fs/namei.c

   1 /*
   2  *  linux/fs/namei.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * Some corrections by tytso.
   9  */
  10
  11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12  * lookup logic.
  13  */
  14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15  */
  16
  17 #include <linux/init.h>
  18 #include <linux/module.h>
  19 #include <linux/slab.h>
  20 #include <linux/fs.h>
  21 #include <linux/namei.h>
  22 #include <linux/quotaops.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/dnotify.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/personality.h>
  27 #include <linux/security.h>
  28 #include <linux/mount.h>
  29 #include <linux/audit.h>
  30 #include <linux/vs_base.h>
  31
  32 #include <asm/namei.h>
  33 #include <asm/uaccess.h>
  34
  35 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
  36
  37 /* [Feb-1997 T. Schoebel-Theuer]
  38  * Fundamental changes in the pathname lookup mechanisms (namei)
  39  * were necessary because of omirr.  The reason is that omirr needs
  40  * to know the _real_ pathname, not the user-supplied one, in case
  41  * of symlinks (and also when transname replacements occur).
  42  *
  43  * The new code replaces the old recursive symlink resolution with
  44  * an iterative one (in case of non-nested symlink chains).  It does
  45  * this with calls to <fs>_follow_link().
  46  * As a side effect, dir_namei(), _namei() and follow_link() are now
  47  * replaced with a single function lookup_dentry() that can handle all
  48  * the special cases of the former code.
  49  *
  50  * With the new dcache, the pathname is stored at each inode, at least as
  51  * long as the refcount of the inode is positive.  As a side effect, the
  52  * size of the dcache depends on the inode cache and thus is dynamic.
  53  *
  54  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  55  * resolution to correspond with current state of the code.
  56  *
  57  * Note that the symlink resolution is not *completely* iterative.
  58  * There is still a significant amount of tail- and mid- recursion in
  59  * the algorithm.  Also, note that <fs>_readlink() is not used in
  60  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  61  * may return different results than <fs>_follow_link().  Many virtual
  62  * filesystems (including /proc) exhibit this behavior.
  63  */
  64
  65 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  66  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  67  * and the name already exists in form of a symlink, try to create the new
  68  * name indicated by the symlink. The old code always complained that the
  69  * name already exists, due to not following the symlink even if its target
  70  * is nonexistent.  The new semantics affects also mknod() and link() when
  71  * the name is a symlink pointing to a non-existant name.
  72  *
  73  * I don't know which semantics is the right one, since I have no access
  74  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  75  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  76  * "old" one. Personally, I think the new semantics is much more logical.
  77  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  78  * file does succeed in both HP-UX and SunOs, but not in Solaris
  79  * and in the old Linux semantics.
  80  */
  81
  82 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  83  * semantics.  See the comments in "open_namei" and "do_link" below.
  84  *
  85  * [10-Sep-98 Alan Modra] Another symlink change.
  86  */
  87
  88 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  89  *      inside the path - always follow.
  90  *      in the last component in creation/removal/renaming - never follow.
  91  *      if LOOKUP_FOLLOW passed - follow.
  92  *      if the pathname has trailing slashes - follow.
  93  *      otherwise - don't follow.
  94  * (applied in that order).
  95  *
  96  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
  97  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
  98  * During the 2.4 we need to fix the userland stuff depending on it -
  99  * hopefully we will be able to get rid of that wart in 2.5. So far only
 100  * XEmacs seems to be relying on it...
 101  */
 102 /*
 103  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 104  * implemented.  Let's see if raised priority of ->s_vfs_rename_sem gives
 105  * any extra contention...
 106  */
 107
 108 /* In order to reduce some races, while at the same time doing additional
 109  * checking and hopefully speeding things up, we copy filenames to the
 110  * kernel data space before using them..
 111  *
 112  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 113  * PATH_MAX includes the nul terminator --RR.
 114  */
 115 static inline int do_getname(const char __user *filename, char *page)
 116 {
 117         int retval;
 118         unsigned long len = PATH_MAX;
 119
 120         if ((unsigned long) filename >= TASK_SIZE) {
 121                 if (!segment_eq(get_fs(), KERNEL_DS))
 122                         return -EFAULT;
 123         } else if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
 124                 len = TASK_SIZE - (unsigned long) filename;
 125
 126         retval = strncpy_from_user((char *)page, filename, len);
 127         if (retval > 0) {
 128                 if (retval < len)
 129                         return 0;
 130                 return -ENAMETOOLONG;
 131         } else if (!retval)
 132                 retval = -ENOENT;
 133         return retval;
 134 }
 135
 136 char * getname(const char __user * filename)
 137 {
 138         char *tmp, *result;
 139
 140         result = ERR_PTR(-ENOMEM);
 141         tmp = __getname();
 142         if (tmp)  {
 143                 int retval = do_getname(filename, tmp);
 144
 145                 result = tmp;
 146                 if (retval < 0) {
 147                         __putname(tmp);
 148                         result = ERR_PTR(retval);
 149                 }
 150         }
 151         if (unlikely(current->audit_context) && !IS_ERR(result) && result)
 152                 audit_getname(result);
 153         return result;
 154 }
 155
 156 /*
 157  *      vfs_permission()
 158  *
 159  * is used to check for read/write/execute permissions on a file.
 160  * We use "fsuid" for this, letting us set arbitrary permissions
 161  * for filesystem access without changing the "normal" uids which
 162  * are used for other things..
 163  */
 164 int vfs_permission(struct inode * inode, int mask)
 165 {
 166         umode_t                 mode = inode->i_mode;
 167
 168         if (mask & MAY_WRITE) {
 169                 /*
 170                  * Nobody gets write access to a read-only fs.
 171                  */
 172                 if (IS_RDONLY(inode) &&
 173                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 174                         return -EROFS;
 175
 176                 /*
 177                  * Nobody gets write access to an immutable file.
 178                  */
 179                 if (IS_IMMUTABLE(inode))
 180                         return -EACCES;
 181         }
 182
 183         if (current->fsuid == inode->i_uid)
 184                 mode >>= 6;
 185         else if (in_group_p(inode->i_gid))
 186                 mode >>= 3;
 187
 188         /*
 189          * If the DACs are ok we don't need any capability check.
 190          */
 191         if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
 192                 return 0;
 193
 194         /*
 195          * Read/write DACs are always overridable.
 196          * Executable DACs are overridable if at least one exec bit is set.
 197          */
 198         if (!(mask & MAY_EXEC) ||
 199             (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
 200                 if (capable(CAP_DAC_OVERRIDE))
 201                         return 0;
 202
 203         /*
 204          * Searching includes executable on directories, else just read.
 205          */
 206         if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
 207                 if (capable(CAP_DAC_READ_SEARCH))
 208                         return 0;
 209
 210         return -EACCES;
 211 }
 212
 213 static inline int xid_permission(struct inode *inode, int mask, struct nameidata *nd)
 214 {
 215         if (inode->i_xid == 0)
 216                 return 0;
 217         if (vx_check(inode->i_xid, VX_ADMIN|VX_WATCH|VX_IDENT))
 218                 return 0;
 219 /*
 220         printk("VSW: xid=%d denied access to %p[#%d,%lu] »%*s«.\n",
 221                 vx_current_xid(), inode, inode->i_xid, inode->i_ino,
 222                 nd->dentry->d_name.len, nd->dentry->d_name.name);
 223 */
 224         return -EACCES;
 225 }
 226
 227 int permission(struct inode * inode,int mask, struct nameidata *nd)
 228 {
 229         int retval;
 230         int submask;
 231
 232         /* Ordinary permission routines do not understand MAY_APPEND. */
 233         submask = mask & ~MAY_APPEND;
 234
 235         if ((retval = xid_permission(inode, mask, nd)))
 236                 return retval;
 237         if (inode->i_op && inode->i_op->permission)
 238                 retval = inode->i_op->permission(inode, submask, nd);
 239         else
 240                 retval = vfs_permission(inode, submask);
 241         if (retval)
 242                 return retval;
 243
 244         return security_inode_permission(inode, mask, nd);
 245 }
 246
 247 /*
 248  * get_write_access() gets write permission for a file.
 249  * put_write_access() releases this write permission.
 250  * This is used for regular files.
 251  * We cannot support write (and maybe mmap read-write shared) accesses and
 252  * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
 253  * can have the following values:
 254  * 0: no writers, no VM_DENYWRITE mappings
 255  * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
 256  * > 0: (i_writecount) users are writing to the file.
 257  *
 258  * Normally we operate on that counter with atomic_{inc,dec} and it's safe
 259  * except for the cases where we don't hold i_writecount yet. Then we need to
 260  * use {get,deny}_write_access() - these functions check the sign and refuse
 261  * to do the change if sign is wrong. Exclusion between them is provided by
 262  * the inode->i_lock spinlock.
 263  */
 264
 265 int get_write_access(struct inode * inode)
 266 {
 267         spin_lock(&inode->i_lock);
 268         if (atomic_read(&inode->i_writecount) < 0) {
 269                 spin_unlock(&inode->i_lock);
 270                 return -ETXTBSY;
 271         }
 272         atomic_inc(&inode->i_writecount);
 273         spin_unlock(&inode->i_lock);
 274
 275         return 0;
 276 }
 277
 278 int deny_write_access(struct file * file)
 279 {
 280         struct inode *inode = file->f_dentry->d_inode;
 281
 282         spin_lock(&inode->i_lock);
 283         if (atomic_read(&inode->i_writecount) > 0) {
 284                 spin_unlock(&inode->i_lock);
 285                 return -ETXTBSY;
 286         }
 287         atomic_dec(&inode->i_writecount);
 288         spin_unlock(&inode->i_lock);
 289
 290         return 0;
 291 }
 292
 293 void path_release(struct nameidata *nd)
 294 {
 295         dput(nd->dentry);
 296         mntput(nd->mnt);
 297 }
 298
 299 /*
 300  * umount() mustn't call path_release()/mntput() as that would clear
 301  * mnt_expiry_mark
 302  */
 303 void path_release_on_umount(struct nameidata *nd)
 304 {
 305         dput(nd->dentry);
 306         _mntput(nd->mnt);
 307 }
 308
 309 /*
 310  * Internal lookup() using the new generic dcache.
