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/config.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/quotaops.h>
  17 #include <linux/acct.h>
  18 #include <linux/module.h>
  19 #include <linux/seq_file.h>
  20 #include <linux/namespace.h>
  21 #include <linux/namei.h>
  22 #include <linux/security.h>
  23 #include <linux/mount.h>
  24 #include <asm/uaccess.h>
  25
  26 extern int __init init_rootfs(void);
  27 extern int __init sysfs_init(void);
  28
  29 static struct list_head *mount_hashtable;
  30 static int hash_mask, hash_bits;
  31 static kmem_cache_t *mnt_cache;
  32
  33 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  34 {
  35         unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
  36         tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
  37         tmp = tmp + (tmp >> hash_bits);
  38         return tmp & hash_mask;
  39 }
  40
  41 struct vfsmount *alloc_vfsmnt(const char *name)
  42 {
  43         struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
  44         if (mnt) {
  45                 memset(mnt, 0, sizeof(struct vfsmount));
  46                 atomic_set(&mnt->mnt_count,1);
  47                 INIT_LIST_HEAD(&mnt->mnt_hash);
  48                 INIT_LIST_HEAD(&mnt->mnt_child);
  49                 INIT_LIST_HEAD(&mnt->mnt_mounts);
  50                 INIT_LIST_HEAD(&mnt->mnt_list);
  51                 if (name) {
  52                         int size = strlen(name)+1;
  53                         char *newname = kmalloc(size, GFP_KERNEL);
  54                         if (newname) {
  55                                 memcpy(newname, name, size);
  56                                 mnt->mnt_devname = newname;
  57                         }
  58                 }
  59         }
  60         return mnt;
  61 }
  62
  63 void free_vfsmnt(struct vfsmount *mnt)
  64 {
  65         kfree(mnt->mnt_devname);
  66         kmem_cache_free(mnt_cache, mnt);
  67 }
  68
  69 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
  70 {
  71         struct list_head * head = mount_hashtable + hash(mnt, dentry);
  72         struct list_head * tmp = head;
  73         struct vfsmount *p;
  74
  75         for (;;) {
  76                 tmp = tmp->next;
  77                 p = NULL;
  78                 if (tmp == head)
  79                         break;
  80                 p = list_entry(tmp, struct vfsmount, mnt_hash);
  81                 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry)
  82                         break;
  83         }
  84         return p;
  85 }
  86
  87 static int check_mnt(struct vfsmount *mnt)
  88 {
  89         spin_lock(&dcache_lock);
  90         while (mnt->mnt_parent != mnt)
  91                 mnt = mnt->mnt_parent;
  92         spin_unlock(&dcache_lock);
  93         return mnt == current->namespace->root;
  94 }
  95
  96 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
  97 {
  98         old_nd->dentry = mnt->mnt_mountpoint;
  99         old_nd->mnt = mnt->mnt_parent;
 100         mnt->mnt_parent = mnt;
 101         mnt->mnt_mountpoint = mnt->mnt_root;
 102         list_del_init(&mnt->mnt_child);
 103         list_del_init(&mnt->mnt_hash);
 104         old_nd->dentry->d_mounted--;
 105 }
 106
 107 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
 108 {
 109         mnt->mnt_parent = mntget(nd->mnt);
 110         mnt->mnt_mountpoint = dget(nd->dentry);
 111         list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
 112         list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
 113         nd->dentry->d_mounted++;
 114 }
 115
 116 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
 117 {
 118         struct list_head *next = p->mnt_mounts.next;
 119         if (next == &p->mnt_mounts) {
 120                 while (1) {
 121                         if (p == root)
 122                                 return NULL;
 123                         next = p->mnt_child.next;
 124                         if (next != &p->mnt_parent->mnt_mounts)
 125                                 break;
 126                         p = p->mnt_parent;
 127                 }
 128         }
 129         return list_entry(next, struct vfsmount, mnt_child);
 130 }
 131
 132 static struct vfsmount *
 133 clone_mnt(struct vfsmount *old, struct dentry *root)
 134 {
 135         struct super_block *sb = old->mnt_sb;
 136         struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
 137
 138         if (mnt) {
 139                 mnt->mnt_flags = old->mnt_flags;
 140                 atomic_inc(&sb->s_active);
 141                 mnt->mnt_sb = sb;
 142                 mnt->mnt_root = dget(root);
 143                 mnt->mnt_mountpoint = mnt->mnt_root;
 144                 mnt->mnt_parent = mnt;
 145         }
 146         return mnt;
 147 }
 148
 149 void __mntput(struct vfsmount *mnt)
 150 {
 151         struct super_block *sb = mnt->mnt_sb;
 152         dput(mnt->mnt_root);
 153         free_vfsmnt(mnt);
 154         deactivate_super(sb);
 155 }
 156
