1 /* 2 * linux/fs/namespace.c 3 * 4 * (C) Copyright Al Viro 2000, 2001 5 * Released under GPL v2. 6 * 7 * Based on code from fs/super.c, copyright Linus Torvalds and others. 8 * Heavily rewritten. 9 */ 10 11 #include <linux/syscalls.h> 12 #include <linux/slab.h> 13 #include <linux/sched.h> 14 #include <linux/spinlock.h> 15 #include <linux/percpu.h> 16 #include <linux/init.h> 17 #include <linux/kernel.h> 18 #include <linux/acct.h> 19 #include <linux/capability.h> 20 #include <linux/cpumask.h> 21 #include <linux/module.h> 22 #include <linux/sysfs.h> 23 #include <linux/seq_file.h> 24 #include <linux/mnt_namespace.h> 25 #include <linux/namei.h> 26 #include <linux/nsproxy.h> 27 #include <linux/security.h> 28 #include <linux/mount.h> 29 #include <linux/ramfs.h> 30 #include <linux/log2.h> 31 #include <linux/idr.h> 32 #include <linux/fs_struct.h> 33 #include <linux/fsnotify.h> 34 #include <asm/uaccess.h> 35 #include <asm/unistd.h> 36 #include "pnode.h" 37 #include "internal.h" 38 39 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 40 #define HASH_SIZE (1UL << HASH_SHIFT) 41 42 static int event; 43 static DEFINE_IDA(mnt_id_ida); 44 static DEFINE_IDA(mnt_group_ida); 45 static DEFINE_SPINLOCK(mnt_id_lock); 46 static int mnt_id_start = 0; 47 static int mnt_group_start = 1; 48 49 static struct list_head *mount_hashtable __read_mostly; 50 static struct kmem_cache *mnt_cache __read_mostly; 51 static struct rw_semaphore namespace_sem; 52 53 /* /sys/fs */ 54 struct kobject *fs_kobj; 55 EXPORT_SYMBOL_GPL(fs_kobj); 56 57 /* 58 * vfsmount lock may be taken for read to prevent changes to the 59 * vfsmount hash, ie. during mountpoint lookups or walking back 60 * up the tree. 61 * 62 * It should be taken for write in all cases where the vfsmount 63 * tree or hash is modified or when a vfsmount structure is modified. 64 */ 65 DEFINE_BRLOCK(vfsmount_lock); 66 67 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 68 { 69 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 70 tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 71 tmp = tmp + (tmp >> HASH_SHIFT); 72 return tmp & (HASH_SIZE - 1); 73 } 74 75 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 76 77 /* 78 * allocation is serialized by namespace_sem, but we need the spinlock to 79 * serialize with freeing. 80 */ 81 static int mnt_alloc_id(struct vfsmount *mnt) 82 { 83 int res; 84 85 retry: 86 ida_pre_get(&mnt_id_ida, GFP_KERNEL); 87 spin_lock(&mnt_id_lock); 88 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 89 if (!res) 90 mnt_id_start = mnt->mnt_id + 1; 91 spin_unlock(&mnt_id_lock); 92 if (res == -EAGAIN) 93 goto retry; 94 95 return res; 96 } 97 98 static void mnt_free_id(struct vfsmount *mnt) 99 { 100 int id = mnt->mnt_id; 101 spin_lock(&mnt_id_lock); 102 ida_remove(&mnt_id_ida, id); 103 if (mnt_id_start > id) 104 mnt_id_start = id; 105 spin_unlock(&mnt_id_lock); 106 } 107 108 /* 109 * Allocate a new peer group ID 110 * 111 * mnt_group_ida is protected by namespace_sem 112 */ 113 static int mnt_alloc_group_id(struct vfsmount *mnt) 114 { 115 int res; 116 117 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 118 return -ENOMEM; 119 120 res = ida_get_new_above(&mnt_group_ida, 121 mnt_group_start, 122 &mnt->mnt_group_id); 123 if (!res) 124 mnt_group_start = mnt->mnt_group_id + 1; 125 126 return res; 127 } 128 129 /* 130 * Release a peer group ID 131 */ 132 void mnt_release_group_id(struct vfsmount *mnt) 133 { 134 int id = mnt->mnt_group_id; 135 ida_remove(&mnt_group_ida, id); 136 if (mnt_group_start > id) 137 mnt_group_start = id; 138 mnt->mnt_group_id = 0; 139 } 140 141 /* 142 * vfsmount lock must be held for read 143 */ 144 static inline void mnt_add_count(struct vfsmount *mnt, int n) 145 { 146 #ifdef CONFIG_SMP 147 this_cpu_add(mnt->mnt_pcp->mnt_count, n); 148 #else 149 preempt_disable(); 150 mnt->mnt_count += n; 151 preempt_enable(); 152 #endif 153 } 154 155 static inline void mnt_set_count(struct vfsmount *mnt, int n) 156 { 157 #ifdef CONFIG_SMP 158 this_cpu_write(mnt->mnt_pcp->mnt_count, n); 159 #else 160 mnt->mnt_count = n; 161 #endif 162 } 163 164 /* 165 * vfsmount lock must be held for read 166 */ 167 static inline void mnt_inc_count(struct vfsmount *mnt) 168 { 169 mnt_add_count(mnt, 1); 170 } 171 172 /* 173 * vfsmount lock must be held for read 174 */ 175 static inline void mnt_dec_count(struct vfsmount *mnt) 176 { 177 mnt_add_count(mnt, -1); 178 } 179 180 /* 181 * vfsmount lock must be held for write 182 */ 183 unsigned int mnt_get_count(struct vfsmount *mnt) 184 { 185 #ifdef CONFIG_SMP 186 unsigned int count = 0; 187 int cpu; 188 189 for_each_possible_cpu(cpu) { 190 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 191 } 192 193 return count; 194 #else 195 return mnt->mnt_count; 196 #endif 197 } 198 199 static struct vfsmount *alloc_vfsmnt(const char *name) 200 { 201 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 202 if (mnt) { 203 int err; 204 205 err = mnt_alloc_id(mnt); 206 if (err) 207 goto out_free_cache; 208 209 if (name) { 210 mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 211 if (!mnt->mnt_devname) 212 goto out_free_id; 213 } 214 215 #ifdef CONFIG_SMP 216 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 217 if (!mnt->mnt_pcp) 218 goto out_free_devname; 219 220 this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 221 #else 222 mnt->mnt_count = 1; 223 mnt->mnt_writers = 0; 224 #endif 225 226 INIT_LIST_HEAD(&mnt->mnt_hash); 227 INIT_LIST_HEAD(&mnt->mnt_child); 228 INIT_LIST_HEAD(&mnt->mnt_mounts); 229 INIT_LIST_HEAD(&mnt->mnt_list); 230 INIT_LIST_HEAD(&mnt->mnt_expire); 231 INIT_LIST_HEAD(&mnt->mnt_share); 232 INIT_LIST_HEAD(&mnt->mnt_slave_list); 233 INIT_LIST_HEAD(&mnt->mnt_slave); 234 #ifdef CONFIG_FSNOTIFY 235 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 236 #endif 237 } 238 return mnt; 239 240 #ifdef CONFIG_SMP 241 out_free_devname: 242 kfree(mnt->mnt_devname); 243 #endif 244 out_free_id: 245 mnt_free_id(mnt); 246 out_free_cache: 247 kmem_cache_free(mnt_cache, mnt); 248 return NULL; 249 } 250 251 /* 252 * Most r/o checks on a fs are for operations that take 253 * discrete amounts of time, like a write() or unlink(). 254 * We must keep track of when those operations start 255 * (for permission checks) and when they end, so that 256 * we can determine when writes are able to occur to 257 * a filesystem. 258 */ 259 /* 260 * __mnt_is_readonly: check whether a mount is read-only 261 * @mnt: the mount to check for its write status 262 * 263 * This shouldn't be used directly ouside of the VFS. 264 * It does not guarantee that the filesystem will stay 265 * r/w, just that it is right *now*. This can not and 266 * should not be used in place of IS_RDONLY(inode). 267 * mnt_want/drop_write() will _keep_ the filesystem 268 * r/w. 269 */ 270 int __mnt_is_readonly(struct vfsmount *mnt) 271 { 272 if (mnt->mnt_flags & MNT_READONLY) 273 return 1; 274 if (mnt->mnt_sb->s_flags & MS_RDONLY) 275 return 1; 276 return 0; 277 } 278 EXPORT_SYMBOL_GPL(__mnt_is_readonly); 279 280 static inline void mnt_inc_writers(struct vfsmount *mnt) 281 { 282 #ifdef CONFIG_SMP 283 this_cpu_inc(mnt->mnt_pcp->mnt_writers); 284 #else 285 mnt->mnt_writers++; 286 #endif 287 } 288 289 static inline void mnt_dec_writers(struct vfsmount *mnt) 290 { 291 #ifdef CONFIG_SMP 292 this_cpu_dec(mnt->mnt_pcp->mnt_writers); 293 #else 294 mnt->mnt_writers--; 295 #endif 296 } 297 298 static unsigned int mnt_get_writers(struct vfsmount *mnt) 299 { 300 #ifdef CONFIG_SMP 301 unsigned int count = 0; 302 int cpu; 303 304 for_each_possible_cpu(cpu) { 305 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 306 } 307 308 return count; 309 #else 310 return mnt->mnt_writers; 311 #endif 312 } 313 314 /* 315 * Most r/o checks on a fs are for operations that take 316 * discrete amounts of time, like a write() or unlink(). 317 * We must keep track of when those operations start 318 * (for permission checks) and when they end, so that 319 * we can determine when writes are able to occur to 320 * a filesystem. 321 */ 322 /** 323 * mnt_want_write - get write access to a mount 324 * @mnt: the mount on which to take a write 325 * 326 * This tells the low-level filesystem that a write is 327 * about to be performed to it, and makes sure that 328 * writes are allowed before returning success. When 329 * the write operation is finished, mnt_drop_write() 330 * must be called. This is effectively a refcount. 331 */ 332 int mnt_want_write(struct vfsmount *mnt) 333 { 334 int ret = 0; 335 336 preempt_disable(); 337 mnt_inc_writers(mnt); 338 /* 339 * The store to mnt_inc_writers must be visible before we pass 340 * MNT_WRITE_HOLD loop below, so that the slowpath can see our 341 * incremented count after it has set MNT_WRITE_HOLD. 342 */ 343 smp_mb(); 344 while (mnt->mnt_flags & MNT_WRITE_HOLD) 345 cpu_relax(); 346 /* 347 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 348 * be set to match its requirements. So we must not load that until 349 * MNT_WRITE_HOLD is cleared. 350 */ 351 smp_rmb(); 352 if (__mnt_is_readonly(mnt)) { 353 mnt_dec_writers(mnt); 354 ret = -EROFS; 355 goto out; 356 } 357 out: 358 preempt_enable(); 359 return ret; 360 } 361 EXPORT_SYMBOL_GPL(mnt_want_write); 362 363 /** 364 * mnt_clone_write - get write access to a mount 365 * @mnt: the mount on which to take a write 366 * 367 * This is effectively like mnt_want_write, except 368 * it must only be used to take an extra write reference 369 * on a mountpoint that we already know has a write reference 370 * on it. This allows some optimisation. 