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/config.h> 12 #include <linux/syscalls.h> 13 #include <linux/slab.h> 14 #include <linux/sched.h> 15 #include <linux/smp_lock.h> 16 #include <linux/init.h> 17 #include <linux/quotaops.h> 18 #include <linux/acct.h> 19 #include <linux/capability.h> 20 #include <linux/module.h> 21 #include <linux/seq_file.h> 22 #include <linux/namespace.h> 23 #include <linux/namei.h> 24 #include <linux/security.h> 25 #include <linux/mount.h> 26 #include <asm/uaccess.h> 27 #include <asm/unistd.h> 28 #include "pnode.h" 29 30 extern int __init init_rootfs(void); 31 32 #ifdef CONFIG_SYSFS 33 extern int __init sysfs_init(void); 34 #else 35 static inline int sysfs_init(void) 36 { 37 return 0; 38 } 39 #endif 40 41 /* spinlock for vfsmount related operations, inplace of dcache_lock */ 42 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 43 44 static int event; 45 46 static struct list_head *mount_hashtable __read_mostly; 47 static int hash_mask __read_mostly, hash_bits __read_mostly; 48 static kmem_cache_t *mnt_cache __read_mostly; 49 static struct rw_semaphore namespace_sem; 50 51 /* /sys/fs */ 52 decl_subsys(fs, NULL, NULL); 53 EXPORT_SYMBOL_GPL(fs_subsys); 54 55 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 56 { 57 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 58 tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 59 tmp = tmp + (tmp >> hash_bits); 60 return tmp & hash_mask; 61 } 62 63 struct vfsmount *alloc_vfsmnt(const char *name) 64 { 65 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); 66 if (mnt) { 67 memset(mnt, 0, sizeof(struct vfsmount)); 68 atomic_set(&mnt->mnt_count, 1); 69 INIT_LIST_HEAD(&mnt->mnt_hash); 70 INIT_LIST_HEAD(&mnt->mnt_child); 71 INIT_LIST_HEAD(&mnt->mnt_mounts); 72 INIT_LIST_HEAD(&mnt->mnt_list); 73 INIT_LIST_HEAD(&mnt->mnt_expire); 74 INIT_LIST_HEAD(&mnt->mnt_share); 75 INIT_LIST_HEAD(&mnt->mnt_slave_list); 76 INIT_LIST_HEAD(&mnt->mnt_slave); 77 if (name) { 78 int size = strlen(name) + 1; 79 char *newname = kmalloc(size, GFP_KERNEL); 80 if (newname) { 81 memcpy(newname, name, size); 82 mnt->mnt_devname = newname; 83 } 84 } 85 } 86 return mnt; 87 } 88 89 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 90 { 91 mnt->mnt_sb = sb; 92 mnt->mnt_root = dget(sb->s_root); 93 return 0; 94 } 95 96 EXPORT_SYMBOL(simple_set_mnt); 97 98 void free_vfsmnt(struct vfsmount *mnt) 99 { 100 kfree(mnt->mnt_devname); 101 kmem_cache_free(mnt_cache, mnt); 102 } 103 104 /* 105 * find the first or last mount at @dentry on vfsmount @mnt depending on 106 * @dir. If @dir is set return the first mount else return the last mount. 107 */ 108 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 109 int dir) 110 { 111 struct list_head *head = mount_hashtable + hash(mnt, dentry); 112 struct list_head *tmp = head; 113 struct vfsmount *p, *found = NULL; 114 115 for (;;) { 116 tmp = dir ? tmp->next : tmp->prev; 117 p = NULL; 118 if (tmp == head) 119 break; 120 p = list_entry(tmp, struct vfsmount, mnt_hash); 121 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 122 found = p; 123 break; 124 } 125 } 126 return found; 127 } 128 129 /* 130 * lookup_mnt increments the ref count before returning 131 * the vfsmount struct. 132 */ 133 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 134 { 135 struct vfsmount *child_mnt; 136 spin_lock(&vfsmount_lock); 137 if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 138 mntget(child_mnt); 139 spin_unlock(&vfsmount_lock); 140 return child_mnt; 141 } 142 143 static inline int check_mnt(struct vfsmount *mnt) 144 { 145 return mnt->mnt_namespace == current->namespace; 146 } 147 148 static void touch_namespace(struct namespace *ns) 149 { 150 if (ns) { 151 ns->event = ++event; 152 wake_up_interruptible(&ns->poll); 153 } 154 } 155 156 static void __touch_namespace(struct namespace *ns) 157 { 158 if (ns && ns->event != event) { 159 ns->event = event; 160 wake_up_interruptible(&ns->poll); 161 } 162 } 163 164 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) 165 { 166 old_nd->dentry = mnt->mnt_mountpoint; 167 old_nd->mnt = mnt->mnt_parent; 168 mnt->mnt_parent = mnt; 169 mnt->mnt_mountpoint = mnt->mnt_root; 170 list_del_init(&mnt->mnt_child); 171 list_del_init(&mnt->mnt_hash); 172 old_nd->dentry->d_mounted--; 173 } 174 175 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 176 struct vfsmount *child_mnt) 177 { 178 child_mnt->mnt_parent = mntget(mnt); 179 child_mnt->mnt_mountpoint = dget(dentry); 180 dentry->d_mounted++; 181 } 182 183 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) 184 { 185 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); 186 list_add_tail(&mnt->mnt_hash, mount_hashtable + 187 hash(nd->mnt, nd->dentry)); 188 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); 189 } 190 191 /* 192 * the caller must hold vfsmount_lock 193 */ 194 static void commit_tree(struct vfsmount *mnt) 195 { 196 struct vfsmount *parent = mnt->mnt_parent; 197 struct vfsmount *m; 198 LIST_HEAD(head); 199 struct namespace *n = parent->mnt_namespace; 200 201 BUG_ON(parent == mnt); 202 203 list_add_tail(&head, &mnt->mnt_list); 204 list_for_each_entry(m, &head, mnt_list) 205 m->mnt_namespace = n; 206 list_splice(&head, n->list.prev); 207 208 list_add_tail(&mnt->mnt_hash, mount_hashtable + 209 hash(parent, mnt->mnt_mountpoint)); 210 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 211 touch_namespace(n); 212 } 213 214 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 215 { 216 struct list_head *next = p->mnt_mounts.next; 217 if (next == &p->mnt_mounts) { 218 while (1) { 219 if (p == root) 220 return NULL; 221 next = p->mnt_child.next; 222 if (next != &p->mnt_parent->mnt_mounts) 223 break; 224 p = p->mnt_parent; 225 } 226 } 227 return list_entry(next, struct vfsmount, mnt_child); 228 } 229 230 static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 231 { 232 struct list_head *prev = p->mnt_mounts.prev; 233 while (prev != &p->mnt_mounts) { 234 p = list_entry(prev, struct vfsmount, mnt_child); 235 prev = p->mnt_mounts.prev; 236 } 237 return p; 238 } 239 240 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 241 int flag) 242 { 243 struct super_block *sb = old->mnt_sb; 244 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 245 246 if (mnt) { 247 mnt->mnt_flags = old->mnt_flags; 248 atomic_inc(&sb->s_active); 249 mnt->mnt_sb = sb; 250 mnt->mnt_root = dget(root); 251 mnt->mnt_mountpoint = mnt->mnt_root; 252 mnt->mnt_parent = mnt; 253 254 if (flag & CL_SLAVE) { 255 list_add(&mnt->mnt_slave, &old->mnt_slave_list); 