 157 /* iterator */
 158 static void *m_start(struct seq_file *m, loff_t *pos)
 159 {
 160         struct namespace *n = m->private;
 161         struct list_head *p;
 162         loff_t l = *pos;
 163
 164         down_read(&n->sem);
 165         list_for_each(p, &n->list)
 166                 if (!l--)
 167                         return list_entry(p, struct vfsmount, mnt_list);
 168         return NULL;
 169 }
 170
 171 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 172 {
 173         struct namespace *n = m->private;
 174         struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
 175         (*pos)++;
 176         return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
 177 }
 178
 179 static void m_stop(struct seq_file *m, void *v)
 180 {
 181         struct namespace *n = m->private;
 182         up_read(&n->sem);
 183 }
 184
 185 static inline void mangle(struct seq_file *m, const char *s)
 186 {
 187         seq_escape(m, s, " \t\n\\");
 188 }
 189
 190 static int show_vfsmnt(struct seq_file *m, void *v)
 191 {
 192         struct vfsmount *mnt = v;
 193         int err = 0;
 194         static struct proc_fs_info {
 195                 int flag;
 196                 char *str;
 197         } fs_info[] = {
 198                 { MS_SYNCHRONOUS, ",sync" },
 199                 { MS_DIRSYNC, ",dirsync" },
 200                 { MS_MANDLOCK, ",mand" },
 201                 { MS_NOATIME, ",noatime" },
 202                 { MS_NODIRATIME, ",nodiratime" },
 203                 { 0, NULL }
 204         };
 205         static struct proc_fs_info mnt_info[] = {
 206                 { MNT_NOSUID, ",nosuid" },
 207                 { MNT_NODEV, ",nodev" },
 208                 { MNT_NOEXEC, ",noexec" },
 209                 { 0, NULL }
 210         };
 211         struct proc_fs_info *fs_infop;
 212
 213         mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 214         seq_putc(m, ' ');
 215         seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
 216         seq_putc(m, ' ');
 217         mangle(m, mnt->mnt_sb->s_type->name);
 218         seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
 219         for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
 220                 if (mnt->mnt_sb->s_flags & fs_infop->flag)
 221                         seq_puts(m, fs_infop->str);
 222         }
 223         for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
 224                 if (mnt->mnt_flags & fs_infop->flag)
 225                         seq_puts(m, fs_infop->str);
 226         }
 227         if (mnt->mnt_sb->s_op->show_options)
 228                 err = mnt->mnt_sb->s_op->show_options(m, mnt);
 229         seq_puts(m, " 0 0\n");
 230         return err;
 231 }
 232
 233 struct seq_operations mounts_op = {
 234         .start  = m_start,
 235         .next   = m_next,
 236         .stop   = m_stop,
 237         .show   = show_vfsmnt
 238 };
 239
 240 /*
 241  * Doesn't take quota and stuff into account. IOW, in some cases it will
 242  * give false negatives. The main reason why it's here is that we need
 243  * a non-destructive way to look for easily umountable filesystems.
 244  */
 245 int may_umount(struct vfsmount *mnt)
 246 {
 247         if (atomic_read(&mnt->mnt_count) > 2)
 248                 return -EBUSY;
 249         return 0;
 250 }
 251
 252 void umount_tree(struct vfsmount *mnt)
 253 {
 254         struct vfsmount *p;
 255         LIST_HEAD(kill);
 256
 257         for (p = mnt; p; p = next_mnt(p, mnt)) {
 258                 list_del(&p->mnt_list);
 259                 list_add(&p->mnt_list, &kill);
 260         }
 261
 262         while (!list_empty(&kill)) {
 263                 mnt = list_entry(kill.next, struct vfsmount, mnt_list);
 264                 list_del_init(&mnt->mnt_list);
 265                 if (mnt->mnt_parent == mnt) {
 266                         spin_unlock(&dcache_lock);
 267                 } else {
 268                         struct nameidata old_nd;
 269                         detach_mnt(mnt, &old_nd);
 270                         spin_unlock(&dcache_lock);
 271                         path_release(&old_nd);
 272                 }
 273                 mntput(mnt);
 274                 spin_lock(&dcache_lock);
 275         }
 276 }
 277
 278 static int do_umount(struct vfsmount *mnt, int flags)
 279 {
 280         struct super_block * sb = mnt->mnt_sb;
 281         int retval;
 282
 283         retval = security_sb_umount(mnt, flags);
 284         if (retval)
 285                 return retval;
 286
 287         /*
 288          * If we may have to abort operations to get out of this
 289          * mount, and they will themselves hold resources we must
 290          * allow the fs to do things. In the Unix tradition of
 291          * 'Gee thats tricky lets do it in userspace' the umount_begin
 292          * might fail to complete on the first run through as other tasks
 293          * must return, and the like. Thats for the mount program to worry
 294          * about for the moment.
