1 /* 2 * linux/fs/super.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 * 6 * super.c contains code to handle: - mount structures 7 * - super-block tables 8 * - filesystem drivers list 9 * - mount system call 10 * - umount system call 11 * - ustat system call 12 * 13 * GK 2/5/95 - Changed to support mounting the root fs via NFS 14 * 15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall 16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96 17 * Added options to /proc/mounts: 18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996. 19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998 20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000 21 */ 22 23 #include <linux/module.h> 24 #include <linux/slab.h> 25 #include <linux/init.h> 26 #include <linux/smp_lock.h> 27 #include <linux/acct.h> 28 #include <linux/blkdev.h> 29 #include <linux/quotaops.h> 30 #include <linux/namei.h> 31 #include <linux/buffer_head.h> /* for fsync_super() */ 32 #include <linux/mount.h> 33 #include <linux/security.h> 34 #include <linux/syscalls.h> 35 #include <linux/vfs.h> 36 #include <linux/writeback.h> /* for the emergency remount stuff */ 37 #include <linux/idr.h> 38 #include <linux/kobject.h> 39 #include <linux/mutex.h> 40 #include <linux/file.h> 41 #include <linux/async.h> 42 #include <asm/uaccess.h> 43 #include "internal.h" 44 45 46 LIST_HEAD(super_blocks); 47 DEFINE_SPINLOCK(sb_lock); 48 49 /** 50 * alloc_super - create new superblock 51 * @type: filesystem type superblock should belong to 52 * 53 * Allocates and initializes a new &struct super_block. alloc_super() 54 * returns a pointer new superblock or %NULL if allocation had failed. 55 */ 56 static struct super_block *alloc_super(struct file_system_type *type) 57 { 58 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); 59 static struct super_operations default_op; 60 61 if (s) { 62 if (security_sb_alloc(s)) { 63 kfree(s); 64 s = NULL; 65 goto out; 66 } 67 INIT_LIST_HEAD(&s->s_dirty); 68 INIT_LIST_HEAD(&s->s_io); 69 INIT_LIST_HEAD(&s->s_more_io); 70 INIT_LIST_HEAD(&s->s_files); 71 INIT_LIST_HEAD(&s->s_instances); 72 INIT_HLIST_HEAD(&s->s_anon); 73 INIT_LIST_HEAD(&s->s_inodes); 74 INIT_LIST_HEAD(&s->s_dentry_lru); 75 INIT_LIST_HEAD(&s->s_async_list); 76 init_rwsem(&s->s_umount); 77 mutex_init(&s->s_lock); 78 lockdep_set_class(&s->s_umount, &type->s_umount_key); 79 /* 80 * The locking rules for s_lock are up to the 81 * filesystem. For example ext3fs has different 82 * lock ordering than usbfs: 83 */ 84 lockdep_set_class(&s->s_lock, &type->s_lock_key); 85 /* 86 * sget() can have s_umount recursion. 87 * 88 * When it cannot find a suitable sb, it allocates a new 89 * one (this one), and tries again to find a suitable old 90 * one. 91 * 92 * In case that succeeds, it will acquire the s_umount 93 * lock of the old one. Since these are clearly distrinct 94 * locks, and this object isn't exposed yet, there's no 95 * risk of deadlocks. 96 * 97 * Annotate this by putting this lock in a different 98 * subclass. 