256 mnt->mnt_master = old; 257 CLEAR_MNT_SHARED(mnt); 258 } else { 259 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 260 list_add(&mnt->mnt_share, &old->mnt_share); 261 if (IS_MNT_SLAVE(old)) 262 list_add(&mnt->mnt_slave, &old->mnt_slave); 263 mnt->mnt_master = old->mnt_master; 264 } 265 if (flag & CL_MAKE_SHARED) 266 set_mnt_shared(mnt); 267 268 /* stick the duplicate mount on the same expiry list 269 * as the original if that was on one */ 270 if (flag & CL_EXPIRE) { 271 spin_lock(&vfsmount_lock); 272 if (!list_empty(&old->mnt_expire)) 273 list_add(&mnt->mnt_expire, &old->mnt_expire); 274 spin_unlock(&vfsmount_lock); 275 } 276 } 277 return mnt; 278 } 279 280 static inline void __mntput(struct vfsmount *mnt) 281 { 282 struct super_block *sb = mnt->mnt_sb; 283 dput(mnt->mnt_root); 284 free_vfsmnt(mnt); 285 deactivate_super(sb); 286 } 287 288 void mntput_no_expire(struct vfsmount *mnt) 289 { 290 repeat: 291 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 292 if (likely(!mnt->mnt_pinned)) { 293 spin_unlock(&vfsmount_lock); 294 __mntput(mnt); 295 return; 296 } 297 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 298 mnt->mnt_pinned = 0; 299 spin_unlock(&vfsmount_lock); 300 acct_auto_close_mnt(mnt); 301 security_sb_umount_close(mnt); 302 goto repeat; 303 } 304 } 305 306 EXPORT_SYMBOL(mntput_no_expire); 307 308 void mnt_pin(struct vfsmount *mnt) 309 { 310 spin_lock(&vfsmount_lock); 311 mnt->mnt_pinned++; 312 spin_unlock(&vfsmount_lock); 313 } 314 315 EXPORT_SYMBOL(mnt_pin); 316 317 void mnt_unpin(struct vfsmount *mnt) 318 { 319 spin_lock(&vfsmount_lock); 320 if (mnt->mnt_pinned) { 321 atomic_inc(&mnt->mnt_count); 322 mnt->mnt_pinned--; 323 } 324 spin_unlock(&vfsmount_lock); 325 } 326 327 EXPORT_SYMBOL(mnt_unpin); 328 329 /* iterator */ 330 static void *m_start(struct seq_file *m, loff_t *pos) 331 { 332 struct namespace *n = m->private; 333 struct list_head *p; 334 loff_t l = *pos; 335 336 down_read(&namespace_sem); 337 list_for_each(p, &n->list) 338 if (!l--) 339 return list_entry(p, struct vfsmount, mnt_list); 340 return NULL; 341 } 342 343 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 344 { 345 struct namespace *n = m->private; 346 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; 347 (*pos)++; 348 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); 349 } 350 351 static void m_stop(struct seq_file *m, void *v) 352 { 353 up_read(&namespace_sem); 354 } 355 356 static inline void mangle(struct seq_file *m, const char *s) 357 { 358 seq_escape(m, s, " \t\n\\"); 359 } 360 361 static int show_vfsmnt(struct seq_file *m, void *v) 362 { 363 struct vfsmount *mnt = v; 364 int err = 0; 365 static struct proc_fs_info { 366 int flag; 367 char *str; 368 } fs_info[] = { 369 { MS_SYNCHRONOUS, ",sync" }, 370 { MS_DIRSYNC, ",dirsync" }, 371 { MS_MANDLOCK, ",mand" }, 372 { 0, NULL } 373 }; 374 static struct proc_fs_info mnt_info[] = { 375 { MNT_NOSUID, ",nosuid" }, 376 { MNT_NODEV, ",nodev" }, 377 { MNT_NOEXEC, ",noexec" }, 378 { MNT_NOATIME, ",noatime" }, 379 { MNT_NODIRATIME, ",nodiratime" }, 380 { 0, NULL } 381 }; 382 struct proc_fs_info *fs_infop; 383 384 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 385 seq_putc(m, ' '); 386 seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 387 seq_putc(m, ' '); 388 mangle(m, mnt->mnt_sb->s_type->name); 389 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 390 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 391 if (mnt->mnt_sb->s_flags & fs_infop->flag) 392 seq_puts(m, fs_infop->str); 393 } 394 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 395 if (mnt->mnt_flags & fs_infop->flag) 396 seq_puts(m, fs_infop->str); 397 } 398 if (mnt->mnt_sb->s_op->show_options) 399 err = mnt->mnt_sb->s_op->show_options(m, mnt); 400 seq_puts(m, " 0 0\n"); 401 return err; 402 } 403 404 struct seq_operations mounts_op = { 405 .start = m_start, 406 .next = m_next, 407 .stop = m_stop, 408 .show = show_vfsmnt 409 }; 410 411 static int show_vfsstat(struct seq_file *m, void *v) 412 { 413 struct vfsmount *mnt = v; 414 int err = 0; 415 416 /* device */ 417 if (mnt->mnt_devname) { 418 seq_puts(m, "device "); 419 mangle(m, mnt->mnt_devname); 420 } else 421 seq_puts(m, "no device"); 422 423 /* mount point */ 424 seq_puts(m, " mounted on "); 425 seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); 426 seq_putc(m, ' '); 427 428 /* file system type */ 429 seq_puts(m, "with fstype "); 430 mangle(m, mnt->mnt_sb->s_type->name); 431 432 /* optional statistics */ 433 if (mnt->mnt_sb->s_op->show_stats) { 434 seq_putc(m, ' '); 435 err = mnt->mnt_sb->s_op->show_stats(m, mnt); 436 } 437 438 seq_putc(m, '\n'); 439 return err; 440 } 441 442 struct seq_operations mountstats_op = { 443 .start = m_start, 444 .next = m_next, 445 .stop = m_stop, 446 .show = show_vfsstat, 447 }; 448 449 /** 450 * may_umount_tree - check if a mount tree is busy 451 * @mnt: root of mount tree 452 * 453 * This is called to check if a tree of mounts has any 454 * open files, pwds, chroots or sub mounts that are 455 * busy. 456 */ 457 int may_umount_tree(struct vfsmount *mnt) 458 { 459 int actual_refs = 0; 460 int minimum_refs = 0; 461 struct vfsmount *p; 462 463 spin_lock(&vfsmount_lock); 464 for (p = mnt; p; p = next_mnt(p, mnt)) { 465 actual_refs += atomic_read(&p->mnt_count); 466 minimum_refs += 2; 467 } 468 spin_unlock(&vfsmount_lock); 469 470 if (actual_refs > minimum_refs) 471 return 0; 472 473 return 1; 474 } 475 476 EXPORT_SYMBOL(may_umount_tree); 477 478 /** 479 * may_umount - check if a mount point is busy 480 * @mnt: root of mount 481 * 482 * This is called to check if a mount point has any 483 * open files, pwds, chroots or sub mounts. If the 484 * mount has sub mounts this will return busy 485 * regardless of whether the sub mounts are busy. 486 * 487 * Doesn't take quota and stuff into account. IOW, in some cases it will 488 * give false negatives. The main reason why it's here is that we need 489 * a non-destructive way to look for easily umountable filesystems. 