 295          */
 296
 297         lock_kernel();
 298         if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
 299                 sb->s_op->umount_begin(sb);
 300         unlock_kernel();
 301
 302         /*
 303          * No sense to grab the lock for this test, but test itself looks
 304          * somewhat bogus. Suggestions for better replacement?
 305          * Ho-hum... In principle, we might treat that as umount + switch
 306          * to rootfs. GC would eventually take care of the old vfsmount.
 307          * Actually it makes sense, especially if rootfs would contain a
 308          * /reboot - static binary that would close all descriptors and
 309          * call reboot(9). Then init(8) could umount root and exec /reboot.
 310          */
 311         if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
 312                 /*
 313                  * Special case for "unmounting" root ...
 314                  * we just try to remount it readonly.
 315                  */
 316                 down_write(&sb->s_umount);
 317                 if (!(sb->s_flags & MS_RDONLY)) {
 318                         lock_kernel();
 319                         retval = do_remount_sb(sb, MS_RDONLY, 0, 0);
 320                         unlock_kernel();
 321                 }
 322                 up_write(&sb->s_umount);
 323                 return retval;
 324         }
 325
 326         down_write(&current->namespace->sem);
 327         spin_lock(&dcache_lock);
 328
 329         if (atomic_read(&sb->s_active) == 1) {
 330                 /* last instance - try to be smart */
 331                 spin_unlock(&dcache_lock);
 332                 lock_kernel();
 333                 DQUOT_OFF(sb);
 334                 acct_auto_close(sb);
 335                 unlock_kernel();
 336                 security_sb_umount_close(mnt);
 337                 spin_lock(&dcache_lock);
 338         }
 339         retval = -EBUSY;
 340         if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
 341                 if (!list_empty(&mnt->mnt_list))
 342                         umount_tree(mnt);
 343                 retval = 0;
 344         }
 345         spin_unlock(&dcache_lock);
 346         if (retval)
 347                 security_sb_umount_busy(mnt);
 348         up_write(&current->namespace->sem);
 349         return retval;
 350 }
 351
 352 /*
 353  * Now umount can handle mount points as well as block devices.
 354  * This is important for filesystems which use unnamed block devices.
 355  *
 356  * We now support a flag for forced unmount like the other 'big iron'
 357  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
 358  */
 359
 360 asmlinkage long sys_umount(char __user * name, int flags)
 361 {
 362         struct nameidata nd;
 363         int retval;
 364
 365         retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
 366         if (retval)
 367                 goto out;
 368         retval = -EINVAL;
 369         if (nd.dentry != nd.mnt->mnt_root)
 370                 goto dput_and_out;
 371         if (!check_mnt(nd.mnt))
 372                 goto dput_and_out;
 373
 374         retval = -EPERM;
 375         if (!capable(CAP_SYS_ADMIN))
 376                 goto dput_and_out;
 377
 378         retval = do_umount(nd.mnt, flags);
 379 dput_and_out:
 380         path_release(&nd);
 381 out:
 382         return retval;
 383 }
 384
 385 /*
 386  *      The 2.0 compatible umount. No flags.