 311  * SMP-safe
 312  */
 313 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
 314 {
 315         struct dentry * dentry = __d_lookup(parent, name);
 316
 317         /* lockess __d_lookup may fail due to concurrent d_move()
 318          * in some unrelated directory, so try with d_lookup
 319          */
 320         if (!dentry)
 321                 dentry = d_lookup(parent, name);
 322
 323         if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
 324                 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
 325                         dput(dentry);
 326                         dentry = NULL;
 327                 }
 328         }
 329         return dentry;
 330 }
 331
 332 /*
 333  * Short-cut version of permission(), for calling by
 334  * path_walk(), when dcache lock is held.  Combines parts
 335  * of permission() and vfs_permission(), and tests ONLY for
 336  * MAY_EXEC permission.
 337  *
 338  * If appropriate, check DAC only.  If not appropriate, or
 339  * short-cut DAC fails, then call permission() to do more
 340  * complete permission check.
 341  */
 342 static inline int exec_permission_lite(struct inode *inode,
 343                                        struct nameidata *nd)
 344 {
 345         umode_t mode = inode->i_mode;
 346
 347         if (inode->i_op && inode->i_op->permission)
 348                 return -EAGAIN;
 349
 350         if (current->fsuid == inode->i_uid)
 351                 mode >>= 6;
 352         else if (in_group_p(inode->i_gid))
 353                 mode >>= 3;
 354
 355         if (mode & MAY_EXEC)
 356                 goto ok;
 357
 358         if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
 359                 goto ok;
 360
 361         if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
 362                 goto ok;
 363
 364         if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
 365                 goto ok;
 366
 367         return -EACCES;
 368 ok:
 369         return security_inode_permission(inode, MAY_EXEC, nd);
 370 }
 371
 372 /*
 373  * This is called when everything else fails, and we actually have
 374  * to go to the low-level filesystem to find out what we should do..
 375  *
 376  * We get the directory semaphore, and after getting that we also
 377  * make sure that nobody added the entry to the dcache in the meantime..
 378  * SMP-safe
 379  */
 380 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
 381 {
 382         struct dentry * result;
 383         struct inode *dir = parent->d_inode;
 384
 385         down(&dir->i_sem);
 386         /*
 387          * First re-do the cached lookup just in case it was created
 388          * while we waited for the directory semaphore..
 389          *
 390          * FIXME! This could use version numbering or similar to
 391          * avoid unnecessary cache lookups.
 392          *
 393          * The "dcache_lock" is purely to protect the RCU list walker
 394          * from concurrent renames at this point (we mustn't get false
 395          * negatives from the RCU list walk here, unlike the optimistic
 396          * fast walk).
 397          *
 398          * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
 399          */
 400         result = d_lookup(parent, name);
 401         if (!result) {
 402                 struct dentry * dentry = d_alloc(parent, name);
 403                 result = ERR_PTR(-ENOMEM);
 404                 if (dentry) {
 405                         result = dir->i_op->lookup(dir, dentry, nd);
 406                         if (result)
 407                                 dput(dentry);
 408                         else
 409                                 result = dentry;
 410                 }
 411                 up(&dir->i_sem);
 412                 return result;
 413         }
 414
 415         /*
 416          * Uhhuh! Nasty case: the cache was re-populated while
 417          * we waited on the semaphore. Need to revalidate.
 418          */
 419         up(&dir->i_sem);
 420         if (result->d_op && result->d_op->d_revalidate) {
 421                 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
 422                         dput(result);
 423                         result = ERR_PTR(-ENOENT);
 424                 }
 425         }
 426         return result;
 427 }
 428
 429 static int __emul_lookup_dentry(const char *, struct nameidata *);
 430
 431 /* SMP-safe */
 432 static inline int
 433 walk_init_root(const char *name, struct nameidata *nd)
 434 {
 435         read_lock(&current->fs->lock);
 436         if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
 437                 nd->mnt = mntget(current->fs->altrootmnt);
 438                 nd->dentry = dget(current->fs->altroot);
 439                 read_unlock(&current->fs->lock);
 440                 if (__emul_lookup_dentry(name,nd))
 441                         return 0;
 442                 read_lock(&current->fs->lock);
 443         }
 444         nd->mnt = mntget(current->fs->rootmnt);
 445         nd->dentry = dget(current->fs->root);
 446         read_unlock(&current->fs->lock);
 447         return 1;
 448 }
 449
 450 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 451 {
 452         int res = 0;
 453         char *name;
 454         if (IS_ERR(link))
 455                 goto fail;
 456
 457         if (*link == '/') {
 458                 path_release(nd);
 459                 if (!walk_init_root(link, nd))
 460                         /* weird __emul_prefix() stuff did it */
 461                         goto out;
 462         }
 463         res = link_path_walk(link, nd);
 464 out:
 465         if (nd->depth || res || nd->last_type!=LAST_NORM)
 466                 return res;
 467         /*
 468          * If it is an iterative symlinks resolution in open_namei() we
 469          * have to copy the last component. And all that crap because of
 470          * bloody create() on broken symlinks. Furrfu...
 471          */
 472         name = __getname();
 473         if (unlikely(!name)) {
 474                 path_release(nd);
 475                 return -ENOMEM;
 476         }
 477         strcpy(name, nd->last.name);
 478         nd->last.name = name;
 479         return 0;
 480 fail:
 481         path_release(nd);
 482         return PTR_ERR(link);
 483 }
 484
 485 /*
 486  * This limits recursive symlink follows to 8, while
 487  * limiting consecutive symlinks to 40.
 488  *
 489  * Without that kind of total limit, nasty chains of consecutive
 490  * symlinks can cause almost arbitrarily long lookups.
 491  */
 492 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
 493 {
 494         int err = -ELOOP;
 495         if (current->link_count >= MAX_NESTED_LINKS)
 496                 goto loop;
 497         if (current->total_link_count >= 40)
 498                 goto loop;
 499         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
 500         cond_resched();
 501         err = security_inode_follow_link(dentry, nd);
 502         if (err)
 503                 goto loop;
 504         current->link_count++;
 505         current->total_link_count++;
 506         nd->depth++;
 507         touch_atime(nd->mnt, dentry);
 508         nd_set_link(nd, NULL);
 509         err = dentry->d_inode->i_op->follow_link(dentry, nd);
 510         if (!err) {
 511                 char *s = nd_get_link(nd);
 512                 if (s)
 513                         err = __vfs_follow_link(nd, s);
 514                 if (dentry->d_inode->i_op->put_link)
 515                         dentry->d_inode->i_op->put_link(dentry, nd);
 516         }
 517         current->link_count--;
 518         nd->depth--;
 519         return err;
 520 loop:
 521         path_release(nd);
 522         return err;
 523 }
 524
 525 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 526 {
 527         struct vfsmount *parent;
 528         struct dentry *mountpoint;
 529         spin_lock(&vfsmount_lock);
 530         parent=(*mnt)->mnt_parent;
 531         if (parent == *mnt) {
 532                 spin_unlock(&vfsmount_lock);
 533                 return 0;
 534         }
 535         mntget(parent);
 536         mountpoint=dget((*mnt)->mnt_mountpoint);
 537         spin_unlock(&vfsmount_lock);
 538         dput(*dentry);
 539         *dentry = mountpoint;
 540         mntput(*mnt);
 541         *mnt = parent;
 542         return 1;
 543 }
 544
 545 /* no need for dcache_lock, as serialization is taken care in
 546  * namespace.c
 547  */
 548 static int follow_mount(struct vfsmount **mnt, struct dentry **dentry)
 549 {
 550         int res = 0;
 551         while (d_mountpoint(*dentry)) {
 552                 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
 553                 if (!mounted)
 554                         break;
 555                 mntput(*mnt);
 556                 *mnt = mounted;
 557                 dput(*dentry);
 558                 *dentry = dget(mounted->mnt_root);
 559                 res = 1;
 560         }
 561         return res;
 562 }
 563
 564 /* no need for dcache_lock, as serialization is taken care in
 565  * namespace.c
 566  */
 567 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
 568 {
 569         struct vfsmount *mounted;
 570
 571         mounted = lookup_mnt(*mnt, *dentry);
 572         if (mounted) {
 573                 mntput(*mnt);
 574                 *mnt = mounted;
 575                 dput(*dentry);
 576                 *dentry = dget(mounted->mnt_root);
 577                 return 1;
 578         }
 579         return 0;
 580 }
 581
 582 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 583 {
 584         return __follow_down(mnt,dentry);
 585 }
 586
 587 static inline void follow_dotdot(struct vfsmount **mnt, struct dentry **dentry)
 588 {
 589         while(1) {
 590                 struct vfsmount *parent;
 591                 struct dentry *old = *dentry;
 592
 593                 read_lock(&current->fs->lock);
 594                 if (*dentry == current->fs->root &&
 595                     *mnt == current->fs->rootmnt) {
 596                         read_unlock(&current->fs->lock);
 597                         break;
 598                 }
 599                 read_unlock(&current->fs->lock);
 600                 spin_lock(&dcache_lock);
 601                 if (*dentry != (*mnt)->mnt_root) {
 602                         *dentry = dget((*dentry)->d_parent);
 603                         spin_unlock(&dcache_lock);
 604                         dput(old);
 605                         break;
 606                 }
 607                 spin_unlock(&dcache_lock);
 608                 spin_lock(&vfsmount_lock);
 609                 parent = (*mnt)->mnt_parent;
 610                 if (parent == *mnt) {
 611                         spin_unlock(&vfsmount_lock);
 612                         break;
 613                 }
 614                 mntget(parent);
 615                 *dentry = dget((*mnt)->mnt_mountpoint);
 616                 spin_unlock(&vfsmount_lock);
 617                 dput(old);
 618                 mntput(*mnt);
 619                 *mnt = parent;
 620         }
 621         follow_mount(mnt, dentry);
 622 }
 623
 624 struct path {
 625         struct vfsmount *mnt;
 626         struct dentry *dentry;
 627 };
 628
 629 /*
 630  *  It's more convoluted than I'd like it to be, but... it's still fairly
 631  *  small and for now I'd prefer to have fast path as straight as possible.
