fs/namespace.c

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