 387  */
 388
 389 asmlinkage long sys_oldumount(char __user * name)
 390 {
 391         return sys_umount(name,0);
 392 }
 393
 394 static int mount_is_safe(struct nameidata *nd)
 395 {
 396         if (capable(CAP_SYS_ADMIN))
 397                 return 0;
 398         return -EPERM;
 399 #ifdef notyet
 400         if (S_ISLNK(nd->dentry->d_inode->i_mode))
 401                 return -EPERM;
 402         if (nd->dentry->d_inode->i_mode & S_ISVTX) {
 403                 if (current->uid != nd->dentry->d_inode->i_uid)
 404                         return -EPERM;
 405         }
 406         if (permission(nd->dentry->d_inode, MAY_WRITE))
 407                 return -EPERM;
 408         return 0;
 409 #endif
 410 }
 411
 412 static int
 413 lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
 414 {
 415         while (1) {
 416                 if (d == dentry)
 417                         return 1;
 418                 if (d == NULL || d == d->d_parent)
 419                         return 0;
 420                 d = d->d_parent;
 421         }
 422 }
 423
 424 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
 425 {
 426         struct vfsmount *res, *p, *q, *r, *s;
 427         struct list_head *h;
 428         struct nameidata nd;
 429
 430         res = q = clone_mnt(mnt, dentry);
 431         if (!q)
 432                 goto Enomem;
 433         q->mnt_mountpoint = mnt->mnt_mountpoint;
 434
 435         p = mnt;
 436         for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) {
 437                 r = list_entry(h, struct vfsmount, mnt_child);
 438                 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
 439                         continue;
 440
 441                 for (s = r; s; s = next_mnt(s, r)) {
 442                         while (p != s->mnt_parent) {
 443                                 p = p->mnt_parent;
 444                                 q = q->mnt_parent;
 445                         }
 446                         p = s;
 447                         nd.mnt = q;
 448                         nd.dentry = p->mnt_mountpoint;
 449                         q = clone_mnt(p, p->mnt_root);
 450                         if (!q)
 451                                 goto Enomem;
 452                         spin_lock(&dcache_lock);
 453                         list_add_tail(&q->mnt_list, &res->mnt_list);
 454                         attach_mnt(q, &nd);
 455                         spin_unlock(&dcache_lock);
 456                 }
 457         }
 458         return res;
 459  Enomem:
 460         if (res) {
 461                 spin_lock(&dcache_lock);
 462                 umount_tree(res);
 463                 spin_unlock(&dcache_lock);
 464         }
 465         return NULL;
 466 }
 467
 468 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
 469 {
 470         int err;
 471         if (mnt->mnt_sb->s_flags & MS_NOUSER)
 472                 return -EINVAL;
 473
 474         if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
 475               S_ISDIR(mnt->mnt_root->d_inode->i_mode))
 476                 return -ENOTDIR;
 477
 478         err = -ENOENT;
 479         down(&nd->dentry->d_inode->i_sem);
 480         if (IS_DEADDIR(nd->dentry->d_inode))
 481                 goto out_unlock;
 482
 483         err = security_sb_check_sb(mnt, nd);
 484         if (err)
 485                 goto out_unlock;
 486
 487         err = -ENOENT;
 488         spin_lock(&dcache_lock);
 489         if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
 490                 struct list_head head;
 491
 492                 attach_mnt(mnt, nd);
 493                 list_add_tail(&head, &mnt->mnt_list);
 494                 list_splice(&head, current->namespace->list.prev);
 495                 mntget(mnt);
 496                 err = 0;
 497         }
 498         spin_unlock(&dcache_lock);
 499 out_unlock:
 500         up(&nd->dentry->d_inode->i_sem);
 501         if (!err)
 502                 security_sb_post_addmount(mnt, nd);
 503         return err;
 504 }
 505
 506 /*
 507  * do loopback mount.
 508  */
 509 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
 510 {
 511         struct nameidata old_nd;
 512         struct vfsmount *mnt = NULL;
 513         int err = mount_is_safe(nd);
 514         if (err)
 515                 return err;
 516         if (!old_name || !*old_name)
 517                 return -EINVAL;
 518         err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
 519         if (err)
 520                 return err;
 521
 522         down_write(&current->namespace->sem);
 523         err = -EINVAL;
 524         if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
 525                 err = -ENOMEM;
 526                 if (recurse)
 527                         mnt = copy_tree(old_nd.mnt, old_nd.dentry);
 528                 else
 529                         mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
 530         }
 531
 532         if (mnt) {
 533                 err = graft_tree(mnt, nd);
 534                 if (err) {
 535                         spin_lock(&dcache_lock);
 536                         umount_tree(mnt);
 537                         spin_unlock(&dcache_lock);
 538                 } else
 539                         mntput(mnt);
 540         }
 541
 542         up_write(&current->namespace->sem);
 543         path_release(&old_nd);
 544         return err;
 545 }
 546
 547 /*
 548  * change filesystem flags. dir should be a physical root of filesystem.
 549  * If you've mounted a non-root directory somewhere and want to do remount
 550  * on it - tough luck.