 632  *  It _is_ time-critical.
 633  */
 634 static int do_lookup(struct nameidata *nd, struct qstr *name,
 635                      struct path *path)
 636 {
 637         struct vfsmount *mnt = nd->mnt;
 638         struct dentry *dentry = __d_lookup(nd->dentry, name);
 639
 640         if (!dentry)
 641                 goto need_lookup;
 642         if (dentry->d_op && dentry->d_op->d_revalidate)
 643                 goto need_revalidate;
 644 done:
 645         path->mnt = mnt;
 646         path->dentry = dentry;
 647         return 0;
 648
 649 need_lookup:
 650         dentry = real_lookup(nd->dentry, name, nd);
 651         if (IS_ERR(dentry))
 652                 goto fail;
 653         goto done;
 654
 655 need_revalidate:
 656         if (dentry->d_op->d_revalidate(dentry, nd))
 657                 goto done;
 658         if (d_invalidate(dentry))
 659                 goto done;
 660         dput(dentry);
 661         goto need_lookup;
 662
 663 fail:
 664         return PTR_ERR(dentry);
 665 }
 666
 667 /*
 668  * Name resolution.
 669  *
 670  * This is the basic name resolution function, turning a pathname
 671  * into the final dentry.
 672  *
 673  * We expect 'base' to be positive and a directory.
 674  */
 675 int fastcall link_path_walk(const char * name, struct nameidata *nd)
 676 {
 677         struct path next;
 678         struct inode *inode;
 679         int err;
 680         unsigned int lookup_flags = nd->flags;
 681
 682         while (*name=='/')
 683                 name++;
 684         if (!*name)
 685                 goto return_reval;
 686
 687         inode = nd->dentry->d_inode;
 688         if (nd->depth)
 689                 lookup_flags = LOOKUP_FOLLOW;
 690
 691         /* At this point we know we have a real path component. */
 692         for(;;) {
 693                 unsigned long hash;
 694                 struct qstr this;
 695                 unsigned int c;
 696
 697                 err = exec_permission_lite(inode, nd);
 698                 if (err == -EAGAIN) {
 699                         err = permission(inode, MAY_EXEC, nd);
 700                 }
 701                 if (err)
 702                         break;
 703
 704                 this.name = name;
 705                 c = *(const unsigned char *)name;
 706
 707                 hash = init_name_hash();
 708                 do {
 709                         name++;
 710                         hash = partial_name_hash(c, hash);
 711                         c = *(const unsigned char *)name;
 712                 } while (c && (c != '/'));
 713                 this.len = name - (const char *) this.name;
 714                 this.hash = end_name_hash(hash);
 715
 716                 /* remove trailing slashes? */
 717                 if (!c)
 718                         goto last_component;
 719                 while (*++name == '/');
 720                 if (!*name)
 721                         goto last_with_slashes;
 722
 723                 /*
 724                  * "." and ".." are special - ".." especially so because it has
 725                  * to be able to know about the current root directory and
 726                  * parent relationships.
 727                  */
 728                 if (this.name[0] == '.') switch (this.len) {
 729                         default:
 730                                 break;
 731                         case 2:
 732                                 if (this.name[1] != '.')
 733                                         break;
 734                                 follow_dotdot(&nd->mnt, &nd->dentry);
 735                                 inode = nd->dentry->d_inode;
 736                                 /* fallthrough */
 737                         case 1:
 738                                 continue;
 739                 }
 740                 /*
 741                  * See if the low-level filesystem might want
 742                  * to use its own hash..
 743                  */
 744                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 745                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 746                         if (err < 0)
 747                                 break;
 748                 }
 749                 nd->flags |= LOOKUP_CONTINUE;
 750                 /* This does the actual lookups.. */
 751                 err = do_lookup(nd, &this, &next);
 752                 if (err)
 753                         break;
 754                 /* Check mountpoints.. */
 755                 follow_mount(&next.mnt, &next.dentry);
 756
 757                 err = -ENOENT;
 758                 inode = next.dentry->d_inode;
 759                 if (!inode)
 760                         goto out_dput;
 761                 err = -ENOTDIR;
 762                 if (!inode->i_op)
 763                         goto out_dput;
 764
 765                 if (inode->i_op->follow_link) {
 766                         mntget(next.mnt);
 767                         err = do_follow_link(next.dentry, nd);
 768                         dput(next.dentry);
 769                         mntput(next.mnt);
 770                         if (err)
 771                                 goto return_err;
 772                         err = -ENOENT;
 773                         inode = nd->dentry->d_inode;
 774                         if (!inode)
 775                                 break;
 776                         err = -ENOTDIR;
 777                         if (!inode->i_op)
 778                                 break;
 779                 } else {
 780                         dput(nd->dentry);
 781                         nd->mnt = next.mnt;
 782                         nd->dentry = next.dentry;
 783                 }
 784                 err = -ENOTDIR;
 785                 if (!inode->i_op->lookup)
 786                         break;
 787                 continue;
 788                 /* here ends the main loop */
 789
 790 last_with_slashes:
 791                 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 792 last_component:
 793                 nd->flags &= ~LOOKUP_CONTINUE;
 794                 if (lookup_flags & LOOKUP_PARENT)
 795                         goto lookup_parent;
 796                 if (this.name[0] == '.') switch (this.len) {
 797                         default:
 798                                 break;
 799                         case 2:
 800                                 if (this.name[1] != '.')
 801                                         break;
 802                                 follow_dotdot(&nd->mnt, &nd->dentry);
 803                                 inode = nd->dentry->d_inode;
 804                                 /* fallthrough */
 805                         case 1:
 806                                 goto return_reval;
 807                 }
 808                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 809                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 810                         if (err < 0)
 811                                 break;
 812                 }
 813                 err = do_lookup(nd, &this, &next);
 814                 if (err)
 815                         break;
 816                 follow_mount(&next.mnt, &next.dentry);
 817                 inode = next.dentry->d_inode;
 818                 if ((lookup_flags & LOOKUP_FOLLOW)
 819                     && inode && inode->i_op && inode->i_op->follow_link) {
 820                         mntget(next.mnt);
 821                         err = do_follow_link(next.dentry, nd);
 822                         dput(next.dentry);
 823                         mntput(next.mnt);
 824                         if (err)
 825                                 goto return_err;
 826                         inode = nd->dentry->d_inode;
 827                 } else {
 828                         dput(nd->dentry);
 829                         nd->mnt = next.mnt;
 830                         nd->dentry = next.dentry;
 831                 }
 832                 err = -ENOENT;
 833                 if (!inode)
 834                         break;
 835                 if (lookup_flags & LOOKUP_DIRECTORY) {
 836                         err = -ENOTDIR;
 837                         if (!inode->i_op || !inode->i_op->lookup)
 838                                 break;
 839                 }
 840                 goto return_base;
 841 lookup_parent:
 842                 nd->last = this;
 843                 nd->last_type = LAST_NORM;
 844                 if (this.name[0] != '.')
 845                         goto return_base;
 846                 if (this.len == 1)
 847                         nd->last_type = LAST_DOT;
 848                 else if (this.len == 2 && this.name[1] == '.')
 849                         nd->last_type = LAST_DOTDOT;
 850                 else
 851                         goto return_base;
 852 return_reval:
 853                 /*
 854                  * We bypassed the ordinary revalidation routines.
 855                  * We may need to check the cached dentry for staleness.