99 */ 100 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); 101 s->s_count = S_BIAS; 102 atomic_set(&s->s_active, 1); 103 mutex_init(&s->s_vfs_rename_mutex); 104 mutex_init(&s->s_dquot.dqio_mutex); 105 mutex_init(&s->s_dquot.dqonoff_mutex); 106 init_rwsem(&s->s_dquot.dqptr_sem); 107 init_waitqueue_head(&s->s_wait_unfrozen); 108 s->s_maxbytes = MAX_NON_LFS; 109 s->dq_op = sb_dquot_ops; 110 s->s_qcop = sb_quotactl_ops; 111 s->s_op = &default_op; 112 s->s_time_gran = 1000000000; 113 } 114 out: 115 return s; 116 } 117 118 /** 119 * destroy_super - frees a superblock 120 * @s: superblock to free 121 * 122 * Frees a superblock. 123 */ 124 static inline void destroy_super(struct super_block *s) 125 { 126 security_sb_free(s); 127 kfree(s->s_subtype); 128 kfree(s->s_options); 129 kfree(s); 130 } 131 132 /* Superblock refcounting */ 133 134 /* 135 * Drop a superblock's refcount. Returns non-zero if the superblock was 136 * destroyed. The caller must hold sb_lock. 137 */ 138 static int __put_super(struct super_block *sb) 139 { 140 int ret = 0; 141 142 if (!--sb->s_count) { 143 destroy_super(sb); 144 ret = 1; 145 } 146 return ret; 147 } 148 149 /* 150 * Drop a superblock's refcount. 151 * Returns non-zero if the superblock is about to be destroyed and 152 * at least is already removed from super_blocks list, so if we are 153 * making a loop through super blocks then we need to restart. 154 * The caller must hold sb_lock. 155 */ 156 int __put_super_and_need_restart(struct super_block *sb) 157 { 158 /* check for race with generic_shutdown_super() */ 159 if (list_empty(&sb->s_list)) { 160 /* super block is removed, need to restart... */ 161 __put_super(sb); 162 return 1; 163 } 164 /* can't be the last, since s_list is still in use */ 165 sb->s_count--; 166 BUG_ON(sb->s_count == 0); 167 return 0; 168 } 169 170 /** 171 * put_super - drop a temporary reference to superblock 172 * @sb: superblock in question 173 * 174 * Drops a temporary reference, frees superblock if there's no 175 * references left. 176 */ 177 static void put_super(struct super_block *sb) 178 { 179 spin_lock(&sb_lock); 180 __put_super(sb); 181 spin_unlock(&sb_lock); 182 } 183 184 185 /** 186 * deactivate_super - drop an active reference to superblock 187 * @s: superblock to deactivate 188 * 189 * Drops an active reference to superblock, acquiring a temprory one if 190 * there is no active references left. In that case we lock superblock, 191 * tell fs driver to shut it down and drop the temporary reference we 192 * had just acquired. 193 */ 194 void deactivate_super(struct super_block *s) 195 { 196 struct file_system_type *fs = s->s_type; 197 if (atomic_dec_and_lock(&s->s_active, &sb_lock)) { 198 s->s_count -= S_BIAS-1; 199 spin_unlock(&sb_lock); 200 DQUOT_OFF(s, 0); 201 down_write(&s->s_umount); 202 fs->kill_sb(s); 203 put_filesystem(fs); 204 put_super(s); 205 } 206 } 207 208 EXPORT_SYMBOL(deactivate_super); 209 210 /** 211 * grab_super - acquire an active reference 212 * @s: reference we are trying to make active 213 * 214 * Tries to acquire an active reference. grab_super() is used when we 215 * had just found a superblock in super_blocks or fs_type->fs_supers 216 * and want to turn it into a full-blown active reference. grab_super() 217 * is called with sb_lock held and drops it. Returns 1 in case of 218 * success, 0 if we had failed (superblock contents was already dead or 219 * dying when grab_super() had been called). 