 551  */
 552
 553 static int do_remount(struct nameidata *nd,int flags,int mnt_flags,void *data)
 554 {
 555         int err;
 556         struct super_block * sb = nd->mnt->mnt_sb;
 557
 558         if (!capable(CAP_SYS_ADMIN))
 559                 return -EPERM;
 560
 561         if (!check_mnt(nd->mnt))
 562                 return -EINVAL;
 563
 564         if (nd->dentry != nd->mnt->mnt_root)
 565                 return -EINVAL;
 566
 567         down_write(&sb->s_umount);
 568         err = do_remount_sb(sb, flags, data, 0);
 569         if (!err)
 570                 nd->mnt->mnt_flags=mnt_flags;
 571         up_write(&sb->s_umount);
 572         if (!err)
 573                 security_sb_post_remount(nd->mnt, flags, data);
 574         return err;
 575 }
 576
 577 static int do_move_mount(struct nameidata *nd, char *old_name)
 578 {
 579         struct nameidata old_nd, parent_nd;
 580         struct vfsmount *p;
 581         int err = 0;
 582         if (!capable(CAP_SYS_ADMIN))
 583                 return -EPERM;
 584         if (!old_name || !*old_name)
 585                 return -EINVAL;
 586         err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
 587         if (err)
 588                 return err;
 589
 590         down_write(&current->namespace->sem);
 591         while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
 592                 ;
 593         err = -EINVAL;
 594         if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
 595                 goto out;
 596
 597         err = -ENOENT;
 598         down(&nd->dentry->d_inode->i_sem);
 599         if (IS_DEADDIR(nd->dentry->d_inode))
 600                 goto out1;
 601
 602         spin_lock(&dcache_lock);
 603         if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
 604                 goto out2;
 605
 606         err = -EINVAL;
 607         if (old_nd.dentry != old_nd.mnt->mnt_root)
 608                 goto out2;
 609
 610         if (old_nd.mnt == old_nd.mnt->mnt_parent)
 611                 goto out2;
 612
 613         if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
 614               S_ISDIR(old_nd.dentry->d_inode->i_mode))
 615                 goto out2;
 616
 617         err = -ELOOP;
 618         for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
 619                 if (p == old_nd.mnt)
 620                         goto out2;
 621         err = 0;
 622
 623         detach_mnt(old_nd.mnt, &parent_nd);
 624         attach_mnt(old_nd.mnt, nd);
 625 out2:
 626         spin_unlock(&dcache_lock);
 627 out1:
 628         up(&nd->dentry->d_inode->i_sem);
 629 out:
 630         up_write(&current->namespace->sem);
 631         if (!err)
 632                 path_release(&parent_nd);
 633         path_release(&old_nd);
 634         return err;
 635 }
 636
 637 static int do_add_mount(struct nameidata *nd, char *type, int flags,
 638                         int mnt_flags, char *name, void *data)
 639 {
 640         struct vfsmount *mnt;
 641         int err;
 642
 643         if (!type || !memchr(type, 0, PAGE_SIZE))
 644                 return -EINVAL;
 645
 646         /* we need capabilities... */
 647         if (!capable(CAP_SYS_ADMIN))
 648                 return -EPERM;
 649
 650         mnt = do_kern_mount(type, flags, name, data);
 651         err = PTR_ERR(mnt);
 652         if (IS_ERR(mnt))
 653                 goto out;
 654
 655         down_write(&current->namespace->sem);
 656         /* Something was mounted here while we slept */
 657         while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
 658                 ;
 659         err = -EINVAL;
 660         if (!check_mnt(nd->mnt))
 661                 goto unlock;
 662
 663         /* Refuse the same filesystem on the same mount point */
 664         err = -EBUSY;
 665         if (nd->mnt->mnt_sb == mnt->mnt_sb && nd->mnt->mnt_root == nd->dentry)
 666                 goto unlock;
 667
 668         mnt->mnt_flags = mnt_flags;
 669         err = graft_tree(mnt, nd);
 670 unlock:
 671         up_write(&current->namespace->sem);
 672         mntput(mnt);
 673 out:
 674         return err;
 675 }
 676
 677 static int copy_mount_options (const void __user *data, unsigned long *where)
 678 {
 679         int i;
 680         unsigned long page;
 681         unsigned long size;
 682
 683         *where = 0;
 684         if (!data)
 685                 return 0;
 686
 687         if (!(page = __get_free_page(GFP_KERNEL)))
 688                 return -ENOMEM;
 689
 690         /* We only care that *some* data at the address the user
 691          * gave us is valid.  Just in case, we'll zero
 692          * the remainder of the page.
 693          */
 694         /* copy_from_user cannot cross TASK_SIZE ! */
 695         size = TASK_SIZE - (unsigned long)data;
 696         if (size > PAGE_SIZE)
 697                 size = PAGE_SIZE;
 698
 699         i = size - copy_from_user((void *)page, data, size);
 700         if (!i) {
 701                 free_page(page);
 702                 return -EFAULT;
 703         }
 704         if (i != PAGE_SIZE)
 705                 memset((char *)page + i, 0, PAGE_SIZE - i);
 706         *where = page;
 707         return 0;
 708 }
 709
 710 /*
 711  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
 712  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
 713  *
 714  * data is a (void *) that can point to any structure up to
 715  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
 716  * information (or be NULL).