 856                  */
 857                 if (nd->dentry && nd->dentry->d_sb &&
 858                     (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
 859                         err = -ESTALE;
 860                         /* Note: we do not d_invalidate() */
 861                         if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
 862                                 break;
 863                 }
 864 return_base:
 865                 return 0;
 866 out_dput:
 867                 dput(next.dentry);
 868                 break;
 869         }
 870         path_release(nd);
 871 return_err:
 872         return err;
 873 }
 874
 875 int fastcall path_walk(const char * name, struct nameidata *nd)
 876 {
 877         current->total_link_count = 0;
 878         return link_path_walk(name, nd);
 879 }
 880
 881 /* SMP-safe */
 882 /* returns 1 if everything is done */
 883 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
 884 {
 885         if (path_walk(name, nd))
 886                 return 0;               /* something went wrong... */
 887
 888         if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
 889                 struct nameidata nd_root;
 890                 /*
 891                  * NAME was not found in alternate root or it's a directory.  Try to find
 892                  * it in the normal root:
 893                  */
 894                 nd_root.last_type = LAST_ROOT;
 895                 nd_root.flags = nd->flags;
 896                 nd_root.depth = 0;
 897                 memcpy(&nd_root.intent, &nd->intent, sizeof(nd_root.intent));
 898                 read_lock(&current->fs->lock);
 899                 nd_root.mnt = mntget(current->fs->rootmnt);
 900                 nd_root.dentry = dget(current->fs->root);
 901                 read_unlock(&current->fs->lock);
 902                 if (path_walk(name, &nd_root))
 903                         return 1;
 904                 if (nd_root.dentry->d_inode) {
 905                         path_release(nd);
 906                         nd->dentry = nd_root.dentry;
 907                         nd->mnt = nd_root.mnt;
 908                         nd->last = nd_root.last;
 909                         return 1;
 910                 }
 911                 path_release(&nd_root);
 912         }
 913         return 1;
 914 }
 915
 916 void set_fs_altroot(void)
 917 {
 918         char *emul = __emul_prefix();
 919         struct nameidata nd;
 920         struct vfsmount *mnt = NULL, *oldmnt;
 921         struct dentry *dentry = NULL, *olddentry;
 922         int err;
 923
 924         if (!emul)
 925                 goto set_it;
 926         err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
 927         if (!err) {
 928                 mnt = nd.mnt;
 929                 dentry = nd.dentry;
 930         }
 931 set_it:
 932         write_lock(&current->fs->lock);
 933         oldmnt = current->fs->altrootmnt;
 934         olddentry = current->fs->altroot;
 935         current->fs->altrootmnt = mnt;
 936         current->fs->altroot = dentry;
 937         write_unlock(&current->fs->lock);
 938         if (olddentry) {
 939                 dput(olddentry);
 940                 mntput(oldmnt);
 941         }
 942 }
 943
 944 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
 945 {
 946         int retval;
 947
 948         nd->last_type = LAST_ROOT; /* if there are only slashes... */
 949         nd->flags = flags;
 950         nd->depth = 0;
 951
 952         read_lock(&current->fs->lock);
 953         if (*name=='/') {
 954                 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
 955                         nd->mnt = mntget(current->fs->altrootmnt);
 956                         nd->dentry = dget(current->fs->altroot);
 957                         read_unlock(&current->fs->lock);
 958                         if (__emul_lookup_dentry(name,nd))
 959                                 return 0;
 960                         read_lock(&current->fs->lock);
 961                 }
 962                 nd->mnt = mntget(current->fs->rootmnt);
 963                 nd->dentry = dget(current->fs->root);
 964         }
 965         else{
 966                 nd->mnt = mntget(current->fs->pwdmnt);
 967                 nd->dentry = dget(current->fs->pwd);
 968         }
 969         read_unlock(&current->fs->lock);
 970         current->total_link_count = 0;
 971         retval = link_path_walk(name, nd);
 972         if (unlikely(current->audit_context
 973                      && nd && nd->dentry && nd->dentry->d_inode))
 974                 audit_inode(name,
 975                             nd->dentry->d_inode->i_ino,
 976                             nd->dentry->d_inode->i_rdev);
 977         return retval;
 978 }
 979
 980 /*
 981  * Restricted form of lookup. Doesn't follow links, single-component only,
 982  * needs parent already locked. Doesn't follow mounts.
 983  * SMP-safe.
 984  */
 985 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
 986 {
 987         struct dentry * dentry;
 988         struct inode *inode;
 989         int err;
 990
 991         inode = base->d_inode;
 992         err = permission(inode, MAY_EXEC, nd);
 993         dentry = ERR_PTR(err);
 994         if (err)
 995                 goto out;
 996
 997         /*
 998          * See if the low-level filesystem might want
 999          * to use its own hash..
1000          */
1001         if (base->d_op && base->d_op->d_hash) {
1002                 err = base->d_op->d_hash(base, name);
1003                 dentry = ERR_PTR(err);
1004                 if (err < 0)
1005                         goto out;
1006         }
1007
1008         dentry = cached_lookup(base, name, nd);
1009         if (!dentry) {
1010                 struct dentry *new = d_alloc(base, name);
1011                 dentry = ERR_PTR(-ENOMEM);
1012                 if (!new)
1013                         goto out;
1014                 dentry = inode->i_op->lookup(inode, new, nd);
1015                 if (!dentry)
1016                         dentry = new;
1017                 else
1018                         dput(new);
1019         }
1020 out:
1021         return dentry;
1022 }
1023
1024 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1025 {
1026         return __lookup_hash(name, base, NULL);
1027 }
1028
1029 /* SMP-safe */
1030 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1031 {
1032         unsigned long hash;
1033         struct qstr this;
1034         unsigned int c;
1035
1036         this.name = name;
1037         this.len = len;
1038         if (!len)
1039                 goto access;
1040
1041         hash = init_name_hash();
1042         while (len--) {
1043                 c = *(const unsigned char *)name++;
1044                 if (c == '/' || c == '\0')
1045                         goto access;
1046                 hash = partial_name_hash(c, hash);
1047         }
1048         this.hash = end_name_hash(hash);
1049
1050         return lookup_hash(&this, base);
1051 access:
1052         return ERR_PTR(-EACCES);
1053 }
1054
1055 /*
1056  *      namei()
1057  *
1058  * is used by most simple commands to get the inode of a specified name.
1059  * Open, link etc use their own routines, but this is enough for things
1060  * like 'chmod' etc.
1061  *
1062  * namei exists in two versions: namei/lnamei. The only difference is
1063  * that namei follows links, while lnamei does not.
1064  * SMP-safe
1065  */
1066 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1067 {
1068         char *tmp = getname(name);
1069         int err = PTR_ERR(tmp);
1070
1071         if (!IS_ERR(tmp)) {
1072                 err = path_lookup(tmp, flags, nd);
1073                 putname(tmp);
1074         }
1075         return err;
1076 }
1077
1078 /*
1079  * It's inline, so penalty for filesystems that don't use sticky bit is
1080  * minimal.
1081  */
1082 static inline int check_sticky(struct inode *dir, struct inode *inode)
1083 {
1084         if (!(dir->i_mode & S_ISVTX))
1085                 return 0;
1086         if (inode->i_uid == current->fsuid)
1087                 return 0;
1088         if (dir->i_uid == current->fsuid)
1089                 return 0;
1090         return !capable(CAP_FOWNER);
1091 }
1092
1093 /*
1094  *      Check whether we can remove a link victim from directory dir, check
1095  *  whether the type of victim is right.
1096  *  1. We can't do it if dir is read-only (done in permission())
1097  *  2. We should have write and exec permissions on dir
1098  *  3. We can't remove anything from append-only dir
1099  *  4. We can't do anything with immutable dir (done in permission())
1100  *  5. If the sticky bit on dir is set we should either
1101  *      a. be owner of dir, or
1102  *      b. be owner of victim, or
1103  *      c. have CAP_FOWNER capability
1104  *  6. If the victim is append-only or immutable we can't do antyhing with
1105  *     links pointing to it.
1106  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1107  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1108  *  9. We can't remove a root or mountpoint.
1109  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1110  *     nfs_async_unlink().
1111  */
1112 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1113 {
1114         int error;
1115         if (!victim->d_inode || victim->d_parent->d_inode != dir)
1116                 return -ENOENT;
1117         error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1118         if (error)
1119                 return error;
1120         if (IS_APPEND(dir))
1121                 return -EPERM;
1122         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1123                 IS_IXORUNLINK(victim->d_inode))
1124                 return -EPERM;
1125         if (isdir) {
1126                 if (!S_ISDIR(victim->d_inode->i_mode))
1127                         return -ENOTDIR;
1128                 if (IS_ROOT(victim))
1129                         return -EBUSY;
1130         } else if (S_ISDIR(victim->d_inode->i_mode))
1131                 return -EISDIR;
1132         if (IS_DEADDIR(dir))
1133                 return -ENOENT;
1134         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1135                 return -EBUSY;
1136         return 0;
1137 }
1138
1139 /*      Check whether we can create an object with dentry child in directory
1140  *  dir.
1141  *  1. We can't do it if child already exists (open has special treatment for
1142  *     this case, but since we are inlined it's OK)
1143  *  2. We can't do it if dir is read-only (done in permission())
1144  *  3. We should have write and exec permissions on dir
1145  *  4. We can't do it if dir is immutable (done in permission())
1146  */
1147 static inline int may_create(struct inode *dir, struct dentry *child,
1148                              struct nameidata *nd)
1149 {
1150         if (child->d_inode)
1151                 return -EEXIST;
1152         if (IS_DEADDIR(dir))
1153                 return -ENOENT;
1154         return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1155 }
1156
1157 /*
1158  * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1159  * reasons.
1160  *
1161  * O_DIRECTORY translates into forcing a directory lookup.
1162  */
1163 static inline int lookup_flags(unsigned int f)
1164 {
1165         unsigned long retval = LOOKUP_FOLLOW;
1166
1167         if (f & O_NOFOLLOW)
1168                 retval &= ~LOOKUP_FOLLOW;
1169
1170         if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1171                 retval &= ~LOOKUP_FOLLOW;
1172
1173         if (f & O_DIRECTORY)
1174                 retval |= LOOKUP_DIRECTORY;
1175
1176         return retval;
1177 }
1178
1179 /*
1180  * p1 and p2 should be directories on the same fs.
