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