 717  *
 718  * Pre-0.97 versions of mount() didn't have a flags word.
 719  * When the flags word was introduced its top half was required
 720  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
 721  * Therefore, if this magic number is present, it carries no information
 722  * and must be discarded.
 723  */
 724 long do_mount(char * dev_name, char * dir_name, char *type_page,
 725                   unsigned long flags, void *data_page)
 726 {
 727         struct nameidata nd;
 728         int retval = 0;
 729         int mnt_flags = 0;
 730
 731         /* Discard magic */
 732         if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
 733                 flags &= ~MS_MGC_MSK;
 734
 735         /* Basic sanity checks */
 736
 737         if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
 738                 return -EINVAL;
 739         if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
 740                 return -EINVAL;
 741
 742         /* Separate the per-mountpoint flags */
 743         if (flags & MS_NOSUID)
 744                 mnt_flags |= MNT_NOSUID;
 745         if (flags & MS_NODEV)
 746                 mnt_flags |= MNT_NODEV;
 747         if (flags & MS_NOEXEC)
 748                 mnt_flags |= MNT_NOEXEC;
 749         flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV);
 750
 751         /* ... and get the mountpoint */
 752         retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
 753         if (retval)
 754                 return retval;
 755
 756         retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
 757         if (retval)
 758                 goto dput_out;
 759
 760         if (flags & MS_REMOUNT)
 761                 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
 762                                     data_page);
 763         else if (flags & MS_BIND)
 764                 retval = do_loopback(&nd, dev_name, flags & MS_REC);
 765         else if (flags & MS_MOVE)
 766                 retval = do_move_mount(&nd, dev_name);
 767         else
 768                 retval = do_add_mount(&nd, type_page, flags, mnt_flags,
 769                                       dev_name, data_page);
 770 dput_out:
 771         path_release(&nd);
 772         return retval;
 773 }
 774
 775 int copy_namespace(int flags, struct task_struct *tsk)
 776 {
 777         struct namespace *namespace = tsk->namespace;
 778         struct namespace *new_ns;
 779         struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
 780         struct fs_struct *fs = tsk->fs;
 781
 782         if (!namespace)
 783                 return 0;
 784
 785         get_namespace(namespace);
 786
 787         if (!(flags & CLONE_NEWNS))
 788                 return 0;
 789
 790         if (!capable(CAP_SYS_ADMIN)) {
 791                 put_namespace(namespace);
 792                 return -EPERM;
 793         }
 794
 795         new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
 796         if (!new_ns)
 797                 goto out;
 798
 799         atomic_set(&new_ns->count, 1);
 800         init_rwsem(&new_ns->sem);
 801         new_ns->root = NULL;
 802         INIT_LIST_HEAD(&new_ns->list);
 803
 804         down_write(&tsk->namespace->sem);
 805         /* First pass: copy the tree topology */
 806         new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
 807         spin_lock(&dcache_lock);
 808         list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
 809         spin_unlock(&dcache_lock);
 810
 811         /* Second pass: switch the tsk->fs->* elements */
 812         if (fs) {
 813                 struct vfsmount *p, *q;
 814                 write_lock(&fs->lock);
 815
 816                 p = namespace->root;
 817                 q = new_ns->root;
 818                 while (p) {
 819                         if (p == fs->rootmnt) {
 820                                 rootmnt = p;
 821                                 fs->rootmnt = mntget(q);
 822                         }
 823                         if (p == fs->pwdmnt) {
 824                                 pwdmnt = p;
 825                                 fs->pwdmnt = mntget(q);
 826                         }
 827                         if (p == fs->altrootmnt) {
 828                                 altrootmnt = p;
 829                                 fs->altrootmnt = mntget(q);
 830                         }
 831                         p = next_mnt(p, namespace->root);
 832                         q = next_mnt(q, new_ns->root);
 833                 }
 834                 write_unlock(&fs->lock);
 835         }
 836         up_write(&tsk->namespace->sem);
 837
 838         tsk->namespace = new_ns;
 839
 840         if (rootmnt)
 841                 mntput(rootmnt);
 842         if (pwdmnt)
 843                 mntput(pwdmnt);
 844         if (altrootmnt)
 845                 mntput(altrootmnt);
 846
 847         put_namespace(namespace);
 848         return 0;
 849
 850 out:
 851         put_namespace(namespace);
 852         return -ENOMEM;
 853 }
 854
 855 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
 856                           char __user * type, unsigned long flags,
 857                           void __user * data)
 858 {
 859         int retval;
 860         unsigned long data_page;
 861         unsigned long type_page;
 862         unsigned long dev_page;
 863         char *dir_page;
 864
 865         retval = copy_mount_options (type, &type_page);
 866         if (retval < 0)
 867                 return retval;
 868
 869         dir_page = getname(dir_name);
 870         retval = PTR_ERR(dir_page);
 871         if (IS_ERR(dir_page))
 872                 goto out1;
 873
 874         retval = copy_mount_options (dev_name, &dev_page);
 875         if (retval < 0)
 876                 goto out2;
 877
 878         retval = copy_mount_options (data, &data_page);
 879         if (retval < 0)
 880                 goto out3;
 881
 882         lock_kernel();
 883         retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
 884                           flags, (void*)data_page);
 885         unlock_kernel();
 886         free_page(data_page);
 887
 888 out3:
 889         free_page(dev_page);
 890 out2:
 891         putname(dir_page);
 892 out1:
 893         free_page(type_page);
 894         return retval;
 895 }
 896
 897 /*
 898  * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
 899  * It can block. Requires the big lock held.