1181  */
1182 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1183 {
1184         struct dentry *p;
1185
1186         if (p1 == p2) {
1187                 down(&p1->d_inode->i_sem);
1188                 return NULL;
1189         }
1190
1191         down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1192
1193         for (p = p1; p->d_parent != p; p = p->d_parent) {
1194                 if (p->d_parent == p2) {
1195                         down(&p2->d_inode->i_sem);
1196                         down(&p1->d_inode->i_sem);
1197                         return p;
1198                 }
1199         }
1200
1201         for (p = p2; p->d_parent != p; p = p->d_parent) {
1202                 if (p->d_parent == p1) {
1203                         down(&p1->d_inode->i_sem);
1204                         down(&p2->d_inode->i_sem);
1205                         return p;
1206                 }
1207         }
1208
1209         down(&p1->d_inode->i_sem);
1210         down(&p2->d_inode->i_sem);
1211         return NULL;
1212 }
1213
1214 void unlock_rename(struct dentry *p1, struct dentry *p2)
1215 {
1216         up(&p1->d_inode->i_sem);
1217         if (p1 != p2) {
1218                 up(&p2->d_inode->i_sem);
1219                 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1220         }
1221 }
1222
1223 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1224                 struct nameidata *nd)
1225 {
1226         int error = may_create(dir, dentry, nd);
1227
1228         if (error)
1229                 return error;
1230
1231         if (!dir->i_op || !dir->i_op->create)
1232                 return -EACCES; /* shouldn't it be ENOSYS? */
1233         mode &= S_IALLUGO;
1234         mode |= S_IFREG;
1235         error = security_inode_create(dir, dentry, mode);
1236         if (error)
1237                 return error;
1238         DQUOT_INIT(dir);
1239         error = dir->i_op->create(dir, dentry, mode, nd);
1240         if (!error) {
1241                 inode_dir_notify(dir, DN_CREATE);
1242                 security_inode_post_create(dir, dentry, mode);
1243         }
1244         return error;
1245 }
1246
1247 int may_open(struct nameidata *nd, int acc_mode, int flag)
1248 {
1249         struct dentry *dentry = nd->dentry;
1250         struct inode *inode = dentry->d_inode;
1251         int error;
1252
1253         if (!inode)
1254                 return -ENOENT;
1255
1256         if (S_ISLNK(inode->i_mode))
1257                 return -ELOOP;
1258
1259         if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1260                 return -EISDIR;
1261
1262         error = permission(inode, acc_mode, nd);
1263         if (error)
1264                 return error;
1265
1266         /*
1267          * FIFO's, sockets and device files are special: they don't
1268          * actually live on the filesystem itself, and as such you
1269          * can write to them even if the filesystem is read-only.
1270          */
1271         if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1272                 flag &= ~O_TRUNC;
1273         } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1274                 if (nd->mnt->mnt_flags & MNT_NODEV)
1275                         return -EACCES;
1276
1277                 flag &= ~O_TRUNC;
1278         } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1279                 return -EROFS;
1280         /*
1281          * An append-only file must be opened in append mode for writing.
1282          */
1283         if (IS_APPEND(inode)) {
1284                 if  ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1285                         return -EPERM;
1286                 if (flag & O_TRUNC)
1287                         return -EPERM;
1288         }
1289
1290         /* O_NOATIME can only be set by the owner or superuser */
1291         if (flag & O_NOATIME)
1292                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1293                         return -EPERM;
1294
1295         /*
1296          * Ensure there are no outstanding leases on the file.
1297          */
1298         error = break_lease(inode, flag);
1299         if (error)
1300                 return error;
1301
1302         if (flag & O_TRUNC) {
1303                 error = get_write_access(inode);
1304                 if (error)
1305                         return error;
1306
1307                 /*
1308                  * Refuse to truncate files with mandatory locks held on them.
1309                  */
1310                 error = locks_verify_locked(inode);
1311                 if (!error) {
1312                         DQUOT_INIT(inode);
1313
1314                         error = do_truncate(dentry, 0);
1315                 }
1316                 put_write_access(inode);
1317                 if (error)
1318                         return error;
1319         } else
1320                 if (flag & FMODE_WRITE)
1321                         DQUOT_INIT(inode);
1322
1323         return 0;
1324 }
1325
1326 /*
1327  *      open_namei()
1328  *
1329  * namei for open - this is in fact almost the whole open-routine.
1330  *
1331  * Note that the low bits of "flag" aren't the same as in the open
1332  * system call - they are 00 - no permissions needed
1333  *                        01 - read permission needed
1334  *                        10 - write permission needed
1335  *                        11 - read/write permissions needed
1336  * which is a lot more logical, and also allows the "no perm" needed
1337  * for symlinks (where the permissions are checked later).
1338  * SMP-safe
1339  */
1340 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1341 {
1342         int acc_mode, error = 0;
1343         struct dentry *dentry;
1344         struct dentry *dir;
1345         int count = 0;
1346
1347         acc_mode = ACC_MODE(flag);
1348
1349         /* Allow the LSM permission hook to distinguish append
1350            access from general write access. */
1351         if (flag & O_APPEND)
1352                 acc_mode |= MAY_APPEND;
1353
1354         /* Fill in the open() intent data */
1355         nd->intent.open.flags = flag;
1356         nd->intent.open.create_mode = mode;
1357
1358         /*
1359          * The simplest case - just a plain lookup.
1360          */
1361         if (!(flag & O_CREAT)) {
1362                 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1363                 if (error)
1364                         return error;
1365                 goto ok;
1366         }
1367
1368         /*
1369          * Create - we need to know the parent.
1370          */
1371         error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1372         if (error)
1373                 return error;
1374
1375         /*
1376          * We have the parent and last component. First of all, check
1377          * that we are not asked to creat(2) an obvious directory - that
1378          * will not do.
1379          */
1380         error = -EISDIR;
1381         if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1382                 goto exit;
1383
1384         dir = nd->dentry;
1385         nd->flags &= ~LOOKUP_PARENT;
1386         down(&dir->d_inode->i_sem);
1387         dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1388
1389 do_last:
1390         error = PTR_ERR(dentry);
1391         if (IS_ERR(dentry)) {
1392                 up(&dir->d_inode->i_sem);
1393                 goto exit;
1394         }
1395
1396         /* Negative dentry, just create the file */
1397         if (!dentry->d_inode) {
1398                 if (!IS_POSIXACL(dir->d_inode))
1399                         mode &= ~current->fs->umask;
1400                 error = vfs_create(dir->d_inode, dentry, mode, nd);
1401                 up(&dir->d_inode->i_sem);
1402                 dput(nd->dentry);
1403                 nd->dentry = dentry;
1404                 if (error)
1405                         goto exit;
1406                 /* Don't check for write permission, don't truncate */
1407                 acc_mode = 0;
1408                 flag &= ~O_TRUNC;
1409                 goto ok;
1410         }
1411
1412         /*
1413          * It already exists.
1414          */
1415         up(&dir->d_inode->i_sem);
1416
1417         error = -EEXIST;
1418         if (flag & O_EXCL)
1419                 goto exit_dput;
1420
1421         if (d_mountpoint(dentry)) {
1422                 error = -ELOOP;
1423                 if (flag & O_NOFOLLOW)
1424                         goto exit_dput;
1425                 while (__follow_down(&nd->mnt,&dentry) && d_mountpoint(dentry));
1426         }
1427         error = -ENOENT;
1428         if (!dentry->d_inode)
1429                 goto exit_dput;
1430         if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1431                 goto do_link;
1432
1433         dput(nd->dentry);
1434         nd->dentry = dentry;
1435         error = -EISDIR;
1436         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1437                 goto exit;
1438 ok:
1439         error = may_open(nd, acc_mode, flag);
1440         if (error)
1441                 goto exit;
1442         return 0;
1443
1444 exit_dput:
1445         dput(dentry);
1446 exit:
1447         path_release(nd);
1448         return error;
1449
1450 do_link:
1451         error = -ELOOP;
1452         if (flag & O_NOFOLLOW)
1453                 goto exit_dput;
1454         /*
1455          * This is subtle. Instead of calling do_follow_link() we do the
1456          * thing by hands. The reason is that this way we have zero link_count
1457          * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1458          * After that we have the parent and last component, i.e.
1459          * we are in the same situation as after the first path_walk().
1460          * Well, almost - if the last component is normal we get its copy
1461          * stored in nd->last.name and we will have to putname() it when we
1462          * are done. Procfs-like symlinks just set LAST_BIND.
1463          */
1464         nd->flags |= LOOKUP_PARENT;
1465         error = security_inode_follow_link(dentry, nd);
1466         if (error)
1467                 goto exit_dput;
1468         touch_atime(nd->mnt, dentry);
1469         nd_set_link(nd, NULL);
1470         error = dentry->d_inode->i_op->follow_link(dentry, nd);
1471         if (!error) {
1472                 char *s = nd_get_link(nd);
1473                 if (s)
1474                         error = __vfs_follow_link(nd, s);
1475                 if (dentry->d_inode->i_op->put_link)
1476                         dentry->d_inode->i_op->put_link(dentry, nd);
1477         }
1478         dput(dentry);
1479         if (error)
1480                 return error;
1481         nd->flags &= ~LOOKUP_PARENT;
1482         if (nd->last_type == LAST_BIND) {
1483                 dentry = nd->dentry;
1484                 goto ok;
1485         }
1486         error = -EISDIR;
1487         if (nd->last_type != LAST_NORM)
1488                 goto exit;
1489         if (nd->last.name[nd->last.len]) {
1490                 putname(nd->last.name);
1491                 goto exit;
1492         }
1493         error = -ELOOP;
1494         if (count++==32) {
1495                 putname(nd->last.name);
1496                 goto exit;
1497         }
1498         dir = nd->dentry;
1499         down(&dir->d_inode->i_sem);
1500         dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1501         putname(nd->last.name);
1502         goto do_last;
1503 }
1504
1505 /**
1506  * lookup_create - lookup a dentry, creating it if it doesn't exist
1507  * @nd: nameidata info
1508  * @is_dir: directory flag
1509  *
1510  * Simple function to lookup and return a dentry and create it
1511  * if it doesn't exist.  Is SMP-safe.
