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/export.h>
  19 #include <linux/kernel.h>
  20 #include <linux/slab.h>
  21 #include <linux/fs.h>
  22 #include <linux/namei.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/personality.h>
  26 #include <linux/security.h>
  27 #include <linux/ima.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/mount.h>
  30 #include <linux/audit.h>
  31 #include <linux/capability.h>
  32 #include <linux/file.h>
  33 #include <linux/fcntl.h>
  34 #include <linux/device_cgroup.h>
  35 #include <linux/fs_struct.h>
  36 #include <linux/posix_acl.h>
  37 #include <asm/uaccess.h>
  38
  39 #include "internal.h"
  40 #include "mount.h"
  41
  42 /* [Feb-1997 T. Schoebel-Theuer]
  43  * Fundamental changes in the pathname lookup mechanisms (namei)
  44  * were necessary because of omirr.  The reason is that omirr needs
  45  * to know the _real_ pathname, not the user-supplied one, in case
  46  * of symlinks (and also when transname replacements occur).
  47  *
  48  * The new code replaces the old recursive symlink resolution with
  49  * an iterative one (in case of non-nested symlink chains).  It does
  50  * this with calls to <fs>_follow_link().
  51  * As a side effect, dir_namei(), _namei() and follow_link() are now
  52  * replaced with a single function lookup_dentry() that can handle all
  53  * the special cases of the former code.
  54  *
  55  * With the new dcache, the pathname is stored at each inode, at least as
  56  * long as the refcount of the inode is positive.  As a side effect, the
  57  * size of the dcache depends on the inode cache and thus is dynamic.
  58  *
  59  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  60  * resolution to correspond with current state of the code.
  61  *
  62  * Note that the symlink resolution is not *completely* iterative.
  63  * There is still a significant amount of tail- and mid- recursion in
  64  * the algorithm.  Also, note that <fs>_readlink() is not used in
  65  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  66  * may return different results than <fs>_follow_link().  Many virtual
  67  * filesystems (including /proc) exhibit this behavior.
  68  */
  69
  70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  71  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  72  * and the name already exists in form of a symlink, try to create the new
  73  * name indicated by the symlink. The old code always complained that the
  74  * name already exists, due to not following the symlink even if its target
  75  * is nonexistent.  The new semantics affects also mknod() and link() when
  76  * the name is a symlink pointing to a non-existent name.
  77  *
  78  * I don't know which semantics is the right one, since I have no access
  79  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  80  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  81  * "old" one. Personally, I think the new semantics is much more logical.
  82  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  83  * file does succeed in both HP-UX and SunOs, but not in Solaris
  84  * and in the old Linux semantics.
  85  */
  86
  87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  88  * semantics.  See the comments in "open_namei" and "do_link" below.
  89  *
  90  * [10-Sep-98 Alan Modra] Another symlink change.
  91  */
  92
  93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  94  *      inside the path - always follow.
  95  *      in the last component in creation/removal/renaming - never follow.
  96  *      if LOOKUP_FOLLOW passed - follow.
  97  *      if the pathname has trailing slashes - follow.
  98  *      otherwise - don't follow.
  99  * (applied in that order).
 100  *
 101  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 102  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 103  * During the 2.4 we need to fix the userland stuff depending on it -
 104  * hopefully we will be able to get rid of that wart in 2.5. So far only
 105  * XEmacs seems to be relying on it...
 106  */
 107 /*
 108  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 109  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 110  * any extra contention...
 111  */
 112
 113 /* In order to reduce some races, while at the same time doing additional
 114  * checking and hopefully speeding things up, we copy filenames to the
 115  * kernel data space before using them..
 116  *
 117  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 118  * PATH_MAX includes the nul terminator --RR.
 119  */
 120 static char *getname_flags(const char __user *filename, int flags, int *empty)
 121 {
 122         char *result = __getname(), *err;
 123         int len;
 124
 125         if (unlikely(!result))
 126                 return ERR_PTR(-ENOMEM);
 127
 128         len = strncpy_from_user(result, filename, PATH_MAX);
 129         err = ERR_PTR(len);
 130         if (unlikely(len < 0))
 131                 goto error;
 132
 133         /* The empty path is special. */
 134         if (unlikely(!len)) {
 135                 if (empty)
 136                         *empty = 1;
 137                 err = ERR_PTR(-ENOENT);
 138                 if (!(flags & LOOKUP_EMPTY))
 139                         goto error;
 140         }
 141
 142         err = ERR_PTR(-ENAMETOOLONG);
 143         if (likely(len < PATH_MAX)) {
 144                 audit_getname(result);
 145                 return result;
 146         }
 147
 148 error:
 149         __putname(result);
 150         return err;
 151 }
 152
 153 char *getname(const char __user * filename)
 154 {
 155         return getname_flags(filename, 0, NULL);
 156 }
 157
 158 #ifdef CONFIG_AUDITSYSCALL
 159 void putname(const char *name)
 160 {
 161         if (unlikely(!audit_dummy_context()))
 162                 audit_putname(name);
 163         else
 164                 __putname(name);
 165 }
 166 EXPORT_SYMBOL(putname);
 167 #endif
 168
 169 static int check_acl(struct inode *inode, int mask)
 170 {
 171 #ifdef CONFIG_FS_POSIX_ACL
 172         struct posix_acl *acl;
 173
 174         if (mask & MAY_NOT_BLOCK) {
 175                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
 176                 if (!acl)
 177                         return -EAGAIN;
 178                 /* no ->get_acl() calls in RCU mode... */
 179                 if (acl == ACL_NOT_CACHED)
 180                         return -ECHILD;
 181                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
 182         }
 183
 184         acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
 185
 186         /*
 187          * A filesystem can force a ACL callback by just never filling the
 188          * ACL cache. But normally you'd fill the cache either at inode
 189          * instantiation time, or on the first ->get_acl call.
 190          *
 191          * If the filesystem doesn't have a get_acl() function at all, we'll
 192          * just create the negative cache entry.
 193          */
 194         if (acl == ACL_NOT_CACHED) {
 195                 if (inode->i_op->get_acl) {
 196                         acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
 197                         if (IS_ERR(acl))
 198                                 return PTR_ERR(acl);
 199                 } else {
 200                         set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
 201                         return -EAGAIN;
 202                 }
 203         }
 204
 205         if (acl) {
 206                 int error = posix_acl_permission(inode, acl, mask);
 207                 posix_acl_release(acl);
 208                 return error;
 209         }
 210 #endif
 211
 212         return -EAGAIN;
 213 }
 214
 215 /*
 216  * This does the basic permission checking
 217  */
 218 static int acl_permission_check(struct inode *inode, int mask)
 219 {
 220         unsigned int mode = inode->i_mode;
 221
 222         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
 223                 mode >>= 6;
 224         else {
 225                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 226                         int error = check_acl(inode, mask);
 227                         if (error != -EAGAIN)
 228                                 return error;
 229                 }
 230
 231                 if (in_group_p(inode->i_gid))
 232                         mode >>= 3;
 233         }
 234
 235         /*
 236          * If the DACs are ok we don't need any capability check.
 237          */
 238         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
 239                 return 0;
 240         return -EACCES;
 241 }
 242
 243 /**
 244  * generic_permission -  check for access rights on a Posix-like filesystem
 245  * @inode:      inode to check access rights for
 246  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
 247  *
 248  * Used to check for read/write/execute permissions on a file.
 249  * We use "fsuid" for this, letting us set arbitrary permissions
 250  * for filesystem access without changing the "normal" uids which
 251  * are used for other things.
 252  *
 253  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 254  * request cannot be satisfied (eg. requires blocking or too much complexity).
 255  * It would then be called again in ref-walk mode.
 256  */
 257 int generic_permission(struct inode *inode, int mask)
 258 {
 259         int ret;
 260
 261         /*
 262          * Do the basic permission checks.
 263          */
 264         ret = acl_permission_check(inode, mask);
 265         if (ret != -EACCES)
 266                 return ret;
 267
 268         if (S_ISDIR(inode->i_mode)) {
 269                 /* DACs are overridable for directories */
 270                 if (inode_capable(inode, CAP_DAC_OVERRIDE))
 271                         return 0;
 272                 if (!(mask & MAY_WRITE))
 273                         if (inode_capable(inode, CAP_DAC_READ_SEARCH))
 274                                 return 0;
 275                 return -EACCES;
 276         }
 277         /*
 278          * Read/write DACs are always overridable.
 279          * Executable DACs are overridable when there is
 280          * at least one exec bit set.
 281          */
 282         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 283                 if (inode_capable(inode, CAP_DAC_OVERRIDE))
 284                         return 0;
 285
 286         /*
 287          * Searching includes executable on directories, else just read.
 288          */
 289         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 290         if (mask == MAY_READ)
 291                 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
 292                         return 0;
 293
 294         return -EACCES;
 295 }
 296
 297 /*
 298  * We _really_ want to just do "generic_permission()" without
 299  * even looking at the inode->i_op values. So we keep a cache
 300  * flag in inode->i_opflags, that says "this has not special
 301  * permission function, use the fast case".
 302  */
 303 static inline int do_inode_permission(struct inode *inode, int mask)
 304 {
 305         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
 306                 if (likely(inode->i_op->permission))
 307                         return inode->i_op->permission(inode, mask);
 308
 309                 /* This gets set once for the inode lifetime */
 310                 spin_lock(&inode->i_lock);
 311                 inode->i_opflags |= IOP_FASTPERM;
 312                 spin_unlock(&inode->i_lock);
 313         }
 314         return generic_permission(inode, mask);
 315 }
 316
 317 /**
 318  * inode_permission  -  check for access rights to a given inode
 319  * @inode:      inode to check permission on
 320  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
 321  *
 322  * Used to check for read/write/execute permissions on an inode.
 323  * We use "fsuid" for this, letting us set arbitrary permissions
 324  * for filesystem access without changing the "normal" uids which
 325  * are used for other things.
 326  *
 327  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
 328  */
 329 int inode_permission(struct inode *inode, int mask)
 330 {
 331         int retval;
 332
 333         if (unlikely(mask & MAY_WRITE)) {
 334                 umode_t mode = inode->i_mode;
 335
 336                 /*
 337                  * Nobody gets write access to a read-only fs.
 338                  */
 339                 if (IS_RDONLY(inode) &&
 340                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 341                         return -EROFS;
 342
 343                 /*
 344                  * Nobody gets write access to an immutable file.
 345                  */
 346                 if (IS_IMMUTABLE(inode))
 347                         return -EACCES;
 348         }
 349
 350         retval = do_inode_permission(inode, mask);
 351         if (retval)
 352                 return retval;
 353
 354         retval = devcgroup_inode_permission(inode, mask);
 355         if (retval)
 356                 return retval;
 357
 358         return security_inode_permission(inode, mask);
 359 }
 360
 361 /**
 362  * path_get - get a reference to a path
 363  * @path: path to get the reference to
 364  *
 365  * Given a path increment the reference count to the dentry and the vfsmount.
 366  */
 367 void path_get(struct path *path)
 368 {
 369         mntget(path->mnt);
 370         dget(path->dentry);
 371 }
 372 EXPORT_SYMBOL(path_get);
 373
 374 /**
 375  * path_put - put a reference to a path
 376  * @path: path to put the reference to
 377  *
 378  * Given a path decrement the reference count to the dentry and the vfsmount.
 379  */
 380 void path_put(struct path *path)
 381 {
 382         dput(path->dentry);
 383         mntput(path->mnt);
 384 }
 385 EXPORT_SYMBOL(path_put);
 386
 387 /*
 388  * Path walking has 2 modes, rcu-walk and ref-walk (see
 389  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 390  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 391  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
 392  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 393  * got stuck, so ref-walk may continue from there. If this is not successful
 394  * (eg. a seqcount has changed), then failure is returned and it's up to caller
 395  * to restart the path walk from the beginning in ref-walk mode.
 396  */
 397
 398 /**
 399  * unlazy_walk - try to switch to ref-walk mode.
 400  * @nd: nameidata pathwalk data
 401  * @dentry: child of nd->path.dentry or NULL
 402  * Returns: 0 on success, -ECHILD on failure
 403  *
 404  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
 405  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
 406  * @nd or NULL.  Must be called from rcu-walk context.
 407  */
 408 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
 409 {
 410         struct fs_struct *fs = current->fs;
 411         struct dentry *parent = nd->path.dentry;
 412         int want_root = 0;
 413
 414         BUG_ON(!(nd->flags & LOOKUP_RCU));
 415         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
 416                 want_root = 1;
 417                 spin_lock(&fs->lock);
 418                 if (nd->root.mnt != fs->root.mnt ||
 419                                 nd->root.dentry != fs->root.dentry)
 420                         goto err_root;
 421         }
 422         spin_lock(&parent->d_lock);
 423         if (!dentry) {
 424                 if (!__d_rcu_to_refcount(parent, nd->seq))
 425                         goto err_parent;
 426                 BUG_ON(nd->inode != parent->d_inode);
 427         } else {
 428                 if (dentry->d_parent != parent)
 429                         goto err_parent;
 430                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 431                 if (!__d_rcu_to_refcount(dentry, nd->seq))
 432                         goto err_child;
 433                 /*
 434                  * If the sequence check on the child dentry passed, then
 435                  * the child has not been removed from its parent. This
 436                  * means the parent dentry must be valid and able to take
 437                  * a reference at this point.
