buf[80];
 457         char c;
 458         int fd;
 459         va_list args;
 460         va_start(args, fmt);
 461         vsprintf(buf, fmt, args);
 462         va_end(args);
 463         fd = sys_open("/dev/root", O_RDWR | O_NDELAY, 0);
 464         if (fd >= 0) {
 465                 sys_ioctl(fd, FDEJECT, 0);
 466                 sys_close(fd);
 467         }
 468         printk(KERN_NOTICE "VFS: Insert %s and press ENTER\n", buf);
 469         fd = sys_open("/dev/console", O_RDWR, 0);
 470         if (fd >= 0) {
 471                 sys_ioctl(fd, TCGETS, (long)&termios);
 472                 termios.c_lflag &= ~ICANON;
 473                 sys_ioctl(fd, TCSETSF, (long)&termios);
 474                 sys_read(fd, &c, 1);
 475                 termios.c_lflag |= ICANON;
 476                 sys_ioctl(fd, TCSETSF, (long)&termios);
 477                 sys_close(fd);
 478         }
 479 }
 480 #endif
 481
 482 void __init mount_root(void)
 483 {
 484 #ifdef CONFIG_ROOT_NFS
 485         if (ROOT_DEV == Root_NFS) {
 486                 if (mount_nfs_root())
 487                         return;
 488
 489                 printk(KERN_ERR "VFS: Unable to mount root fs via NFS, trying floppy.\n");
 490                 ROOT_DEV = Root_FD0;
 491         }
 492 #endif
 493 #ifdef CONFIG_BLK_DEV_FD
 494         if (MAJOR(ROOT_DEV) == FLOPPY_MAJOR) {
 495                 /* rd_doload is 2 for a dual initrd/ramload setup */
 496                 if (rd_doload==2) {
 497                         if (rd_load_disk(1)) {
 498                                 ROOT_DEV = Root_RAM1;
 499                                 root_device_name = NULL;
 500                         }
 501                 } else
 502                         change_floppy("root floppy");
 503         }
 504 #endif
 505 #ifdef CONFIG_BLOCK
 506         create_dev("/dev/root", ROOT_DEV);
 507         mount_block_root("/dev/root", root_mountflags);
 508 #endif
 509 }
 510
 511 /*
 512  * Prepare the namespace - decide what/where to mount, load ramdisks, etc.
 513  */
 514 void __init prepare_namespace(void)
 515 {
 516         int is_floppy;
 517
 518         if (root_delay) {
 519                 printk(KERN_INFO "Waiting %dsec before mounting root device...\n",
 520                        root_delay);
 521                 ssleep(root_delay);
 522         }
 523
 524         /*
 525          * wait for the known devices to complete their probing
 526          *
 527          * Note: this is a potential source of long boot delays.
 528          * For example, it is not atypical to wait 5 seconds here
 529          * for the touchpad of a laptop to initialize.
 530          */
 531         wait_for_device_probe();
 532
 533         md_run_setup();
 534
 535         if (saved_root_name[0]) {
 536                 root_device_name = saved_root_name;
 537                 if (!strncmp(root_device_name, "mtd", 3) ||
 538                     !strncmp(root_device_name, "ubi", 3)) {
 539                         mount_block_root(root_device_name, root_mountflags);
 540                         goto out;
 541                 }
 542                 ROOT_DEV = name_to_dev_t(root_device_name);
 543                 if (strncmp(root_device_name, "/dev/", 5) == 0)
 544                         root_device_name += 5;
 545         }
 546
 547         if (initrd_load())
 548                 goto out;
 549
 550         /* wait for any asynchronous scanning to complete */
 551         if ((ROOT_DEV == 0) && root_wait) {
 552                 printk(KERN_INFO "Waiting for root device %s...\n",
 553                         saved_root_name);
 554                 while (driver_probe_done() != 0 ||
 555                         (ROOT_DEV = name_to_dev_t(saved_root_name)) == 0)
 556                         msleep(100);
 557                 async_synchronize_full();
 558         }
 559
 560         is_floppy = MAJOR(ROOT_DEV) == FLOPPY_MAJOR;
 561
 562         if (is_floppy && rd_doload && rd_load_disk(0))
 563                 ROOT_DEV = Root_RAM0;
 564
 565         mount_root();
 566 out:
 567         devtmpfs_mount("dev");
 568         sys_mount(".", "/", NULL, MS_MOVE, NULL);
 569         sys_chroot(".");
 570 }