 900  */
 901 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
 902                  struct dentry *dentry)
 903 {
 904         struct dentry *old_root;
 905         struct vfsmount *old_rootmnt;
 906         write_lock(&fs->lock);
 907         old_root = fs->root;
 908         old_rootmnt = fs->rootmnt;
 909         fs->rootmnt = mntget(mnt);
 910         fs->root = dget(dentry);
 911         write_unlock(&fs->lock);
 912         if (old_root) {
 913                 dput(old_root);
 914                 mntput(old_rootmnt);
 915         }
 916 }
 917
 918 /*
 919  * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
 920  * It can block. Requires the big lock held.
 921  */
 922 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
 923                 struct dentry *dentry)
 924 {
 925         struct dentry *old_pwd;
 926         struct vfsmount *old_pwdmnt;
 927
 928         write_lock(&fs->lock);
 929         old_pwd = fs->pwd;
 930         old_pwdmnt = fs->pwdmnt;
 931         fs->pwdmnt = mntget(mnt);
 932         fs->pwd = dget(dentry);
 933         write_unlock(&fs->lock);
 934
 935         if (old_pwd) {
 936                 dput(old_pwd);
 937                 mntput(old_pwdmnt);
 938         }
 939 }
 940
 941 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
 942 {
 943         struct task_struct *g, *p;
 944         struct fs_struct *fs;
 945
 946         read_lock(&tasklist_lock);
 947         do_each_thread(g, p) {
 948                 task_lock(p);
 949                 fs = p->fs;
 950                 if (fs) {
 951                         atomic_inc(&fs->count);
 952                         task_unlock(p);
 953                         if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
 954                                 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
 955                         if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
 956                                 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
 957                         put_fs_struct(fs);
 958                 } else
 959                         task_unlock(p);
 960         } while_each_thread(g, p);
 961         read_unlock(&tasklist_lock);
 962 }
 963
 964 /*
 965  * Moves the current root to put_root, and sets root/cwd of all processes
 966  * which had them on the old root to new_root.
 967  *
 968  * Note:
 969  *  - we don't move root/cwd if they are not at the root (reason: if something
 970  *    cared enough to change them, it's probably wrong to force them elsewhere)
 971  *  - it's okay to pick a root that isn't the root of a file system, e.g.
 972  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
 973  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
 974  *    first.