 438                  */
 439                 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
 440                 BUG_ON(!parent->d_count);
 441                 parent->d_count++;
 442                 spin_unlock(&dentry->d_lock);
 443         }
 444         spin_unlock(&parent->d_lock);
 445         if (want_root) {
 446                 path_get(&nd->root);
 447                 spin_unlock(&fs->lock);
 448         }
 449         mntget(nd->path.mnt);
 450
 451         rcu_read_unlock();
 452         br_read_unlock(vfsmount_lock);
 453         nd->flags &= ~LOOKUP_RCU;
 454         return 0;
 455
 456 err_child:
 457         spin_unlock(&dentry->d_lock);
 458 err_parent:
 459         spin_unlock(&parent->d_lock);
 460 err_root:
 461         if (want_root)
 462                 spin_unlock(&fs->lock);
 463         return -ECHILD;
 464 }
 465
 466 /**
 467  * release_open_intent - free up open intent resources
 468  * @nd: pointer to nameidata
 469  */
 470 void release_open_intent(struct nameidata *nd)
 471 {
 472         struct file *file = nd->intent.open.file;
 473
 474         if (file && !IS_ERR(file)) {
 475                 if (file->f_path.dentry == NULL)
 476                         put_filp(file);
 477                 else
 478                         fput(file);
 479         }
 480 }
 481
 482 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
 483 {
 484         return dentry->d_op->d_revalidate(dentry, nd);
 485 }
 486
 487 /**
 488  * complete_walk - successful completion of path walk
 489  * @nd:  pointer nameidata
 490  *
 491  * If we had been in RCU mode, drop out of it and legitimize nd->path.
 492  * Revalidate the final result, unless we'd already done that during
 493  * the path walk or the filesystem doesn't ask for it.  Return 0 on
 494  * success, -error on failure.  In case of failure caller does not
 495  * need to drop nd->path.
 496  */
 497 static int complete_walk(struct nameidata *nd)
 498 {
 499         struct dentry *dentry = nd->path.dentry;
 500         int status;
 501
 502         if (nd->flags & LOOKUP_RCU) {
 503                 nd->flags &= ~LOOKUP_RCU;
 504                 if (!(nd->flags & LOOKUP_ROOT))
 505                         nd->root.mnt = NULL;
 506                 spin_lock(&dentry->d_lock);
 507                 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
 508                         spin_unlock(&dentry->d_lock);
 509                         rcu_read_unlock();
 510                         br_read_unlock(vfsmount_lock);
 511                         return -ECHILD;
 512                 }
 513                 BUG_ON(nd->inode != dentry->d_inode);
 514                 spin_unlock(&dentry->d_lock);
 515                 mntget(nd->path.mnt);
 516                 rcu_read_unlock();
 517                 br_read_unlock(vfsmount_lock);
 518         }
 519
 520         if (likely(!(nd->flags & LOOKUP_JUMPED)))
 521                 return 0;
 522
 523         if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
 524                 return 0;
 525
 526         if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
 527                 return 0;
 528
 529         /* Note: we do not d_invalidate() */
 530         status = d_revalidate(dentry, nd);
 531         if (status > 0)
 532                 return 0;
 533
 534         if (!status)
 535                 status = -ESTALE;
 536
 537         path_put(&nd->path);
 538         return status;
 539 }
 540
 541 static __always_inline void set_root(struct nameidata *nd)
 542 {
 543         if (!nd->root.mnt)
 544                 get_fs_root(current->fs, &nd->root);
 545 }
 546
 547 static int link_path_walk(const char *, struct nameidata *);
 548
 549 static __always_inline void set_root_rcu(struct nameidata *nd)
 550 {
 551         if (!nd->root.mnt) {
 552                 struct fs_struct *fs = current->fs;
 553                 unsigned seq;
 554
 555                 do {
 556                         seq = read_seqcount_begin(&fs->seq);
 557                         nd->root = fs->root;
 558                         nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 559                 } while (read_seqcount_retry(&fs->seq, seq));
 560         }
 561 }
 562
 563 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 564 {
 565         int ret;
 566
 567         if (IS_ERR(link))
 568                 goto fail;
 569
 570         if (*link == '/') {
 571                 set_root(nd);
 572                 path_put(&nd->path);
 573                 nd->path = nd->root;
 574                 path_get(&nd->root);
 575                 nd->flags |= LOOKUP_JUMPED;
 576         }
 577         nd->inode = nd->path.dentry->d_inode;
 578
 579         ret = link_path_walk(link, nd);
 580         return ret;
 581 fail:
 582         path_put(&nd->path);
 583         return PTR_ERR(link);
 584 }
 585
 586 static void path_put_conditional(struct path *path, struct nameidata *nd)
 587 {
 588         dput(path->dentry);
 589         if (path->mnt != nd->path.mnt)
 590                 mntput(path->mnt);
 591 }
 592
 593 static inline void path_to_nameidata(const struct path *path,
 594                                         struct nameidata *nd)
 595 {
 596         if (!(nd->flags & LOOKUP_RCU)) {
 597                 dput(nd->path.dentry);
 598                 if (nd->path.mnt != path->mnt)
 599                         mntput(nd->path.mnt);
 600         }
 601         nd->path.mnt = path->mnt;
 602         nd->path.dentry = path->dentry;
 603 }
 604
 605 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
 606 {
 607         struct inode *inode = link->dentry->d_inode;
 608         if (!IS_ERR(cookie) && inode->i_op->put_link)
 609                 inode->i_op->put_link(link->dentry, nd, cookie);
 610         path_put(link);
 611 }
 612
 613 static __always_inline int
 614 follow_link(struct path *link, struct nameidata *nd, void **p)
 615 {
 616         int error;
 617         struct dentry *dentry = link->dentry;
 618
 619         BUG_ON(nd->flags & LOOKUP_RCU);
 620
 621         if (link->mnt == nd->path.mnt)
 622                 mntget(link->mnt);
 623
 624         if (unlikely(current->total_link_count >= 40)) {
 625                 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
 626                 path_put(&nd->path);
 627                 return -ELOOP;
 628         }
 629         cond_resched();
 630         current->total_link_count++;
 631
 632         touch_atime(link);
 633         nd_set_link(nd, NULL);
 634
 635         error = security_inode_follow_link(link->dentry, nd);
 636         if (error) {
 637                 *p = ERR_PTR(error); /* no ->put_link(), please */
 638                 path_put(&nd->path);
 639                 return error;
 640         }
 641
 642         nd->last_type = LAST_BIND;
 643         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
 644         error = PTR_ERR(*p);
 645         if (!IS_ERR(*p)) {
 646                 char *s = nd_get_link(nd);
 647                 error = 0;
 648                 if (s)
 649                         error = __vfs_follow_link(nd, s);
 650                 else if (nd->last_type == LAST_BIND) {
 651                         nd->flags |= LOOKUP_JUMPED;
 652                         nd->inode = nd->path.dentry->d_inode;
 653                         if (nd->inode->i_op->follow_link) {
 654                                 /* stepped on a _really_ weird one */
 655                                 path_put(&nd->path);
 656                                 error = -ELOOP;
 657                         }
 658                 }
 659         }
 660         return error;
 661 }
 662
 663 static int follow_up_rcu(struct path *path)
 664 {
 665         struct mount *mnt = real_mount(path->mnt);
 666         struct mount *parent;
 667         struct dentry *mountpoint;
 668
 669         parent = mnt->mnt_parent;
 670         if (&parent->mnt == path->mnt)
 671                 return 0;
 672         mountpoint = mnt->mnt_mountpoint;
 673         path->dentry = mountpoint;
 674         path->mnt = &parent->mnt;
 675         return 1;
 676 }
 677
 678 int follow_up(struct path *path)
 679 {
 680         struct mount *mnt = real_mount(path->mnt);
 681         struct mount *parent;
 682         struct dentry *mountpoint;
 683
 684         br_read_lock(vfsmount_lock);
 685         parent = mnt->mnt_parent;
 686         if (&parent->mnt == path->mnt) {
 687                 br_read_unlock(vfsmount_lock);
 688                 return 0;
 689         }
 690         mntget(&parent->mnt);
 691         mountpoint = dget(mnt->mnt_mountpoint);
 692         br_read_unlock(vfsmount_lock);
 693         dput(path->dentry);
 694         path->dentry = mountpoint;
 695         mntput(path->mnt);
 696         path->mnt = &parent->mnt;
 697         return 1;
 698 }
 699
 700 /*
 701  * Perform an automount
 702  * - return -EISDIR to tell follow_managed() to stop and return the path we
 703  *   were called with.
 704  */
 705 static int follow_automount(struct path *path, unsigned flags,
 706                             bool *need_mntput)
 707 {
 708         struct vfsmount *mnt;
 709         int err;
 710
 711         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
 712                 return -EREMOTE;
 713
 714         /* We don't want to mount if someone's just doing a stat -
 715          * unless they're stat'ing a directory and appended a '/' to
 716          * the name.
 717          *
 718          * We do, however, want to mount if someone wants to open or
 719          * create a file of any type under the mountpoint, wants to
 720          * traverse through the mountpoint or wants to open the
 721          * mounted directory.  Also, autofs may mark negative dentries
 722          * as being automount points.  These will need the attentions
 723          * of the daemon to instantiate them before they can be used.
 724          */
 725         if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
 726                      LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
 727             path->dentry->d_inode)
 728                 return -EISDIR;
 729
 730         current->total_link_count++;
 731         if (current->total_link_count >= 40)
 732                 return -ELOOP;
 733
 734         mnt = path->dentry->d_op->d_automount(path);
 735         if (IS_ERR(mnt)) {
 736                 /*
 737                  * The filesystem is allowed to return -EISDIR here to indicate
 738                  * it doesn't want to automount.  For instance, autofs would do
 739                  * this so that its userspace daemon can mount on this dentry.
 740                  *
 741                  * However, we can only permit this if it's a terminal point in
 742                  * the path being looked up; if it wasn't then the remainder of
 743                  * the path is inaccessible and we should say so.
 744                  */
 745                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
 746                         return -EREMOTE;
 747                 return PTR_ERR(mnt);
 748         }
 749
 750         if (!mnt) /* mount collision */
 751                 return 0;
 752
 753         if (!*need_mntput) {
 754                 /* lock_mount() may release path->mnt on error */
 755                 mntget(path->mnt);
 756                 *need_mntput = true;
 757         }
 758         err = finish_automount(mnt, path);
 759
 760         switch (err) {
 761         case -EBUSY:
 762                 /* Someone else made a mount here whilst we were busy */
 763                 return 0;
 764         case 0:
 765                 path_put(path);
 766                 path->mnt = mnt;
 767                 path->dentry = dget(mnt->mnt_root);
 768                 return 0;
 769         default:
 770                 return err;
 771         }
 772
 773 }
 774
 775 /*
 776  * Handle a dentry that is managed in some way.
 777  * - Flagged for transit management (autofs)
 778  * - Flagged as mountpoint
 779  * - Flagged as automount point
 780  *
 781  * This may only be called in refwalk mode.
 782  *
 783  * Serialization is taken care of in namespace.c
 784  */
 785 static int follow_managed(struct path *path, unsigned flags)
 786 {
 787         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
 788         unsigned managed;
 789         bool need_mntput = false;
 790         int ret = 0;
 791
 792         /* Given that we're not holding a lock here, we retain the value in a
 793          * local variable for each dentry as we look at it so that we don't see
 794          * the components of that value change under us */
 795         while (managed = ACCESS_ONCE(path->dentry->d_flags),
 796                managed &= DCACHE_MANAGED_DENTRY,
 797                unlikely(managed != 0)) {
 798                 /* Allow the filesystem to manage the transit without i_mutex
 799                  * being held. */
 800                 if (managed & DCACHE_MANAGE_TRANSIT) {
 801                         BUG_ON(!path->dentry->d_op);
 802                         BUG_ON(!path->dentry->d_op->d_manage);
 803                         ret = path->dentry->d_op->d_manage(path->dentry, false);
 804                         if (ret < 0)
 805                                 break;
 806                 }
 807
 808                 /* Transit to a mounted filesystem. */
 809                 if (managed & DCACHE_MOUNTED) {
 810                         struct vfsmount *mounted = lookup_mnt(path);
 811                         if (mounted) {
 812                                 dput(path->dentry);
 813                                 if (need_mntput)
 814                                         mntput(path->mnt);
 815                                 path->mnt = mounted;
 816                                 path->dentry = dget(mounted->mnt_root);
 817                                 need_mntput = true;
 818                                 continue;
 819                         }
 820
 821                         /* Something is mounted on this dentry in another
 822                          * namespace and/or whatever was mounted there in this
 823                          * namespace got unmounted before we managed to get the
 824                          * vfsmount_lock */
 825                 }
 826
 827                 /* Handle an automount point */
 828                 if (managed & DCACHE_NEED_AUTOMOUNT) {
 829                         ret = follow_automount(path, flags, &need_mntput);
 830                         if (ret < 0)
 831                                 break;
 832                         continue;
 833                 }
 834
 835                 /* We didn't change the current path point */
 836                 break;
 837         }
 838
 839         if (need_mntput && path->mnt == mnt)
 840                 mntput(path->mnt);
 841         if (ret == -EISDIR)
 842                 ret = 0;
 843         return ret < 0 ? ret : need_mntput;
 844 }
 845
 846 int follow_down_one(struct path *path)
 847 {
 848         struct vfsmount *mounted;
 849
 850         mounted = lookup_mnt(path);
 851         if (mounted) {
 852                 dput(path->dentry);
 853                 mntput(path->mnt);
 854                 path->mnt = mounted;
 855                 path->dentry = dget(mounted->mnt_root);
 856                 return 1;
 857         }
 858         return 0;
 859 }
 860
 861 static inline bool managed_dentry_might_block(struct dentry *dentry)
 862 {
 863         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
 864                 dentry->d_op->d_manage(dentry, true) < 0);
 865 }
 866
 867 /*
 868  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
 869  * we meet a managed dentry that would need blocking.
