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