371 * 372 * After finished, mnt_drop_write must be called as usual to 373 * drop the reference. 374 */ 375 int mnt_clone_write(struct vfsmount *mnt) 376 { 377 /* superblock may be r/o */ 378 if (__mnt_is_readonly(mnt)) 379 return -EROFS; 380 preempt_disable(); 381 mnt_inc_writers(mnt); 382 preempt_enable(); 383 return 0; 384 } 385 EXPORT_SYMBOL_GPL(mnt_clone_write); 386 387 /** 388 * mnt_want_write_file - get write access to a file's mount 389 * @file: the file who's mount on which to take a write 390 * 391 * This is like mnt_want_write, but it takes a file and can 392 * do some optimisations if the file is open for write already 393 */ 394 int mnt_want_write_file(struct file *file) 395 { 396 struct inode *inode = file->f_dentry->d_inode; 397 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 398 return mnt_want_write(file->f_path.mnt); 399 else 400 return mnt_clone_write(file->f_path.mnt); 401 } 402 EXPORT_SYMBOL_GPL(mnt_want_write_file); 403 404 /** 405 * mnt_drop_write - give up write access to a mount 406 * @mnt: the mount on which to give up write access 407 * 408 * Tells the low-level filesystem that we are done 409 * performing writes to it. Must be matched with 410 * mnt_want_write() call above. 411 */ 412 void mnt_drop_write(struct vfsmount *mnt) 413 { 414 preempt_disable(); 415 mnt_dec_writers(mnt); 416 preempt_enable(); 417 } 418 EXPORT_SYMBOL_GPL(mnt_drop_write); 419 420 static int mnt_make_readonly(struct vfsmount *mnt) 421 { 422 int ret = 0; 423 424 br_write_lock(vfsmount_lock); 425 mnt->mnt_flags |= MNT_WRITE_HOLD; 426 /* 427 * After storing MNT_WRITE_HOLD, we'll read the counters. This store 428 * should be visible before we do. 429 */ 430 smp_mb(); 431 432 /* 433 * With writers on hold, if this value is zero, then there are 434 * definitely no active writers (although held writers may subsequently 435 * increment the count, they'll have to wait, and decrement it after 436 * seeing MNT_READONLY). 437 * 438 * It is OK to have counter incremented on one CPU and decremented on 439 * another: the sum will add up correctly. The danger would be when we 440 * sum up each counter, if we read a counter before it is incremented, 441 * but then read another CPU's count which it has been subsequently 442 * decremented from -- we would see more decrements than we should. 443 * MNT_WRITE_HOLD protects against this scenario, because 444 * mnt_want_write first increments count, then smp_mb, then spins on 445 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 446 * we're counting up here. 447 */ 448 if (mnt_get_writers(mnt) > 0) 449 ret = -EBUSY; 450 else 451 mnt->mnt_flags |= MNT_READONLY; 452 /* 453 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 454 * that become unheld will see MNT_READONLY. 455 */ 456 smp_wmb(); 457 mnt->mnt_flags &= ~MNT_WRITE_HOLD; 458 br_write_unlock(vfsmount_lock); 459 return ret; 460 } 461 462 static void __mnt_unmake_readonly(struct vfsmount *mnt) 463 { 464 br_write_lock(vfsmount_lock); 465 mnt->mnt_flags &= ~MNT_READONLY; 466 br_write_unlock(vfsmount_lock); 467 } 468 469 static void free_vfsmnt(struct vfsmount *mnt) 470 { 471 kfree(mnt->mnt_devname); 472 mnt_free_id(mnt); 473 #ifdef CONFIG_SMP 474 free_percpu(mnt->mnt_pcp); 475 #endif 476 kmem_cache_free(mnt_cache, mnt); 477 } 478 479 /* 480 * find the first or last mount at @dentry on vfsmount @mnt depending on 481 * @dir. If @dir is set return the first mount else return the last mount. 482 * vfsmount_lock must be held for read or write. 483 */ 484 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 485 int dir) 486 { 487 struct list_head *head = mount_hashtable + hash(mnt, dentry); 488 struct list_head *tmp = head; 489 struct vfsmount *p, *found = NULL; 490 491 for (;;) { 492 tmp = dir ? tmp->next : tmp->prev; 493 p = NULL; 494 if (tmp == head) 495 break; 496 p = list_entry(tmp, struct vfsmount, mnt_hash); 497 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 498 found = p; 499 break; 500 } 501 } 502 return found; 503 } 504 505 /* 506 * lookup_mnt increments the ref count before returning 507 * the vfsmount struct. 508 */ 509 struct vfsmount *lookup_mnt(struct path *path) 510 { 511 struct vfsmount *child_mnt; 512 513 br_read_lock(vfsmount_lock); 514 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) 515 mntget(child_mnt); 516 br_read_unlock(vfsmount_lock); 517 return child_mnt; 518 } 519 520 static inline int check_mnt(struct vfsmount *mnt) 521 { 522 return mnt->mnt_ns == current->nsproxy->mnt_ns; 523 } 524 525 /* 526 * vfsmount lock must be held for write 527 */ 528 static void touch_mnt_namespace(struct mnt_namespace *ns) 529 { 530 if (ns) { 531 ns->event = ++event; 532 wake_up_interruptible(&ns->poll); 533 } 534 } 535 536 /* 537 * vfsmount lock must be held for write 538 */ 539 static void __touch_mnt_namespace(struct mnt_namespace *ns) 540 { 541 if (ns && ns->event != event) { 542 ns->event = event; 543 wake_up_interruptible(&ns->poll); 544 } 545 } 546 547 /* 548 * Clear dentry's mounted state if it has no remaining mounts. 549 * vfsmount_lock must be held for write. 550 */ 551 static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry) 552 { 553 unsigned u; 554 555 for (u = 0; u < HASH_SIZE; u++) { 556 struct vfsmount *p; 557 558 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { 559 if (p->mnt_mountpoint == dentry) 560 return; 561 } 562 } 563 spin_lock(&dentry->d_lock); 564 dentry->d_flags &= ~DCACHE_MOUNTED; 565 spin_unlock(&dentry->d_lock); 566 } 567 568 /* 569 * vfsmount lock must be held for write 570 */ 571 static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 572 { 573 old_path->dentry = mnt->mnt_mountpoint; 574 old_path->mnt = mnt->mnt_parent; 575 mnt->mnt_parent = mnt; 576 mnt->mnt_mountpoint = mnt->mnt_root; 577 list_del_init(&mnt->mnt_child); 578 list_del_init(&mnt->mnt_hash); 579 dentry_reset_mounted(old_path->mnt, old_path->dentry); 580 } 581 582 /* 583 * vfsmount lock must be held for write 584 */ 585 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 586 struct vfsmount *child_mnt) 587 { 588 child_mnt->mnt_parent = mntget(mnt); 589 child_mnt->mnt_mountpoint = dget(dentry); 590 spin_lock(&dentry->d_lock); 591 dentry->d_flags |= DCACHE_MOUNTED; 592 spin_unlock(&dentry->d_lock); 593 } 594 595 /* 596 * vfsmount lock must be held for write 597 */ 598 static void attach_mnt(struct vfsmount *mnt, struct path *path) 599 { 600 mnt_set_mountpoint(path->mnt, path->dentry, mnt); 601 list_add_tail(&mnt->mnt_hash, mount_hashtable + 602 hash(path->mnt, path->dentry)); 603 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 604 } 605 606 static inline void __mnt_make_longterm(struct vfsmount *mnt) 607 { 608 #ifdef CONFIG_SMP 609 atomic_inc(&mnt->mnt_longterm); 610 #endif 611 } 612 613 /* needs vfsmount lock for write */ 614 static inline void __mnt_make_shortterm(struct vfsmount *mnt) 615 { 616 #ifdef CONFIG_SMP 617 atomic_dec(&mnt->mnt_longterm); 618 #endif 619 } 620 621 /* 622 * vfsmount lock must be held for write 623 */ 624 static void commit_tree(struct vfsmount *mnt) 625 { 626 struct vfsmount *parent = mnt->mnt_parent; 627 struct vfsmount *m; 628 LIST_HEAD(head); 629 struct mnt_namespace *n = parent->mnt_ns; 630 631 BUG_ON(parent == mnt); 632 633 list_add_tail(&head, &mnt->mnt_list); 634 list_for_each_entry(m, &head, mnt_list) { 635 m->mnt_ns = n; 636 __mnt_make_longterm(m); 637 } 638 639 list_splice(&head, n->list.prev); 640 641 list_add_tail(&mnt->mnt_hash, mount_hashtable + 642 hash(parent, mnt->mnt_mountpoint)); 643 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 644 touch_mnt_namespace(n); 645 } 646 647 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 648 { 649 struct list_head *next = p->mnt_mounts.next; 650 if (next == &p->mnt_mounts) { 651 while (1) { 652 if (p == root) 653 return NULL; 654 next = p->mnt_child.next; 655 if (next != &p->mnt_parent->mnt_mounts) 656 break; 657 p = p->mnt_parent; 658 } 659 } 660 return list_entry(next, struct vfsmount, mnt_child); 661 } 662 663 static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 664 { 665 struct list_head *prev = p->mnt_mounts.prev; 666 while (prev != &p->mnt_mounts) { 667 p = list_entry(prev, struct vfsmount, mnt_child); 668 prev = p->mnt_mounts.prev; 669 } 670 return p; 671 } 672 673 struct vfsmount * 674 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 675 { 676 struct vfsmount *mnt; 677 struct dentry *root; 678 679 if (!type) 680 return ERR_PTR(-ENODEV); 681 682 mnt = alloc_vfsmnt(name); 683 if (!mnt) 684 return ERR_PTR(-ENOMEM); 685 686 if (flags & MS_KERNMOUNT) 687 mnt->mnt_flags = MNT_INTERNAL; 688 689 root = mount_fs(type, flags, name, data); 690 if (IS_ERR(root)) { 691 free_vfsmnt(mnt); 692 return ERR_CAST(root); 693 } 694 695 mnt->mnt_root = root; 696 mnt->mnt_sb = root->d_sb; 697 mnt->mnt_mountpoint = mnt->mnt_root; 698 mnt->mnt_parent = mnt; 699 return mnt; 700 } 701 EXPORT_SYMBOL_GPL(vfs_kern_mount); 702 703 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 704 int flag) 705 { 706 struct super_block *sb = old->mnt_sb; 707 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 708 709 if (mnt) { 710 if (flag & (CL_SLAVE | CL_PRIVATE)) 711 mnt->mnt_group_id = 0; /* not a peer of original */ 712 else 713 mnt->mnt_group_id = old->mnt_group_id; 714 715 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 716 int err = mnt_alloc_group_id(mnt); 717 if (err) 718 goto out_free; 719 } 720 721 mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD; 722 atomic_inc(&sb->s_active); 723 mnt->mnt_sb = sb; 724 mnt->mnt_root = dget(root); 725 mnt->mnt_mountpoint = mnt->mnt_root; 726 mnt->mnt_parent = mnt; 727 728 if (flag & CL_SLAVE) { 729 list_add(&mnt->mnt_slave, &old->mnt_slave_list); 730 mnt->mnt_master = old; 731 CLEAR_MNT_SHARED(mnt); 732 } else if (!