 870  */
 871 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
 872                                struct inode **inode)
 873 {
 874         for (;;) {
 875                 struct mount *mounted;
 876                 /*
 877                  * Don't forget we might have a non-mountpoint managed dentry
 878                  * that wants to block transit.
 879                  */
 880                 if (unlikely(managed_dentry_might_block(path->dentry)))
 881                         return false;
 882
 883                 if (!d_mountpoint(path->dentry))
 884                         break;
 885
 886                 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
 887                 if (!mounted)
 888                         break;
 889                 path->mnt = &mounted->mnt;
 890                 path->dentry = mounted->mnt.mnt_root;
 891                 nd->flags |= LOOKUP_JUMPED;
 892                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
 893                 /*
 894                  * Update the inode too. We don't need to re-check the
 895                  * dentry sequence number here after this d_inode read,
 896                  * because a mount-point is always pinned.
 897                  */
 898                 *inode = path->dentry->d_inode;
 899         }
 900         return true;
 901 }
 902
 903 static void follow_mount_rcu(struct nameidata *nd)
 904 {
 905         while (d_mountpoint(nd->path.dentry)) {
 906                 struct mount *mounted;
 907                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
 908                 if (!mounted)
 909                         break;
 910                 nd->path.mnt = &mounted->mnt;
 911                 nd->path.dentry = mounted->mnt.mnt_root;
 912                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
 913         }
 914 }
 915
 916 static int follow_dotdot_rcu(struct nameidata *nd)
 917 {
 918         set_root_rcu(nd);
 919
 920         while (1) {
 921                 if (nd->path.dentry == nd->root.dentry &&
 922                     nd->path.mnt == nd->root.mnt) {
 923                         break;
 924                 }
 925                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
 926                         struct dentry *old = nd->path.dentry;
 927                         struct dentry *parent = old->d_parent;
 928                         unsigned seq;
 929
 930                         seq = read_seqcount_begin(&parent->d_seq);
 931                         if (read_seqcount_retry(&old->d_seq, nd->seq))
 932                                 goto failed;
 933                         nd->path.dentry = parent;
 934                         nd->seq = seq;
 935                         break;
 936                 }
 937                 if (!follow_up_rcu(&nd->path))
 938                         break;
 939                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
 940         }
 941         follow_mount_rcu(nd);
 942         nd->inode = nd->path.dentry->d_inode;
 943         return 0;
 944
 945 failed:
 946         nd->flags &= ~LOOKUP_RCU;
 947         if (!(nd->flags & LOOKUP_ROOT))
 948                 nd->root.mnt = NULL;
 949         rcu_read_unlock();
 950         br_read_unlock(vfsmount_lock);
 951         return -ECHILD;
 952 }
 953
 954 /*
 955  * Follow down to the covering mount currently visible to userspace.  At each
 956  * point, the filesystem owning that dentry may be queried as to whether the
 957  * caller is permitted to proceed or not.
 958  */
 959 int follow_down(struct path *path)
 960 {
 961         unsigned managed;
 962         int ret;
 963
 964         while (managed = ACCESS_ONCE(path->dentry->d_flags),
 965                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
 966                 /* Allow the filesystem to manage the transit without i_mutex
 967                  * being held.
 968                  *
 969                  * We indicate to the filesystem if someone is trying to mount
 970                  * something here.  This gives autofs the chance to deny anyone
 971                  * other than its daemon the right to mount on its
 972                  * superstructure.
 973                  *
 974                  * The filesystem may sleep at this point.
 975                  */
 976                 if (managed & DCACHE_MANAGE_TRANSIT) {
 977                         BUG_ON(!path->dentry->d_op);
 978                         BUG_ON(!path->dentry->d_op->d_manage);
 979                         ret = path->dentry->d_op->d_manage(
 980                                 path->dentry, false);
 981                         if (ret < 0)
 982                                 return ret == -EISDIR ? 0 : ret;
 983                 }
 984
 985                 /* Transit to a mounted filesystem. */
 986                 if (managed & DCACHE_MOUNTED) {
 987                         struct vfsmount *mounted = lookup_mnt(path);
 988                         if (!mounted)
 989                                 break;
 990                         dput(path->dentry);
 991                         mntput(path->mnt);
 992                         path->mnt = mounted;
 993                         path->dentry = dget(mounted->mnt_root);
 994                         continue;
 995                 }
 996
 997                 /* Don't handle automount points here */
 998                 break;
 999         }
1000         return 0;
1001 }
1002
1003 /*
1004  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1005  */
1006 static void follow_mount(struct path *path)
1007 {
1008         while (d_mountpoint(path->dentry)) {
1009                 struct vfsmount *mounted = lookup_mnt(path);
1010                 if (!mounted)
1011                         break;
1012                 dput(path->dentry);
1013                 mntput(path->mnt);
1014                 path->mnt = mounted;
1015                 path->dentry = dget(mounted->mnt_root);
1016         }
1017 }
1018
1019 static void follow_dotdot(struct nameidata *nd)
1020 {
1021         set_root(nd);
1022
1023         while(1) {
1024                 struct dentry *old = nd->path.dentry;
1025
1026                 if (nd->path.dentry == nd->root.dentry &&
1027                     nd->path.mnt == nd->root.mnt) {
1028                         break;
1029                 }
1030                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1031                         /* rare case of legitimate dget_parent()... */
1032                         nd->path.dentry = dget_parent(nd->path.dentry);
1033                         dput(old);
1034                         break;
1035                 }
1036                 if (!follow_up(&nd->path))
1037                         break;
1038         }
1039         follow_mount(&nd->path);
1040         nd->inode = nd->path.dentry->d_inode;
1041 }
1042
1043 /*
1044  * This looks up the name in dcache, possibly revalidates the old dentry and
1045  * allocates a new one if not found or not valid.  In the need_lookup argument
1046  * returns whether i_op->lookup is necessary.
1047  *
1048  * dir->d_inode->i_mutex must be held
1049  */
1050 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1051                                     struct nameidata *nd, bool *need_lookup)
1052 {
1053         struct dentry *dentry;
1054         int error;
1055
1056         *need_lookup = false;
1057         dentry = d_lookup(dir, name);
1058         if (dentry) {
1059                 if (d_need_lookup(dentry)) {
1060                         *need_lookup = true;
1061                 } else if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1062                         error = d_revalidate(dentry, nd);
1063                         if (unlikely(error <= 0)) {
1064                                 if (error < 0) {
1065                                         dput(dentry);
1066                                         return ERR_PTR(error);
1067                                 } else if (!d_invalidate(dentry)) {
1068                                         dput(dentry);
1069                                         dentry = NULL;
1070                                 }
1071                         }
1072                 }
1073         }
1074
1075         if (!dentry) {
1076                 dentry = d_alloc(dir, name);
1077                 if (unlikely(!dentry))
1078                         return ERR_PTR(-ENOMEM);
1079
1080                 *need_lookup = true;
1081         }
1082         return dentry;
1083 }
1084
1085 /*
1086  * Call i_op->lookup on the dentry.  The dentry must be negative but may be
1087  * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1088  *
1089  * dir->d_inode->i_mutex must be held
1090  */
1091 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1092                                   struct nameidata *nd)
1093 {
1094         struct dentry *old;
1095
1096         /* Don't create child dentry for a dead directory. */
1097         if (unlikely(IS_DEADDIR(dir))) {
1098                 dput(dentry);
1099                 return ERR_PTR(-ENOENT);
1100         }
1101
1102         old = dir->i_op->lookup(dir, dentry, nd);
1103         if (unlikely(old)) {
1104                 dput(dentry);
1105                 dentry = old;
1106         }
1107         return dentry;
1108 }
1109
1110 static struct dentry *__lookup_hash(struct qstr *name,
1111                 struct dentry *base, struct nameidata *nd)
1112 {
1113         bool need_lookup;
1114         struct dentry *dentry;
1115
1116         dentry = lookup_dcache(name, base, nd, &need_lookup);
1117         if (!need_lookup)
1118                 return dentry;
1119
1120         return lookup_real(base->d_inode, dentry, nd);
1121 }
1122
1123 /*
1124  *  It's more convoluted than I'd like it to be, but... it's still fairly
1125  *  small and for now I'd prefer to have fast path as straight as possible.
1126  *  It _is_ time-critical.
1127  */
1128 static int do_lookup(struct nameidata *nd, struct qstr *name,
1129                         struct path *path, struct inode **inode)
1130 {
1131         struct vfsmount *mnt = nd->path.mnt;
1132         struct dentry *dentry, *parent = nd->path.dentry;
1133         int need_reval = 1;
1134         int status = 1;
1135         int err;
1136
1137         /*
1138          * Rename seqlock is not required here because in the off chance
1139          * of a false negative due to a concurrent rename, we're going to
1140          * do the non-racy lookup, below.
1141          */
1142         if (nd->flags & LOOKUP_RCU) {
1143                 unsigned seq;
1144                 dentry = __d_lookup_rcu(parent, name, &seq, nd->inode);
1145                 if (!dentry)
1146                         goto unlazy;
1147
1148                 /*
1149                  * This sequence count validates that the inode matches
1150                  * the dentry name information from lookup.
1151                  */
1152                 *inode = dentry->d_inode;
1153                 if (read_seqcount_retry(&dentry->d_seq, seq))
1154                         return -ECHILD;
1155
1156                 /*
1157                  * This sequence count validates that the parent had no
1158                  * changes while we did the lookup of the dentry above.
1159                  *
1160                  * The memory barrier in read_seqcount_begin of child is
1161                  *  enough, we can use __read_seqcount_retry here.
1162                  */
1163                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1164                         return -ECHILD;
1165                 nd->seq = seq;
1166
1167                 if (unlikely(d_need_lookup(dentry)))
1168                         goto unlazy;
1169                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1170                         status = d_revalidate(dentry, nd);
1171                         if (unlikely(status <= 0)) {
1172                                 if (status != -ECHILD)
1173                                         need_reval = 0;
1174                                 goto unlazy;
1175                         }
1176                 }
1177                 path->mnt = mnt;
1178                 path->dentry = dentry;
1179                 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1180                         goto unlazy;
1181                 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1182                         goto unlazy;
1183                 return 0;
1184 unlazy:
1185                 if (unlazy_walk(nd, dentry))
1186                         return -ECHILD;
1187         } else {
1188                 dentry = __d_lookup(parent, name);
1189         }
1190
1191         if (unlikely(!dentry))
1192                 goto need_lookup;
1193
1194         if (unlikely(d_need_lookup(dentry))) {
1195                 dput(dentry);
1196                 goto need_lookup;
1197         }
1198
1199         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1200                 status = d_revalidate(dentry, nd);
1201         if (unlikely(status <= 0)) {
1202                 if (status < 0) {
1203                         dput(dentry);
1204                         return status;
1205                 }
1206                 if (!d_invalidate(dentry)) {
1207                         dput(dentry);
1208                         goto need_lookup;
1209                 }
1210         }
1211 done:
1212         path->mnt = mnt;
1213         path->dentry = dentry;
1214         err = follow_managed(path, nd->flags);
1215         if (unlikely(err < 0)) {
1216                 path_put_conditional(path, nd);
1217                 return err;
1218         }
1219         if (err)
1220                 nd->flags |= LOOKUP_JUMPED;
1221         *inode = path->dentry->d_inode;
1222         return 0;
1223
1224 need_lookup:
1225         BUG_ON(nd->inode != parent->d_inode);
1226
1227         mutex_lock(&parent->d_inode->i_mutex);
1228         dentry = __lookup_hash(name, parent, nd);
1229         mutex_unlock(&parent->d_inode->i_mutex);
1230         if (IS_ERR(dentry))
1231                 return PTR_ERR(dentry);
1232         goto done;
1233 }
1234
1235 static inline int may_lookup(struct nameidata *nd)
1236 {
1237         if (nd->flags & LOOKUP_RCU) {
1238                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1239                 if (err != -ECHILD)
1240                         return err;
1241                 if (unlazy_walk(nd, NULL))
1242                         return -ECHILD;
1243         }
1244         return inode_permission(nd->inode, MAY_EXEC);
1245 }
1246
1247 static inline int handle_dots(struct nameidata *nd, int type)
1248 {
1249         if (type == LAST_DOTDOT) {
1250                 if (nd->flags & LOOKUP_RCU) {
1251                         if (follow_dotdot_rcu(nd))
1252                                 return -ECHILD;
1253                 } else
1254                         follow_dotdot(nd);
1255         }
1256         return 0;
1257 }
1258
1259 static void terminate_walk(struct nameidata *nd)
1260 {
1261         if (!(nd->flags & LOOKUP_RCU)) {
1262                 path_put(&nd->path);
1263         } else {
1264                 nd->flags &= ~LOOKUP_RCU;
1265                 if (!(nd->flags & LOOKUP_ROOT))
1266                         nd->root.mnt = NULL;
1267                 rcu_read_unlock();
1268                 br_read_unlock(vfsmount_lock);
1269         }
1270 }
1271
1272 /*
1273  * Do we need to follow links? We _really_ want to be able
1274  * to do this check without having to look at inode->i_op,
1275  * so we keep a cache of "no, this doesn't need follow_link"
1276  * for the common case.
