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