220 */ 221 static int grab_super(struct super_block *s) __releases(sb_lock) 222 { 223 s->s_count++; 224 spin_unlock(&sb_lock); 225 down_write(&s->s_umount); 226 if (s->s_root) { 227 spin_lock(&sb_lock); 228 if (s->s_count > S_BIAS) { 229 atomic_inc(&s->s_active); 230 s->s_count--; 231 spin_unlock(&sb_lock); 232 return 1; 233 } 234 spin_unlock(&sb_lock); 235 } 236 up_write(&s->s_umount); 237 put_super(s); 238 yield(); 239 return 0; 240 } 241 242 /* 243 * Superblock locking. We really ought to get rid of these two. 244 */ 245 void lock_super(struct super_block * sb) 246 { 247 get_fs_excl(); 248 mutex_lock(&sb->s_lock); 249 } 250 251 void unlock_super(struct super_block * sb) 252 { 253 put_fs_excl(); 254 mutex_unlock(&sb->s_lock); 255 } 256 257 EXPORT_SYMBOL(lock_super); 258 EXPORT_SYMBOL(unlock_super); 259 260 /* 261 * Write out and wait upon all dirty data associated with this 262 * superblock. Filesystem data as well as the underlying block 263 * device. Takes the superblock lock. Requires a second blkdev 264 * flush by the caller to complete the operation. 265 */ 266 void __fsync_super(struct super_block *sb) 267 { 268 sync_inodes_sb(sb, 0); 269 DQUOT_SYNC(sb); 270 lock_super(sb); 271 if (sb->s_dirt && sb->s_op->write_super) 272 sb->s_op->write_super(sb); 273 unlock_super(sb); 274 if (sb->s_op->sync_fs) 275 sb->s_op->sync_fs(sb, 1); 276 sync_blockdev(sb->s_bdev); 277 sync_inodes_sb(sb, 1); 278 } 279 280 /* 281 * Write out and wait upon all dirty data associated with this 282 * superblock. Filesystem data as well as the underlying block 283 * device. Takes the superblock lock. 284 */ 285 int fsync_super(struct super_block *sb) 286 { 287 __fsync_super(sb); 288 return sync_blockdev(sb->s_bdev); 289 } 290 291 /** 292 * generic_shutdown_super - common helper for ->kill_sb() 293 * @sb: superblock to kill 294 * 295 * generic_shutdown_super() does all fs-independent work on superblock 296 * shutdown. Typical ->kill_sb() should pick all fs-specific objects 297 * that need destruction out of superblock, call generic_shutdown_super() 298 * and release aforementioned objects. Note: dentries and inodes _are_ 299 * taken care of and do not need specific handling. 300 * 301 * Upon calling this function, the filesystem may no longer alter or 302 * rearrange the set of dentries belonging to this super_block, nor may it 303 * change the attachments of dentries to inodes. 304 */ 305 void generic_shutdown_super(struct super_block *sb) 306 { 307 const struct super_operations *sop = sb->s_op; 308 309 310 if (sb->s_root) { 311 shrink_dcache_for_umount(sb); 312 fsync_super(sb); 313 lock_super(sb); 314 sb->s_flags &= ~MS_ACTIVE; 315 316 /* 317 * wait for asynchronous fs operations to finish before going further 318 */ 319 async_synchronize_full_domain(&sb->s_async_list); 320 321 /* bad name - it should be evict_inodes() */ 322 invalidate_inodes(sb); 323 lock_kernel(); 324 325 if (sop->write_super && sb->s_dirt) 326 sop->write_super(sb); 327 if (sop->put_super) 328 sop->put_super(sb); 329 330 /* Forget any remaining inodes */ 331 if (invalidate_inodes(sb)) { 332 printk("VFS: Busy inodes after unmount of %s. " 333 "Self-destruct in 5 seconds. Have a nice day...