1277  */
1278 static inline int should_follow_link(struct inode *inode, int follow)
1279 {
1280         if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1281                 if (likely(inode->i_op->follow_link))
1282                         return follow;
1283
1284                 /* This gets set once for the inode lifetime */
1285                 spin_lock(&inode->i_lock);
1286                 inode->i_opflags |= IOP_NOFOLLOW;
1287                 spin_unlock(&inode->i_lock);
1288         }
1289         return 0;
1290 }
1291
1292 static inline int walk_component(struct nameidata *nd, struct path *path,
1293                 struct qstr *name, int type, int follow)
1294 {
1295         struct inode *inode;
1296         int err;
1297         /*
1298          * "." and ".." are special - ".." especially so because it has
1299          * to be able to know about the current root directory and
1300          * parent relationships.
1301          */
1302         if (unlikely(type != LAST_NORM))
1303                 return handle_dots(nd, type);
1304         err = do_lookup(nd, name, path, &inode);
1305         if (unlikely(err)) {
1306                 terminate_walk(nd);
1307                 return err;
1308         }
1309         if (!inode) {
1310                 path_to_nameidata(path, nd);
1311                 terminate_walk(nd);
1312                 return -ENOENT;
1313         }
1314         if (should_follow_link(inode, follow)) {
1315                 if (nd->flags & LOOKUP_RCU) {
1316                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1317                                 terminate_walk(nd);
1318                                 return -ECHILD;
1319                         }
1320                 }
1321                 BUG_ON(inode != path->dentry->d_inode);
1322                 return 1;
1323         }
1324         path_to_nameidata(path, nd);
1325         nd->inode = inode;
1326         return 0;
1327 }
1328
1329 /*
1330  * This limits recursive symlink follows to 8, while
1331  * limiting consecutive symlinks to 40.
1332  *
1333  * Without that kind of total limit, nasty chains of consecutive
1334  * symlinks can cause almost arbitrarily long lookups.
1335  */
1336 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1337 {
1338         int res;
1339
1340         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1341                 path_put_conditional(path, nd);
1342                 path_put(&nd->path);
1343                 return -ELOOP;
1344         }
1345         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1346
1347         nd->depth++;
1348         current->link_count++;
1349
1350         do {
1351                 struct path link = *path;
1352                 void *cookie;
1353
1354                 res = follow_link(&link, nd, &cookie);
1355                 if (!res)
1356                         res = walk_component(nd, path, &nd->last,
1357                                              nd->last_type, LOOKUP_FOLLOW);
1358                 put_link(nd, &link, cookie);
1359         } while (res > 0);
1360
1361         current->link_count--;
1362         nd->depth--;
1363         return res;
1364 }
1365
1366 /*
1367  * We really don't want to look at inode->i_op->lookup
1368  * when we don't have to. So we keep a cache bit in
1369  * the inode ->i_opflags field that says "yes, we can
1370  * do lookup on this inode".
1371  */
1372 static inline int can_lookup(struct inode *inode)
1373 {
1374         if (likely(inode->i_opflags & IOP_LOOKUP))
1375                 return 1;
1376         if (likely(!inode->i_op->lookup))
1377                 return 0;
1378
1379         /* We do this once for the lifetime of the inode */
1380         spin_lock(&inode->i_lock);
1381         inode->i_opflags |= IOP_LOOKUP;
1382         spin_unlock(&inode->i_lock);
1383         return 1;
1384 }
1385
1386 /*
1387  * We can do the critical dentry name comparison and hashing
1388  * operations one word at a time, but we are limited to:
1389  *
1390  * - Architectures with fast unaligned word accesses. We could
1391  *   do a "get_unaligned()" if this helps and is sufficiently
1392  *   fast.
1393  *
1394  * - Little-endian machines (so that we can generate the mask
1395  *   of low bytes efficiently). Again, we *could* do a byte
1396  *   swapping load on big-endian architectures if that is not
1397  *   expensive enough to make the optimization worthless.
1398  *
1399  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1400  *   do not trap on the (extremely unlikely) case of a page
1401  *   crossing operation.
1402  *
1403  * - Furthermore, we need an efficient 64-bit compile for the
1404  *   64-bit case in order to generate the "number of bytes in
1405  *   the final mask". Again, that could be replaced with a
1406  *   efficient population count instruction or similar.
1407  */
1408 #ifdef CONFIG_DCACHE_WORD_ACCESS
1409
1410 #include <asm/word-at-a-time.h>
1411
1412 #ifdef CONFIG_64BIT
1413
1414 static inline unsigned int fold_hash(unsigned long hash)
1415 {
1416         hash += hash >> (8*sizeof(int));
1417         return hash;
1418 }
1419
1420 #else   /* 32-bit case */
1421
1422 #define fold_hash(x) (x)
1423
1424 #endif
1425
1426 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1427 {
1428         unsigned long a, mask;
1429         unsigned long hash = 0;
1430
1431         for (;;) {
1432                 a = load_unaligned_zeropad(name);
1433                 if (len < sizeof(unsigned long))
1434                         break;
1435                 hash += a;
1436                 hash *= 9;
1437                 name += sizeof(unsigned long);
1438                 len -= sizeof(unsigned long);
1439                 if (!len)
1440                         goto done;
1441         }
1442         mask = ~(~0ul << len*8);
1443         hash += mask & a;
1444 done:
1445         return fold_hash(hash);
1446 }
1447 EXPORT_SYMBOL(full_name_hash);
1448
1449 /*
1450  * Calculate the length and hash of the path component, and
1451  * return the length of the component;
1452  */
1453 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1454 {
1455         unsigned long a, b, adata, bdata, mask, hash, len;
1456         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1457
1458         hash = a = 0;
1459         len = -sizeof(unsigned long);
1460         do {
1461                 hash = (hash + a) * 9;
1462                 len += sizeof(unsigned long);
1463                 a = load_unaligned_zeropad(name+len);
1464                 b = a ^ REPEAT_BYTE('/');
1465         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1466
1467         adata = prep_zero_mask(a, adata, &constants);
1468         bdata = prep_zero_mask(b, bdata, &constants);
1469
1470         mask = create_zero_mask(adata | bdata);
1471
1472         hash += a & zero_bytemask(mask);
1473         *hashp = fold_hash(hash);
1474
1475         return len + find_zero(mask);
1476 }
1477
1478 #else
1479
1480 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1481 {
1482         unsigned long hash = init_name_hash();
1483         while (len--)
1484                 hash = partial_name_hash(*name++, hash);
1485         return end_name_hash(hash);
1486 }
1487 EXPORT_SYMBOL(full_name_hash);
1488
1489 /*
1490  * We know there's a real path component here of at least
1491  * one character.
1492  */
1493 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1494 {
1495         unsigned long hash = init_name_hash();
1496         unsigned long len = 0, c;
1497
1498         c = (unsigned char)*name;
1499         do {
1500                 len++;
1501                 hash = partial_name_hash(c, hash);
1502                 c = (unsigned char)name[len];
1503         } while (c && c != '/');
1504         *hashp = end_name_hash(hash);
1505         return len;
1506 }
1507
1508 #endif
1509
1510 /*
1511  * Name resolution.
1512  * This is the basic name resolution function, turning a pathname into
1513  * the final dentry. We expect 'base' to be positive and a directory.
1514  *
1515  * Returns 0 and nd will have valid dentry and mnt on success.
1516  * Returns error and drops reference to input namei data on failure.
1517  */
1518 static int link_path_walk(const char *name, struct nameidata *nd)
1519 {
1520         struct path next;
1521         int err;
1522
1523         while (*name=='/')
1524                 name++;
1525         if (!*name)
1526                 return 0;
1527
1528         /* At this point we know we have a real path component. */
1529         for(;;) {
1530                 struct qstr this;
1531                 long len;
1532                 int type;
1533
1534                 err = may_lookup(nd);
1535                 if (err)
1536                         break;
1537
1538                 len = hash_name(name, &this.hash);
1539                 this.name = name;
1540                 this.len = len;
1541
1542                 type = LAST_NORM;
1543                 if (name[0] == '.') switch (len) {
1544                         case 2:
1545                                 if (name[1] == '.') {
1546                                         type = LAST_DOTDOT;
1547                                         nd->flags |= LOOKUP_JUMPED;
1548                                 }
1549                                 break;
1550                         case 1:
1551                                 type = LAST_DOT;
1552                 }
1553                 if (likely(type == LAST_NORM)) {
1554                         struct dentry *parent = nd->path.dentry;
1555                         nd->flags &= ~LOOKUP_JUMPED;
1556                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1557                                 err = parent->d_op->d_hash(parent, nd->inode,
1558                                                            &this);
1559                                 if (err < 0)
1560                                         break;
1561                         }
1562                 }
1563
1564                 if (!name[len])
1565                         goto last_component;
1566                 /*
1567                  * If it wasn't NUL, we know it was '/'. Skip that
1568                  * slash, and continue until no more slashes.
1569                  */
1570                 do {
1571                         len++;
1572                 } while (unlikely(name[len] == '/'));
1573                 if (!name[len])
1574                         goto last_component;
1575                 name += len;
1576
1577                 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1578                 if (err < 0)
1579                         return err;
1580
1581                 if (err) {
1582                         err = nested_symlink(&next, nd);
1583                         if (err)
1584                                 return err;
1585                 }
1586                 if (can_lookup(nd->inode))
1587                         continue;
1588                 err = -ENOTDIR;
1589                 break;
1590                 /* here ends the main loop */
1591
1592 last_component:
1593                 nd->last = this;
1594                 nd->last_type = type;
1595                 return 0;
1596         }
1597         terminate_walk(nd);
1598         return err;
1599 }
1600
1601 static int path_init(int dfd, const char *name, unsigned int flags,
1602                      struct nameidata *nd, struct file **fp)
1603 {
1604         int retval = 0;
1605         int fput_needed;
1606         struct file *file;
1607
1608         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1609         nd->flags = flags | LOOKUP_JUMPED;
1610         nd->depth = 0;
1611         if (flags & LOOKUP_ROOT) {
1612                 struct inode *inode = nd->root.dentry->d_inode;
1613                 if (*name) {
1614                         if (!inode->i_op->lookup)
1615                                 return -ENOTDIR;
1616                         retval = inode_permission(inode, MAY_EXEC);
1617                         if (retval)
1618                                 return retval;
1619                 }
1620                 nd->path = nd->root;
1621                 nd->inode = inode;
1622                 if (flags & LOOKUP_RCU) {
1623                         br_read_lock(vfsmount_lock);
1624                         rcu_read_lock();
1625                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1626                 } else {
1627                         path_get(&nd->path);
1628                 }
1629                 return 0;
1630         }
1631
1632         nd->root.mnt = NULL;
1633
1634         if (*name=='/') {
1635                 if (flags & LOOKUP_RCU) {
1636                         br_read_lock(vfsmount_lock);
1637                         rcu_read_lock();
1638                         set_root_rcu(nd);
1639                 } else {
1640                         set_root(nd);
1641                         path_get(&nd->root);
1642                 }
1643                 nd->path = nd->root;
1644         } else if (dfd == AT_FDCWD) {
1645                 if (flags & LOOKUP_RCU) {
1646                         struct fs_struct *fs = current->fs;
1647                         unsigned seq;
1648
1649                         br_read_lock(vfsmount_lock);
1650                         rcu_read_lock();
1651
1652                         do {
1653                                 seq = read_seqcount_begin(&fs->seq);
1654                                 nd->path = fs->pwd;
1655                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1656                         } while (read_seqcount_retry(&fs->seq, seq));
1657                 } else {
1658                         get_fs_pwd(current->fs, &nd->path);
1659                 }
1660         } else {
1661                 struct dentry *dentry;
1662
1663                 file = fget_raw_light(dfd, &fput_needed);
1664                 retval = -EBADF;
1665                 if (!file)
1666                         goto out_fail;
1667
1668                 dentry = file->f_path.dentry;
1669
1670                 if (*name) {
1671                         retval = -ENOTDIR;
1672                         if (!S_ISDIR(dentry->d_inode->i_mode))
1673                                 goto fput_fail;
1674
1675                         retval = inode_permission(dentry->d_inode, MAY_EXEC);
1676                         if (retval)
1677                                 goto fput_fail;
1678                 }
1679
1680                 nd->path = file->f_path;
1681                 if (flags & LOOKUP_RCU) {
1682                         if (fput_needed)
1683                                 *fp = file;
1684                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1685                         br_read_lock(vfsmount_lock);
1686                         rcu_read_lock();
1687                 } else {
1688                         path_get(&file->f_path);
1689                         fput_light(file, fput_needed);
1690                 }
1691         }
1692
1693         nd->inode = nd->path.dentry->d_inode;
1694         return 0;
1695
1696 fput_fail:
1697         fput_light(file, fput_needed);
1698 out_fail:
1699         return retval;
1700 }
1701
1702 static inline int lookup_last(struct nameidata *nd, struct path *path)
1703 {
1704         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1705                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1706
1707         nd->flags &= ~LOOKUP_PARENT;
1708         return walk_component(nd, path, &nd->last, nd->last_type,
1709                                         nd->flags & LOOKUP_FOLLOW);
1710 }
1711
1712 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1713 static int path_lookupat(int dfd, const char *name,
1714                                 unsigned int flags, struct nameidata *nd)
1715 {
1716         struct file *base = NULL;
1717         struct path path;
1718         int err;
1719
1720         /*
1721          * Path walking is largely split up into 2 different synchronisation
1722          * schemes, rcu-walk and ref-walk (explained in
1723          * Documentation/filesystems/path-lookup.txt). These share much of the
1724          * path walk code, but some things particularly setup, cleanup, and
1725          * following mounts are sufficiently divergent that functions are
1726          * duplicated. Typically there is a function foo(), and its RCU
1727          * analogue, foo_rcu().