 975  */
 976
 977 asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
 978 {
 979         struct vfsmount *tmp;
 980         struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
 981         int error;
 982
 983         if (!capable(CAP_SYS_ADMIN))
 984                 return -EPERM;
 985
 986         lock_kernel();
 987
 988         error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
 989         if (error)
 990                 goto out0;
 991         error = -EINVAL;
 992         if (!check_mnt(new_nd.mnt))
 993                 goto out1;
 994
 995         error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
 996         if (error)
 997                 goto out1;
 998
 999         error = security_sb_pivotroot(&old_nd, &new_nd);
1000         if (error) {
1001                 path_release(&old_nd);
1002                 goto out1;
1003         }
1004
1005         read_lock(&current->fs->lock);
1006         user_nd.mnt = mntget(current->fs->rootmnt);
1007         user_nd.dentry = dget(current->fs->root);
1008         read_unlock(&current->fs->lock);
1009         down_write(&current->namespace->sem);
1010         down(&old_nd.dentry->d_inode->i_sem);
1011         error = -EINVAL;
1012         if (!check_mnt(user_nd.mnt))
1013                 goto out2;
1014         error = -ENOENT;
1015         if (IS_DEADDIR(new_nd.dentry->d_inode))
1016                 goto out2;
1017         if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1018                 goto out2;
1019         if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1020                 goto out2;
1021         error = -EBUSY;
1022         if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1023                 goto out2; /* loop */
1024         error = -EINVAL;
1025         if (user_nd.mnt->mnt_root != user_nd.dentry)
1026                 goto out2;
1027         if (new_nd.mnt->mnt_root != new_nd.dentry)
1028                 goto out2; /* not a mountpoint */
1029         tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1030         spin_lock(&dcache_lock);
1031         if (tmp != new_nd.mnt) {
1032                 for (;;) {
1033                         if (tmp->mnt_parent == tmp)
1034                                 goto out3;
1035                         if (tmp->mnt_parent == new_nd.mnt)
1036                                 break;
1037                         tmp = tmp->mnt_parent;
1038                 }
1039                 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1040                         goto out3;
1041         } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1042                 goto out3;
1043         detach_mnt(new_nd.mnt, &parent_nd);
1044         detach_mnt(user_nd.mnt, &root_parent);
1045         attach_mnt(user_nd.mnt, &old_nd);
1046         attach_mnt(new_nd.mnt, &root_parent);
1047         spin_unlock(&dcache_lock);
1048         chroot_fs_refs(&user_nd, &new_nd);
1049         security_sb_post_pivotroot(&user_nd, &new_nd);
1050         error = 0;
1051         path_release(&root_parent);
1052         path_release(&parent_nd);
1053 out2:
1054         up(&old_nd.dentry->d_inode->i_sem);
1055         up_write(&current->namespace->sem);
1056         path_release(&user_nd);
1057         path_release(&old_nd);
1058 out1:
1059         path_release(&new_nd);
1060 out0:
1061         unlock_kernel();
1062         return error;
1063 out3:
1064         spin_unlock(&dcache_lock);
1065         goto out2;
1066 }
1067
1068 static void __init init_mount_tree(void)
1069 {
1070         struct vfsmount *mnt;
1071         struct namespace *namespace;
1072         struct task_struct *g, *p;
1073
1074         mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1075         if (IS_ERR(mnt))
1076                 panic("Can't create rootfs");
1077         namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1078         if (!namespace)
1079                 panic("Can't allocate initial namespace");
1080         atomic_set(&namespace->count, 1);
1081         INIT_LIST_HEAD(&namespace->list);
1082         init_rwsem(&namespace->sem);
1083         list_add(&mnt->mnt_list, &namespace->list);
1084         namespace->root = mnt;
1085
1086         init_task.namespace = namespace;
1087         read_lock(&tasklist_lock);
1088         do_each_thread(g, p) {
1089                 get_namespace(namespace);
1090                 p->namespace = namespace;
1091         } while_each_thread(g, p);
1092         read_unlock(&tasklist_lock);
1093
1094         set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1095         set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1096 }
1097
1098 void __init mnt_init(unsigned long mempages)
1099 {
1100         struct list_head *d;
1101         unsigned long order;
1102         unsigned int nr_hash;
1103         int i;
1104
1105         mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1106                                         0, SLAB_HWCACHE_ALIGN, NULL, NULL);
1107         if (!mnt_cache)
1108                 panic("Cannot create vfsmount cache");
1109
1110         order = 0;
1111         mount_hashtable = (struct list_head *)
1112                 __get_free_pages(GFP_ATOMIC, order);
1113
1114         if (!mount_hashtable)
1115                 panic("Failed to allocate mount hash table\n");
1116
1117         /*
1118          * Find the power-of-two list-heads that can fit into the allocation..
1119          * We don't guarantee that "sizeof(struct list_head)" is necessarily
1120          * a power-of-two.
1121          */
1122         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct list_head);
1123         hash_bits = 0;
1124         do {
1125                 hash_bits++;
1126         } while ((nr_hash >> hash_bits) != 0);
1127         hash_bits--;
1128
1129         /*
1130          * Re-calculate the actual number of entries and the mask
1131          * from the number of bits we can fit.
1132          */
1133         nr_hash = 1UL << hash_bits;
1134         hash_mask = nr_hash-1;
1135
1136         printk("Mount-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1137                         nr_hash, order, (PAGE_SIZE << order));
1138
1139         /* And initialize the newly allocated array */
1140         d = mount_hashtable;
1141         i = nr_hash;
1142         do {
1143                 INIT_LIST_HEAD(d);
1144                 d++;
1145                 i--;
1146         } while (i);
1147         sysfs_init();
1148         init_rootfs();
1149         init_mount_tree();
1150 }