490 */ 491 int may_umount(struct vfsmount *mnt) 492 { 493 int ret = 1; 494 spin_lock(&vfsmount_lock); 495 if (propagate_mount_busy(mnt, 2)) 496 ret = 0; 497 spin_unlock(&vfsmount_lock); 498 return ret; 499 } 500 501 EXPORT_SYMBOL(may_umount); 502 503 void release_mounts(struct list_head *head) 504 { 505 struct vfsmount *mnt; 506 while (!list_empty(head)) { 507 mnt = list_entry(head->next, struct vfsmount, mnt_hash); 508 list_del_init(&mnt->mnt_hash); 509 if (mnt->mnt_parent != mnt) { 510 struct dentry *dentry; 511 struct vfsmount *m; 512 spin_lock(&vfsmount_lock); 513 dentry = mnt->mnt_mountpoint; 514 m = mnt->mnt_parent; 515 mnt->mnt_mountpoint = mnt->mnt_root; 516 mnt->mnt_parent = mnt; 517 spin_unlock(&vfsmount_lock); 518 dput(dentry); 519 mntput(m); 520 } 521 mntput(mnt); 522 } 523 } 524 525 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 526 { 527 struct vfsmount *p; 528 529 for (p = mnt; p; p = next_mnt(p, mnt)) 530 list_move(&p->mnt_hash, kill); 531 532 if (propagate) 533 propagate_umount(kill); 534 535 list_for_each_entry(p, kill, mnt_hash) { 536 list_del_init(&p->mnt_expire); 537 list_del_init(&p->mnt_list); 538 __touch_namespace(p->mnt_namespace); 539 p->mnt_namespace = NULL; 540 list_del_init(&p->mnt_child); 541 if (p->mnt_parent != p) 542 p->mnt_mountpoint->d_mounted--; 543 change_mnt_propagation(p, MS_PRIVATE); 544 } 545 } 546 547 static int do_umount(struct vfsmount *mnt, int flags) 548 { 549 struct super_block *sb = mnt->mnt_sb; 550 int retval; 551 LIST_HEAD(umount_list); 552 553 retval = security_sb_umount(mnt, flags); 554 if (retval) 555 return retval; 556 557 /* 558 * Allow userspace to request a mountpoint be expired rather than 559 * unmounting unconditionally. Unmount only happens if: 560 * (1) the mark is already set (the mark is cleared by mntput()) 561 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 562 */ 563 if (flags & MNT_EXPIRE) { 564 if (mnt == current->fs->rootmnt || 565 flags & (MNT_FORCE | MNT_DETACH)) 566 return -EINVAL; 567 568 if (atomic_read(&mnt->mnt_count) != 2) 569 return -EBUSY; 570 571 if (!xchg(&mnt->mnt_expiry_mark, 1)) 572 return -EAGAIN; 573 } 574 575 /* 576 * If we may have to abort operations to get out of this 577 * mount, and they will themselves hold resources we must 578 * allow the fs to do things. In the Unix tradition of 579 * 'Gee thats tricky lets do it in userspace' the umount_begin 580 * might fail to complete on the first run through as other tasks 581 * must return, and the like. Thats for the mount program to worry 582 * about for the moment. 583 */ 584 585 lock_kernel(); 586 if (sb->s_op->umount_begin) 587 sb->s_op->umount_begin(mnt, flags); 588 unlock_kernel(); 589 590 /* 591 * No sense to grab the lock for this test, but test itself looks 592 * somewhat bogus. Suggestions for better replacement? 593 * Ho-hum... In principle, we might treat that as umount + switch 594 * to rootfs. GC would eventually take care of the old vfsmount. 595 * Actually it makes sense, especially if rootfs would contain a 596 * /reboot - static binary that would close all descriptors and 597 * call reboot(9). Then init(8) could umount root and exec /reboot. 598 */ 599 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { 600 /* 601 * Special case for "unmounting" root ... 602 * we just try to remount it readonly. 603 */ 604 down_write(&sb->s_umount); 605 if (!(sb->s_flags & MS_RDONLY)) { 606 lock_kernel(); 607 DQUOT_OFF(sb); 608 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 609 unlock_kernel(); 610 } 611 up_write(&sb->s_umount); 612 return retval; 613 } 614 615 down_write(&namespace_sem); 616 spin_lock(&vfsmount_lock); 617 event++; 618 619 retval = -EBUSY; 620 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 621 if (!list_empty(&mnt->mnt_list)) 622 umount_tree(mnt, 1, &umount_list); 623 retval = 0; 624 } 625 spin_unlock(&vfsmount_lock); 626 if (retval) 627 security_sb_umount_busy(mnt); 628 up_write(&namespace_sem); 629 release_mounts(&umount_list); 630 return retval; 631 } 632 633 /* 634 * Now umount can handle mount points as well as block devices. 635 * This is important for filesystems which use unnamed block devices. 636 * 637 * We now support a flag for forced unmount like the other 'big iron' 638 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 639 */ 640 641 asmlinkage long sys_umount(char __user * name, int flags) 642 { 643 struct nameidata nd; 644 int retval; 645 646 retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 647 if (retval) 648 goto out; 649 retval = -EINVAL; 650 if (nd.dentry != nd.mnt->mnt_root) 651 goto dput_and_out; 652 if (!check_mnt(nd.mnt)) 653 goto dput_and_out; 654 655 retval = -EPERM; 656 if (!capable(CAP_SYS_ADMIN)) 657 goto dput_and_out; 658 659 retval = do_umount(nd.mnt, flags); 660 dput_and_out: 661 path_release_on_umount(&nd); 662 out: 663 return retval; 664 } 665 666 #ifdef __ARCH_WANT_SYS_OLDUMOUNT 667 668 /* 669 * The 2.0 compatible umount. No flags. 670 */ 671 asmlinkage long sys_oldumount(char __user * name) 672 { 673 return sys_umount(name, 0); 674 } 675 676 #endif 677 678 static int mount_is_safe(struct nameidata *nd) 679 { 680 if (capable(CAP_SYS_ADMIN)) 681 return 0; 682 return -EPERM; 683 #ifdef notyet 684 if (S_ISLNK(nd->dentry->d_inode->i_mode)) 685 return -EPERM; 686 if (nd->dentry->d_inode->i_mode & S_ISVTX) { 687 if (current->uid != nd->dentry->d_inode->i_uid) 688 return -EPERM; 689 } 690 if (vfs_permission(nd, MAY_WRITE)) 691 return -EPERM; 692 return 0; 693 #endif 694 } 695 696 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 697 { 698 while (1) { 699 if (d == dentry) 700 return 1; 701 if (d == NULL || d == d->d_parent) 702 return 0; 703 d = d->d_parent; 704 } 705 } 706 707 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 708 int flag) 709 { 710 struct vfsmount *res, *p, *q, *r, *s; 711 struct nameidata nd; 712 713 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 714 return NULL; 715 716 res = q = clone_mnt(mnt, dentry, flag); 717 if (!q) 718 goto Enomem; 719 q->mnt_mountpoint = mnt->mnt_mountpoint; 720 721 p = mnt; 722 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 723 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 724 continue; 725 726 for (s = r; s; s = next_mnt(s, r)) { 727 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 728 s = skip_mnt_tree(s); 729 continue; 730 } 731 while (p != s->mnt_parent) { 732 p = p->mnt_parent; 733 q = q->mnt_parent; 734 } 735 p = s; 736 nd.mnt = q; 737 nd.dentry = p->mnt_mountpoint; 738 q = clone_mnt(p, p->mnt_root, flag); 739 if (!q) 740 goto Enomem; 741 spin_lock(&vfsmount_lock); 742 list_add_tail(&q->mnt_list, &res->mnt_list); 743 attach_mnt(q, &nd); 744 spin_unlock(&vfsmount_lock); 745 } 746 } 747 return res; 748 Enomem: 749 if (res) { 750 LIST_HEAD(umount_list); 751 spin_lock(&vfsmount_lock); 752 umount_tree(res, 0, &umount_list); 753 spin_unlock(&vfsmount_lock); 754 release_mounts(&umount_list); 755 } 756 return NULL; 757 } 758 759 /* 760 * @source_mnt : mount tree to be attached 761 * @nd : place the mount tree @source_mnt is attached 762 * @parent_nd : if non-null, detach the source_mnt from its parent and 763 * store the parent mount and mountpoint dentry. 