1728          *
1729          * -ECHILD is the error number of choice (just to avoid clashes) that
1730          * is returned if some aspect of an rcu-walk fails. Such an error must
1731          * be handled by restarting a traditional ref-walk (which will always
1732          * be able to complete).
1733          */
1734         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1735
1736         if (unlikely(err))
1737                 return err;
1738
1739         current->total_link_count = 0;
1740         err = link_path_walk(name, nd);
1741
1742         if (!err && !(flags & LOOKUP_PARENT)) {
1743                 err = lookup_last(nd, &path);
1744                 while (err > 0) {
1745                         void *cookie;
1746                         struct path link = path;
1747                         nd->flags |= LOOKUP_PARENT;
1748                         err = follow_link(&link, nd, &cookie);
1749                         if (!err)
1750                                 err = lookup_last(nd, &path);
1751                         put_link(nd, &link, cookie);
1752                 }
1753         }
1754
1755         if (!err)
1756                 err = complete_walk(nd);
1757
1758         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1759                 if (!nd->inode->i_op->lookup) {
1760                         path_put(&nd->path);
1761                         err = -ENOTDIR;
1762                 }
1763         }
1764
1765         if (base)
1766                 fput(base);
1767
1768         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1769                 path_put(&nd->root);
1770                 nd->root.mnt = NULL;
1771         }
1772         return err;
1773 }
1774
1775 static int do_path_lookup(int dfd, const char *name,
1776                                 unsigned int flags, struct nameidata *nd)
1777 {
1778         int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1779         if (unlikely(retval == -ECHILD))
1780                 retval = path_lookupat(dfd, name, flags, nd);
1781         if (unlikely(retval == -ESTALE))
1782                 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1783
1784         if (likely(!retval)) {
1785                 if (unlikely(!audit_dummy_context())) {
1786                         if (nd->path.dentry && nd->inode)
1787                                 audit_inode(name, nd->path.dentry);
1788                 }
1789         }
1790         return retval;
1791 }
1792
1793 int kern_path_parent(const char *name, struct nameidata *nd)
1794 {
1795         return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1796 }
1797
1798 int kern_path(const char *name, unsigned int flags, struct path *path)
1799 {
1800         struct nameidata nd;
1801         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1802         if (!res)
1803                 *path = nd.path;
1804         return res;
1805 }
1806
1807 /**
1808  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1809  * @dentry:  pointer to dentry of the base directory
1810  * @mnt: pointer to vfs mount of the base directory
1811  * @name: pointer to file name
1812  * @flags: lookup flags
1813  * @path: pointer to struct path to fill
1814  */
1815 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1816                     const char *name, unsigned int flags,
1817                     struct path *path)
1818 {
1819         struct nameidata nd;
1820         int err;
1821         nd.root.dentry = dentry;
1822         nd.root.mnt = mnt;
1823         BUG_ON(flags & LOOKUP_PARENT);
1824         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1825         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1826         if (!err)
1827                 *path = nd.path;
1828         return err;
1829 }
1830
1831 /*
1832  * Restricted form of lookup. Doesn't follow links, single-component only,
1833  * needs parent already locked. Doesn't follow mounts.
1834  * SMP-safe.
1835  */
1836 static struct dentry *lookup_hash(struct nameidata *nd)
1837 {
1838         return __lookup_hash(&nd->last, nd->path.dentry, nd);
1839 }
1840
1841 /**
1842  * lookup_one_len - filesystem helper to lookup single pathname component
1843  * @name:       pathname component to lookup
1844  * @base:       base directory to lookup from
1845  * @len:        maximum length @len should be interpreted to
1846  *
1847  * Note that this routine is purely a helper for filesystem usage and should
1848  * not be called by generic code.  Also note that by using this function the
1849  * nameidata argument is passed to the filesystem methods and a filesystem
1850  * using this helper needs to be prepared for that.
1851  */
1852 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1853 {
1854         struct qstr this;
1855         unsigned int c;
1856         int err;
1857
1858         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1859
1860         this.name = name;
1861         this.len = len;
1862         this.hash = full_name_hash(name, len);
1863         if (!len)
1864                 return ERR_PTR(-EACCES);
1865
1866         while (len--) {
1867                 c = *(const unsigned char *)name++;
1868                 if (c == '/' || c == '\0')
1869                         return ERR_PTR(-EACCES);
1870         }
1871         /*
1872          * See if the low-level filesystem might want
1873          * to use its own hash..
1874          */
1875         if (base->d_flags & DCACHE_OP_HASH) {
1876                 int err = base->d_op->d_hash(base, base->d_inode, &this);
1877                 if (err < 0)
1878                         return ERR_PTR(err);
1879         }
1880
1881         err = inode_permission(base->d_inode, MAY_EXEC);
1882         if (err)
1883                 return ERR_PTR(err);
1884
1885         return __lookup_hash(&this, base, NULL);
1886 }
1887
1888 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1889                  struct path *path, int *empty)
1890 {
1891         struct nameidata nd;
1892         char *tmp = getname_flags(name, flags, empty);
1893         int err = PTR_ERR(tmp);
1894         if (!IS_ERR(tmp)) {
1895
1896                 BUG_ON(flags & LOOKUP_PARENT);
1897
1898                 err = do_path_lookup(dfd, tmp, flags, &nd);
1899                 putname(tmp);
1900                 if (!err)
1901                         *path = nd.path;
1902         }
1903         return err;
1904 }
1905
1906 int user_path_at(int dfd, const char __user *name, unsigned flags,
1907                  struct path *path)
1908 {
1909         return user_path_at_empty(dfd, name, flags, path, NULL);
1910 }
1911
1912 static int user_path_parent(int dfd, const char __user *path,
1913                         struct nameidata *nd, char **name)
1914 {
1915         char *s = getname(path);
1916         int error;
1917
1918         if (IS_ERR(s))
1919                 return PTR_ERR(s);
1920
1921         error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1922         if (error)
1923                 putname(s);
1924         else
1925                 *name = s;
1926
1927         return error;
1928 }
1929
1930 /*
1931  * It's inline, so penalty for filesystems that don't use sticky bit is
1932  * minimal.
1933  */
1934 static inline int check_sticky(struct inode *dir, struct inode *inode)
1935 {
1936         kuid_t fsuid = current_fsuid();
1937
1938         if (!(dir->i_mode & S_ISVTX))
1939                 return 0;
1940         if (uid_eq(inode->i_uid, fsuid))
1941                 return 0;
1942         if (uid_eq(dir->i_uid, fsuid))
1943                 return 0;
1944         return !inode_capable(inode, CAP_FOWNER);
1945 }
1946
1947 /*
1948  *      Check whether we can remove a link victim from directory dir, check
1949  *  whether the type of victim is right.
1950  *  1. We can't do it if dir is read-only (done in permission())
1951  *  2. We should have write and exec permissions on dir
1952  *  3. We can't remove anything from append-only dir
1953  *  4. We can't do anything with immutable dir (done in permission())
1954  *  5. If the sticky bit on dir is set we should either
1955  *      a. be owner of dir, or
1956  *      b. be owner of victim, or
1957  *      c. have CAP_FOWNER capability
1958  *  6. If the victim is append-only or immutable we can't do antyhing with
1959  *     links pointing to it.
1960  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1961  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1962  *  9. We can't remove a root or mountpoint.
1963  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1964  *     nfs_async_unlink().
1965  */
1966 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1967 {
1968         int error;
1969
1970         if (!victim->d_inode)
1971                 return -ENOENT;
1972
1973         BUG_ON(victim->d_parent->d_inode != dir);
1974         audit_inode_child(victim, dir);
1975
1976         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1977         if (error)
1978                 return error;
1979         if (IS_APPEND(dir))
1980                 return -EPERM;
1981         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1982             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1983                 return -EPERM;
1984         if (isdir) {
1985                 if (!S_ISDIR(victim->d_inode->i_mode))
1986                         return -ENOTDIR;
1987                 if (IS_ROOT(victim))
1988                         return -EBUSY;
1989         } else if (S_ISDIR(victim->d_inode->i_mode))
1990                 return -EISDIR;
1991         if (IS_DEADDIR(dir))
1992                 return -ENOENT;
1993         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1994                 return -EBUSY;
1995         return 0;
1996 }
1997
1998 /*      Check whether we can create an object with dentry child in directory
1999  *  dir.
2000  *  1. We can't do it if child already exists (open has special treatment for
2001  *     this case, but since we are inlined it's OK)
2002  *  2. We can't do it if dir is read-only (done in permission())
2003  *  3. We should have write and exec permissions on dir
2004  *  4. We can't do it if dir is immutable (done in permission())
2005  */
2006 static inline int may_create(struct inode *dir, struct dentry *child)
2007 {
2008         if (child->d_inode)
2009                 return -EEXIST;
2010         if (IS_DEADDIR(dir))
2011                 return -ENOENT;
2012         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2013 }
2014
2015 /*
2016  * p1 and p2 should be directories on the same fs.
2017  */
2018 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2019 {
2020         struct dentry *p;
2021
2022         if (p1 == p2) {
2023                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2024                 return NULL;
2025         }
2026
2027         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2028
2029         p = d_ancestor(p2, p1);
2030         if (p) {
2031                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2032                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2033                 return p;
2034         }
2035
2036         p = d_ancestor(p1, p2);
2037         if (p) {
2038                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2039                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2040                 return p;
2041         }
2042
2043         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2044         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2045         return NULL;
2046 }
2047
2048 void unlock_rename(struct dentry *p1, struct dentry *p2)
2049 {
2050         mutex_unlock(&p1->d_inode->i_mutex);
2051         if (p1 != p2) {
2052                 mutex_unlock(&p2->d_inode->i_mutex);
2053                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2054         }
2055 }
2056
2057 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2058                 struct nameidata *nd)
2059 {
2060         int error = may_create(dir, dentry);
2061
2062         if (error)
2063                 return error;
2064
2065         if (!dir->i_op->create)
2066                 return -EACCES; /* shouldn't it be ENOSYS? */
2067         mode &= S_IALLUGO;
2068         mode |= S_IFREG;
2069         error = security_inode_create(dir, dentry, mode);
2070         if (error)
2071                 return error;
2072         error = dir->i_op->create(dir, dentry, mode, nd);
2073         if (!error)
2074                 fsnotify_create(dir, dentry);
2075         return error;
2076 }
2077
2078 static int may_open(struct path *path, int acc_mode, int flag)
2079 {
2080         struct dentry *dentry = path->dentry;
2081         struct inode *inode = dentry->d_inode;
2082         int error;
2083
2084         /* O_PATH? */
2085         if (!acc_mode)
2086                 return 0;
2087
2088         if (!inode)
2089                 return -ENOENT;
2090
2091         switch (inode->i_mode & S_IFMT) {
2092         case S_IFLNK:
2093                 return -ELOOP;
2094         case S_IFDIR:
2095                 if (acc_mode & MAY_WRITE)
2096                         return -EISDIR;
2097                 break;
2098         case S_IFBLK:
2099         case S_IFCHR:
2100                 if (path->mnt->mnt_flags & MNT_NODEV)
2101                         return -EACCES;
2102                 /*FALLTHRU*/
2103         case S_IFIFO:
2104         case S_IFSOCK:
2105                 flag &= ~O_TRUNC;
2106                 break;
2107         }
2108
2109         error = inode_permission(inode, acc_mode);
2110         if (error)
2111                 return error;
2112
2113         /*
2114          * An append-only file must be opened in append mode for writing.
2115          */
2116         if (IS_APPEND(inode)) {
2117                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2118                         return -EPERM;
2119                 if (flag & O_TRUNC)
2120                         return -EPERM;
2121         }
2122
2123         /* O_NOATIME can only be set by the owner or superuser */
2124         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2125                 return -EPERM;
2126
2127         return 0;
2128 }
2129
2130 static int handle_truncate(struct file *filp)
2131 {
2132         struct path *path = &filp->f_path;
2133         struct inode *inode = path->dentry->d_inode;
2134         int error = get_write_access(inode);
2135         if (error)
2136                 return error;
2137         /*
2138          * Refuse to truncate files with mandatory locks held on them.