(flag & CL_PRIVATE)) { 733 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 734 list_add(&mnt->mnt_share, &old->mnt_share); 735 if (IS_MNT_SLAVE(old)) 736 list_add(&mnt->mnt_slave, &old->mnt_slave); 737 mnt->mnt_master = old->mnt_master; 738 } 739 if (flag & CL_MAKE_SHARED) 740 set_mnt_shared(mnt); 741 742 /* stick the duplicate mount on the same expiry list 743 * as the original if that was on one */ 744 if (flag & CL_EXPIRE) { 745 if (!list_empty(&old->mnt_expire)) 746 list_add(&mnt->mnt_expire, &old->mnt_expire); 747 } 748 } 749 return mnt; 750 751 out_free: 752 free_vfsmnt(mnt); 753 return NULL; 754 } 755 756 static inline void mntfree(struct vfsmount *mnt) 757 { 758 struct super_block *sb = mnt->mnt_sb; 759 760 /* 761 * This probably indicates that somebody messed 762 * up a mnt_want/drop_write() pair. If this 763 * happens, the filesystem was probably unable 764 * to make r/w->r/o transitions. 765 */ 766 /* 767 * The locking used to deal with mnt_count decrement provides barriers, 768 * so mnt_get_writers() below is safe. 769 */ 770 WARN_ON(mnt_get_writers(mnt)); 771 fsnotify_vfsmount_delete(mnt); 772 dput(mnt->mnt_root); 773 free_vfsmnt(mnt); 774 deactivate_super(sb); 775 } 776 777 static void mntput_no_expire(struct vfsmount *mnt) 778 { 779 put_again: 780 #ifdef CONFIG_SMP 781 br_read_lock(vfsmount_lock); 782 if (likely(atomic_read(&mnt->mnt_longterm))) { 783 mnt_dec_count(mnt); 784 br_read_unlock(vfsmount_lock); 785 return; 786 } 787 br_read_unlock(vfsmount_lock); 788 789 br_write_lock(vfsmount_lock); 790 mnt_dec_count(mnt); 791 if (mnt_get_count(mnt)) { 792 br_write_unlock(vfsmount_lock); 793 return; 794 } 795 #else 796 mnt_dec_count(mnt); 797 if (likely(mnt_get_count(mnt))) 798 return; 799 br_write_lock(vfsmount_lock); 800 #endif 801 if (unlikely(mnt->mnt_pinned)) { 802 mnt_add_count(mnt, mnt->mnt_pinned + 1); 803 mnt->mnt_pinned = 0; 804 br_write_unlock(vfsmount_lock); 805 acct_auto_close_mnt(mnt); 806 goto put_again; 807 } 808 br_write_unlock(vfsmount_lock); 809 mntfree(mnt); 810 } 811 812 void mntput(struct vfsmount *mnt) 813 { 814 if (mnt) { 815 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 816 if (unlikely(mnt->mnt_expiry_mark)) 817 mnt->mnt_expiry_mark = 0; 818 mntput_no_expire(mnt); 819 } 820 } 821 EXPORT_SYMBOL(mntput); 822 823 struct vfsmount *mntget(struct vfsmount *mnt) 824 { 825 if (mnt) 826 mnt_inc_count(mnt); 827 return mnt; 828 } 829 EXPORT_SYMBOL(mntget); 830 831 void mnt_pin(struct vfsmount *mnt) 832 { 833 br_write_lock(vfsmount_lock); 834 mnt->mnt_pinned++; 835 br_write_unlock(vfsmount_lock); 836 } 837 EXPORT_SYMBOL(mnt_pin); 838 839 void mnt_unpin(struct vfsmount *mnt) 840 { 841 br_write_lock(vfsmount_lock); 842 if (mnt->mnt_pinned) { 843 mnt_inc_count(mnt); 844 mnt->mnt_pinned--; 845 } 846 br_write_unlock(vfsmount_lock); 847 } 848 EXPORT_SYMBOL(mnt_unpin); 849 850 static inline void mangle(struct seq_file *m, const char *s) 851 { 852 seq_escape(m, s, " \t\n\\"); 853 } 854 855 /* 856 * Simple .show_options callback for filesystems which don't want to 857 * implement more complex mount option showing. 858 * 859 * See also save_mount_options(). 860 */ 861 int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 862 { 863 const char *options; 864 865 rcu_read_lock(); 866 options = rcu_dereference(mnt->mnt_sb->s_options); 867 868 if (options != NULL && options[0]) { 869 seq_putc(m, ','); 870 mangle(m, options); 871 } 872 rcu_read_unlock(); 873 874 return 0; 875 } 876 EXPORT_SYMBOL(generic_show_options); 877 878 /* 879 * If filesystem uses generic_show_options(), this function should be 880 * called from the fill_super() callback. 881 * 882 * The .remount_fs callback usually needs to be handled in a special 883 * way, to make sure, that previous options are not overwritten if the 884 * remount fails. 885 * 886 * Also note, that if the filesystem's .remount_fs function doesn't 887 * reset all options to their default value, but changes only newly 888 * given options, then the displayed options will not reflect reality 889 * any more. 890 */ 891 void save_mount_options(struct super_block *sb, char *options) 892 { 893 BUG_ON(sb->s_options); 894 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 895 } 896 EXPORT_SYMBOL(save_mount_options); 897 898 void replace_mount_options(struct super_block *sb, char *options) 899 { 900 char *old = sb->s_options; 901 rcu_assign_pointer(sb->s_options, options); 902 if (old) { 903 synchronize_rcu(); 904 kfree(old); 905 } 906 } 907 EXPORT_SYMBOL(replace_mount_options); 908 909 #ifdef CONFIG_PROC_FS 910 /* iterator */ 911 static void *m_start(struct seq_file *m, loff_t *pos) 912 { 913 struct proc_mounts *p = m->private; 914 915 down_read(&namespace_sem); 916 return seq_list_start(&p->ns->list, *pos); 917 } 918 919 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 920 { 921 struct proc_mounts *p = m->private; 922 923 return seq_list_next(v, &p->ns->list, pos); 924 } 925 926 static void m_stop(struct seq_file *m, void *v) 927 { 928 up_read(&namespace_sem); 929 } 930 931 int mnt_had_events(struct proc_mounts *p) 932 { 933 struct mnt_namespace *ns = p->ns; 934 int res = 0; 935 936 br_read_lock(vfsmount_lock); 937 if (p->event != ns->event) { 938 p->event = ns->event; 939 res = 1; 940 } 941 br_read_unlock(vfsmount_lock); 942 943 return res; 944 } 945 946 struct proc_fs_info { 947 int flag; 948 const char *str; 949 }; 950 951 static int show_sb_opts(struct seq_file *m, struct super_block *sb) 952 { 953 static const struct proc_fs_info fs_info[] = { 954 { MS_SYNCHRONOUS, ",sync" }, 955 { MS_DIRSYNC, ",dirsync" }, 956 { MS_MANDLOCK, ",mand" }, 957 { 0, NULL } 958 }; 959 const struct proc_fs_info *fs_infop; 960 961 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 962 if (sb->s_flags & fs_infop->flag) 963 seq_puts(m, fs_infop->str); 964 } 965 966 return security_sb_show_options(m, sb); 967 } 968 969 static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 970 { 971 static const struct proc_fs_info mnt_info[] = { 972 { MNT_NOSUID, ",nosuid" }, 973 { MNT_NODEV, ",nodev" }, 974 { MNT_NOEXEC, ",noexec" }, 975 { MNT_NOATIME, ",noatime" }, 976 { MNT_NODIRATIME, ",nodiratime" }, 977 { MNT_RELATIME, ",relatime" }, 978 { 0, NULL } 979 }; 980 const struct proc_fs_info *fs_infop; 981 982 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 983 if (mnt->mnt_flags & fs_infop->flag) 984 seq_puts(m, fs_infop->str); 985 } 986 } 987 988 static void show_type(struct seq_file *m, struct super_block *sb) 989 { 990 mangle(m, sb->s_type->name); 991 if (sb->s_subtype && sb->s_subtype[0]) { 992 seq_putc(m, '.'); 993 mangle(m, sb->s_subtype); 994 } 995 } 996 997 static int show_vfsmnt(struct seq_file *m, void *v) 998 { 999 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1000 int err = 0; 1001 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1002 1003 if (mnt->mnt_sb->s_op->show_devname) { 1004 err = mnt->mnt_sb->s_op->show_devname(m, mnt); 1005 if (err) 1006 goto out; 1007 } else { 1008 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 1009 } 1010 seq_putc(m, ' '); 1011 seq_path(m, &mnt_path, " \t\n\\"); 1012 seq_putc(m, ' '); 1013 show_type(m, mnt->mnt_sb); 1014 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 1015 err = show_sb_opts(m, mnt->mnt_sb); 1016 if (err) 1017 goto out; 1018 show_mnt_opts(m, mnt); 1019 if (mnt->mnt_sb->s_op->show_options) 1020 err = mnt->mnt_sb->s_op->show_options(m, mnt); 1021 seq_puts(m, " 0 0\n"); 1022 out: 1023 return err; 1024 } 1025 1026 const struct seq_operations mounts_op = { 1027 .start = m_start, 1028 .next = m_next, 1029 .stop = m_stop, 1030 .show = show_vfsmnt 1031 }; 1032 1033 static int uuid_is_nil(u8 *uuid) 1034 { 1035 int i; 1036 u8 *cp = (u8 *)uuid; 1037 1038 for (i = 0; i < 16; i++) { 1039 if (*cp++) 1040 return 0; 1041 } 1042 return 1; 1043 } 1044 1045 static int show_mountinfo(struct seq_file *m, void *v) 1046 { 1047 struct proc_mounts *p = m->private; 1048 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1049 struct super_block *sb = mnt->mnt_sb; 1050 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1051 struct path root = p->root; 1052 int err = 0; 1053 1054 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 1055 MAJOR(sb->s_dev), MINOR(sb->s_dev)); 1056 if (sb->s_op->show_path) 1057 err = sb->s_op->show_path(m, mnt); 1058 else 1059 seq_dentry(m, mnt->mnt_root, " \t\n\\"); 1060 if (err) 1061 goto out; 1062 seq_putc(m, ' '); 1063 seq_path_root(m, &mnt_path, &root, " \t\n\\"); 1064 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 1065 /* 1066 * Mountpoint is outside root, discard that one. Ugly, 1067 * but less so than trying to do that in iterator in a 1068 * race-free way (due to renames). 