764 * (done when source_mnt is moved) 765 * 766 * NOTE: in the table below explains the semantics when a source mount 767 * of a given type is attached to a destination mount of a given type. 768 * --------------------------------------------------------------------------- 769 * | BIND MOUNT OPERATION | 770 * |************************************************************************** 771 * | source-->| shared | private | slave | unbindable | 772 * | dest | | | | | 773 * | | | | | | | 774 * | v | | | | | 775 * |************************************************************************** 776 * | shared | shared (++) | shared (+) | shared(+++)| invalid | 777 * | | | | | | 778 * |non-shared| shared (+) | private | slave (*) | invalid | 779 * *************************************************************************** 780 * A bind operation clones the source mount and mounts the clone on the 781 * destination mount. 782 * 783 * (++) the cloned mount is propagated to all the mounts in the propagation 784 * tree of the destination mount and the cloned mount is added to 785 * the peer group of the source mount. 786 * (+) the cloned mount is created under the destination mount and is marked 787 * as shared. The cloned mount is added to the peer group of the source 788 * mount. 789 * (+++) the mount is propagated to all the mounts in the propagation tree 790 * of the destination mount and the cloned mount is made slave 791 * of the same master as that of the source mount. The cloned mount 792 * is marked as 'shared and slave'. 793 * (*) the cloned mount is made a slave of the same master as that of the 794 * source mount. 795 * 796 * --------------------------------------------------------------------------- 797 * | MOVE MOUNT OPERATION | 798 * |************************************************************************** 799 * | source-->| shared | private | slave | unbindable | 800 * | dest | | | | | 801 * | | | | | | | 802 * | v | | | | | 803 * |************************************************************************** 804 * | shared | shared (+) | shared (+) | shared(+++) | invalid | 805 * | | | | | | 806 * |non-shared| shared (+*) | private | slave (*) | unbindable | 807 * *************************************************************************** 808 * 809 * (+) the mount is moved to the destination. And is then propagated to 810 * all the mounts in the propagation tree of the destination mount. 811 * (+*) the mount is moved to the destination. 812 * (+++) the mount is moved to the destination and is then propagated to 813 * all the mounts belonging to the destination mount's propagation tree. 814 * the mount is marked as 'shared and slave'. 815 * (*) the mount continues to be a slave at the new location. 816 * 817 * if the source mount is a tree, the operations explained above is 818 * applied to each mount in the tree. 819 * Must be called without spinlocks held, since this function can sleep 820 * in allocations. 821 */ 822 static int attach_recursive_mnt(struct vfsmount *source_mnt, 823 struct nameidata *nd, struct nameidata *parent_nd) 824 { 825 LIST_HEAD(tree_list); 826 struct vfsmount *dest_mnt = nd->mnt; 827 struct dentry *dest_dentry = nd->dentry; 828 struct vfsmount *child, *p; 829 830 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 831 return -EINVAL; 832 833 if (IS_MNT_SHARED(dest_mnt)) { 834 for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 835 set_mnt_shared(p); 836 } 837 838 spin_lock(&vfsmount_lock); 839 if (parent_nd) { 840 detach_mnt(source_mnt, parent_nd); 841 attach_mnt(source_mnt, nd); 842 touch_namespace(current->namespace); 843 } else { 844 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 845 commit_tree(source_mnt); 846 } 847 848 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 849 list_del_init(&child->mnt_hash); 850 commit_tree(child); 851 } 852 spin_unlock(&vfsmount_lock); 853 return 0; 854 } 855 856 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 857 { 858 int err; 859 if (mnt->mnt_sb->s_flags & MS_NOUSER) 860 return -EINVAL; 861 862 if (S_ISDIR(nd->dentry->d_inode->i_mode) != 863 S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 864 return -ENOTDIR; 865 866 err = -ENOENT; 867 mutex_lock(&nd->dentry->d_inode->i_mutex); 868 if (IS_DEADDIR(nd->dentry->d_inode)) 869 goto out_unlock; 870 871 err = security_sb_check_sb(mnt, nd); 872 if (err) 873 goto out_unlock; 874 875 err = -ENOENT; 876 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) 877 err = attach_recursive_mnt(mnt, nd, NULL); 878 out_unlock: 879 mutex_unlock(&nd->dentry->d_inode->i_mutex); 880 if (!err) 881 security_sb_post_addmount(mnt, nd); 882 return err; 883 } 884 885 /* 886 * recursively change the type of the mountpoint. 887 */ 888 static int do_change_type(struct nameidata *nd, int flag) 889 { 890 struct vfsmount *m, *mnt = nd->mnt; 891 int recurse = flag & MS_REC; 892 int type = flag & ~MS_REC; 893 894 if (nd->dentry != nd->mnt->mnt_root) 895 return -EINVAL; 896 897 down_write(&namespace_sem); 898 spin_lock(&vfsmount_lock); 899 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 900 change_mnt_propagation(m, type); 901 spin_unlock(&vfsmount_lock); 902 up_write(&namespace_sem); 903 return 0; 904 } 905 906 /* 907 * do loopback mount. 908 */ 909 static int do_loopback(struct nameidata *nd, char *old_name, int recurse) 910 { 911 struct nameidata old_nd; 912 struct vfsmount *mnt = NULL; 913 int err = mount_is_safe(nd); 914 if (err) 915 return err; 916 if (!old_name || !*old_name) 917 return -EINVAL; 918 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 919 if (err) 920 return err; 921 922 down_write(&namespace_sem); 923 err = -EINVAL; 924 if (IS_MNT_UNBINDABLE(old_nd.mnt)) 925 goto out; 926 927 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 928 goto out; 929 930 err = -ENOMEM; 931 if (recurse) 932 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); 933 else 934 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); 935 936 if (!mnt) 937 goto out; 938 939 err = graft_tree(mnt, nd); 940 if (err) { 941 LIST_HEAD(umount_list); 942 spin_lock(&vfsmount_lock); 943 umount_tree(mnt, 0, &umount_list); 944 spin_unlock(&vfsmount_lock); 945 release_mounts(&umount_list); 946 } 947 948 out: 949 up_write(&namespace_sem); 950 path_release(&old_nd); 951 return err; 952 } 953 954 /* 955 * change filesystem flags. dir should be a physical root of filesystem. 