2139          */
2140         error = locks_verify_locked(inode);
2141         if (!error)
2142                 error = security_path_truncate(path);
2143         if (!error) {
2144                 error = do_truncate(path->dentry, 0,
2145                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2146                                     filp);
2147         }
2148         put_write_access(inode);
2149         return error;
2150 }
2151
2152 static inline int open_to_namei_flags(int flag)
2153 {
2154         if ((flag & O_ACCMODE) == 3)
2155                 flag--;
2156         return flag;
2157 }
2158
2159 /*
2160  * Handle the last step of open()
2161  */
2162 static struct file *do_last(struct nameidata *nd, struct path *path,
2163                             const struct open_flags *op, const char *pathname)
2164 {
2165         struct dentry *dir = nd->path.dentry;
2166         struct dentry *dentry;
2167         int open_flag = op->open_flag;
2168         int will_truncate = open_flag & O_TRUNC;
2169         int want_write = 0;
2170         int acc_mode = op->acc_mode;
2171         struct file *filp;
2172         int error;
2173
2174         nd->flags &= ~LOOKUP_PARENT;
2175         nd->flags |= op->intent;
2176
2177         switch (nd->last_type) {
2178         case LAST_DOTDOT:
2179         case LAST_DOT:
2180                 error = handle_dots(nd, nd->last_type);
2181                 if (error)
2182                         return ERR_PTR(error);
2183                 /* fallthrough */
2184         case LAST_ROOT:
2185                 error = complete_walk(nd);
2186                 if (error)
2187                         return ERR_PTR(error);
2188                 audit_inode(pathname, nd->path.dentry);
2189                 if (open_flag & O_CREAT) {
2190                         error = -EISDIR;
2191                         goto exit;
2192                 }
2193                 goto ok;
2194         case LAST_BIND:
2195                 error = complete_walk(nd);
2196                 if (error)
2197                         return ERR_PTR(error);
2198                 audit_inode(pathname, dir);
2199                 goto ok;
2200         }
2201
2202         if (!(open_flag & O_CREAT)) {
2203                 int symlink_ok = 0;
2204                 if (nd->last.name[nd->last.len])
2205                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2206                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2207                         symlink_ok = 1;
2208                 /* we _can_ be in RCU mode here */
2209                 error = walk_component(nd, path, &nd->last, LAST_NORM,
2210                                         !symlink_ok);
2211                 if (error < 0)
2212                         return ERR_PTR(error);
2213                 if (error) /* symlink */
2214                         return NULL;
2215                 /* sayonara */
2216                 error = complete_walk(nd);
2217                 if (error)
2218                         return ERR_PTR(error);
2219
2220                 error = -ENOTDIR;
2221                 if (nd->flags & LOOKUP_DIRECTORY) {
2222                         if (!nd->inode->i_op->lookup)
2223                                 goto exit;
2224                 }
2225                 audit_inode(pathname, nd->path.dentry);
2226                 goto ok;
2227         }
2228
2229         /* create side of things */
2230         /*
2231          * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2232          * cleared when we got to the last component we are about to look up
2233          */
2234         error = complete_walk(nd);
2235         if (error)
2236                 return ERR_PTR(error);
2237
2238         audit_inode(pathname, dir);
2239         error = -EISDIR;
2240         /* trailing slashes? */
2241         if (nd->last.name[nd->last.len])
2242                 goto exit;
2243
2244         mutex_lock(&dir->d_inode->i_mutex);
2245
2246         dentry = lookup_hash(nd);
2247         error = PTR_ERR(dentry);
2248         if (IS_ERR(dentry)) {
2249                 mutex_unlock(&dir->d_inode->i_mutex);
2250                 goto exit;
2251         }
2252
2253         path->dentry = dentry;
2254         path->mnt = nd->path.mnt;
2255
2256         /* Negative dentry, just create the file */
2257         if (!dentry->d_inode) {
2258                 umode_t mode = op->mode;
2259                 if (!IS_POSIXACL(dir->d_inode))
2260                         mode &= ~current_umask();
2261                 /*
2262                  * This write is needed to ensure that a
2263                  * rw->ro transition does not occur between
2264                  * the time when the file is created and when
2265                  * a permanent write count is taken through
2266                  * the 'struct file' in nameidata_to_filp().
2267                  */
2268                 error = mnt_want_write(nd->path.mnt);
2269                 if (error)
2270                         goto exit_mutex_unlock;
2271                 want_write = 1;
2272                 /* Don't check for write permission, don't truncate */
2273                 open_flag &= ~O_TRUNC;
2274                 will_truncate = 0;
2275                 acc_mode = MAY_OPEN;
2276                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2277                 if (error)
2278                         goto exit_mutex_unlock;
2279                 error = vfs_create(dir->d_inode, dentry, mode, nd);
2280                 if (error)
2281                         goto exit_mutex_unlock;
2282                 mutex_unlock(&dir->d_inode->i_mutex);
2283                 dput(nd->path.dentry);
2284                 nd->path.dentry = dentry;
2285                 goto common;
2286         }
2287
2288         /*
2289          * It already exists.
2290          */
2291         mutex_unlock(&dir->d_inode->i_mutex);
2292         audit_inode(pathname, path->dentry);
2293
2294         error = -EEXIST;
2295         if (open_flag & O_EXCL)
2296                 goto exit_dput;
2297
2298         error = follow_managed(path, nd->flags);
2299         if (error < 0)
2300                 goto exit_dput;
2301
2302         if (error)
2303                 nd->flags |= LOOKUP_JUMPED;
2304
2305         error = -ENOENT;
2306         if (!path->dentry->d_inode)
2307                 goto exit_dput;
2308
2309         if (path->dentry->d_inode->i_op->follow_link)
2310                 return NULL;
2311
2312         path_to_nameidata(path, nd);
2313         nd->inode = path->dentry->d_inode;
2314         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
2315         error = complete_walk(nd);
2316         if (error)
2317                 return ERR_PTR(error);
2318         error = -EISDIR;
2319         if (S_ISDIR(nd->inode->i_mode))
2320                 goto exit;
2321 ok:
2322         if (!S_ISREG(nd->inode->i_mode))
2323                 will_truncate = 0;
2324
2325         if (will_truncate) {
2326                 error = mnt_want_write(nd->path.mnt);
2327                 if (error)
2328                         goto exit;
2329                 want_write = 1;
2330         }
2331 common:
2332         error = may_open(&nd->path, acc_mode, open_flag);
2333         if (error)
2334                 goto exit;
2335         filp = nameidata_to_filp(nd);
2336         if (!IS_ERR(filp)) {
2337                 error = ima_file_check(filp, op->acc_mode);
2338                 if (error) {
2339                         fput(filp);
2340                         filp = ERR_PTR(error);
2341                 }
2342         }
2343         if (!IS_ERR(filp)) {
2344                 if (will_truncate) {
2345                         error = handle_truncate(filp);
2346                         if (error) {
2347                                 fput(filp);
2348                                 filp = ERR_PTR(error);
2349                         }
2350                 }
2351         }
2352 out:
2353         if (want_write)
2354                 mnt_drop_write(nd->path.mnt);
2355         path_put(&nd->path);
2356         return filp;
2357
2358 exit_mutex_unlock:
2359         mutex_unlock(&dir->d_inode->i_mutex);
2360 exit_dput:
2361         path_put_conditional(path, nd);
2362 exit:
2363         filp = ERR_PTR(error);
2364         goto out;
2365 }
2366
2367 static struct file *path_openat(int dfd, const char *pathname,
2368                 struct nameidata *nd, const struct open_flags *op, int flags)
2369 {
2370         struct file *base = NULL;
2371         struct file *filp;
2372         struct path path;
2373         int error;
2374
2375         filp = get_empty_filp();
2376         if (!filp)
2377                 return ERR_PTR(-ENFILE);
2378
2379         filp->f_flags = op->open_flag;
2380         nd->intent.open.file = filp;
2381         nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2382         nd->intent.open.create_mode = op->mode;
2383
2384         error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2385         if (unlikely(error))
2386                 goto out_filp;
2387
2388         current->total_link_count = 0;
2389         error = link_path_walk(pathname, nd);
2390         if (unlikely(error))
2391                 goto out_filp;
2392
2393         filp = do_last(nd, &path, op, pathname);
2394         while (unlikely(!filp)) { /* trailing symlink */
2395                 struct path link = path;
2396                 void *cookie;
2397                 if (!(nd->flags & LOOKUP_FOLLOW)) {
2398                         path_put_conditional(&path, nd);
2399                         path_put(&nd->path);
2400                         filp = ERR_PTR(-ELOOP);
2401                         break;
2402                 }
2403                 nd->flags |= LOOKUP_PARENT;
2404                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2405                 error = follow_link(&link, nd, &cookie);
2406                 if (unlikely(error))
2407                         filp = ERR_PTR(error);
2408                 else
2409                         filp = do_last(nd, &path, op, pathname);
2410                 put_link(nd, &link, cookie);
2411         }
2412 out:
2413         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2414                 path_put(&nd->root);
2415         if (base)
2416                 fput(base);
2417         release_open_intent(nd);
2418         return filp;
2419
2420 out_filp:
2421         filp = ERR_PTR(error);
2422         goto out;
2423 }
2424
2425 struct file *do_filp_open(int dfd, const char *pathname,
2426                 const struct open_flags *op, int flags)
2427 {
2428         struct nameidata nd;
2429         struct file *filp;
2430
2431         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2432         if (unlikely(filp == ERR_PTR(-ECHILD)))
2433                 filp = path_openat(dfd, pathname, &nd, op, flags);
2434         if (unlikely(filp == ERR_PTR(-ESTALE)))
2435                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2436         return filp;
2437 }
2438
2439 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2440                 const char *name, const struct open_flags *op, int flags)
2441 {
2442         struct nameidata nd;
2443         struct file *file;
2444
2445         nd.root.mnt = mnt;
2446         nd.root.dentry = dentry;
2447
2448         flags |= LOOKUP_ROOT;
2449
2450         if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2451                 return ERR_PTR(-ELOOP);
2452
2453         file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2454         if (unlikely(file == ERR_PTR(-ECHILD)))
2455                 file = path_openat(-1, name, &nd, op, flags);
2456         if (unlikely(file == ERR_PTR(-ESTALE)))
2457                 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2458         return file;
2459 }
2460
2461 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2462 {
2463         struct dentry *dentry = ERR_PTR(-EEXIST);
2464         struct nameidata nd;
2465         int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2466         if (error)
2467                 return ERR_PTR(error);
2468
2469         /*
2470          * Yucky last component or no last component at all?
2471          * (foo/., foo/.., /////)
2472          */
2473         if (nd.last_type != LAST_NORM)
2474                 goto out;
2475         nd.flags &= ~LOOKUP_PARENT;
2476         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2477         nd.intent.open.flags = O_EXCL;
2478
2479         /*
2480          * Do the final lookup.
2481          */
2482         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2483         dentry = lookup_hash(&nd);
2484         if (IS_ERR(dentry))
2485                 goto fail;
2486
2487         if (dentry->d_inode)
2488                 goto eexist;
2489         /*
2490          * Special case - lookup gave negative, but... we had foo/bar/
2491          * From the vfs_mknod() POV we just have a negative dentry -
2492          * all is fine. Let's be bastards - you had / on the end, you've
2493          * been asking for (non-existent) directory. -ENOENT for you.
2494          */
2495         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2496                 dput(dentry);
2497                 dentry = ERR_PTR(-ENOENT);
2498                 goto fail;
2499         }
2500         *path = nd.path;
2501         return dentry;
2502 eexist:
2503         dput(dentry);
2504         dentry = ERR_PTR(-EEXIST);
2505 fail:
2506         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2507 out:
2508         path_put(&nd.path);
2509         return dentry;
2510 }
2511 EXPORT_SYMBOL(kern_path_create);
2512
2513 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2514 {
2515         char *tmp = getname(pathname);
2516         struct dentry *res;
2517         if (IS_ERR(tmp))
2518                 return ERR_CAST(tmp);
2519         res = kern_path_create(dfd, tmp, path, is_dir);
2520         putname(tmp);
2521         return res;
2522 }
2523 EXPORT_SYMBOL(user_path_create);
2524
2525 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2526 {
2527         int error = may_create(dir, dentry);
2528
2529         if (error)
2530                 return error;
2531
2532         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2533                 return -EPERM;
2534
2535         if (!dir->i_op->mknod)
2536                 return -EPERM;
2537
2538         error = devcgroup_inode_mknod(mode, dev);
2539         if (error)
2540                 return error;
2541
2542         error = security_inode_mknod(dir, dentry, mode, dev);
2543         if (error)
2544                 return error;
2545
2546         error = dir->i_op->mknod(dir, dentry, mode, dev);
2547         if (!error)
2548                 fsnotify_create(dir, dentry);
2549         return error;
2550 }
2551
2552 static int may_mknod(umode_t mode)
2553 {
2554         switch (mode & S_IFMT) {
2555         case S_IFREG:
2556         case S_IFCHR:
2557         case S_IFBLK:
2558         case S_IFIFO:
2559         case S_IFSOCK:
2560         case 0: /* zero mode translates to S_IFREG */
2561                 return 0;
2562         case S_IFDIR:
2563                 return -EPERM;
2564         default:
2565                 return -EINVAL;
2566         }
2567 }
2568
2569 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2570                 unsigned, dev)
2571 {
2572         struct dentry *dentry;
2573         struct path path;
2574         int error;
2575
2576         if (S_ISDIR(mode))
2577                 return -EPERM;
2578
2579         dentry = user_path_create(dfd, filename, &path, 0);
2580         if (IS_ERR(dentry))
2581                 return PTR_ERR(dentry);
2582
2583         if (!IS_POSIXACL(path.dentry->d_inode))
2584                 mode &= ~current_umask();
2585         error = may_mknod(mode);
2586         if (error)
2587                 goto out_dput;
2588         error = mnt_want_write(path.mnt);
2589         if (error)
2590                 goto out_dput;
2591         error = security_path_mknod(&path, dentry, mode, dev);
2592         if (error)
2593                 goto out_drop_write;
2594         switch (mode & S_IFMT) {
2595                 case 0: case S_IFREG:
2596                         error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2597                         break;
2598                 case S_IFCHR: case S_IFBLK:
2599                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2600                                         new_decode_dev(dev));
2601                         break;
2602                 case S_IFIFO: case S_IFSOCK:
2603                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2604                         break;
2605         }
2606 out_drop_write:
2607         mnt_drop_write(path.mnt);
2608 out_dput:
2609         dput(dentry);
2610         mutex_unlock(&path.dentry->d_inode->i_mutex);
2611         path_put(&path);
2612
2613         return error;
2614 }
2615
2616 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2617 {
2618         return sys_mknodat(AT_FDCWD, filename, mode, dev);
2619 }
2620
2621 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2622 {
2623         int error = may_create(dir, dentry);
2624         unsigned max_links = dir->i_sb->s_max_links;
2625
2626         if (error)
2627                 return error;
2628
2629         if (!dir->i_op->mkdir)
2630                 return -EPERM;
2631
2632         mode &= (S_IRWXUGO|S_ISVTX);
2633         error = security_inode_mkdir(dir, dentry, mode);
2634         if (error)
2635                 return error;
2636
2637         if (max_links && dir->i_nlink >= max_links)
2638                 return -EMLINK;
2639
2640         error = dir->i_op->mkdir(dir, dentry, mode);
2641         if (!error)
2642                 fsnotify_mkdir(dir, dentry);
2643         return error;
2644 }
2645
2646 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2647 {
2648         struct dentry *dentry;
2649         struct path path;
2650         int error;
2651
2652         dentry = user_path_create(dfd, pathname, &path, 1);
2653         if (IS_ERR(dentry))
2654                 return PTR_ERR(dentry);
2655
2656         if (!IS_POSIXACL(path.dentry->d_inode))
2657                 mode &= ~current_umask();
2658         error = mnt_want_write(path.mnt);
2659         if (error)
2660                 goto out_dput;
2661         error = security_path_mkdir(&path, dentry, mode);
2662         if (error)
2663                 goto out_drop_write;
2664         error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2665 out_drop_write:
2666         mnt_drop_write(path.mnt);
2667 out_dput:
2668         dput(dentry);
2669         mutex_unlock(&path.dentry->d_inode->i_mutex);
2670         path_put(&path);
2671         return error;
2672 }
2673
2674 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2675 {
2676         return sys_mkdirat(AT_FDCWD, pathname, mode);
2677 }
2678
2679 /*
2680  * The dentry_unhash() helper will try to drop the dentry early: we
2681  * should have a usage count of 1 if we're the only user of this
2682  * dentry, and if that is true (possibly after pruning the dcache),
2683  * then we drop the dentry now.