1069 */ 1070 return SEQ_SKIP; 1071 } 1072 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 1073 show_mnt_opts(m, mnt); 1074 1075 /* Tagged fields ("foo:X" or "bar") */ 1076 if (IS_MNT_SHARED(mnt)) 1077 seq_printf(m, " shared:%i", mnt->mnt_group_id); 1078 if (IS_MNT_SLAVE(mnt)) { 1079 int master = mnt->mnt_master->mnt_group_id; 1080 int dom = get_dominating_id(mnt, &p->root); 1081 seq_printf(m, " master:%i", master); 1082 if (dom && dom != master) 1083 seq_printf(m, " propagate_from:%i", dom); 1084 } 1085 if (IS_MNT_UNBINDABLE(mnt)) 1086 seq_puts(m, " unbindable"); 1087 1088 if (!uuid_is_nil(mnt->mnt_sb->s_uuid)) 1089 /* print the uuid */ 1090 seq_printf(m, " uuid:%pU", mnt->mnt_sb->s_uuid); 1091 1092 /* Filesystem specific data */ 1093 seq_puts(m, " - "); 1094 show_type(m, sb); 1095 seq_putc(m, ' '); 1096 if (sb->s_op->show_devname) 1097 err = sb->s_op->show_devname(m, mnt); 1098 else 1099 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 1100 if (err) 1101 goto out; 1102 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 1103 err = show_sb_opts(m, sb); 1104 if (err) 1105 goto out; 1106 if (sb->s_op->show_options) 1107 err = sb->s_op->show_options(m, mnt); 1108 seq_putc(m, '\n'); 1109 out: 1110 return err; 1111 } 1112 1113 const struct seq_operations mountinfo_op = { 1114 .start = m_start, 1115 .next = m_next, 1116 .stop = m_stop, 1117 .show = show_mountinfo, 1118 }; 1119 1120 static int show_vfsstat(struct seq_file *m, void *v) 1121 { 1122 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1123 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1124 int err = 0; 1125 1126 /* device */ 1127 if (mnt->mnt_sb->s_op->show_devname) { 1128 err = mnt->mnt_sb->s_op->show_devname(m, mnt); 1129 } else { 1130 if (mnt->mnt_devname) { 1131 seq_puts(m, "device "); 1132 mangle(m, mnt->mnt_devname); 1133 } else 1134 seq_puts(m, "no device"); 1135 } 1136 1137 /* mount point */ 1138 seq_puts(m, " mounted on "); 1139 seq_path(m, &mnt_path, " \t\n\\"); 1140 seq_putc(m, ' '); 1141 1142 /* file system type */ 1143 seq_puts(m, "with fstype "); 1144 show_type(m, mnt->mnt_sb); 1145 1146 /* optional statistics */ 1147 if (mnt->mnt_sb->s_op->show_stats) { 1148 seq_putc(m, ' '); 1149 if (!err) 1150 err = mnt->mnt_sb->s_op->show_stats(m, mnt); 1151 } 1152 1153 seq_putc(m, '\n'); 1154 return err; 1155 } 1156 1157 const struct seq_operations mountstats_op = { 1158 .start = m_start, 1159 .next = m_next, 1160 .stop = m_stop, 1161 .show = show_vfsstat, 1162 }; 1163 #endif /* CONFIG_PROC_FS */ 1164 1165 /** 1166 * may_umount_tree - check if a mount tree is busy 1167 * @mnt: root of mount tree 1168 * 1169 * This is called to check if a tree of mounts has any 1170 * open files, pwds, chroots or sub mounts that are 1171 * busy. 1172 */ 1173 int may_umount_tree(struct vfsmount *mnt) 1174 { 1175 int actual_refs = 0; 1176 int minimum_refs = 0; 1177 struct vfsmount *p; 1178 1179 /* write lock needed for mnt_get_count */ 1180 br_write_lock(vfsmount_lock); 1181 for (p = mnt; p; p = next_mnt(p, mnt)) { 1182 actual_refs += mnt_get_count(p); 1183 minimum_refs += 2; 1184 } 1185 br_write_unlock(vfsmount_lock); 1186 1187 if (actual_refs > minimum_refs) 1188 return 0; 1189 1190 return 1; 1191 } 1192 1193 EXPORT_SYMBOL(may_umount_tree); 1194 1195 /** 1196 * may_umount - check if a mount point is busy 1197 * @mnt: root of mount 1198 * 1199 * This is called to check if a mount point has any 1200 * open files, pwds, chroots or sub mounts. If the 1201 * mount has sub mounts this will return busy 1202 * regardless of whether the sub mounts are busy. 1203 * 1204 * Doesn't take quota and stuff into account. IOW, in some cases it will 1205 * give false negatives. The main reason why it's here is that we need 1206 * a non-destructive way to look for easily umountable filesystems. 1207 */ 1208 int may_umount(struct vfsmount *mnt) 1209 { 1210 int ret = 1; 1211 down_read(&namespace_sem); 1212 br_write_lock(vfsmount_lock); 1213 if (propagate_mount_busy(mnt, 2)) 1214 ret = 0; 1215 br_write_unlock(vfsmount_lock); 1216 up_read(&namespace_sem); 1217 return ret; 1218 } 1219 1220 EXPORT_SYMBOL(may_umount); 1221 1222 void release_mounts(struct list_head *head) 1223 { 1224 struct vfsmount *mnt; 1225 while (!list_empty(head)) { 1226 mnt = list_first_entry(head, struct vfsmount, mnt_hash); 1227 list_del_init(&mnt->mnt_hash); 1228 if (mnt->mnt_parent != mnt) { 1229 struct dentry *dentry; 1230 struct vfsmount *m; 1231 1232 br_write_lock(vfsmount_lock); 1233 dentry = mnt->mnt_mountpoint; 1234 m = mnt->mnt_parent; 1235 mnt->mnt_mountpoint = mnt->mnt_root; 1236 mnt->mnt_parent = mnt; 1237 m->mnt_ghosts--; 1238 br_write_unlock(vfsmount_lock); 1239 dput(dentry); 1240 mntput(m); 1241 } 1242 mntput(mnt); 1243 } 1244 } 1245 1246 /* 1247 * vfsmount lock must be held for write 1248 * namespace_sem must be held for write 1249 */ 1250 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 1251 { 1252 LIST_HEAD(tmp_list); 1253 struct vfsmount *p; 1254 1255 for (p = mnt; p; p = next_mnt(p, mnt)) 1256 list_move(&p->mnt_hash, &tmp_list); 1257 1258 if (propagate) 1259 propagate_umount(&tmp_list); 1260 1261 list_for_each_entry(p, &tmp_list, mnt_hash) { 1262 list_del_init(&p->mnt_expire); 1263 list_del_init(&p->mnt_list); 1264 __touch_mnt_namespace(p->mnt_ns); 1265 p->mnt_ns = NULL; 1266 __mnt_make_shortterm(p); 1267 list_del_init(&p->mnt_child); 1268 if (p->mnt_parent != p) { 1269 p->mnt_parent->mnt_ghosts++; 1270 dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint); 1271 } 1272 change_mnt_propagation(p, MS_PRIVATE); 1273 } 1274 list_splice(&tmp_list, kill); 1275 } 1276 1277 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1278 1279 static int do_umount(struct vfsmount *mnt, int flags) 1280 { 1281 struct super_block *sb = mnt->mnt_sb; 1282 int retval; 1283 LIST_HEAD(umount_list); 1284 1285 retval = security_sb_umount(mnt, flags); 1286 if (retval) 1287 return retval; 1288 1289 /* 1290 * Allow userspace to request a mountpoint be expired rather than 1291 * unmounting unconditionally. Unmount only happens if: 1292 * (1) the mark is already set (the mark is cleared by mntput()) 1293 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 1294 */ 1295 if (flags & MNT_EXPIRE) { 1296 if (mnt == current->fs->root.mnt || 1297 flags & (MNT_FORCE | MNT_DETACH)) 1298 return -EINVAL; 1299 1300 /* 1301 * probably don't strictly need the lock here if we examined 1302 * all race cases, but it's a slowpath. 1303 */ 1304 br_write_lock(vfsmount_lock); 1305 if (mnt_get_count(mnt) != 2) { 1306 br_write_unlock(vfsmount_lock); 1307 return -EBUSY; 1308 } 1309 br_write_unlock(vfsmount_lock); 1310 1311 if (!xchg(&mnt->mnt_expiry_mark, 1)) 1312 return -EAGAIN; 1313 } 1314 1315 /* 1316 * If we may have to abort operations to get out of this 1317 * mount, and they will themselves hold resources we must 1318 * allow the fs to do things. In the Unix tradition of 1319 * 'Gee thats tricky lets do it in userspace' the umount_begin 1320 * might fail to complete on the first run through as other tasks 1321 * must return, and the like. Thats for the mount program to worry 1322 * about for the moment. 1323 */ 1324 1325 if (flags & MNT_FORCE && sb->s_op->umount_begin) { 1326 sb->s_op->umount_begin(sb); 1327 } 1328 1329 /* 1330 * No sense to grab the lock for this test, but test itself looks 1331 * somewhat bogus. Suggestions for better replacement? 1332 * Ho-hum... In principle, we might treat that as umount + switch 1333 * to rootfs. GC would eventually take care of the old vfsmount. 1334 * Actually it makes sense, especially if rootfs would contain a 1335 * /reboot - static binary that would close all descriptors and 1336 * call reboot(9). Then init(8) could umount root and exec /reboot. 1337 */ 1338 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 1339 /* 1340 * Special case for "unmounting" root ... 1341 * we just try to remount it readonly. 1342 */ 1343 down_write(&sb->s_umount); 1344 if (!(sb->s_flags & MS_RDONLY)) 1345 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 1346 up_write(&sb->s_umount); 1347 return retval; 1348 } 1349 1350 down_write(&namespace_sem); 1351 br_write_lock(vfsmount_lock); 1352 event++; 1353 1354 if (!(flags & MNT_DETACH)) 1355 shrink_submounts(mnt, &umount_list); 1356 1357 retval = -EBUSY; 1358 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 1359 if (!list_empty(&mnt->mnt_list)) 1360 umount_tree(mnt, 1, &umount_list); 1361 retval = 0; 1362 } 1363 br_write_unlock(vfsmount_lock); 1364 up_write(&namespace_sem); 1365 release_mounts(&umount_list); 1366 return retval; 1367 } 1368 1369 /* 1370 * Now umount can handle mount points as well as block devices. 1371 * This is important for filesystems which use unnamed block devices. 1372 * 1373 * We now support a flag for forced unmount like the other 'big iron' 1374 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 1375 */ 1376 1377 SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 1378 { 1379 struct path path; 1380 int retval; 1381 int lookup_flags = 0; 1382 1383 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1384 return -EINVAL; 1385 1386 if (!