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