2684  *
2685  * A low-level filesystem can, if it choses, legally
2686  * do a
2687  *
2688  *      if (!d_unhashed(dentry))
2689  *              return -EBUSY;
2690  *
2691  * if it cannot handle the case of removing a directory
2692  * that is still in use by something else..
2693  */
2694 void dentry_unhash(struct dentry *dentry)
2695 {
2696         shrink_dcache_parent(dentry);
2697         spin_lock(&dentry->d_lock);
2698         if (dentry->d_count == 1)
2699                 __d_drop(dentry);
2700         spin_unlock(&dentry->d_lock);
2701 }
2702
2703 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2704 {
2705         int error = may_delete(dir, dentry, 1);
2706
2707         if (error)
2708                 return error;
2709
2710         if (!dir->i_op->rmdir)
2711                 return -EPERM;
2712
2713         dget(dentry);
2714         mutex_lock(&dentry->d_inode->i_mutex);
2715
2716         error = -EBUSY;
2717         if (d_mountpoint(dentry))
2718                 goto out;
2719
2720         error = security_inode_rmdir(dir, dentry);
2721         if (error)
2722                 goto out;
2723
2724         shrink_dcache_parent(dentry);
2725         error = dir->i_op->rmdir(dir, dentry);
2726         if (error)
2727                 goto out;
2728
2729         dentry->d_inode->i_flags |= S_DEAD;
2730         dont_mount(dentry);
2731
2732 out:
2733         mutex_unlock(&dentry->d_inode->i_mutex);
2734         dput(dentry);
2735         if (!error)
2736                 d_delete(dentry);
2737         return error;
2738 }
2739
2740 static long do_rmdir(int dfd, const char __user *pathname)
2741 {
2742         int error = 0;
2743         char * name;
2744         struct dentry *dentry;
2745         struct nameidata nd;
2746
2747         error = user_path_parent(dfd, pathname, &nd, &name);
2748         if (error)
2749                 return error;
2750
2751         switch(nd.last_type) {
2752         case LAST_DOTDOT:
2753                 error = -ENOTEMPTY;
2754                 goto exit1;
2755         case LAST_DOT:
2756                 error = -EINVAL;
2757                 goto exit1;
2758         case LAST_ROOT:
2759                 error = -EBUSY;
2760                 goto exit1;
2761         }
2762
2763         nd.flags &= ~LOOKUP_PARENT;
2764
2765         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2766         dentry = lookup_hash(&nd);
2767         error = PTR_ERR(dentry);
2768         if (IS_ERR(dentry))
2769                 goto exit2;
2770         if (!dentry->d_inode) {
2771                 error = -ENOENT;
2772                 goto exit3;
2773         }
2774         error = mnt_want_write(nd.path.mnt);
2775         if (error)
2776                 goto exit3;
2777         error = security_path_rmdir(&nd.path, dentry);
2778         if (error)
2779                 goto exit4;
2780         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2781 exit4:
2782         mnt_drop_write(nd.path.mnt);
2783 exit3:
2784         dput(dentry);
2785 exit2:
2786         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2787 exit1:
2788         path_put(&nd.path);
2789         putname(name);
2790         return error;
2791 }
2792
2793 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2794 {
2795         return do_rmdir(AT_FDCWD, pathname);
2796 }
2797
2798 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2799 {
2800         int error = may_delete(dir, dentry, 0);
2801
2802         if (error)
2803                 return error;
2804
2805         if (!dir->i_op->unlink)
2806                 return -EPERM;
2807
2808         mutex_lock(&dentry->d_inode->i_mutex);
2809         if (d_mountpoint(dentry))
2810                 error = -EBUSY;
2811         else {
2812                 error = security_inode_unlink(dir, dentry);
2813                 if (!error) {
2814                         error = dir->i_op->unlink(dir, dentry);
2815                         if (!error)
2816                                 dont_mount(dentry);
2817                 }
2818         }
2819         mutex_unlock(&dentry->d_inode->i_mutex);
2820
2821         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2822         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2823                 fsnotify_link_count(dentry->d_inode);
2824                 d_delete(dentry);
2825         }
2826
2827         return error;
2828 }
2829
2830 /*
2831  * Make sure that the actual truncation of the file will occur outside its
2832  * directory's i_mutex.  Truncate can take a long time if there is a lot of
2833  * writeout happening, and we don't want to prevent access to the directory
2834  * while waiting on the I/O.
2835  */
2836 static long do_unlinkat(int dfd, const char __user *pathname)
2837 {
2838         int error;
2839         char *name;
2840         struct dentry *dentry;
2841         struct nameidata nd;
2842         struct inode *inode = NULL;
2843
2844         error = user_path_parent(dfd, pathname, &nd, &name);
2845         if (error)
2846                 return error;
2847
2848         error = -EISDIR;
2849         if (nd.last_type != LAST_NORM)
2850                 goto exit1;
2851
2852         nd.flags &= ~LOOKUP_PARENT;
2853
2854         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2855         dentry = lookup_hash(&nd);
2856         error = PTR_ERR(dentry);
2857         if (!IS_ERR(dentry)) {
2858                 /* Why not before? Because we want correct error value */
2859                 if (nd.last.name[nd.last.len])
2860                         goto slashes;
2861                 inode = dentry->d_inode;
2862                 if (!inode)
2863                         goto slashes;
2864                 ihold(inode);
2865                 error = mnt_want_write(nd.path.mnt);
2866                 if (error)
2867                         goto exit2;
2868                 error = security_path_unlink(&nd.path, dentry);
2869                 if (error)
2870                         goto exit3;
2871                 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2872 exit3:
2873                 mnt_drop_write(nd.path.mnt);
2874         exit2:
2875                 dput(dentry);
2876         }
2877         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2878         if (inode)
2879                 iput(inode);    /* truncate the inode here */
2880 exit1:
2881         path_put(&nd.path);
2882         putname(name);
2883         return error;
2884
2885 slashes:
2886         error = !dentry->d_inode ? -ENOENT :
2887                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2888         goto exit2;
2889 }
2890
2891 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2892 {
2893         if ((flag & ~AT_REMOVEDIR) != 0)
2894                 return -EINVAL;
2895
2896         if (flag & AT_REMOVEDIR)
2897                 return do_rmdir(dfd, pathname);
2898
2899         return do_unlinkat(dfd, pathname);
2900 }
2901
2902 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2903 {
2904         return do_unlinkat(AT_FDCWD, pathname);
2905 }
2906
2907 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2908 {
2909         int error = may_create(dir, dentry);
2910
2911         if (error)
2912                 return error;
2913
2914         if (!dir->i_op->symlink)
2915                 return -EPERM;
2916
2917         error = security_inode_symlink(dir, dentry, oldname);
2918         if (error)
2919                 return error;
2920
2921         error = dir->i_op->symlink(dir, dentry, oldname);
2922         if (!error)
2923                 fsnotify_create(dir, dentry);
2924         return error;
2925 }
2926
2927 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2928                 int, newdfd, const char __user *, newname)
2929 {
2930         int error;
2931         char *from;
2932         struct dentry *dentry;
2933         struct path path;
2934
2935         from = getname(oldname);
2936         if (IS_ERR(from))
2937                 return PTR_ERR(from);
2938
2939         dentry = user_path_create(newdfd, newname, &path, 0);
2940         error = PTR_ERR(dentry);
2941         if (IS_ERR(dentry))
2942                 goto out_putname;
2943
2944         error = mnt_want_write(path.mnt);
2945         if (error)
2946                 goto out_dput;
2947         error = security_path_symlink(&path, dentry, from);
2948         if (error)
2949                 goto out_drop_write;
2950         error = vfs_symlink(path.dentry->d_inode, dentry, from);
2951 out_drop_write:
2952         mnt_drop_write(path.mnt);
2953 out_dput:
2954         dput(dentry);
2955         mutex_unlock(&path.dentry->d_inode->i_mutex);
2956         path_put(&path);
2957 out_putname:
2958         putname(from);
2959         return error;
2960 }
2961
2962 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2963 {
2964         return sys_symlinkat(oldname, AT_FDCWD, newname);
2965 }
2966
2967 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2968 {
2969         struct inode *inode = old_dentry->d_inode;
2970         unsigned max_links = dir->i_sb->s_max_links;
2971         int error;
2972
2973         if (!inode)
2974                 return -ENOENT;
2975
2976         error = may_create(dir, new_dentry);
2977         if (error)
2978                 return error;
2979
2980         if (dir->i_sb != inode->i_sb)
2981                 return -EXDEV;
2982
2983         /*
2984          * A link to an append-only or immutable file cannot be created.
2985          */
2986         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2987                 return -EPERM;
2988         if (!dir->i_op->link)
2989                 return -EPERM;
2990         if (S_ISDIR(inode->i_mode))
2991                 return -EPERM;
2992
2993         error = security_inode_link(old_dentry, dir, new_dentry);
2994         if (error)
2995                 return error;
2996
2997         mutex_lock(&inode->i_mutex);
2998         /* Make sure we don't allow creating hardlink to an unlinked file */
2999         if (inode->i_nlink == 0)
3000                 error =  -ENOENT;
3001         else if (max_links && inode->i_nlink >= max_links)
3002                 error = -EMLINK;
3003         else
3004                 error = dir->i_op->link(old_dentry, dir, new_dentry);
3005         mutex_unlock(&inode->i_mutex);
3006         if (!error)
3007                 fsnotify_link(dir, inode, new_dentry);
3008         return error;
3009 }
3010
3011 /*
3012  * Hardlinks are often used in delicate situations.  We avoid
3013  * security-related surprises by not following symlinks on the
3014  * newname.  --KAB
3015  *
3016  * We don't follow them on the oldname either to be compatible
3017  * with linux 2.0, and to avoid hard-linking to directories
3018  * and other special files.  --ADM
3019  */
3020 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3021                 int, newdfd, const char __user *, newname, int, flags)
3022 {
3023         struct dentry *new_dentry;
3024         struct path old_path, new_path;
3025         int how = 0;
3026         int error;
3027
3028         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3029                 return -EINVAL;
3030         /*
3031          * To use null names we require CAP_DAC_READ_SEARCH
3032          * This ensures that not everyone will be able to create
3033          * handlink using the passed filedescriptor.