(flags & UMOUNT_NOFOLLOW)) 1387 lookup_flags |= LOOKUP_FOLLOW; 1388 1389 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 1390 if (retval) 1391 goto out; 1392 retval = -EINVAL; 1393 if (path.dentry != path.mnt->mnt_root) 1394 goto dput_and_out; 1395 if (!check_mnt(path.mnt)) 1396 goto dput_and_out; 1397 1398 retval = -EPERM; 1399 if (!capable(CAP_SYS_ADMIN)) 1400 goto dput_and_out; 1401 1402 retval = do_umount(path.mnt, flags); 1403 dput_and_out: 1404 /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 1405 dput(path.dentry); 1406 mntput_no_expire(path.mnt); 1407 out: 1408 return retval; 1409 } 1410 1411 #ifdef __ARCH_WANT_SYS_OLDUMOUNT 1412 1413 /* 1414 * The 2.0 compatible umount. No flags. 1415 */ 1416 SYSCALL_DEFINE1(oldumount, char __user *, name) 1417 { 1418 return sys_umount(name, 0); 1419 } 1420 1421 #endif 1422 1423 static int mount_is_safe(struct path *path) 1424 { 1425 if (capable(CAP_SYS_ADMIN)) 1426 return 0; 1427 return -EPERM; 1428 #ifdef notyet 1429 if (S_ISLNK(path->dentry->d_inode->i_mode)) 1430 return -EPERM; 1431 if (path->dentry->d_inode->i_mode & S_ISVTX) { 1432 if (current_uid() != path->dentry->d_inode->i_uid) 1433 return -EPERM; 1434 } 1435 if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 1436 return -EPERM; 1437 return 0; 1438 #endif 1439 } 1440 1441 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 1442 int flag) 1443 { 1444 struct vfsmount *res, *p, *q, *r, *s; 1445 struct path path; 1446 1447 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 1448 return NULL; 1449 1450 res = q = clone_mnt(mnt, dentry, flag); 1451 if (!q) 1452 goto Enomem; 1453 q->mnt_mountpoint = mnt->mnt_mountpoint; 1454 1455 p = mnt; 1456 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1457 if (!is_subdir(r->mnt_mountpoint, dentry)) 1458 continue; 1459 1460 for (s = r; s; s = next_mnt(s, r)) { 1461 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 1462 s = skip_mnt_tree(s); 1463 continue; 1464 } 1465 while (p != s->mnt_parent) { 1466 p = p->mnt_parent; 1467 q = q->mnt_parent; 1468 } 1469 p = s; 1470 path.mnt = q; 1471 path.dentry = p->mnt_mountpoint; 1472 q = clone_mnt(p, p->mnt_root, flag); 1473 if (!q) 1474 goto Enomem; 1475 br_write_lock(vfsmount_lock); 1476 list_add_tail(&q->mnt_list, &res->mnt_list); 1477 attach_mnt(q, &path); 1478 br_write_unlock(vfsmount_lock); 1479 } 1480 } 1481 return res; 1482 Enomem: 1483 if (res) { 1484 LIST_HEAD(umount_list); 1485 br_write_lock(vfsmount_lock); 1486 umount_tree(res, 0, &umount_list); 1487 br_write_unlock(vfsmount_lock); 1488 release_mounts(&umount_list); 1489 } 1490 return NULL; 1491 } 1492 1493 struct vfsmount *collect_mounts(struct path *path) 1494 { 1495 struct vfsmount *tree; 1496 down_write(&namespace_sem); 1497 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 1498 up_write(&namespace_sem); 1499 return tree; 1500 } 1501 1502 void drop_collected_mounts(struct vfsmount *mnt) 1503 { 1504 LIST_HEAD(umount_list); 1505 down_write(&namespace_sem); 1506 br_write_lock(vfsmount_lock); 1507 umount_tree(mnt, 0, &umount_list); 1508 br_write_unlock(vfsmount_lock); 1509 up_write(&namespace_sem); 1510 release_mounts(&umount_list); 1511 } 1512 1513 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 1514 struct vfsmount *root) 1515 { 1516 struct vfsmount *mnt; 1517 int res = f(root, arg); 1518 if (res) 1519 return res; 1520 list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 1521 res = f(mnt, arg); 1522 if (res) 1523 return res; 1524 } 1525 return 0; 1526 } 1527 1528 static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1529 { 1530 struct vfsmount *p; 1531 1532 for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1533 if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1534 mnt_release_group_id(p); 1535 } 1536 } 1537 1538 static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1539 { 1540 struct vfsmount *p; 1541 1542 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1543 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1544 int err = mnt_alloc_group_id(p); 1545 if (err) { 1546 cleanup_group_ids(mnt, p); 1547 return err; 1548 } 1549 } 1550 } 1551 1552 return 0; 1553 } 1554 1555 /* 1556 * @source_mnt : mount tree to be attached 1557 * @nd : place the mount tree @source_mnt is attached 1558 * @parent_nd : if non-null, detach the source_mnt from its parent and 1559 * store the parent mount and mountpoint dentry. 1560 * (done when source_mnt is moved) 1561 * 1562 * NOTE: in the table below explains the semantics when a source mount 1563 * of a given type is attached to a destination mount of a given type. 1564 * --------------------------------------------------------------------------- 1565 * | BIND MOUNT OPERATION | 1566 * |************************************************************************** 1567 * | source-->| shared | private | slave | unbindable | 1568 * | dest | | | | | 1569 * | | | | | | | 1570 * | v | | | | | 1571 * |************************************************************************** 1572 * | shared | shared (++) | shared (+) | shared(+++)| invalid | 1573 * | | | | | | 1574 * |non-shared| shared (+) | private | slave (*) | invalid | 1575 * *************************************************************************** 1576 * A bind operation clones the source mount and mounts the clone on the 1577 * destination mount. 1578 * 1579 * (++) the cloned mount is propagated to all the mounts in the propagation 1580 * tree of the destination mount and the cloned mount is added to 1581 * the peer group of the source mount. 1582 * (+) the cloned mount is created under the destination mount and is marked 1583 * as shared. The cloned mount is added to the peer group of the source 1584 * mount. 1585 * (+++) the mount is propagated to all the mounts in the propagation tree 1586 * of the destination mount and the cloned mount is made slave 1587 * of the same master as that of the source mount. The cloned mount 1588 * is marked as 'shared and slave'. 1589 * (*) the cloned mount is made a slave of the same master as that of the 1590 * source mount. 1591 * 1592 * --------------------------------------------------------------------------- 1593 * | MOVE MOUNT OPERATION | 1594 * |************************************************************************** 1595 * | source-->| shared | private | slave | unbindable | 1596 * | dest | | | | | 1597 * | | | | | | | 1598 * | v | | | | | 1599 * |************************************************************************** 1600 * | shared | shared (+) | shared (+) | shared(+++) | invalid | 1601 * | | | | | | 1602 * |non-shared| shared (+*) | private | slave (*) | unbindable | 1603 * *************************************************************************** 1604 * 1605 * (+) the mount is moved to the destination. And is then propagated to 1606 * all the mounts in the propagation tree of the destination mount. 1607 * (+*) the mount is moved to the destination. 1608 * (+++) the mount is moved to the destination and is then propagated to 1609 * all the mounts belonging to the destination mount's propagation tree. 1610 * the mount is marked as 'shared and slave'. 1611 * (*) the mount continues to be a slave at the new location. 1612 * 1613 * if the source mount is a tree, the operations explained above is 1614 * applied to each mount in the tree. 1615 * Must be called without spinlocks held, since this function can sleep 1616 * in allocations. 1617 */ 1618 static int attach_recursive_mnt(struct vfsmount *source_mnt, 1619 struct path *path, struct path *parent_path) 1620 { 1621 LIST_HEAD(tree_list); 1622 struct vfsmount *dest_mnt = path->mnt; 1623 struct dentry *dest_dentry = path->dentry; 1624 struct vfsmount *child, *p; 1625 int err; 1626 1627 if (IS_MNT_SHARED(dest_mnt)) { 1628 err = invent_group_ids(source_mnt, true); 1629 if (err) 1630 goto out; 1631 } 1632 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1633 if (err) 1634 goto out_cleanup_ids; 1635 1636 br_write_lock(vfsmount_lock); 1637 1638 if (IS_MNT_SHARED(dest_mnt)) { 1639 for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1640 set_mnt_shared(p); 1641 } 1642 if (parent_path) { 1643 detach_mnt(source_mnt, parent_path); 1644 attach_mnt(source_mnt, path); 1645 touch_mnt_namespace(parent_path->mnt->mnt_ns); 1646 } else { 1647 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1648 commit_tree(source_mnt); 1649 } 1650 1651 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1652 list_del_init(&child->mnt_hash); 1653 commit_tree(child); 1654 } 1655 br_write_unlock(vfsmount_lock); 1656 1657 return 0; 1658 1659 out_cleanup_ids: 1660 if (IS_MNT_SHARED(dest_mnt)) 1661 cleanup_group_ids(source_mnt, NULL); 1662 out: 1663 return err; 1664 } 1665 1666 static int lock_mount(struct path *path) 1667 { 1668 struct vfsmount *mnt; 1669 retry: 1670 mutex_lock(&path->dentry->d_inode->i_mutex); 1671 if (unlikely(cant_mount(path->dentry))) { 1672 mutex_unlock(&path->dentry->d_inode->i_mutex); 1673 return -ENOENT; 1674 } 1675 down_write(&namespace_sem); 1676 mnt = lookup_mnt(path); 1677 if (likely(!mnt)) 1678 return 0; 1679 up_write(&namespace_sem); 1680 mutex_unlock(&path->dentry->d_inode->i_mutex); 1681 path_put(path); 1682 path->mnt = mnt; 1683 path->dentry = dget(mnt->mnt_root); 1684 goto retry; 1685 } 1686 1687 static void unlock_mount(struct path *path) 1688 { 1689 up_write(&namespace_sem); 1690 mutex_unlock(&path->dentry->d_inode->i_mutex); 1691 } 1692 1693 static int graft_tree(struct vfsmount *mnt, struct path *path) 1694 { 1695 if (mnt->mnt_sb->s_flags & MS_NOUSER) 1696 return -EINVAL; 1697 1698 if (S_ISDIR(path->dentry->d_inode->i_mode) != 1699 S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 1700 return -ENOTDIR; 1701 1702 if (d_unlinked(path->dentry)) 1703 return -ENOENT; 1704 1705 return attach_recursive_mnt(mnt, path, NULL); 1706 } 1707 1708 /* 1709 * Sanity check the flags to change_mnt_propagation. 