956 * If you've mounted a non-root directory somewhere and want to do remount 957 * on it - tough luck. 958 */ 959 static int do_remount(struct nameidata *nd, int flags, int mnt_flags, 960 void *data) 961 { 962 int err; 963 struct super_block *sb = nd->mnt->mnt_sb; 964 965 if (!capable(CAP_SYS_ADMIN)) 966 return -EPERM; 967 968 if (!check_mnt(nd->mnt)) 969 return -EINVAL; 970 971 if (nd->dentry != nd->mnt->mnt_root) 972 return -EINVAL; 973 974 down_write(&sb->s_umount); 975 err = do_remount_sb(sb, flags, data, 0); 976 if (!err) 977 nd->mnt->mnt_flags = mnt_flags; 978 up_write(&sb->s_umount); 979 if (!err) 980 security_sb_post_remount(nd->mnt, flags, data); 981 return err; 982 } 983 984 static inline int tree_contains_unbindable(struct vfsmount *mnt) 985 { 986 struct vfsmount *p; 987 for (p = mnt; p; p = next_mnt(p, mnt)) { 988 if (IS_MNT_UNBINDABLE(p)) 989 return 1; 990 } 991 return 0; 992 } 993 994 static int do_move_mount(struct nameidata *nd, char *old_name) 995 { 996 struct nameidata old_nd, parent_nd; 997 struct vfsmount *p; 998 int err = 0; 999 if (!capable(CAP_SYS_ADMIN)) 1000 return -EPERM; 1001 if (!old_name || !*old_name) 1002 return -EINVAL; 1003 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 1004 if (err) 1005 return err; 1006 1007 down_write(&namespace_sem); 1008 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 1009 ; 1010 err = -EINVAL; 1011 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) 1012 goto out; 1013 1014 err = -ENOENT; 1015 mutex_lock(&nd->dentry->d_inode->i_mutex); 1016 if (IS_DEADDIR(nd->dentry->d_inode)) 1017 goto out1; 1018 1019 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) 1020 goto out1; 1021 1022 err = -EINVAL; 1023 if (old_nd.dentry != old_nd.mnt->mnt_root) 1024 goto out1; 1025 1026 if (old_nd.mnt == old_nd.mnt->mnt_parent) 1027 goto out1; 1028 1029 if (S_ISDIR(nd->dentry->d_inode->i_mode) != 1030 S_ISDIR(old_nd.dentry->d_inode->i_mode)) 1031 goto out1; 1032 /* 1033 * Don't move a mount residing in a shared parent. 1034 */ 1035 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) 1036 goto out1; 1037 /* 1038 * Don't move a mount tree containing unbindable mounts to a destination 1039 * mount which is shared. 1040 */ 1041 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt)) 1042 goto out1; 1043 err = -ELOOP; 1044 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) 1045 if (p == old_nd.mnt) 1046 goto out1; 1047 1048 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) 1049 goto out1; 1050 1051 spin_lock(&vfsmount_lock); 1052 /* if the mount is moved, it should no longer be expire 1053 * automatically */ 1054 list_del_init(&old_nd.mnt->mnt_expire); 1055 spin_unlock(&vfsmount_lock); 1056 out1: 1057 mutex_unlock(&nd->dentry->d_inode->i_mutex); 1058 out: 1059 up_write(&namespace_sem); 1060 if (!err) 1061 path_release(&parent_nd); 1062 path_release(&old_nd); 1063 return err; 1064 } 1065 1066 /* 1067 * create a new mount for userspace and request it to be added into the 1068 * namespace's tree 1069 */ 1070 static int do_new_mount(struct nameidata *nd, char *type, int flags, 1071 int mnt_flags, char *name, void *data) 1072 { 1073 struct vfsmount *mnt; 1074 1075 if (!type || !memchr(type, 0, PAGE_SIZE)) 1076 return -EINVAL; 1077 1078 /* we need capabilities... */ 1079 if (!capable(CAP_SYS_ADMIN)) 1080 return -EPERM; 1081 1082 mnt = do_kern_mount(type, flags, name, data); 1083 if (IS_ERR(mnt)) 1084 return PTR_ERR(mnt); 1085 1086 return do_add_mount(mnt, nd, mnt_flags, NULL); 1087 } 1088 1089 /* 1090 * add a mount into a namespace's mount tree 1091 * - provide the option of adding the new mount to an expiration list 1092 */ 1093 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 1094 int mnt_flags, struct list_head *fslist) 1095 { 1096 int err; 1097 1098 down_write(&namespace_sem); 1099 /* Something was mounted here while we slept */ 1100 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) 1101 ; 1102 err = -EINVAL; 1103 if (!check_mnt(nd->mnt)) 1104 goto unlock; 1105 1106 /* Refuse the same filesystem on the same mount point */ 1107 err = -EBUSY; 1108 if (nd->mnt->mnt_sb == newmnt->mnt_sb && 1109 nd->mnt->mnt_root == nd->dentry) 1110 goto unlock; 1111 1112 err = -EINVAL; 1113 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 1114 goto unlock; 1115 1116 newmnt->mnt_flags = mnt_flags; 1117 if ((err = graft_tree(newmnt, nd))) 1118 goto unlock; 1119 1120 if (fslist) { 1121 /* add to the specified expiration list */ 1122 spin_lock(&vfsmount_lock); 1123 list_add_tail(&newmnt->mnt_expire, fslist); 1124 spin_unlock(&vfsmount_lock); 1125 } 1126 up_write(&namespace_sem); 1127 return 0; 1128 1129 unlock: 1130 up_write(&namespace_sem); 1131 mntput(newmnt); 1132 return err; 1133 } 1134 1135 EXPORT_SYMBOL_GPL(do_add_mount); 1136 1137 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, 1138 struct list_head *umounts) 1139 { 1140 spin_lock(&vfsmount_lock); 1141 1142 /* 1143 * Check if mount is still attached, if not, let whoever holds it deal 1144 * with the sucker 1145 */ 1146 if (mnt->mnt_parent == mnt) { 1147 spin_unlock(&vfsmount_lock); 1148 return; 1149 } 1150 1151 /* 1152 * Check that it is still dead: the count should now be 2 - as 1153 * contributed by the vfsmount parent and the mntget above 1154 */ 1155 if (!propagate_mount_busy(mnt, 2)) { 1156 /* delete from the namespace */ 1157 touch_namespace(mnt->mnt_namespace); 1158 list_del_init(&mnt->mnt_list); 1159 mnt->mnt_namespace = NULL; 1160 umount_tree(mnt, 1, umounts); 1161 spin_unlock(&vfsmount_lock); 1162 } else { 1163 /* 1164 * Someone brought it back to life whilst we didn't have any 1165 * locks held so return it to the expiration list 1166 */ 1167 list_add_tail(&mnt->mnt_expire, mounts); 1168 spin_unlock(&vfsmount_lock); 1169 } 1170 } 1171 1172 /* 1173 * go through the vfsmounts we've just consigned to the graveyard to 1174 * - check that they're still dead 1175 * - delete the vfsmount from the appropriate namespace under lock 1176 * - dispose of the corpse 1177 */ 1178 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts) 1179 { 1180 struct namespace *namespace; 1181 struct vfsmount *mnt; 1182 1183 while (!list_empty(graveyard)) { 1184 LIST_HEAD(umounts); 1185 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire); 1186 list_del_init(&mnt->mnt_expire); 1187 1188 /* don't do anything if the namespace is dead - all the 1189 * vfsmounts from it are going away anyway */ 1190 namespace = mnt->mnt_namespace; 1191 if (!namespace || !namespace->root) 1192 continue; 