\n", 334 sb->s_id); 335 } 336 337 unlock_kernel(); 338 unlock_super(sb); 339 } 340 spin_lock(&sb_lock); 341 /* should be initialized for __put_super_and_need_restart() */ 342 list_del_init(&sb->s_list); 343 list_del(&sb->s_instances); 344 spin_unlock(&sb_lock); 345 up_write(&sb->s_umount); 346 } 347 348 EXPORT_SYMBOL(generic_shutdown_super); 349 350 /** 351 * sget - find or create a superblock 352 * @type: filesystem type superblock should belong to 353 * @test: comparison callback 354 * @set: setup callback 355 * @data: argument to each of them 356 */ 357 struct super_block *sget(struct file_system_type *type, 358 int (*test)(struct super_block *,void *), 359 int (*set)(struct super_block *,void *), 360 void *data) 361 { 362 struct super_block *s = NULL; 363 struct super_block *old; 364 int err; 365 366 retry: 367 spin_lock(&sb_lock); 368 if (test) { 369 list_for_each_entry(old, &type->fs_supers, s_instances) { 370 if (!test(old, data)) 371 continue; 372 if (!grab_super(old)) 373 goto retry; 374 if (s) 375 destroy_super(s); 376 return old; 377 } 378 } 379 if (!s) { 380 spin_unlock(&sb_lock); 381 s = alloc_super(type); 382 if (!s) 383 return ERR_PTR(-ENOMEM); 384 goto retry; 385 } 386 387 err = set(s, data); 388 if (err) { 389 spin_unlock(&sb_lock); 390 destroy_super(s); 391 return ERR_PTR(err); 392 } 393 s->s_type = type; 394 strlcpy(s->s_id, type->name, sizeof(s->s_id)); 395 list_add_tail(&s->s_list, &super_blocks); 396 list_add(&s->s_instances, &type->fs_supers); 397 spin_unlock(&sb_lock); 398 get_filesystem(type); 399 return s; 400 } 401 402 EXPORT_SYMBOL(sget); 403 404 void drop_super(struct super_block *sb) 405 { 406 up_read(&sb->s_umount); 407 put_super(sb); 408 } 409 410 EXPORT_SYMBOL(drop_super); 411 412 static inline void write_super(struct super_block *sb) 413 { 414 lock_super(sb); 415 if (sb->s_root && sb->s_dirt) 416 if (sb->s_op->write_super) 417 sb->s_op->write_super(sb); 418 unlock_super(sb); 419 } 420 421 /* 422 * Note: check the dirty flag before waiting, so we don't 423 * hold up the sync while mounting a device. (The newly 424 * mounted device won't need syncing.) 425 */ 426 void sync_supers(void) 427 { 428 struct super_block *sb; 429 430 spin_lock(&sb_lock); 431 restart: 432 list_for_each_entry(sb, &super_blocks, s_list) { 433 if (sb->s_dirt) { 434 sb->s_count++; 435 spin_unlock(&sb_lock); 436 down_read(&sb->s_umount); 437 write_super(sb); 438 up_read(&sb->s_umount); 439 spin_lock(&sb_lock); 440 if (__put_super_and_need_restart(sb)) 441 goto restart; 442 } 443 } 444 spin_unlock(&sb_lock); 445 } 446 447 /* 448 * Call the ->sync_fs super_op against all filesystems which are r/w and 449 * which implement it. 450 * 451 * This operation is careful to avoid the livelock which could easily happen 452 * if two or more filesystems are being continuously dirtied. s_need_sync_fs 453 * is used only here. We set it against all filesystems and then clear it as 454 * we sync them. So redirtied filesystems are skipped. 455 * 456 * But if process A is currently running sync_filesystems and then process B 457 * calls sync_filesystems as well, process B will set all the s_need_sync_fs 458 * flags again, which will cause process A to resync everything. Fix that with 459 * a local mutex. 