1512  */
1513 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1514 {
1515         struct dentry *dentry;
1516
1517         down(&nd->dentry->d_inode->i_sem);
1518         dentry = ERR_PTR(-EEXIST);
1519         if (nd->last_type != LAST_NORM)
1520                 goto fail;
1521         nd->flags &= ~LOOKUP_PARENT;
1522         dentry = lookup_hash(&nd->last, nd->dentry);
1523         if (IS_ERR(dentry))
1524                 goto fail;
1525         if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1526                 goto enoent;
1527         return dentry;
1528 enoent:
1529         dput(dentry);
1530         dentry = ERR_PTR(-ENOENT);
1531 fail:
1532         return dentry;
1533 }
1534
1535 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1536 {
1537         int error = may_create(dir, dentry, NULL);
1538
1539         if (error)
1540                 return error;
1541
1542         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1543                 return -EPERM;
1544
1545         if (!dir->i_op || !dir->i_op->mknod)
1546                 return -EPERM;
1547
1548         error = security_inode_mknod(dir, dentry, mode, dev);
1549         if (error)
1550                 return error;
1551
1552         DQUOT_INIT(dir);
1553         error = dir->i_op->mknod(dir, dentry, mode, dev);
1554         if (!error) {
1555                 inode_dir_notify(dir, DN_CREATE);
1556                 security_inode_post_mknod(dir, dentry, mode, dev);
1557         }
1558         return error;
1559 }
1560
1561 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1562 {
1563         int error = 0;
1564         char * tmp;
1565         struct dentry * dentry;
1566         struct nameidata nd;
1567
1568         if (S_ISDIR(mode))
1569                 return -EPERM;
1570         tmp = getname(filename);
1571         if (IS_ERR(tmp))
1572                 return PTR_ERR(tmp);
1573
1574         error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1575         if (error)
1576                 goto out;
1577         dentry = lookup_create(&nd, 0);
1578         error = PTR_ERR(dentry);
1579
1580         if (!IS_POSIXACL(nd.dentry->d_inode))
1581                 mode &= ~current->fs->umask;
1582         if (!IS_ERR(dentry)) {
1583                 switch (mode & S_IFMT) {
1584                 case 0: case S_IFREG:
1585                         error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1586                         break;
1587                 case S_IFCHR: case S_IFBLK:
1588                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1589                                         new_decode_dev(dev));
1590                         break;
1591                 case S_IFIFO: case S_IFSOCK:
1592                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1593                         break;
1594                 case S_IFDIR:
1595                         error = -EPERM;
1596                         break;
1597                 default:
1598                         error = -EINVAL;
1599                 }
1600                 dput(dentry);
1601         }
1602         up(&nd.dentry->d_inode->i_sem);
1603         path_release(&nd);
1604 out:
1605         putname(tmp);
1606
1607         return error;
1608 }
1609
1610 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1611 {
1612         int error = may_create(dir, dentry, NULL);
1613
1614         if (error)
1615                 return error;
1616
1617         if (!dir->i_op || !dir->i_op->mkdir)
1618                 return -EPERM;
1619
1620         mode &= (S_IRWXUGO|S_ISVTX);
1621         error = security_inode_mkdir(dir, dentry, mode);
1622         if (error)
1623                 return error;
1624
1625         DQUOT_INIT(dir);
1626         error = dir->i_op->mkdir(dir, dentry, mode);
1627         if (!error) {
1628                 inode_dir_notify(dir, DN_CREATE);
1629                 security_inode_post_mkdir(dir,dentry, mode);
1630         }
1631         return error;
1632 }
1633
1634 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1635 {
1636         int error = 0;
1637         char * tmp;
1638
1639         tmp = getname(pathname);
1640         error = PTR_ERR(tmp);
1641         if (!IS_ERR(tmp)) {
1642                 struct dentry *dentry;
1643                 struct nameidata nd;
1644
1645                 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1646                 if (error)
1647                         goto out;
1648                 dentry = lookup_create(&nd, 1);
1649                 error = PTR_ERR(dentry);
1650                 if (!IS_ERR(dentry)) {
1651                         if (!IS_POSIXACL(nd.dentry->d_inode))
1652                                 mode &= ~current->fs->umask;
1653                         error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1654                         dput(dentry);
1655                 }
1656                 up(&nd.dentry->d_inode->i_sem);
1657                 path_release(&nd);
1658 out:
1659                 putname(tmp);
1660         }
1661
1662         return error;
1663 }
1664
1665 /*
1666  * We try to drop the dentry early: we should have
1667  * a usage count of 2 if we're the only user of this
1668  * dentry, and if that is true (possibly after pruning
1669  * the dcache), then we drop the dentry now.
1670  *
1671  * A low-level filesystem can, if it choses, legally
1672  * do a
1673  *
1674  *      if (!d_unhashed(dentry))
1675  *              return -EBUSY;
1676  *
1677  * if it cannot handle the case of removing a directory
1678  * that is still in use by something else..
1679  */
1680 static void d_unhash(struct dentry *dentry)
1681 {
1682         dget(dentry);
1683         spin_lock(&dcache_lock);
1684         switch (atomic_read(&dentry->d_count)) {
1685         default:
1686                 spin_unlock(&dcache_lock);
1687                 shrink_dcache_parent(dentry);
1688                 spin_lock(&dcache_lock);
1689                 if (atomic_read(&dentry->d_count) != 2)
1690                         break;
1691         case 2:
1692                 __d_drop(dentry);
1693         }
1694         spin_unlock(&dcache_lock);
1695 }
1696
1697 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1698 {
1699         int error = may_delete(dir, dentry, 1);
1700
1701         if (error)
1702                 return error;
1703
1704         if (!dir->i_op || !dir->i_op->rmdir)
1705                 return -EPERM;
1706
1707         DQUOT_INIT(dir);
1708
1709         down(&dentry->d_inode->i_sem);
1710         d_unhash(dentry);
1711         if (d_mountpoint(dentry))
1712                 error = -EBUSY;
1713         else {
1714                 error = security_inode_rmdir(dir, dentry);
1715                 if (!error) {
1716                         error = dir->i_op->rmdir(dir, dentry);
1717                         if (!error)
1718                                 dentry->d_inode->i_flags |= S_DEAD;
1719                 }
1720         }
1721         up(&dentry->d_inode->i_sem);
1722         if (!error) {
1723                 inode_dir_notify(dir, DN_DELETE);
1724                 d_delete(dentry);
1725         }
1726         dput(dentry);
1727
1728         return error;
1729 }
1730
1731 asmlinkage long sys_rmdir(const char __user * pathname)
1732 {
1733         int error = 0;
1734         char * name;
1735         struct dentry *dentry;
1736         struct nameidata nd;
1737
1738         name = getname(pathname);
1739         if(IS_ERR(name))
1740                 return PTR_ERR(name);
1741
1742         error = path_lookup(name, LOOKUP_PARENT, &nd);
1743         if (error)
1744                 goto exit;
1745
1746         switch(nd.last_type) {
1747                 case LAST_DOTDOT:
1748                         error = -ENOTEMPTY;
1749                         goto exit1;
1750                 case LAST_DOT:
1751                         error = -EINVAL;
1752                         goto exit1;
1753                 case LAST_ROOT:
1754                         error = -EBUSY;
1755                         goto exit1;
1756         }
1757         down(&nd.dentry->d_inode->i_sem);
1758         dentry = lookup_hash(&nd.last, nd.dentry);
1759         error = PTR_ERR(dentry);
1760         if (!IS_ERR(dentry)) {
1761                 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1762                 dput(dentry);
1763         }
1764         up(&nd.dentry->d_inode->i_sem);
1765 exit1:
1766         path_release(&nd);
1767 exit:
1768         putname(name);
1769         return error;
1770 }
1771
1772 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1773 {
1774         int error = may_delete(dir, dentry, 0);
1775
1776         if (error)
1777                 return error;
1778
1779         if (!dir->i_op || !dir->i_op->unlink)
1780                 return -EPERM;
1781
1782         DQUOT_INIT(dir);
1783
1784         down(&dentry->d_inode->i_sem);
1785         if (d_mountpoint(dentry))
1786                 error = -EBUSY;
1787         else {
1788                 error = security_inode_unlink(dir, dentry);
1789                 if (!error)
1790                         error = dir->i_op->unlink(dir, dentry);
1791         }
1792         up(&dentry->d_inode->i_sem);
1793
1794         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1795         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1796                 d_delete(dentry);
1797                 inode_dir_notify(dir, DN_DELETE);
1798         }
1799         return error;
1800 }
1801
1802 /*
1803  * Make sure that the actual truncation of the file will occur outside its
1804  * directory's i_sem.  Truncate can take a long time if there is a lot of
1805  * writeout happening, and we don't want to prevent access to the directory
1806  * while waiting on the I/O.