1193 get_namespace(namespace); 1194 1195 spin_unlock(&vfsmount_lock); 1196 down_write(&namespace_sem); 1197 expire_mount(mnt, mounts, &umounts); 1198 up_write(&namespace_sem); 1199 release_mounts(&umounts); 1200 mntput(mnt); 1201 put_namespace(namespace); 1202 spin_lock(&vfsmount_lock); 1203 } 1204 } 1205 1206 /* 1207 * process a list of expirable mountpoints with the intent of discarding any 1208 * mountpoints that aren't in use and haven't been touched since last we came 1209 * here 1210 */ 1211 void mark_mounts_for_expiry(struct list_head *mounts) 1212 { 1213 struct vfsmount *mnt, *next; 1214 LIST_HEAD(graveyard); 1215 1216 if (list_empty(mounts)) 1217 return; 1218 1219 spin_lock(&vfsmount_lock); 1220 1221 /* extract from the expiration list every vfsmount that matches the 1222 * following criteria: 1223 * - only referenced by its parent vfsmount 1224 * - still marked for expiry (marked on the last call here; marks are 1225 * cleared by mntput()) 1226 */ 1227 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 1228 if (!xchg(&mnt->mnt_expiry_mark, 1) || 1229 atomic_read(&mnt->mnt_count) != 1) 1230 continue; 1231 1232 mntget(mnt); 1233 list_move(&mnt->mnt_expire, &graveyard); 1234 } 1235 1236 expire_mount_list(&graveyard, mounts); 1237 1238 spin_unlock(&vfsmount_lock); 1239 } 1240 1241 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 1242 1243 /* 1244 * Ripoff of 'select_parent()' 1245 * 1246 * search the list of submounts for a given mountpoint, and move any 1247 * shrinkable submounts to the 'graveyard' list. 1248 */ 1249 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 1250 { 1251 struct vfsmount *this_parent = parent; 1252 struct list_head *next; 1253 int found = 0; 1254 1255 repeat: 1256 next = this_parent->mnt_mounts.next; 1257 resume: 1258 while (next != &this_parent->mnt_mounts) { 1259 struct list_head *tmp = next; 1260 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 1261 1262 next = tmp->next; 1263 if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 1264 continue; 1265 /* 1266 * Descend a level if the d_mounts list is non-empty. 1267 */ 1268 if (!list_empty(&mnt->mnt_mounts)) { 1269 this_parent = mnt; 1270 goto repeat; 1271 } 1272 1273 if (!propagate_mount_busy(mnt, 1)) { 1274 mntget(mnt); 1275 list_move_tail(&mnt->mnt_expire, graveyard); 1276 found++; 1277 } 1278 } 1279 /* 1280 * All done at this level ... ascend and resume the search 1281 */ 1282 if (this_parent != parent) { 1283 next = this_parent->mnt_child.next; 1284 this_parent = this_parent->mnt_parent; 1285 goto resume; 1286 } 1287 return found; 1288 } 1289 1290 /* 1291 * process a list of expirable mountpoints with the intent of discarding any 1292 * submounts of a specific parent mountpoint 1293 */ 1294 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts) 1295 { 1296 LIST_HEAD(graveyard); 1297 int found; 1298 1299 spin_lock(&vfsmount_lock); 1300 1301 /* extract submounts of 'mountpoint' from the expiration list */ 1302 while ((found = select_submounts(mountpoint, &graveyard)) != 0) 1303 expire_mount_list(&graveyard, mounts); 1304 1305 spin_unlock(&vfsmount_lock); 1306 } 1307 1308 EXPORT_SYMBOL_GPL(shrink_submounts); 1309 1310 /* 1311 * Some copy_from_user() implementations do not return the exact number of 1312 * bytes remaining to copy on a fault. But copy_mount_options() requires that. 1313 * Note that this function differs from copy_from_user() in that it will oops 1314 * on bad values of `to', rather than returning a short copy. 1315 */ 1316 static long exact_copy_from_user(void *to, const void __user * from, 1317 unsigned long n) 1318 { 1319 char *t = to; 1320 const char __user *f = from; 1321 char c; 1322 1323 if (!access_ok(VERIFY_READ, from, n)) 1324 return n; 1325 1326 while (n) { 1327 if (__get_user(c, f)) { 1328 memset(t, 0, n); 1329 break; 1330 } 1331 *t++ = c; 1332 f++; 1333 n--; 1334 } 1335 return n; 1336 } 1337 1338 int copy_mount_options(const void __user * data, unsigned long *where) 1339 { 1340 int i; 1341 unsigned long page; 1342 unsigned long size; 1343 1344 *where = 0; 1345 if (!data) 1346 return 0; 1347 1348 if (!(page = __get_free_page(GFP_KERNEL))) 1349 return -ENOMEM; 1350 1351 /* We only care that *some* data at the address the user 1352 * gave us is valid. Just in case, we'll zero 1353 * the remainder of the page. 1354 */ 1355 /* copy_from_user cannot cross TASK_SIZE ! */ 1356 size = TASK_SIZE - (unsigned long)data; 1357 if (size > PAGE_SIZE) 1358 size = PAGE_SIZE; 1359 1360 i = size - exact_copy_from_user((void *)page, data, size); 1361 if (!i) { 1362 free_page(page); 1363 return -EFAULT; 1364 } 1365 if (i != PAGE_SIZE) 1366 memset((char *)page + i, 0, PAGE_SIZE - i); 1367 *where = page; 1368 return 0; 1369 } 1370 1371 /* 1372 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 1373 * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 1374 * 1375 * data is a (void *) that can point to any structure up to 1376 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 1377 * information (or be NULL). 1378 * 1379 * Pre-0.97 versions of mount() didn't have a flags word. 1380 * When the flags word was introduced its top half was required 1381 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 1382 * Therefore, if this magic number is present, it carries no information 1383 * and must be discarded. 1384 */ 1385 long do_mount(char *dev_name, char *dir_name, char *type_page, 1386 unsigned long flags, void *data_page) 1387 { 1388 struct nameidata nd; 1389 int retval = 0; 1390 int mnt_flags = 0; 1391 1392 /* Discard magic */ 1393 if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 1394 flags &= ~MS_MGC_MSK; 1395 1396 /* Basic sanity checks */ 1397 1398 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 1399 return -EINVAL; 1400 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 1401 return -EINVAL; 1402 1403 if (data_page) 1404 ((char *)data_page)[PAGE_SIZE - 1] = 0; 1405 1406 /* Separate the per-mountpoint flags */ 1407 if (flags & MS_NOSUID) 1408 mnt_flags |= MNT_NOSUID; 1409 if (flags & MS_NODEV) 1410 mnt_flags |= MNT_NODEV; 1411 if (flags & MS_NOEXEC) 1412 mnt_flags |= MNT_NOEXEC; 1413 if (flags & MS_NOATIME) 1414 mnt_flags |= MNT_NOATIME; 1415 if (flags & MS_NODIRATIME) 1416 mnt_flags |= MNT_NODIRATIME; 1417 1418 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 1419 MS_NOATIME | MS_NODIRATIME); 1420 1421 /* ... and get the mountpoint */ 1422 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 1423 if (retval) 1424 return retval; 1425 1426 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 1427 if (retval) 1428 goto dput_out; 1429 1430 if (flags & MS_REMOUNT) 1431 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 1432 data_page); 1433 else if (flags & MS_BIND) 1434 retval = do_loopback(&nd, dev_name, flags & MS_REC); 1435 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 1436 retval = do_change_type(&nd, flags); 1437 else if (flags & MS_MOVE) 1438 retval = do_move_mount(&nd, dev_name); 1439 else 1440 retval = do_new_mount(&nd, type_page, flags, mnt_flags, 1441 dev_name, data_page); 1442 dput_out: 1443 path_release(&nd); 1444 return retval; 1445 } 1446 1447 /* 1448 * Allocate a new namespace structure and populate it with contents 1449 * copied from the namespace of the passed in task structure. 