460 * 461 * (Fabian) Avoid sync_fs with clean fs & wait mode 0 462 */ 463 void sync_filesystems(int wait) 464 { 465 struct super_block *sb; 466 static DEFINE_MUTEX(mutex); 467 468 mutex_lock(&mutex); /* Could be down_interruptible */ 469 spin_lock(&sb_lock); 470 list_for_each_entry(sb, &super_blocks, s_list) { 471 if (!sb->s_op->sync_fs) 472 continue; 473 if (sb->s_flags & MS_RDONLY) 474 continue; 475 sb->s_need_sync_fs = 1; 476 } 477 478 restart: 479 list_for_each_entry(sb, &super_blocks, s_list) { 480 if (!sb->s_need_sync_fs) 481 continue; 482 sb->s_need_sync_fs = 0; 483 if (sb->s_flags & MS_RDONLY) 484 continue; /* hm. Was remounted r/o meanwhile */ 485 sb->s_count++; 486 spin_unlock(&sb_lock); 487 down_read(&sb->s_umount); 488 async_synchronize_full_domain(&sb->s_async_list); 489 if (sb->s_root && (wait || sb->s_dirt)) 490 sb->s_op->sync_fs(sb, wait); 491 up_read(&sb->s_umount); 492 /* restart only when sb is no longer on the list */ 493 spin_lock(&sb_lock); 494 if (__put_super_and_need_restart(sb)) 495 goto restart; 496 } 497 spin_unlock(&sb_lock); 498 mutex_unlock(&mutex); 499 } 500 501 /** 502 * get_super - get the superblock of a device 503 * @bdev: device to get the superblock for 504 * 505 * Scans the superblock list and finds the superblock of the file system 506 * mounted on the device given. %NULL is returned if no match is found. 507 */ 508 509 struct super_block * get_super(struct block_device *bdev) 510 { 511 struct super_block *sb; 512 513 if (!bdev) 514 return NULL; 515 516 spin_lock(&sb_lock); 517 rescan: 518 list_for_each_entry(sb, &super_blocks, s_list) { 519 if (sb->s_bdev == bdev) { 520 sb->s_count++; 521 spin_unlock(&sb_lock); 522 down_read(&sb->s_umount); 523 if (sb->s_root) 524 return sb; 525 up_read(&sb->s_umount); 526 /* restart only when sb is no longer on the list */ 527 spin_lock(&sb_lock); 528 if (__put_super_and_need_restart(sb)) 529 goto rescan; 530 } 531 } 532 spin_unlock(&sb_lock); 533 return NULL; 534 } 535 536 EXPORT_SYMBOL(get_super); 537 538 struct super_block * user_get_super(dev_t dev) 539 { 540 struct super_block *sb; 541 542 spin_lock(&sb_lock); 543 rescan: 544 list_for_each_entry(sb, &super_blocks, s_list) { 545 if (sb->s_dev == dev) { 546 sb->s_count++; 547 spin_unlock(&sb_lock); 548 down_read(&sb->s_umount); 549 if (sb->s_root) 550 return sb; 551 up_read(&sb->s_umount); 552 /* restart only when sb is no longer on the list */ 553 spin_lock(&sb_lock); 554 if (__put_super_and_need_restart(sb)) 555 goto rescan; 556 } 557 } 558 spin_unlock(&sb_lock); 559 return NULL; 560 } 561 562 SYSCALL_DEFINE2(ustat, unsigned, dev, struct ustat __user *, ubuf) 563 { 564 struct super_block *s; 565 struct ustat tmp; 566 struct kstatfs sbuf; 567 int err = -EINVAL; 568 569 s = user_get_super(new_decode_dev(dev)); 570 if (s == NULL) 571 goto out; 572 err = vfs_statfs(s->s_root, &sbuf); 573 drop_super(s); 574 if (err) 575 goto out; 576 577 memset(&tmp,0,sizeof(struct ustat)); 578 tmp.f_tfree = sbuf.f_bfree; 579 tmp.f_tinode = sbuf.f_ffree; 580 581 err = copy_to_user(ubuf,&tmp,sizeof(struct ustat)) ? -EFAULT : 0; 582 out: 583 return err; 584 } 585 586 /** 587 * mark_files_ro - mark all files read-only 588 * @sb: superblock in question 589 * 590 * All files are marked read-only. We don't care about pending 591 * delete files so this should be used in 'force' mode only. 592 */ 593 594 static void mark_files_ro(struct super_block *sb) 595 { 596 struct file *f; 597 598 retry: 599 file_list_lock(); 600 list_for_each_entry(f, &sb->s_files, f_u.fu_list) { 601 struct vfsmount *mnt; 602 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode)) 603 continue; 604 if (!file_count(f)) 605 continue; 606 if (!