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