1450 */ 1451 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs) 1452 { 1453 struct namespace *namespace = tsk->namespace; 1454 struct namespace *new_ns; 1455 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 1456 struct vfsmount *p, *q; 1457 1458 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); 1459 if (!new_ns) 1460 return NULL; 1461 1462 atomic_set(&new_ns->count, 1); 1463 INIT_LIST_HEAD(&new_ns->list); 1464 init_waitqueue_head(&new_ns->poll); 1465 new_ns->event = 0; 1466 1467 down_write(&namespace_sem); 1468 /* First pass: copy the tree topology */ 1469 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, 1470 CL_COPY_ALL | CL_EXPIRE); 1471 if (!new_ns->root) { 1472 up_write(&namespace_sem); 1473 kfree(new_ns); 1474 return NULL; 1475 } 1476 spin_lock(&vfsmount_lock); 1477 list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 1478 spin_unlock(&vfsmount_lock); 1479 1480 /* 1481 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 1482 * as belonging to new namespace. We have already acquired a private 1483 * fs_struct, so tsk->fs->lock is not needed. 1484 */ 1485 p = namespace->root; 1486 q = new_ns->root; 1487 while (p) { 1488 q->mnt_namespace = new_ns; 1489 if (fs) { 1490 if (p == fs->rootmnt) { 1491 rootmnt = p; 1492 fs->rootmnt = mntget(q); 1493 } 1494 if (p == fs->pwdmnt) { 1495 pwdmnt = p; 1496 fs->pwdmnt = mntget(q); 1497 } 1498 if (p == fs->altrootmnt) { 1499 altrootmnt = p; 1500 fs->altrootmnt = mntget(q); 1501 } 1502 } 1503 p = next_mnt(p, namespace->root); 1504 q = next_mnt(q, new_ns->root); 1505 } 1506 up_write(&namespace_sem); 1507 1508 if (rootmnt) 1509 mntput(rootmnt); 1510 if (pwdmnt) 1511 mntput(pwdmnt); 1512 if (altrootmnt) 1513 mntput(altrootmnt); 1514 1515 return new_ns; 1516 } 1517 1518 int copy_namespace(int flags, struct task_struct *tsk) 1519 { 1520 struct namespace *namespace = tsk->namespace; 1521 struct namespace *new_ns; 1522 int err = 0; 1523 1524 if (!namespace) 1525 return 0; 1526 1527 get_namespace(namespace); 1528 1529 if (!(flags & CLONE_NEWNS)) 1530 return 0; 1531 1532 if (!capable(CAP_SYS_ADMIN)) { 1533 err = -EPERM; 1534 goto out; 1535 } 1536 1537 new_ns = dup_namespace(tsk, tsk->fs); 1538 if (!new_ns) { 1539 err = -ENOMEM; 1540 goto out; 1541 } 1542 1543 tsk->namespace = new_ns; 1544 1545 out: 1546 put_namespace(namespace); 1547 return err; 1548 } 1549 1550 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 1551 char __user * type, unsigned long flags, 1552 void __user * data) 1553 { 1554 int retval; 1555 unsigned long data_page; 1556 unsigned long type_page; 1557 unsigned long dev_page; 1558 char *dir_page; 1559 1560 retval = copy_mount_options(type, &type_page); 1561 if (retval < 0) 1562 return retval; 1563 1564 dir_page = getname(dir_name); 1565 retval = PTR_ERR(dir_page); 1566 if (IS_ERR(dir_page)) 1567 goto out1; 1568 1569 retval = copy_mount_options(dev_name, &dev_page); 1570 if (retval < 0) 1571 goto out2; 1572 1573 retval = copy_mount_options(data, &data_page); 1574 if (retval < 0) 1575 goto out3; 1576 1577 lock_kernel(); 1578 retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 1579 flags, (void *)data_page); 1580 unlock_kernel(); 1581 free_page(data_page); 1582 1583 out3: 1584 free_page(dev_page); 1585 out2: 1586 putname(dir_page); 1587 out1: 1588 free_page(type_page); 1589 return retval; 1590 } 1591 1592 /* 1593 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 1594 * It can block. Requires the big lock held. 1595 */ 1596 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, 1597 struct dentry *dentry) 1598 { 1599 struct dentry *old_root; 1600 struct vfsmount *old_rootmnt; 1601 write_lock(&fs->lock); 1602 old_root = fs->root; 1603 old_rootmnt = fs->rootmnt; 1604 fs->rootmnt = mntget(mnt); 1605 fs->root = dget(dentry); 1606 write_unlock(&fs->lock); 1607 if (old_root) { 1608 dput(old_root); 1609 mntput(old_rootmnt); 1610 } 1611 } 1612 1613 /* 1614 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 1615 * It can block. Requires the big lock held. 1616 */ 1617 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, 1618 struct dentry *dentry) 1619 { 1620 struct dentry *old_pwd; 1621 struct vfsmount *old_pwdmnt; 1622 1623 write_lock(&fs->lock); 1624 old_pwd = fs->pwd; 1625 old_pwdmnt = fs->pwdmnt; 1626 fs->pwdmnt = mntget(mnt); 1627 fs->pwd = dget(dentry); 1628 write_unlock(&fs->lock); 1629 1630 if (old_pwd) { 1631 dput(old_pwd); 1632 mntput(old_pwdmnt); 1633 } 1634 } 1635 1636 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) 1637 { 1638 struct task_struct *g, *p; 1639 struct fs_struct *fs; 1640 1641 read_lock(&tasklist_lock); 1642 do_each_thread(g, p) { 1643 task_lock(p); 1644 fs = p->fs; 1645 if (fs) { 1646 atomic_inc(&fs->count); 1647 task_unlock(p); 1648 if (fs->root == old_nd->dentry 1649 && fs->rootmnt == old_nd->mnt) 1650 set_fs_root(fs, new_nd->mnt, new_nd->dentry); 1651 if (fs->pwd == old_nd->dentry 1652 && fs->pwdmnt == old_nd->mnt) 1653 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); 1654 put_fs_struct(fs); 1655 } else 1656 task_unlock(p); 1657 } while_each_thread(g, p); 1658 read_unlock(&tasklist_lock); 1659 } 1660 1661 /* 1662 * pivot_root Semantics: 1663 * Moves the root file system of the current process to the directory put_old, 1664 * makes new_root as the new root file system of the current process, and sets 1665 * root/cwd of all processes which had them on the current root to new_root. 1666 * 1667 * Restrictions: 1668 * The new_root and put_old must be directories, and must not be on the 1669 * same file system as the current process root. The put_old must be 1670 * underneath new_root, i.e. adding a non-zero number of /.. to the string 1671 * pointed to by put_old must yield the same directory as new_root. No other 1672 * file system may be mounted on put_old. After all, new_root is a mountpoint. 