(f->f_mode & FMODE_WRITE)) 607 continue; 608 f->f_mode &= ~FMODE_WRITE; 609 if (file_check_writeable(f) != 0) 610 continue; 611 file_release_write(f); 612 mnt = mntget(f->f_path.mnt); 613 file_list_unlock(); 614 /* 615 * This can sleep, so we can't hold 616 * the file_list_lock() spinlock. 617 */ 618 mnt_drop_write(mnt); 619 mntput(mnt); 620 goto retry; 621 } 622 file_list_unlock(); 623 } 624 625 /** 626 * do_remount_sb - asks filesystem to change mount options. 627 * @sb: superblock in question 628 * @flags: numeric part of options 629 * @data: the rest of options 630 * @force: whether or not to force the change 631 * 632 * Alters the mount options of a mounted file system. 633 */ 634 int do_remount_sb(struct super_block *sb, int flags, void *data, int force) 635 { 636 int retval; 637 int remount_rw; 638 639 #ifdef CONFIG_BLOCK 640 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev)) 641 return -EACCES; 642 #endif 643 if (flags & MS_RDONLY) 644 acct_auto_close(sb); 645 shrink_dcache_sb(sb); 646 fsync_super(sb); 647 648 /* If we are remounting RDONLY and current sb is read/write, 649 make sure there are no rw files opened */ 650 if ((flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY)) { 651 if (force) 652 mark_files_ro(sb); 653 else if (!fs_may_remount_ro(sb)) 654 return -EBUSY; 655 retval = DQUOT_OFF(sb, 1); 656 if (retval < 0 && retval != -ENOSYS) 657 return -EBUSY; 658 } 659 remount_rw = !(flags & MS_RDONLY) && (sb->s_flags & MS_RDONLY); 660 661 if (sb->s_op->remount_fs) { 662 lock_super(sb); 663 retval = sb->s_op->remount_fs(sb, &flags, data); 664 unlock_super(sb); 665 if (retval) 666 return retval; 667 } 668 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK); 669 if (remount_rw) 670 DQUOT_ON_REMOUNT(sb); 671 return 0; 672 } 673 674 static void do_emergency_remount(unsigned long foo) 675 { 676 struct super_block *sb; 677 678 spin_lock(&sb_lock); 679 list_for_each_entry(sb, &super_blocks, s_list) { 680 sb->s_count++; 681 spin_unlock(&sb_lock); 682 down_read(&sb->s_umount); 683 if (sb->s_root && sb->s_bdev && !(sb->s_flags & MS_RDONLY)) { 684 /* 685 * ->remount_fs needs lock_kernel(). 686 * 687 * What lock protects sb->s_flags?? 688 */ 689 lock_kernel(); 690 do_remount_sb(sb, MS_RDONLY, NULL, 1); 691 unlock_kernel(); 692 } 693 drop_super(sb); 694 spin_lock(&sb_lock); 695 } 696 spin_unlock(&sb_lock); 697 printk("Emergency Remount complete\n"); 698 } 699 700 void emergency_remount(void) 701 { 702 pdflush_operation(do_emergency_remount, 0); 703 } 704 705 /* 706 * Unnamed block devices are dummy devices used by virtual 707 * filesystems which don't use real block-devices. -- jrs 708 */ 709 710 static DEFINE_IDA(unnamed_dev_ida); 711 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ 712 713 int set_anon_super(struct super_block *s, void *data) 714 { 715 int dev; 716 int error; 717 718 retry: 719 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) 720 return -ENOMEM; 721 spin_lock(&unnamed_dev_lock); 722 error = ida_get_new(&unnamed_dev_ida, &dev); 723 spin_unlock(&unnamed_dev_lock); 724 if (error == -EAGAIN) 725 /* We raced and lost with another CPU. */ 726 goto retry; 727 else if (error) 728 return -EAGAIN; 729 730 if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) { 731 