1807  */
1808 asmlinkage long sys_unlink(const char __user * pathname)
1809 {
1810         int error = 0;
1811         char * name;
1812         struct dentry *dentry;
1813         struct nameidata nd;
1814         struct inode *inode = NULL;
1815
1816         name = getname(pathname);
1817         if(IS_ERR(name))
1818                 return PTR_ERR(name);
1819
1820         error = path_lookup(name, LOOKUP_PARENT, &nd);
1821         if (error)
1822                 goto exit;
1823         error = -EISDIR;
1824         if (nd.last_type != LAST_NORM)
1825                 goto exit1;
1826         down(&nd.dentry->d_inode->i_sem);
1827         dentry = lookup_hash(&nd.last, nd.dentry);
1828         error = PTR_ERR(dentry);
1829         if (!IS_ERR(dentry)) {
1830                 /* Why not before? Because we want correct error value */
1831                 if (nd.last.name[nd.last.len])
1832                         goto slashes;
1833                 inode = dentry->d_inode;
1834                 if (inode)
1835                         atomic_inc(&inode->i_count);
1836                 error = vfs_unlink(nd.dentry->d_inode, dentry);
1837         exit2:
1838                 dput(dentry);
1839         }
1840         up(&nd.dentry->d_inode->i_sem);
1841 exit1:
1842         path_release(&nd);
1843 exit:
1844         putname(name);
1845
1846         if (inode)
1847                 iput(inode);    /* truncate the inode here */
1848         return error;
1849
1850 slashes:
1851         error = !dentry->d_inode ? -ENOENT :
1852                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1853         goto exit2;
1854 }
1855
1856 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1857 {
1858         int error = may_create(dir, dentry, NULL);
1859
1860         if (error)
1861                 return error;
1862
1863         if (!dir->i_op || !dir->i_op->symlink)
1864                 return -EPERM;
1865
1866         error = security_inode_symlink(dir, dentry, oldname);
1867         if (error)
1868                 return error;
1869
1870         DQUOT_INIT(dir);
1871         error = dir->i_op->symlink(dir, dentry, oldname);
1872         if (!error) {
1873                 inode_dir_notify(dir, DN_CREATE);
1874                 security_inode_post_symlink(dir, dentry, oldname);
1875         }
1876         return error;
1877 }
1878
1879 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1880 {
1881         int error = 0;
1882         char * from;
1883         char * to;
1884
1885         from = getname(oldname);
1886         if(IS_ERR(from))
1887                 return PTR_ERR(from);
1888         to = getname(newname);
1889         error = PTR_ERR(to);
1890         if (!IS_ERR(to)) {
1891                 struct dentry *dentry;
1892                 struct nameidata nd;
1893
1894                 error = path_lookup(to, LOOKUP_PARENT, &nd);
1895                 if (error)
1896                         goto out;
1897                 dentry = lookup_create(&nd, 0);
1898                 error = PTR_ERR(dentry);
1899                 if (!IS_ERR(dentry)) {
1900                         error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1901                         dput(dentry);
1902                 }
1903                 up(&nd.dentry->d_inode->i_sem);
1904                 path_release(&nd);
1905 out:
1906                 putname(to);
1907         }
1908         putname(from);
1909         return error;
1910 }
1911
1912 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1913 {
1914         struct inode *inode = old_dentry->d_inode;
1915         int error;
1916
1917         if (!inode)
1918                 return -ENOENT;
1919
1920         error = may_create(dir, new_dentry, NULL);
1921         if (error)
1922                 return error;
1923
1924         if (dir->i_sb != inode->i_sb)
1925                 return -EXDEV;
1926
1927         /*
1928          * A link to an append-only or immutable file cannot be created.
1929          */
1930         if (IS_APPEND(inode) || IS_IXORUNLINK(inode))
1931                 return -EPERM;
1932         if (!dir->i_op || !dir->i_op->link)
1933                 return -EPERM;
1934         if (S_ISDIR(old_dentry->d_inode->i_mode))
1935                 return -EPERM;
1936
1937         error = security_inode_link(old_dentry, dir, new_dentry);
1938         if (error)
1939                 return error;
1940
1941         down(&old_dentry->d_inode->i_sem);
1942         DQUOT_INIT(dir);
1943         error = dir->i_op->link(old_dentry, dir, new_dentry);
1944         up(&old_dentry->d_inode->i_sem);
1945         if (!error) {
1946                 inode_dir_notify(dir, DN_CREATE);
1947                 security_inode_post_link(old_dentry, dir, new_dentry);
1948         }
1949         return error;
1950 }
1951
1952 /*
1953  * Hardlinks are often used in delicate situations.  We avoid
1954  * security-related surprises by not following symlinks on the
1955  * newname.  --KAB
1956  *
1957  * We don't follow them on the oldname either to be compatible
1958  * with linux 2.0, and to avoid hard-linking to directories
1959  * and other special files.  --ADM
1960  */
1961 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
1962 {
1963         struct dentry *new_dentry;
1964         struct nameidata nd, old_nd;
1965         int error;
1966         char * to;
1967
1968         to = getname(newname);
1969         if (IS_ERR(to))
1970                 return PTR_ERR(to);
1971
1972         error = __user_walk(oldname, 0, &old_nd);
1973         if (error)
1974                 goto exit;
1975         error = path_lookup(to, LOOKUP_PARENT, &nd);
1976         if (error)
1977                 goto out;
1978         error = -EXDEV;
1979         if (old_nd.mnt != nd.mnt)
1980                 goto out_release;
1981         new_dentry = lookup_create(&nd, 0);
1982         error = PTR_ERR(new_dentry);
1983         if (!IS_ERR(new_dentry)) {
1984                 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1985                 dput(new_dentry);
1986         }
1987         up(&nd.dentry->d_inode->i_sem);
1988 out_release:
1989         path_release(&nd);
1990 out:
1991         path_release(&old_nd);
1992 exit:
1993         putname(to);
1994
1995         return error;
1996 }
1997
1998 /*
1999  * The worst of all namespace operations - renaming directory. "Perverted"
2000  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2001  * Problems:
2002  *      a) we can get into loop creation. Check is done in is_subdir().
2003  *      b) race potential - two innocent renames can create a loop together.
2004  *         That's where 4.4 screws up. Current fix: serialization on
2005  *         sb->s_vfs_rename_sem. We might be more accurate, but that's another
2006  *         story.
2007  *      c) we have to lock _three_ objects - parents and victim (if it exists).
2008  *         And that - after we got ->i_sem on parents (until then we don't know
2009  *         whether the target exists).  Solution: try to be smart with locking
2010  *         order for inodes.  We rely on the fact that tree topology may change
2011  *         only under ->s_vfs_rename_sem _and_ that parent of the object we
2012  *         move will be locked.  Thus we can rank directories by the tree
2013  *         (ancestors first) and rank all non-directories after them.
2014  *         That works since everybody except rename does "lock parent, lookup,
2015  *         lock child" and rename is under ->s_vfs_rename_sem.
2016  *         HOWEVER, it relies on the assumption that any object with ->lookup()
2017  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
2018  *         we'd better make sure that there's no link(2) for them.
2019  *      d) some filesystems don't support opened-but-unlinked directories,
2020  *         either because of layout or because they are not ready to deal with
2021  *         all cases correctly. The latter will be fixed (taking this sort of
2022  *         stuff into VFS), but the former is not going away. Solution: the same
2023  *         trick as in rmdir().
2024  *      e) conversion from fhandle to dentry may come in the wrong moment - when
2025  *         we are removing the target. Solution: we will have to grab ->i_sem
2026  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2027  *         ->i_sem on parents, which works but leads to some truely excessive
2028  *         locking].
2029  */
2030 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2031                struct inode *new_dir, struct dentry *new_dentry)
2032 {
2033         int error = 0;
2034         struct inode *target;
2035
2036         /*
2037          * If we are going to change the parent - check write permissions,
2038          * we'll need to flip '..'.