1673 * 1674 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 1675 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 1676 * in this situation. 1677 * 1678 * Notes: 1679 * - we don't move root/cwd if they are not at the root (reason: if something 1680 * cared enough to change them, it's probably wrong to force them elsewhere) 1681 * - it's okay to pick a root that isn't the root of a file system, e.g. 1682 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 1683 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 1684 * first. 1685 */ 1686 asmlinkage long sys_pivot_root(const char __user * new_root, 1687 const char __user * put_old) 1688 { 1689 struct vfsmount *tmp; 1690 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; 1691 int error; 1692 1693 if (!capable(CAP_SYS_ADMIN)) 1694 return -EPERM; 1695 1696 lock_kernel(); 1697 1698 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1699 &new_nd); 1700 if (error) 1701 goto out0; 1702 error = -EINVAL; 1703 if (!check_mnt(new_nd.mnt)) 1704 goto out1; 1705 1706 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 1707 if (error) 1708 goto out1; 1709 1710 error = security_sb_pivotroot(&old_nd, &new_nd); 1711 if (error) { 1712 path_release(&old_nd); 1713 goto out1; 1714 } 1715 1716 read_lock(¤t->fs->lock); 1717 user_nd.mnt = mntget(current->fs->rootmnt); 1718 user_nd.dentry = dget(current->fs->root); 1719 read_unlock(¤t->fs->lock); 1720 down_write(&namespace_sem); 1721 mutex_lock(&old_nd.dentry->d_inode->i_mutex); 1722 error = -EINVAL; 1723 if (IS_MNT_SHARED(old_nd.mnt) || 1724 IS_MNT_SHARED(new_nd.mnt->mnt_parent) || 1725 IS_MNT_SHARED(user_nd.mnt->mnt_parent)) 1726 goto out2; 1727 if (!check_mnt(user_nd.mnt)) 1728 goto out2; 1729 error = -ENOENT; 1730 if (IS_DEADDIR(new_nd.dentry->d_inode)) 1731 goto out2; 1732 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) 1733 goto out2; 1734 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) 1735 goto out2; 1736 error = -EBUSY; 1737 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) 1738 goto out2; /* loop, on the same file system */ 1739 error = -EINVAL; 1740 if (user_nd.mnt->mnt_root != user_nd.dentry) 1741 goto out2; /* not a mountpoint */ 1742 if (user_nd.mnt->mnt_parent == user_nd.mnt) 1743 goto out2; /* not attached */ 1744 if (new_nd.mnt->mnt_root != new_nd.dentry) 1745 goto out2; /* not a mountpoint */ 1746 if (new_nd.mnt->mnt_parent == new_nd.mnt) 1747 goto out2; /* not attached */ 1748 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ 1749 spin_lock(&vfsmount_lock); 1750 if (tmp != new_nd.mnt) { 1751 for (;;) { 1752 if (tmp->mnt_parent == tmp) 1753 goto out3; /* already mounted on put_old */ 1754 if (tmp->mnt_parent == new_nd.mnt) 1755 break; 1756 tmp = tmp->mnt_parent; 1757 } 1758 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) 1759 goto out3; 1760 } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) 1761 goto out3; 1762 detach_mnt(new_nd.mnt, &parent_nd); 1763 detach_mnt(user_nd.mnt, &root_parent); 1764 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ 1765 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ 1766 touch_namespace(current->namespace); 1767 spin_unlock(&vfsmount_lock); 1768 chroot_fs_refs(&user_nd, &new_nd); 1769 security_sb_post_pivotroot(&user_nd, &new_nd); 1770 error = 0; 1771 path_release(&root_parent); 1772 path_release(&parent_nd); 1773 out2: 1774 mutex_unlock(&old_nd.dentry->d_inode->i_mutex); 1775 up_write(&namespace_sem); 1776 path_release(&user_nd); 1777 path_release(&old_nd); 1778 out1: 1779 path_release(&new_nd); 1780 out0: 1781 unlock_kernel(); 1782 return error; 1783 out3: 1784 spin_unlock(&vfsmount_lock); 1785 goto out2; 1786 } 1787 1788 static void __init init_mount_tree(void) 1789 { 1790 struct vfsmount *mnt; 1791 struct namespace *namespace; 1792 struct task_struct *g, *p; 1793 1794 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 1795 if (IS_ERR(mnt)) 1796 panic("Can't create rootfs"); 1797 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); 1798 if (!namespace) 1799 panic("Can't allocate initial namespace"); 1800 atomic_set(&namespace->count, 1); 1801 INIT_LIST_HEAD(&namespace->list); 1802 init_waitqueue_head(&namespace->poll); 1803 namespace->event = 0; 1804 list_add(&mnt->mnt_list, &namespace->list); 1805 namespace->root = mnt; 1806 mnt->mnt_namespace = namespace; 1807 1808 init_task.namespace = namespace; 1809 read_lock(&tasklist_lock); 1810 do_each_thread(g, p) { 1811 get_namespace(namespace); 1812 p->namespace = namespace; 1813 } while_each_thread(g, p); 1814 read_unlock(&tasklist_lock); 1815 1816 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); 1817 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); 1818 } 1819 1820 void __init mnt_init(unsigned long mempages) 1821 { 1822 struct list_head *d; 1823 unsigned int nr_hash; 1824 int i; 1825 1826 init_rwsem(&namespace_sem); 1827 1828 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 1829 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); 1830 1831 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 1832 1833 if (!mount_hashtable) 1834 panic("Failed to allocate mount hash table\n"); 1835 1836 /* 1837 * Find the power-of-two list-heads that can fit into the allocation.. 1838 * We don't guarantee that "sizeof(struct list_head)" is necessarily 1839 * a power-of-two. 1840 */ 1841 nr_hash = PAGE_SIZE / sizeof(struct list_head); 1842 hash_bits = 0; 1843 do { 1844 hash_bits++; 1845 } while ((nr_hash >> hash_bits) != 0); 1846 hash_bits--; 1847 1848 /* 1849 * Re-calculate the actual number of entries and the mask 1850 * from the number of bits we can fit. 1851 */ 1852 nr_hash = 1UL << hash_bits; 1853 hash_mask = nr_hash - 1; 1854 1855 printk("Mount-cache hash table entries: %d\n", nr_hash); 1856 1857 /* And initialize the newly allocated array */ 1858 d = mount_hashtable; 1859 i = nr_hash; 1860 do { 1861 INIT_LIST_HEAD(d); 1862 d++; 1863 i--; 1864 } while (i); 1865 sysfs_init(); 1866 subsystem_register(&fs_subsys); 1867 init_rootfs(); 1868 init_mount_tree(); 1869 } 1870 1871 void __put_namespace(struct namespace *namespace) 1872 { 1873 struct vfsmount *root = namespace->root; 1874 LIST_HEAD(umount_list); 1875 namespace->root = NULL; 1876 spin_unlock(&vfsmount_lock); 1877 down_write(&namespace_sem); 1878 spin_lock(&vfsmount_lock); 1879 umount_tree(root, 0, &umount_list); 1880 spin_unlock(&vfsmount_lock); 1881 up_write(&namespace_sem); 1882 release_mounts(&umount_list); 1883 kfree(namespace); 1884 } 1885