1710 */ 1711 1712 static int flags_to_propagation_type(int flags) 1713 { 1714 int type = flags & ~MS_REC; 1715 1716 /* Fail if any non-propagation flags are set */ 1717 if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 1718 return 0; 1719 /* Only one propagation flag should be set */ 1720 if (!is_power_of_2(type)) 1721 return 0; 1722 return type; 1723 } 1724 1725 /* 1726 * recursively change the type of the mountpoint. 1727 */ 1728 static int do_change_type(struct path *path, int flag) 1729 { 1730 struct vfsmount *m, *mnt = path->mnt; 1731 int recurse = flag & MS_REC; 1732 int type; 1733 int err = 0; 1734 1735 if (!capable(CAP_SYS_ADMIN)) 1736 return -EPERM; 1737 1738 if (path->dentry != path->mnt->mnt_root) 1739 return -EINVAL; 1740 1741 type = flags_to_propagation_type(flag); 1742 if (!type) 1743 return -EINVAL; 1744 1745 down_write(&namespace_sem); 1746 if (type == MS_SHARED) { 1747 err = invent_group_ids(mnt, recurse); 1748 if (err) 1749 goto out_unlock; 1750 } 1751 1752 br_write_lock(vfsmount_lock); 1753 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 1754 change_mnt_propagation(m, type); 1755 br_write_unlock(vfsmount_lock); 1756 1757 out_unlock: 1758 up_write(&namespace_sem); 1759 return err; 1760 } 1761 1762 /* 1763 * do loopback mount. 1764 */ 1765 static int do_loopback(struct path *path, char *old_name, 1766 int recurse) 1767 { 1768 LIST_HEAD(umount_list); 1769 struct path old_path; 1770 struct vfsmount *mnt = NULL; 1771 int err = mount_is_safe(path); 1772 if (err) 1773 return err; 1774 if (!old_name || !*old_name) 1775 return -EINVAL; 1776 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 1777 if (err) 1778 return err; 1779 1780 err = lock_mount(path); 1781 if (err) 1782 goto out; 1783 1784 err = -EINVAL; 1785 if (IS_MNT_UNBINDABLE(old_path.mnt)) 1786 goto out2; 1787 1788 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1789 goto out2; 1790 1791 err = -ENOMEM; 1792 if (recurse) 1793 mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 1794 else 1795 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 1796 1797 if (!mnt) 1798 goto out2; 1799 1800 err = graft_tree(mnt, path); 1801 if (err) { 1802 br_write_lock(vfsmount_lock); 1803 umount_tree(mnt, 0, &umount_list); 1804 br_write_unlock(vfsmount_lock); 1805 } 1806 out2: 1807 unlock_mount(path); 1808 release_mounts(&umount_list); 1809 out: 1810 path_put(&old_path); 1811 return err; 1812 } 1813 1814 static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 1815 { 1816 int error = 0; 1817 int readonly_request = 0; 1818 1819 if (ms_flags & MS_RDONLY) 1820 readonly_request = 1; 1821 if (readonly_request == __mnt_is_readonly(mnt)) 1822 return 0; 1823 1824 if (readonly_request) 1825 error = mnt_make_readonly(mnt); 1826 else 1827 __mnt_unmake_readonly(mnt); 1828 return error; 1829 } 1830 1831 /* 1832 * change filesystem flags. dir should be a physical root of filesystem. 1833 * If you've mounted a non-root directory somewhere and want to do remount 1834 * on it - tough luck. 1835 */ 1836 static int do_remount(struct path *path, int flags, int mnt_flags, 1837 void *data) 1838 { 1839 int err; 1840 struct super_block *sb = path->mnt->mnt_sb; 1841 1842 if (!capable(CAP_SYS_ADMIN)) 1843 return -EPERM; 1844 1845 if (!check_mnt(path->mnt)) 1846 return -EINVAL; 1847 1848 if (path->dentry != path->mnt->mnt_root) 1849 return -EINVAL; 1850 1851 err = security_sb_remount(sb, data); 1852 if (err) 1853 return err; 1854 1855 down_write(&sb->s_umount); 1856 if (flags & MS_BIND) 1857 err = change_mount_flags(path->mnt, flags); 1858 else 1859 err = do_remount_sb(sb, flags, data, 0); 1860 if (!err) { 1861 br_write_lock(vfsmount_lock); 1862 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 1863 path->mnt->mnt_flags = mnt_flags; 1864 br_write_unlock(vfsmount_lock); 1865 } 1866 up_write(&sb->s_umount); 1867 if (!err) { 1868 br_write_lock(vfsmount_lock); 1869 touch_mnt_namespace(path->mnt->mnt_ns); 1870 br_write_unlock(vfsmount_lock); 1871 } 1872 return err; 1873 } 1874 1875 static inline int tree_contains_unbindable(struct vfsmount *mnt) 1876 { 1877 struct vfsmount *p; 1878 for (p = mnt; p; p = next_mnt(p, mnt)) { 1879 if (IS_MNT_UNBINDABLE(p)) 1880 return 1; 1881 } 1882 return 0; 1883 } 1884 1885 static int do_move_mount(struct path *path, char *old_name) 1886 { 1887 struct path old_path, parent_path; 1888 struct vfsmount *p; 1889 int err = 0; 1890 if (!capable(CAP_SYS_ADMIN)) 1891 return -EPERM; 1892 if (!old_name || !*old_name) 1893 return -EINVAL; 1894 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 1895 if (err) 1896 return err; 1897 1898 err = lock_mount(path); 1899 if (err < 0) 1900 goto out; 1901 1902 err = -EINVAL; 1903 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1904 goto out1; 1905 1906 if (d_unlinked(path->dentry)) 1907 goto out1; 1908 1909 err = -EINVAL; 1910 if (old_path.dentry != old_path.mnt->mnt_root) 1911 goto out1; 1912 1913 if (old_path.mnt == old_path.mnt->mnt_parent) 1914 goto out1; 1915 1916 if (S_ISDIR(path->dentry->d_inode->i_mode) != 1917 S_ISDIR(old_path.dentry->d_inode->i_mode)) 1918 goto out1; 1919 /* 1920 * Don't move a mount residing in a shared parent. 1921 */ 1922 if (old_path.mnt->mnt_parent && 1923 IS_MNT_SHARED(old_path.mnt->mnt_parent)) 1924 goto out1; 1925 /* 1926 * Don't move a mount tree containing unbindable mounts to a destination 1927 * mount which is shared. 1928 */ 1929 if (IS_MNT_SHARED(path->mnt) && 1930 tree_contains_unbindable(old_path.mnt)) 1931 goto out1; 1932 err = -ELOOP; 1933 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 1934 if (p == old_path.mnt) 1935 goto out1; 1936 1937 err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 1938 if (err) 1939 goto out1; 1940 1941 /* if the mount is moved, it should no longer be expire 1942 * automatically */ 1943 list_del_init(&old_path.mnt->mnt_expire); 1944 out1: 1945 unlock_mount(path); 1946 out: 1947 if (!err) 1948 path_put(&parent_path); 1949 path_put(&old_path); 1950 return err; 1951 } 1952 1953 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 1954 { 1955 int err; 1956 const char *subtype = strchr(fstype, '.'); 1957 if (subtype) { 1958 subtype++; 1959 err = -EINVAL; 1960 if (!subtype[0]) 1961 goto err; 1962 } else 1963 subtype = ""; 1964 1965 mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 1966 err = -ENOMEM; 1967 if (!mnt->mnt_sb->s_subtype) 1968 goto err; 1969 return mnt; 1970 1971 err: 1972 mntput(mnt); 1973 return ERR_PTR(err); 1974 } 1975 1976 struct vfsmount * 1977 do_kern_mount(const char *fstype, int flags, const char *name, void *data) 1978 { 1979 struct file_system_type *type = get_fs_type(fstype); 1980 struct vfsmount *mnt; 1981 if (!type) 1982 return ERR_PTR(-ENODEV); 1983 mnt = vfs_kern_mount(type, flags, name, data); 1984 if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 1985 !mnt->mnt_sb->s_subtype) 1986 mnt = fs_set_subtype(mnt, fstype); 1987 put_filesystem(type); 1988 return mnt; 1989 } 1990 EXPORT_SYMBOL_GPL(do_kern_mount); 1991 1992 /* 1993 * add a mount into a namespace's mount tree 1994 */ 1995 static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags) 1996 { 1997 int err; 1998 1999 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 2000 2001 err = lock_mount(path); 2002 if (err) 2003 return err; 2004 2005 err = -EINVAL; 2006 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 2007 goto unlock; 2008 2009 /* Refuse the same filesystem on the same mount point */ 2010 err = -EBUSY; 2011 if (path->mnt->mnt_sb == newmnt->mnt_sb && 2012 path->mnt->mnt_root == path->dentry) 2013 goto unlock; 2014 2015 err = -EINVAL; 2016 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 2017 goto unlock; 2018 2019 newmnt->mnt_flags = mnt_flags; 2020 err = graft_tree(newmnt, path); 2021 2022 unlock: 2023 unlock_mount(path); 2024 return err; 2025 } 2026 2027 /* 2028 * create a new mount for userspace and request it to be added into the 2029 * namespace's tree 2030 */ 2031 static int do_new_mount(struct path *path, char *type, int flags, 2032 int mnt_flags, char *name, void *data) 2033 { 2034 struct vfsmount *mnt; 2035 int err; 2036 2037 if (!type) 2038 return -EINVAL; 2039 2040 /* we need capabilities... */ 2041 if (!capable(CAP_SYS_ADMIN)) 2042 return -EPERM; 2043 2044 mnt = do_kern_mount(type, flags, name, data); 2045 if (IS_ERR(mnt)) 2046 return PTR_ERR(mnt); 2047 2048 err = do_add_mount(mnt, path, mnt_flags); 2049 if (err) 2050 mntput(mnt); 2051 return err; 2052 } 2053 2054 int finish_automount(struct vfsmount *m, struct path *path) 2055 { 2056 int err; 2057 /* The new mount record should have at least 2 refs to prevent it being 2058 * expired before we get a chance to add it 2059 */ 2060 BUG_ON(mnt_get_count(m) < 2); 2061 2062 if (m->mnt_sb == path->mnt->mnt_sb && 2063 m->mnt_root == path->dentry) { 2064 err = -ELOOP; 2065 goto fail; 2066 } 2067 2068 err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2069 if (!err) 2070 return 0; 2071 fail: 2072 /* remove