spin_lock(&unnamed_dev_lock); 732 ida_remove(&unnamed_dev_ida, dev); 733 spin_unlock(&unnamed_dev_lock); 734 return -EMFILE; 735 } 736 s->s_dev = MKDEV(0, dev & MINORMASK); 737 return 0; 738 } 739 740 EXPORT_SYMBOL(set_anon_super); 741 742 void kill_anon_super(struct super_block *sb) 743 { 744 int slot = MINOR(sb->s_dev); 745 746 generic_shutdown_super(sb); 747 spin_lock(&unnamed_dev_lock); 748 ida_remove(&unnamed_dev_ida, slot); 749 spin_unlock(&unnamed_dev_lock); 750 } 751 752 EXPORT_SYMBOL(kill_anon_super); 753 754 void kill_litter_super(struct super_block *sb) 755 { 756 if (sb->s_root) 757 d_genocide(sb->s_root); 758 kill_anon_super(sb); 759 } 760 761 EXPORT_SYMBOL(kill_litter_super); 762 763 #ifdef CONFIG_BLOCK 764 static int set_bdev_super(struct super_block *s, void *data) 765 { 766 s->s_bdev = data; 767 s->s_dev = s->s_bdev->bd_dev; 768 return 0; 769 } 770 771 static int test_bdev_super(struct super_block *s, void *data) 772 { 773 return (void *)s->s_bdev == data; 774 } 775 776 int get_sb_bdev(struct file_system_type *fs_type, 777 int flags, const char *dev_name, void *data, 778 int (*fill_super)(struct super_block *, void *, int), 779 struct vfsmount *mnt) 780 { 781 struct block_device *bdev; 782 struct super_block *s; 783 fmode_t mode = FMODE_READ; 784 int error = 0; 785 786 if (!(flags & MS_RDONLY)) 787 mode |= FMODE_WRITE; 788 789 bdev = open_bdev_exclusive(dev_name, mode, fs_type); 790 if (IS_ERR(bdev)) 791 return PTR_ERR(bdev); 792 793 /* 794 * once the super is inserted into the list by sget, s_umount 795 * will protect the lockfs code from trying to start a snapshot 796 * while we are mounting 797 */ 798 down(&bdev->bd_mount_sem); 799 s = sget(fs_type, test_bdev_super, set_bdev_super, bdev); 800 up(&bdev->bd_mount_sem); 801 if (IS_ERR(s)) 802 goto error_s; 803 804 if (s->s_root) { 805 if ((flags ^ s->s_flags) & MS_RDONLY) { 806 up_write(&s->s_umount); 807 deactivate_super(s); 808 error = -EBUSY; 809 goto error_bdev; 810 } 811 812 close_bdev_exclusive(bdev, mode); 813 } else { 814 char b[BDEVNAME_SIZE]; 815 816 s->s_flags = flags; 817 s->s_mode = mode; 818 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); 819 sb_set_blocksize(s, block_size(bdev)); 820 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 821 if (error) { 822 up_write(&s->s_umount); 823 deactivate_super(s); 824 goto error; 825 } 826 827 s->s_flags |= MS_ACTIVE; 828 bdev->bd_super = s; 829 } 830 831 return simple_set_mnt(mnt, s); 832 833 error_s: 834 error = PTR_ERR(s); 835 error_bdev: 836 close_bdev_exclusive(bdev, mode); 837 error: 838 return error; 839 } 840 841 EXPORT_SYMBOL(get_sb_bdev); 842 843 void kill_block_super(struct super_block *sb) 844 { 845 struct block_device *bdev = sb->s_bdev; 846 fmode_t mode = sb->s_mode; 847 848 bdev->bd_super = 0; 849 generic_shutdown_super(sb); 850 sync_blockdev(bdev); 851 close_bdev_exclusive(bdev, mode); 852 } 853 854 EXPORT_SYMBOL(kill_block_super); 855 #endif 856 857 int get_sb_nodev(struct file_system_type *fs_type, 858 int flags, void *data, 859 int (*fill_super)(struct super_block *, void *, int), 860 struct vfsmount *mnt) 861 { 862 int error; 863 struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL); 864 865 if (IS_ERR(s)) 866 return PTR_ERR(s); 867 868 s->s_flags = flags; 869 870 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 