3034          */
3035         if (flags & AT_EMPTY_PATH) {
3036                 if (!capable(CAP_DAC_READ_SEARCH))
3037                         return -ENOENT;
3038                 how = LOOKUP_EMPTY;
3039         }
3040
3041         if (flags & AT_SYMLINK_FOLLOW)
3042                 how |= LOOKUP_FOLLOW;
3043
3044         error = user_path_at(olddfd, oldname, how, &old_path);
3045         if (error)
3046                 return error;
3047
3048         new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3049         error = PTR_ERR(new_dentry);
3050         if (IS_ERR(new_dentry))
3051                 goto out;
3052
3053         error = -EXDEV;
3054         if (old_path.mnt != new_path.mnt)
3055                 goto out_dput;
3056         error = mnt_want_write(new_path.mnt);
3057         if (error)
3058                 goto out_dput;
3059         error = security_path_link(old_path.dentry, &new_path, new_dentry);
3060         if (error)
3061                 goto out_drop_write;
3062         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3063 out_drop_write:
3064         mnt_drop_write(new_path.mnt);
3065 out_dput:
3066         dput(new_dentry);
3067         mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3068         path_put(&new_path);
3069 out:
3070         path_put(&old_path);
3071
3072         return error;
3073 }
3074
3075 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3076 {
3077         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3078 }
3079
3080 /*
3081  * The worst of all namespace operations - renaming directory. "Perverted"
3082  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3083  * Problems:
3084  *      a) we can get into loop creation. Check is done in is_subdir().
3085  *      b) race potential - two innocent renames can create a loop together.
3086  *         That's where 4.4 screws up. Current fix: serialization on
3087  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3088  *         story.
3089  *      c) we have to lock _three_ objects - parents and victim (if it exists).
3090  *         And that - after we got ->i_mutex on parents (until then we don't know
3091  *         whether the target exists).  Solution: try to be smart with locking
3092  *         order for inodes.  We rely on the fact that tree topology may change
3093  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
3094  *         move will be locked.  Thus we can rank directories by the tree
3095  *         (ancestors first) and rank all non-directories after them.
3096  *         That works since everybody except rename does "lock parent, lookup,
3097  *         lock child" and rename is under ->s_vfs_rename_mutex.
3098  *         HOWEVER, it relies on the assumption that any object with ->lookup()
3099  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
3100  *         we'd better make sure that there's no link(2) for them.
3101  *      d) conversion from fhandle to dentry may come in the wrong moment - when
3102  *         we are removing the target. Solution: we will have to grab ->i_mutex
3103  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3104  *         ->i_mutex on parents, which works but leads to some truly excessive
3105  *         locking].
3106  */
3107 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3108                           struct inode *new_dir, struct dentry *new_dentry)
3109 {
3110         int error = 0;
3111         struct inode *target = new_dentry->d_inode;
3112         unsigned max_links = new_dir->i_sb->s_max_links;
3113
3114         /*
3115          * If we are going to change the parent - check write permissions,
3116          * we'll need to flip '..'.
3117          */
3118         if (new_dir != old_dir) {
3119                 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3120                 if (error)
3121                         return error;
3122         }
3123
3124         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3125         if (error)
3126                 return error;
3127
3128         dget(new_dentry);
3129         if (target)
3130                 mutex_lock(&target->i_mutex);
3131
3132         error = -EBUSY;
3133         if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3134                 goto out;
3135
3136         error = -EMLINK;
3137         if (max_links && !target && new_dir != old_dir &&
3138             new_dir->i_nlink >= max_links)
3139                 goto out;
3140
3141         if (target)
3142                 shrink_dcache_parent(new_dentry);
3143         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3144         if (error)
3145                 goto out;
3146
3147         if (target) {
3148                 target->i_flags |= S_DEAD;
3149                 dont_mount(new_dentry);
3150         }
3151 out:
3152         if (target)
3153                 mutex_unlock(&target->i_mutex);
3154         dput(new_dentry);
3155         if (!error)
3156                 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3157                         d_move(old_dentry,new_dentry);
3158         return error;
3159 }
3160
3161 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3162                             struct inode *new_dir, struct dentry *new_dentry)
3163 {
3164         struct inode *target = new_dentry->d_inode;
3165         int error;
3166
3167         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3168         if (error)
3169                 return error;
3170
3171         dget(new_dentry);
3172         if (target)
3173                 mutex_lock(&target->i_mutex);
3174
3175         error = -EBUSY;
3176         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3177                 goto out;
3178
3179         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3180         if (error)
3181                 goto out;
3182
3183         if (target)
3184                 dont_mount(new_dentry);
3185         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3186                 d_move(old_dentry, new_dentry);
3187 out:
3188         if (target)
3189                 mutex_unlock(&target->i_mutex);
3190         dput(new_dentry);
3191         return error;
3192 }
3193
3194 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3195                struct inode *new_dir, struct dentry *new_dentry)
3196 {
3197         int error;
3198         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3199         const unsigned char *old_name;
3200
3201         if (old_dentry->d_inode == new_dentry->d_inode)
3202                 return 0;
3203
3204         error = may_delete(old_dir, old_dentry, is_dir);
3205         if (error)
3206                 return error;
3207
3208         if (!new_dentry->d_inode)
3209                 error = may_create(new_dir, new_dentry);
3210         else
3211                 error = may_delete(new_dir, new_dentry, is_dir);
3212         if (error)
3213                 return error;
3214
3215         if (!old_dir->i_op->rename)
3216                 return -EPERM;
3217
3218         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3219
3220         if (is_dir)
3221                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3222         else
3223                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3224         if (!error)
3225                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3226                               new_dentry->d_inode, old_dentry);
3227         fsnotify_oldname_free(old_name);
3228
3229         return error;
3230 }
3231
3232 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3233                 int, newdfd, const char __user *, newname)
3234 {
3235         struct dentry *old_dir, *new_dir;
3236         struct dentry *old_dentry, *new_dentry;
3237         struct dentry *trap;
3238         struct nameidata oldnd, newnd;
3239         char *from;
3240         char *to;
3241         int error;
3242
3243         error = user_path_parent(olddfd, oldname, &oldnd, &from);
3244         if (error)
3245                 goto exit;
3246
3247         error = user_path_parent(newdfd, newname, &newnd, &to);
3248         if (error)
3249                 goto exit1;
3250
3251         error = -EXDEV;
3252         if (oldnd.path.mnt != newnd.path.mnt)
3253                 goto exit2;
3254
3255         old_dir = oldnd.path.dentry;
3256         error = -EBUSY;
3257         if (oldnd.last_type != LAST_NORM)
3258                 goto exit2;
3259
3260         new_dir = newnd.path.dentry;
3261         if (newnd.last_type != LAST_NORM)
3262                 goto exit2;
3263
3264         oldnd.flags &= ~LOOKUP_PARENT;
3265         newnd.flags &= ~LOOKUP_PARENT;
3266         newnd.flags |= LOOKUP_RENAME_TARGET;
3267
3268         trap = lock_rename(new_dir, old_dir);
3269
3270         old_dentry = lookup_hash(&oldnd);
3271         error = PTR_ERR(old_dentry);
3272         if (IS_ERR(old_dentry))
3273                 goto exit3;
3274         /* source must exist */
3275         error = -ENOENT;
3276         if (!old_dentry->d_inode)
3277                 goto exit4;
3278         /* unless the source is a directory trailing slashes give -ENOTDIR */
3279         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3280                 error = -ENOTDIR;
3281                 if (oldnd.last.name[oldnd.last.len])
3282                         goto exit4;
3283                 if (newnd.last.name[newnd.last.len])
3284                         goto exit4;
3285         }
3286         /* source should not be ancestor of target */
3287         error = -EINVAL;
3288         if (old_dentry == trap)
3289                 goto exit4;
3290         new_dentry = lookup_hash(&newnd);
3291         error = PTR_ERR(new_dentry);
3292         if (IS_ERR(new_dentry))
3293                 goto exit4;
3294         /* target should not be an ancestor of source */
3295         error = -ENOTEMPTY;
3296         if (new_dentry == trap)
3297                 goto exit5;
3298
3299         error = mnt_want_write(oldnd.path.mnt);
3300         if (error)
3301                 goto exit5;
3302         error = security_path_rename(&oldnd.path, old_dentry,
3303                                      &newnd.path, new_dentry);
3304         if (error)
3305                 goto exit6;
3306         error = vfs_rename(old_dir->d_inode, old_dentry,
3307                                    new_dir->d_inode, new_dentry);
3308 exit6:
3309         mnt_drop_write(oldnd.path.mnt);
3310 exit5:
3311         dput(new_dentry);
3312 exit4:
3313         dput(old_dentry);
3314 exit3:
3315         unlock_rename(new_dir, old_dir);
3316 exit2:
3317         path_put(&newnd.path);
3318         putname(to);
3319 exit1:
3320         path_put(&oldnd.path);
3321         putname(from);
3322 exit:
3323         return error;
3324 }
3325
3326 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3327 {
3328         return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3329 }
3330
3331 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3332 {
3333         int len;
3334
3335         len = PTR_ERR(link);
3336         if (IS_ERR(link))
3337                 goto out;
3338
3339         len = strlen(link);
3340         if (len > (unsigned) buflen)
3341                 len = buflen;
3342         if (copy_to_user(buffer, link, len))
3343                 len = -EFAULT;
3344 out:
3345         return len;
3346 }
3347
3348 /*
3349  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
3350  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
3351  * using) it for any given inode is up to filesystem.
3352  */
3353 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3354 {
3355         struct nameidata nd;
3356         void *cookie;
3357         int res;
3358
3359         nd.depth = 0;
3360         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3361         if (IS_ERR(cookie))
3362                 return PTR_ERR(cookie);
3363
3364         res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3365         if (dentry->d_inode->i_op->put_link)
3366                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3367         return res;
3368 }
3369
3370 int vfs_follow_link(struct nameidata *nd, const char *link)
3371 {
3372         return __vfs_follow_link(nd, link);
3373 }
3374
3375 /* get the link contents into pagecache */
3376 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3377 {
3378         char *kaddr;
3379         struct page *page;
3380         struct address_space *mapping = dentry->d_inode->i_mapping;
3381         page = read_mapping_page(mapping, 0, NULL);
3382         if (IS_ERR(page))
3383                 return (char*)page;
3384         *ppage = page;
3385         kaddr = kmap(page);
3386         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3387         return kaddr;
3388 }
3389
3390 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3391 {
3392         struct page *page = NULL;
3393         char *s = page_getlink(dentry, &page);
3394         int res = vfs_readlink(dentry,buffer,buflen,s);
3395         if (page) {
3396                 kunmap(page);
3397                 page_cache_release(page);
3398         }
3399         return res;
3400 }
3401
3402 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3403 {
3404         struct page *page = NULL;
3405         nd_set_link(nd, page_getlink(dentry, &page));
3406         return page;
3407 }
3408
3409 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3410 {
3411         struct page *page = cookie;
3412
3413         if (page) {
3414                 kunmap(page);
3415                 page_cache_release(page);
3416         }
3417 }
3418
3419 /*
3420  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3421  */
3422 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3423 {
3424         struct address_space *mapping = inode->i_mapping;
3425         struct page *page;
3426         void *fsdata;
3427         int err;
3428         char *kaddr;
3429         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3430         if (nofs)
3431                 flags |= AOP_FLAG_NOFS;
3432
3433 retry:
3434         err = pagecache_write_begin(NULL, mapping, 0, len-1,
3435                                 flags, &page, &fsdata);
3436         if (err)
3437                 goto fail;
3438
3439         kaddr = kmap_atomic(page);
3440         memcpy(kaddr, symname, len-1);
3441         kunmap_atomic(kaddr);
3442
3443         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3444                                                         page, fsdata);
3445         if (err < 0)
3446                 goto fail;
3447         if (err < len-1)
3448                 goto retry;
3449
3450         mark_inode_dirty(inode);
3451         return 0;
3452 fail:
3453         return err;
3454 }
3455
3456 int page_symlink(struct inode *inode, const char *symname, int len)
3457 {
3458         return __page_symlink(inode, symname, len,
3459                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3460 }
3461
3462 const struct inode_operations page_symlink_inode_operations = {
3463         .readlink       = generic_readlink,
3464         .follow_link    = page_follow_link_light,
3465         .put_link       = page_put_link,
3466 };
3467
3468 EXPORT_SYMBOL(user_path_at);
3469 EXPORT_SYMBOL(follow_down_one);
3470 EXPORT_SYMBOL(follow_down);
3471 EXPORT_SYMBOL(follow_up);
3472 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3473 EXPORT_SYMBOL(getname);
3474 EXPORT_SYMBOL(lock_rename);
3475 EXPORT_SYMBOL(lookup_one_len);
3476 EXPORT_SYMBOL(page_follow_link_light);
3477 EXPORT_SYMBOL(page_put_link);
3478 EXPORT_SYMBOL(page_readlink);
3479 EXPORT_SYMBOL(__page_symlink);
3480 EXPORT_SYMBOL(page_symlink);
3481 EXPORT_SYMBOL(page_symlink_inode_operations);
3482 EXPORT_SYMBOL(kern_path);
3483 EXPORT_SYMBOL(vfs_path_lookup);
3484 EXPORT_SYMBOL(inode_permission);
3485 EXPORT_SYMBOL(unlock_rename);
3486 EXPORT_SYMBOL(vfs_create);
3487 EXPORT_SYMBOL(vfs_follow_link);
3488 EXPORT_SYMBOL(vfs_link);
3489 EXPORT_SYMBOL(vfs_mkdir);
3490 EXPORT_SYMBOL(vfs_mknod);
3491 EXPORT_SYMBOL(generic_permission);
3492 EXPORT_SYMBOL(vfs_readlink);
3493 EXPORT_SYMBOL(vfs_rename);
3494 EXPORT_SYMBOL(vfs_rmdir);
3495 EXPORT_SYMBOL(vfs_symlink);
3496 EXPORT_SYMBOL(vfs_unlink);
3497 EXPORT_SYMBOL(dentry_unhash);
3498 EXPORT_SYMBOL(generic_readlink);