m from any expiration list it may be on */ 2073 if (!list_empty(&m->mnt_expire)) { 2074 down_write(&namespace_sem); 2075 br_write_lock(vfsmount_lock); 2076 list_del_init(&m->mnt_expire); 2077 br_write_unlock(vfsmount_lock); 2078 up_write(&namespace_sem); 2079 } 2080 mntput(m); 2081 mntput(m); 2082 return err; 2083 } 2084 2085 /** 2086 * mnt_set_expiry - Put a mount on an expiration list 2087 * @mnt: The mount to list. 2088 * @expiry_list: The list to add the mount to. 2089 */ 2090 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2091 { 2092 down_write(&namespace_sem); 2093 br_write_lock(vfsmount_lock); 2094 2095 list_add_tail(&mnt->mnt_expire, expiry_list); 2096 2097 br_write_unlock(vfsmount_lock); 2098 up_write(&namespace_sem); 2099 } 2100 EXPORT_SYMBOL(mnt_set_expiry); 2101 2102 /* 2103 * process a list of expirable mountpoints with the intent of discarding any 2104 * mountpoints that aren't in use and haven't been touched since last we came 2105 * here 2106 */ 2107 void mark_mounts_for_expiry(struct list_head *mounts) 2108 { 2109 struct vfsmount *mnt, *next; 2110 LIST_HEAD(graveyard); 2111 LIST_HEAD(umounts); 2112 2113 if (list_empty(mounts)) 2114 return; 2115 2116 down_write(&namespace_sem); 2117 br_write_lock(vfsmount_lock); 2118 2119 /* extract from the expiration list every vfsmount that matches the 2120 * following criteria: 2121 * - only referenced by its parent vfsmount 2122 * - still marked for expiry (marked on the last call here; marks are 2123 * cleared by mntput()) 2124 */ 2125 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2126 if (!xchg(&mnt->mnt_expiry_mark, 1) || 2127 propagate_mount_busy(mnt, 1)) 2128 continue; 2129 list_move(&mnt->mnt_expire, &graveyard); 2130 } 2131 while (!list_empty(&graveyard)) { 2132 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 2133 touch_mnt_namespace(mnt->mnt_ns); 2134 umount_tree(mnt, 1, &umounts); 2135 } 2136 br_write_unlock(vfsmount_lock); 2137 up_write(&namespace_sem); 2138 2139 release_mounts(&umounts); 2140 } 2141 2142 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 2143 2144 /* 2145 * Ripoff of 'select_parent()' 2146 * 2147 * search the list of submounts for a given mountpoint, and move any 2148 * shrinkable submounts to the 'graveyard' list. 2149 */ 2150 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 2151 { 2152 struct vfsmount *this_parent = parent; 2153 struct list_head *next; 2154 int found = 0; 2155 2156 repeat: 2157 next = this_parent->mnt_mounts.next; 2158 resume: 2159 while (next != &this_parent->mnt_mounts) { 2160 struct list_head *tmp = next; 2161 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 2162 2163 next = tmp->next; 2164 if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 2165 continue; 2166 /* 2167 * Descend a level if the d_mounts list is non-empty. 2168 */ 2169 if (!list_empty(&mnt->mnt_mounts)) { 2170 this_parent = mnt; 2171 goto repeat; 2172 } 2173 2174 if (!propagate_mount_busy(mnt, 1)) { 2175 list_move_tail(&mnt->mnt_expire, graveyard); 2176 found++; 2177 } 2178 } 2179 /* 2180 * All done at this level ... ascend and resume the search 2181 */ 2182 if (this_parent != parent) { 2183 next = this_parent->mnt_child.next; 2184 this_parent = this_parent->mnt_parent; 2185 goto resume; 2186 } 2187 return found; 2188 } 2189 2190 /* 2191 * process a list of expirable mountpoints with the intent of discarding any 2192 * submounts of a specific parent mountpoint 2193 * 2194 * vfsmount_lock must be held for write 2195 */ 2196 static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 2197 { 2198 LIST_HEAD(graveyard); 2199 struct vfsmount *m; 2200 2201 /* extract submounts of 'mountpoint' from the expiration list */ 2202 while (select_submounts(mnt, &graveyard)) { 2203 while (!list_empty(&graveyard)) { 2204 m = list_first_entry(&graveyard, struct vfsmount, 2205 mnt_expire); 2206 touch_mnt_namespace(m->mnt_ns); 2207 umount_tree(m, 1, umounts); 2208 } 2209 } 2210 } 2211 2212 /* 2213 * Some copy_from_user() implementations do not return the exact number of 2214 * bytes remaining to copy on a fault. But copy_mount_options() requires that. 2215 * Note that this function differs from copy_from_user() in that it will oops 2216 * on bad values of `to', rather than returning a short copy. 2217 */ 2218 static long exact_copy_from_user(void *to, const void __user * from, 2219 unsigned long n) 2220 { 2221 char *t = to; 2222 const char __user *f = from; 2223 char c; 2224 2225 if (!access_ok(VERIFY_READ, from, n)) 2226 return n; 2227 2228 while (n) { 2229 if (__get_user(c, f)) { 2230 memset(t, 0, n); 2231 break; 2232 } 2233 *t++ = c; 2234 f++; 2235 n--; 2236 } 2237 return n; 2238 } 2239 2240 int copy_mount_options(const void __user * data, unsigned long *where) 2241 { 2242 int i; 2243 unsigned long page; 2244 unsigned long size; 2245 2246 *where = 0; 2247 if (!data) 2248 return 0; 2249 2250 if (!(page = __get_free_page(GFP_KERNEL))) 2251 return -ENOMEM; 2252 2253 /* We only care that *some* data at the address the user 2254 * gave us is valid. Just in case, we'll zero 2255 * the remainder of the page. 2256 */ 2257 /* copy_from_user cannot cross TASK_SIZE ! */ 2258 size = TASK_SIZE - (unsigned long)data; 2259 if (size > PAGE_SIZE) 2260 size = PAGE_SIZE; 2261 2262 i = size - exact_copy_from_user((void *)page, data, size); 2263 if (!i) { 2264 free_page(page); 2265 return -EFAULT; 2266 } 2267 if (i != PAGE_SIZE) 2268 memset((char *)page + i, 0, PAGE_SIZE - i); 2269 *where = page; 2270 return 0; 2271 } 2272 2273 int copy_mount_string(const void __user *data, char **where) 2274 { 2275 char *tmp; 2276 2277 if (!data) { 2278 *where = NULL; 2279 return 0; 2280 } 2281 2282 tmp = strndup_user(data, PAGE_SIZE); 2283 if (IS_ERR(tmp)) 2284 return PTR_ERR(tmp); 2285 2286 *where = tmp; 2287 return 0; 2288 } 2289 2290 /* 2291 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 2292 * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 2293 * 2294 * data is a (void *) that can point to any structure up to 2295 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 2296 * information (or be NULL). 2297 * 2298 * Pre-0.97 versions of mount() didn't have a flags word. 2299 * When the flags word was introduced its top half was required 2300 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 2301 * Therefore, if this magic number is present, it carries no information 2302 * and must be discarded. 2303 */ 2304 long do_mount(char *dev_name, char *dir_name, char *type_page, 2305 unsigned long flags, void *data_page) 2306 { 2307 struct path path; 2308 int retval = 0; 2309 int mnt_flags = 0; 2310 2311 /* Discard magic */ 2312 if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 2313 flags &= ~MS_MGC_MSK; 2314 2315 /* Basic sanity checks */ 2316 2317 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 2318 return -EINVAL; 2319 2320 if (data_page) 2321 ((char *)data_page)[PAGE_SIZE - 1] = 0; 2322 2323 /* ... and get the mountpoint */ 2324 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2325 if (retval) 2326 return retval; 2327 2328 retval = security_sb_mount(dev_name, &path, 2329 type_page, flags, data_page); 2330 if (retval) 2331 goto dput_out; 2332 2333 /* Default to relatime unless overriden */ 2334 if (!(flags & MS_NOATIME)) 2335 mnt_flags |= MNT_RELATIME; 2336 2337 /* Separate the per-mountpoint flags */ 2338 if (flags & MS_NOSUID) 2339 mnt_flags |= MNT_NOSUID; 2340 if (flags & MS_NODEV) 2341 mnt_flags |= MNT_NODEV; 2342 if (flags & MS_NOEXEC) 2343 mnt_flags |= MNT_NOEXEC; 2344 if (flags & MS_NOATIME) 2345 mnt_flags |= MNT_NOATIME; 2346 if (flags & MS_NODIRATIME) 2347 mnt_flags |= MNT_NODIRATIME; 2348 if (flags & MS_STRICTATIME) 2349 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2350 if (flags & MS_RDONLY) 2351 mnt_flags |= MNT_READONLY; 2352 2353 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2354 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2355 MS_STRICTATIME); 2356 2357 if (flags & MS_REMOUNT) 2358 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 2359 data_page); 2360 else if (flags & MS_BIND) 2361 retval = do_loopback(&path, dev_name, flags & MS_REC); 2362 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 2363 retval = do_change_type(&path, flags); 2364 else if (flags & MS_MOVE) 2365 retval = do_move_mount(&path, dev_name); 2366 else 2367 retval = do_new_mount(&path, type_page, flags, mnt_flags, 2368 dev_name, data_page); 2369 dput_out: 2370 path_put(&path); 2371 return retval; 2372 } 2373 2374 static struct mnt_namespace *alloc_mnt_ns(void) 2375 { 2376 struct mnt_namespace *new_ns; 2377 2378 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2379 if (!new_ns) 2380 return ERR_PTR(-ENOMEM); 2381 atomic_set(&new_ns->count, 1); 2382 new_ns->root = NULL; 2383 INIT_LIST_HEAD(&new_ns->list); 2384 init_waitqueue_head(&new_ns->poll); 2385 new_ns->event = 0; 2386 return new_ns; 2387 } 2388 2389 void mnt_make_longterm(struct vfsmount *mnt) 2390 { 2391 __mnt_make_longterm(mnt); 2392 } 2393 2394 void mnt_make_shortterm(struct vfsmount *mnt) 2395 { 2396 #ifdef CONFIG_SMP 2397 if (atomic_add_unless(&mnt->mnt_longterm, -1, 1)) 2398 return; 2399 br_write_lock(vfsmount_lock); 2400 atomic_dec(&mnt->mnt_longterm); 2401 br_write_unlock(vfsmount_lock); 2402 #endif 2403 } 2404 2405 /* 2406 * Allocate a new namespace structure and populate it with contents 2407 * copied from the namespace of the passed in task structure. 