871 if (error) { 872 up_write(&s->s_umount); 873 deactivate_super(s); 874 return error; 875 } 876 s->s_flags |= MS_ACTIVE; 877 return simple_set_mnt(mnt, s); 878 } 879 880 EXPORT_SYMBOL(get_sb_nodev); 881 882 static int compare_single(struct super_block *s, void *p) 883 { 884 return 1; 885 } 886 887 int get_sb_single(struct file_system_type *fs_type, 888 int flags, void *data, 889 int (*fill_super)(struct super_block *, void *, int), 890 struct vfsmount *mnt) 891 { 892 struct super_block *s; 893 int error; 894 895 s = sget(fs_type, compare_single, set_anon_super, NULL); 896 if (IS_ERR(s)) 897 return PTR_ERR(s); 898 if (!s->s_root) { 899 s->s_flags = flags; 900 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 901 if (error) { 902 up_write(&s->s_umount); 903 deactivate_super(s); 904 return error; 905 } 906 s->s_flags |= MS_ACTIVE; 907 } 908 do_remount_sb(s, flags, data, 0); 909 return simple_set_mnt(mnt, s); 910 } 911 912 EXPORT_SYMBOL(get_sb_single); 913 914 struct vfsmount * 915 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 916 { 917 struct vfsmount *mnt; 918 char *secdata = NULL; 919 int error; 920 921 if (!type) 922 return ERR_PTR(-ENODEV); 923 924 error = -ENOMEM; 925 mnt = alloc_vfsmnt(name); 926 if (!mnt) 927 goto out; 928 929 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { 930 secdata = alloc_secdata(); 931 if (!secdata) 932 goto out_mnt; 933 934 error = security_sb_copy_data(data, secdata); 935 if (error) 936 goto out_free_secdata; 937 } 938 939 error = type->get_sb(type, flags, name, data, mnt); 940 if (error < 0) 941 goto out_free_secdata; 942 BUG_ON(!mnt->mnt_sb); 943 944 error = security_sb_kern_mount(mnt->mnt_sb, flags, secdata); 945 if (error) 946 goto out_sb; 947 948 mnt->mnt_mountpoint = mnt->mnt_root; 949 mnt->mnt_parent = mnt; 950 up_write(&mnt->mnt_sb->s_umount); 951 free_secdata(secdata); 952 return mnt; 953 out_sb: 954 dput(mnt->mnt_root); 955 up_write(&mnt->mnt_sb->s_umount); 956 deactivate_super(mnt->mnt_sb); 957 out_free_secdata: 958 free_secdata(secdata); 959 out_mnt: 960 free_vfsmnt(mnt); 961 out: 962 return ERR_PTR(error); 963 } 964 965 EXPORT_SYMBOL_GPL(vfs_kern_mount); 966 967 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 968 { 969 int err; 970 const char *subtype = strchr(fstype, '.'); 971 if (subtype) { 972 subtype++; 973 err = -EINVAL; 974 if (!subtype[0]) 975 goto err; 976 } else 977 subtype = ""; 978 979 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 980 err = -ENOMEM; 981 if (!mnt->mnt_sb->s_subtype) 982 goto err; 983 return mnt; 984 985 err: 986 mntput(mnt); 987 return ERR_PTR(err); 988 } 989 990 struct vfsmount * 991 do_kern_mount(const char *fstype, int flags, const char *name, void *data) 992 { 993 struct file_system_type *type = get_fs_type(fstype); 994 struct vfsmount *mnt; 995 if (!type) 996 return ERR_PTR(-ENODEV); 997 mnt = vfs_kern_mount(type, flags, name, data); 998 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 999 !mnt->mnt_sb->s_subtype) 1000 mnt = fs_set_subtype(mnt, fstype); 1001 put_filesystem(type); 1002 return mnt; 1003 } 1004 EXPORT_SYMBOL_GPL(do_kern_mount); 1005 1006 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 1007 { 1008 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 1009 } 1010 1011 EXPORT_SYMBOL_GPL(kern_mount_data); 1012