2039          */
2040         if (new_dir != old_dir) {
2041                 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2042                 if (error)
2043                         return error;
2044         }
2045
2046         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2047         if (error)
2048                 return error;
2049
2050         target = new_dentry->d_inode;
2051         if (target) {
2052                 down(&target->i_sem);
2053                 d_unhash(new_dentry);
2054         }
2055         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2056                 error = -EBUSY;
2057         else
2058                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2059         if (target) {
2060                 if (!error)
2061                         target->i_flags |= S_DEAD;
2062                 up(&target->i_sem);
2063                 if (d_unhashed(new_dentry))
2064                         d_rehash(new_dentry);
2065                 dput(new_dentry);
2066         }
2067         if (!error) {
2068                 d_move(old_dentry,new_dentry);
2069                 security_inode_post_rename(old_dir, old_dentry,
2070                                            new_dir, new_dentry);
2071         }
2072         return error;
2073 }
2074
2075 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2076                struct inode *new_dir, struct dentry *new_dentry)
2077 {
2078         struct inode *target;
2079         int error;
2080
2081         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2082         if (error)
2083                 return error;
2084
2085         dget(new_dentry);
2086         target = new_dentry->d_inode;
2087         if (target)
2088                 down(&target->i_sem);
2089         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2090                 error = -EBUSY;
2091         else
2092                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2093         if (!error) {
2094                 /* The following d_move() should become unconditional */
2095                 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2096                         d_move(old_dentry, new_dentry);
2097                 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2098         }
2099         if (target)
2100                 up(&target->i_sem);
2101         dput(new_dentry);
2102         return error;
2103 }
2104
2105 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2106                struct inode *new_dir, struct dentry *new_dentry)
2107 {
2108         int error;
2109         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2110
2111         if (old_dentry->d_inode == new_dentry->d_inode)
2112                 return 0;
2113
2114         error = may_delete(old_dir, old_dentry, is_dir);
2115         if (error)
2116                 return error;
2117
2118         if (!new_dentry->d_inode)
2119                 error = may_create(new_dir, new_dentry, NULL);
2120         else
2121                 error = may_delete(new_dir, new_dentry, is_dir);
2122         if (error)
2123                 return error;
2124
2125         if (!old_dir->i_op || !old_dir->i_op->rename)
2126                 return -EPERM;
2127
2128         DQUOT_INIT(old_dir);
2129         DQUOT_INIT(new_dir);
2130
2131         if (is_dir)
2132                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2133         else
2134                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2135         if (!error) {
2136                 if (old_dir == new_dir)
2137                         inode_dir_notify(old_dir, DN_RENAME);
2138                 else {
2139                         inode_dir_notify(old_dir, DN_DELETE);
2140                         inode_dir_notify(new_dir, DN_CREATE);
2141                 }
2142         }
2143         return error;
2144 }
2145
2146 static inline int do_rename(const char * oldname, const char * newname)
2147 {
2148         int error = 0;
2149         struct dentry * old_dir, * new_dir;
2150         struct dentry * old_dentry, *new_dentry;
2151         struct dentry * trap;
2152         struct nameidata oldnd, newnd;
2153
2154         error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2155         if (error)
2156                 goto exit;
2157
2158         error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2159         if (error)
2160                 goto exit1;
2161
2162         error = -EXDEV;
2163         if (oldnd.mnt != newnd.mnt)
2164                 goto exit2;
2165
2166         old_dir = oldnd.dentry;
2167         error = -EBUSY;
2168         if (oldnd.last_type != LAST_NORM)
2169                 goto exit2;
2170
2171         new_dir = newnd.dentry;
2172         if (newnd.last_type != LAST_NORM)
2173                 goto exit2;
2174
2175         trap = lock_rename(new_dir, old_dir);
2176
2177         old_dentry = lookup_hash(&oldnd.last, old_dir);
2178         error = PTR_ERR(old_dentry);
2179         if (IS_ERR(old_dentry))
2180                 goto exit3;
2181         /* source must exist */
2182         error = -ENOENT;
2183         if (!old_dentry->d_inode)
2184                 goto exit4;
2185         /* unless the source is a directory trailing slashes give -ENOTDIR */
2186         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2187                 error = -ENOTDIR;
2188                 if (oldnd.last.name[oldnd.last.len])
2189                         goto exit4;
2190                 if (newnd.last.name[newnd.last.len])
2191                         goto exit4;
2192         }
2193         /* source should not be ancestor of target */
2194         error = -EINVAL;
2195         if (old_dentry == trap)
2196                 goto exit4;
2197         new_dentry = lookup_hash(&newnd.last, new_dir);
2198         error = PTR_ERR(new_dentry);
2199         if (IS_ERR(new_dentry))
2200                 goto exit4;
2201         /* target should not be an ancestor of source */
2202         error = -ENOTEMPTY;
2203         if (new_dentry == trap)
2204                 goto exit5;
2205
2206         error = vfs_rename(old_dir->d_inode, old_dentry,
2207                                    new_dir->d_inode, new_dentry);
2208 exit5:
2209         dput(new_dentry);
2210 exit4:
2211         dput(old_dentry);
2212 exit3:
2213         unlock_rename(new_dir, old_dir);
2214 exit2:
2215         path_release(&newnd);
2216 exit1:
2217         path_release(&oldnd);
2218 exit:
2219         return error;
2220 }
2221
2222 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2223 {
2224         int error;
2225         char * from;
2226         char * to;
2227
2228         from = getname(oldname);
2229         if(IS_ERR(from))
2230                 return PTR_ERR(from);
2231         to = getname(newname);
2232         error = PTR_ERR(to);
2233         if (!IS_ERR(to)) {
2234                 error = do_rename(from,to);
2235                 putname(to);
2236         }
2237         putname(from);
2238         return error;
2239 }
2240
2241 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2242 {
2243         int len;
2244
2245         len = PTR_ERR(link);
2246         if (IS_ERR(link))
2247                 goto out;
2248
2249         len = strlen(link);
2250         if (len > (unsigned) buflen)
2251                 len = buflen;
2252         if (copy_to_user(buffer, link, len))
2253                 len = -EFAULT;
2254 out:
2255         return len;
2256 }
2257
2258 /*
2259  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
2260  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
2261  * using) it for any given inode is up to filesystem.
2262  */
2263 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2264 {
2265         struct nameidata nd;
2266         int res;
2267         nd.depth = 0;
2268         res = dentry->d_inode->i_op->follow_link(dentry, &nd);
2269         if (!res) {
2270                 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2271                 if (dentry->d_inode->i_op->put_link)
2272                         dentry->d_inode->i_op->put_link(dentry, &nd);
2273         }
2274         return res;
2275 }
2276
2277 int vfs_follow_link(struct nameidata *nd, const char *link)
2278 {
2279         return __vfs_follow_link(nd, link);
2280 }
2281
2282 /* get the link contents into pagecache */
2283 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2284 {
2285         struct page * page;
2286         struct address_space *mapping = dentry->d_inode->i_mapping;
2287         page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2288                                 NULL);
2289         if (IS_ERR(page))
2290                 goto sync_fail;
2291         wait_on_page_locked(page);
2292         if (!PageUptodate(page))
2293                 goto async_fail;
2294         *ppage = page;
2295         return kmap(page);
2296
2297 async_fail:
2298         page_cache_release(page);
2299         return ERR_PTR(-EIO);
2300
2301 sync_fail:
2302         return (char*)page;
2303 }
2304
2305 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2306 {
2307         struct page *page = NULL;
2308         char *s = page_getlink(dentry, &page);
2309         int res = vfs_readlink(dentry,buffer,buflen,s);
2310         if (page) {
2311                 kunmap(page);
2312                 page_cache_release(page);
2313         }
2314         return res;
2315 }
2316
2317 int page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2318 {
2319         struct page *page;
2320         char *s = page_getlink(dentry, &page);
2321         if (!IS_ERR(s)) {
2322                 nd_set_link(nd, s);
2323                 s = NULL;
2324         }
2325         return PTR_ERR(s);
2326 }
2327
2328 void page_put_link(struct dentry *dentry, struct nameidata *nd)
2329 {
2330         if (!IS_ERR(nd_get_link(nd))) {
2331                 struct page *page;
2332                 page = find_get_page(dentry->d_inode->i_mapping, 0);
2333                 if (!page)
2334                         BUG();
2335                 kunmap(page);
2336                 page_cache_release(page);
2337                 page_cache_release(page);
2338         }
2339 }
2340
2341 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
2342 {
2343         struct page *page = NULL;
2344         char *s = page_getlink(dentry, &page);
2345         int res = __vfs_follow_link(nd, s);
2346         if (page) {
2347                 kunmap(page);
2348                 page_cache_release(page);
2349         }
2350         return res;
2351 }
2352
2353 int page_symlink(struct inode *inode, const char *symname, int len)
2354 {
2355         struct address_space *mapping = inode->i_mapping;
2356         struct page *page = grab_cache_page(mapping, 0);
2357         int err = -ENOMEM;
2358         char *kaddr;
2359
2360         if (!page)
2361                 goto fail;
2362         err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2363         if (err)
2364                 goto fail_map;
2365         kaddr = kmap_atomic(page, KM_USER0);
2366         memcpy(kaddr, symname, len-1);
2367         kunmap_atomic(kaddr, KM_USER0);
2368         mapping->a_ops->commit_write(NULL, page, 0, len-1);
2369         /*
2370          * Notice that we are _not_ going to block here - end of page is
2371          * unmapped, so this will only try to map the rest of page, see
2372          * that it is unmapped (typically even will not look into inode -
2373          * ->i_size will be enough for everything) and zero it out.
2374          * OTOH it's obviously correct and should make the page up-to-date.
2375          */
2376         if (!PageUptodate(page)) {
2377                 err = mapping->a_ops->readpage(NULL, page);
2378                 wait_on_page_locked(page);
2379         } else {
2380                 unlock_page(page);
2381         }
2382         page_cache_release(page);
2383         if (err < 0)
2384                 goto fail;
2385         mark_inode_dirty(inode);
2386         return 0;
2387 fail_map:
2388         unlock_page(page);
2389         page_cache_release(page);
2390 fail:
2391         return err;
2392 }
2393
2394 struct inode_operations page_symlink_inode_operations = {
2395         .readlink       = generic_readlink,
2396         .follow_link    = page_follow_link_light,
2397         .put_link       = page_put_link,
2398 };
2399
2400 EXPORT_SYMBOL(__user_walk);
2401 EXPORT_SYMBOL(follow_down);
2402 EXPORT_SYMBOL(follow_up);
2403 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2404 EXPORT_SYMBOL(getname);
2405 EXPORT_SYMBOL(lock_rename);
2406 EXPORT_SYMBOL(lookup_create);
2407 EXPORT_SYMBOL(lookup_hash);
2408 EXPORT_SYMBOL(lookup_one_len);
2409 EXPORT_SYMBOL(page_follow_link);
2410 EXPORT_SYMBOL(page_follow_link_light);
2411 EXPORT_SYMBOL(page_put_link);
2412 EXPORT_SYMBOL(page_readlink);
2413 EXPORT_SYMBOL(page_symlink);
2414 EXPORT_SYMBOL(page_symlink_inode_operations);
2415 EXPORT_SYMBOL(path_lookup);
2416 EXPORT_SYMBOL(path_release);
2417 EXPORT_SYMBOL(path_walk);
2418 EXPORT_SYMBOL(permission);
2419 EXPORT_SYMBOL(unlock_rename);
2420 EXPORT_SYMBOL(vfs_create);
2421 EXPORT_SYMBOL(vfs_follow_link);
2422 EXPORT_SYMBOL(vfs_link);
2423 EXPORT_SYMBOL(vfs_mkdir);
2424 EXPORT_SYMBOL(vfs_mknod);
2425 EXPORT_SYMBOL(vfs_permission);
2426 EXPORT_SYMBOL(vfs_readlink);
2427 EXPORT_SYMBOL(vfs_rename);
2428 EXPORT_SYMBOL(vfs_rmdir);
2429 EXPORT_SYMBOL(vfs_symlink);
2430 EXPORT_SYMBOL(vfs_unlink);
2431 EXPORT_SYMBOL(generic_readlink);