2408 */ 2409 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 2410 struct fs_struct *fs) 2411 { 2412 struct mnt_namespace *new_ns; 2413 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2414 struct vfsmount *p, *q; 2415 2416 new_ns = alloc_mnt_ns(); 2417 if (IS_ERR(new_ns)) 2418 return new_ns; 2419 2420 down_write(&namespace_sem); 2421 /* First pass: copy the tree topology */ 2422 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 2423 CL_COPY_ALL | CL_EXPIRE); 2424 if (!new_ns->root) { 2425 up_write(&namespace_sem); 2426 kfree(new_ns); 2427 return ERR_PTR(-ENOMEM); 2428 } 2429 br_write_lock(vfsmount_lock); 2430 list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 2431 br_write_unlock(vfsmount_lock); 2432 2433 /* 2434 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 2435 * as belonging to new namespace. We have already acquired a private 2436 * fs_struct, so tsk->fs->lock is not needed. 2437 */ 2438 p = mnt_ns->root; 2439 q = new_ns->root; 2440 while (p) { 2441 q->mnt_ns = new_ns; 2442 __mnt_make_longterm(q); 2443 if (fs) { 2444 if (p == fs->root.mnt) { 2445 fs->root.mnt = mntget(q); 2446 __mnt_make_longterm(q); 2447 mnt_make_shortterm(p); 2448 rootmnt = p; 2449 } 2450 if (p == fs->pwd.mnt) { 2451 fs->pwd.mnt = mntget(q); 2452 __mnt_make_longterm(q); 2453 mnt_make_shortterm(p); 2454 pwdmnt = p; 2455 } 2456 } 2457 p = next_mnt(p, mnt_ns->root); 2458 q = next_mnt(q, new_ns->root); 2459 } 2460 up_write(&namespace_sem); 2461 2462 if (rootmnt) 2463 mntput(rootmnt); 2464 if (pwdmnt) 2465 mntput(pwdmnt); 2466 2467 return new_ns; 2468 } 2469 2470 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2471 struct fs_struct *new_fs) 2472 { 2473 struct mnt_namespace *new_ns; 2474 2475 BUG_ON(!ns); 2476 get_mnt_ns(ns); 2477 2478 if (!(flags & CLONE_NEWNS)) 2479 return ns; 2480 2481 new_ns = dup_mnt_ns(ns, new_fs); 2482 2483 put_mnt_ns(ns); 2484 return new_ns; 2485 } 2486 2487 /** 2488 * create_mnt_ns - creates a private namespace and adds a root filesystem 2489 * @mnt: pointer to the new root filesystem mountpoint 2490 */ 2491 struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2492 { 2493 struct mnt_namespace *new_ns; 2494 2495 new_ns = alloc_mnt_ns(); 2496 if (!IS_ERR(new_ns)) { 2497 mnt->mnt_ns = new_ns; 2498 __mnt_make_longterm(mnt); 2499 new_ns->root = mnt; 2500 list_add(&new_ns->list, &new_ns->root->mnt_list); 2501 } 2502 return new_ns; 2503 } 2504 EXPORT_SYMBOL(create_mnt_ns); 2505 2506 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2507 char __user *, type, unsigned long, flags, void __user *, data) 2508 { 2509 int ret; 2510 char *kernel_type; 2511 char *kernel_dir; 2512 char *kernel_dev; 2513 unsigned long data_page; 2514 2515 ret = copy_mount_string(type, &kernel_type); 2516 if (ret < 0) 2517 goto out_type; 2518 2519 kernel_dir = getname(dir_name); 2520 if (IS_ERR(kernel_dir)) { 2521 ret = PTR_ERR(kernel_dir); 2522 goto out_dir; 2523 } 2524 2525 ret = copy_mount_string(dev_name, &kernel_dev); 2526 if (ret < 0) 2527 goto out_dev; 2528 2529 ret = copy_mount_options(data, &data_page); 2530 if (ret < 0) 2531 goto out_data; 2532 2533 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2534 (void *) data_page); 2535 2536 free_page(data_page); 2537 out_data: 2538 kfree(kernel_dev); 2539 out_dev: 2540 putname(kernel_dir); 2541 out_dir: 2542 kfree(kernel_type); 2543 out_type: 2544 return ret; 2545 } 2546 2547 /* 2548 * pivot_root Semantics: 2549 * Moves the root file system of the current process to the directory put_old, 2550 * makes new_root as the new root file system of the current process, and sets 2551 * root/cwd of all processes which had them on the current root to new_root. 2552 * 2553 * Restrictions: 2554 * The new_root and put_old must be directories, and must not be on the 2555 * same file system as the current process root. The put_old must be 2556 * underneath new_root, i.e. adding a non-zero number of /.. to the string 2557 * pointed to by put_old must yield the same directory as new_root. No other 2558 * file system may be mounted on put_old. After all, new_root is a mountpoint. 2559 * 2560 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 2561 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 2562 * in this situation. 2563 * 2564 * Notes: 2565 * - we don't move root/cwd if they are not at the root (reason: if something 2566 * cared enough to change them, it's probably wrong to force them elsewhere) 2567 * - it's okay to pick a root that isn't the root of a file system, e.g. 2568 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 2569 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 2570 * first. 2571 */ 2572 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 2573 const char __user *, put_old) 2574 { 2575 struct vfsmount *tmp; 2576 struct path new, old, parent_path, root_parent, root; 2577 int error; 2578 2579 if (!capable(CAP_SYS_ADMIN)) 2580 return -EPERM; 2581 2582 error = user_path_dir(new_root, &new); 2583 if (error) 2584 goto out0; 2585 2586 error = user_path_dir(put_old, &old); 2587 if (error) 2588 goto out1; 2589 2590 error = security_sb_pivotroot(&old, &new); 2591 if (error) 2592 goto out2; 2593 2594 get_fs_root(current->fs, &root); 2595 error = lock_mount(&old); 2596 if (error) 2597 goto out3; 2598 2599 error = -EINVAL; 2600 if (IS_MNT_SHARED(old.mnt) || 2601 IS_MNT_SHARED(new.mnt->mnt_parent) || 2602 IS_MNT_SHARED(root.mnt->mnt_parent)) 2603 goto out4; 2604 if (!check_mnt(root.mnt) || !check_mnt(new.mnt)) 2605 goto out4; 2606 error = -ENOENT; 2607 if (d_unlinked(new.dentry)) 2608 goto out4; 2609 if (d_unlinked(old.dentry)) 2610 goto out4; 2611 error = -EBUSY; 2612 if (new.mnt == root.mnt || 2613 old.mnt == root.mnt) 2614 goto out4; /* loop, on the same file system */ 2615 error = -EINVAL; 2616 if (root.mnt->mnt_root != root.dentry) 2617 goto out4; /* not a mountpoint */ 2618 if (root.mnt->mnt_parent == root.mnt) 2619 goto out4; /* not attached */ 2620 if (new.mnt->mnt_root != new.dentry) 2621 goto out4; /* not a mountpoint */ 2622 if (new.mnt->mnt_parent == new.mnt) 2623 goto out4; /* not attached */ 2624 /* make sure we can reach put_old from new_root */ 2625 tmp = old.mnt; 2626 if (tmp != new.mnt) { 2627 for (;;) { 2628 if (tmp->mnt_parent == tmp) 2629 goto out4; /* already mounted on put_old */ 2630 if (tmp->mnt_parent == new.mnt) 2631 break; 2632 tmp = tmp->mnt_parent; 2633 } 2634 if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 2635 goto out4; 2636 } else if (!is_subdir(old.dentry, new.dentry)) 2637 goto out4; 2638 br_write_lock(vfsmount_lock); 2639 detach_mnt(new.mnt, &parent_path); 2640 detach_mnt(root.mnt, &root_parent); 2641 /* mount old root on put_old */ 2642 attach_mnt(root.mnt, &old); 2643 /* mount new_root on / */ 2644 attach_mnt(new.mnt, &root_parent); 2645 touch_mnt_namespace(current->nsproxy->mnt_ns); 2646 br_write_unlock(vfsmount_lock); 2647 chroot_fs_refs(&root, &new); 2648 error = 0; 2649 out4: 2650 unlock_mount(&old); 2651 if (!error) { 2652 path_put(&root_parent); 2653 path_put(&parent_path); 2654 } 2655 out3: 2656 path_put(&root); 2657 out2: 2658 path_put(&old); 2659 out1: 2660 path_put(&new); 2661 out0: 2662 return error; 2663 } 2664 2665 static void __init init_mount_tree(void) 2666 { 2667 struct vfsmount *mnt; 2668 struct mnt_namespace *ns; 2669 struct path root; 2670 2671 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 2672 if (IS_ERR(mnt)) 2673 panic("Can't create rootfs"); 2674 2675 ns = create_mnt_ns(mnt); 2676 if (IS_ERR(ns)) 2677 panic("Can't allocate initial namespace"); 2678 2679 init_task.nsproxy->mnt_ns = ns; 2680 get_mnt_ns(ns); 2681 2682 root.mnt = ns->root; 2683 root.dentry = ns->root->mnt_root; 2684 2685 set_fs_pwd(current->fs, &root); 2686 set_fs_root(current->fs, &root); 2687 } 2688 2689 void __init mnt_init(void) 2690 { 2691 unsigned u; 2692 int err; 2693 2694 init_rwsem(&namespace_sem); 2695 2696 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 2697 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 2698 2699 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 2700 2701 if (!mount_hashtable) 2702 panic("Failed to allocate mount hash table\n"); 2703 2704 printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); 2705 2706 for (u = 0; u < HASH_SIZE; u++) 2707 INIT_LIST_HEAD(&mount_hashtable[u]); 2708 2709 br_lock_init(vfsmount_lock); 2710 2711 err = sysfs_init(); 2712 if (err) 2713 printk(KERN_WARNING "%s: sysfs_init error: %d\n", 2714 __func__, err); 2715 fs_kobj = kobject_create_and_add("fs", NULL); 2716 if (!fs_kobj) 2717 printk(KERN_WARNING "%s: kobj create error\n", __func__); 2718 init_rootfs(); 2719 init_mount_tree(); 2720 } 2721 2722 void put_mnt_ns(struct mnt_namespace *ns) 2723 { 2724 LIST_HEAD(umount_list); 2725 2726 if (!atomic_dec_and_test(&ns->count)) 2727 return; 2728 down_write(&namespace_sem); 2729 br_write_lock(vfsmount_lock); 2730 umount_tree(ns->root, 0, &umount_list); 2731 br_write_unlock(vfsmount_lock); 2732 up_write(&namespace_sem); 2733 release_mounts(&umount_list); 2734 kfree(ns); 2735 } 2736 EXPORT_SYMBOL(put_mnt_ns); 2737 2738 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 2739 { 2740 return vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 2741 } 2742 EXPORT_SYMBOL_GPL(kern_mount_data); 2743