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 lookup_fast(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
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         return 1;
1226 }
1227
1228 /* Fast lookup failed, do it the slow way */
1229 static int lookup_slow(struct nameidata *nd, struct qstr *name,
1230                        struct path *path)
1231 {
1232         struct dentry *dentry, *parent;
1233         int err;
1234
1235         parent = nd->path.dentry;
1236         BUG_ON(nd->inode != parent->d_inode);
1237
1238         mutex_lock(&parent->d_inode->i_mutex);
1239         dentry = __lookup_hash(name, parent, nd);
1240         mutex_unlock(&parent->d_inode->i_mutex);
1241         if (IS_ERR(dentry))
1242                 return PTR_ERR(dentry);
1243         path->mnt = nd->path.mnt;
1244         path->dentry = dentry;
1245         err = follow_managed(path, nd->flags);
1246         if (unlikely(err < 0)) {
1247                 path_put_conditional(path, nd);
1248                 return err;
1249         }
1250         if (err)
1251                 nd->flags |= LOOKUP_JUMPED;
1252         return 0;
1253 }
1254
1255 static inline int may_lookup(struct nameidata *nd)
1256 {
1257         if (nd->flags & LOOKUP_RCU) {
1258                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1259                 if (err != -ECHILD)
1260                         return err;
1261                 if (unlazy_walk(nd, NULL))
1262                         return -ECHILD;
1263         }
1264         return inode_permission(nd->inode, MAY_EXEC);
1265 }
1266
1267 static inline int handle_dots(struct nameidata *nd, int type)
1268 {
1269         if (type == LAST_DOTDOT) {
1270                 if (nd->flags & LOOKUP_RCU) {
1271                         if (follow_dotdot_rcu(nd))
1272                                 return -ECHILD;
1273                 } else
1274                         follow_dotdot(nd);
1275         }
1276         return 0;
1277 }
1278
1279 static void terminate_walk(struct nameidata *nd)
1280 {
1281         if (!(nd->flags & LOOKUP_RCU)) {
1282                 path_put(&nd->path);
1283         } else {
1284                 nd->flags &= ~LOOKUP_RCU;
1285                 if (!(nd->flags & LOOKUP_ROOT))
1286                         nd->root.mnt = NULL;
1287                 rcu_read_unlock();
1288                 br_read_unlock(&vfsmount_lock);
1289         }
1290 }
1291
1292 /*
1293  * Do we need to follow links? We _really_ want to be able
1294  * to do this check without having to look at inode->i_op,
1295  * so we keep a cache of "no, this doesn't need follow_link"
1296  * for the common case.
1297  */
1298 static inline int should_follow_link(struct inode *inode, int follow)
1299 {
1300         if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1301                 if (likely(inode->i_op->follow_link))
1302                         return follow;
1303
1304                 /* This gets set once for the inode lifetime */
1305                 spin_lock(&inode->i_lock);
1306                 inode->i_opflags |= IOP_NOFOLLOW;
1307                 spin_unlock(&inode->i_lock);
1308         }
1309         return 0;
1310 }
1311
1312 static inline int walk_component(struct nameidata *nd, struct path *path,
1313                 struct qstr *name, int type, int follow)
1314 {
1315         struct inode *inode;
1316         int err;
1317         /*
1318          * "." and ".." are special - ".." especially so because it has
1319          * to be able to know about the current root directory and
1320          * parent relationships.
1321          */
1322         if (unlikely(type != LAST_NORM))
1323                 return handle_dots(nd, type);
1324         err = lookup_fast(nd, name, path, &inode);
1325         if (unlikely(err)) {
1326                 if (err < 0)
1327                         goto out_err;
1328
1329                 err = lookup_slow(nd, name, path);
1330                 if (err < 0)
1331                         goto out_err;
1332
1333                 inode = path->dentry->d_inode;
1334         }
1335         err = -ENOENT;
1336         if (!inode)
1337                 goto out_path_put;
1338
1339         if (should_follow_link(inode, follow)) {
1340                 if (nd->flags & LOOKUP_RCU) {
1341                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1342                                 err = -ECHILD;
1343                                 goto out_err;
1344                         }
1345                 }
1346                 BUG_ON(inode != path->dentry->d_inode);
1347                 return 1;
1348         }
1349         path_to_nameidata(path, nd);
1350         nd->inode = inode;
1351         return 0;
1352
1353 out_path_put:
1354         path_to_nameidata(path, nd);
1355 out_err:
1356         terminate_walk(nd);
1357         return err;
1358 }
1359
1360 /*
1361  * This limits recursive symlink follows to 8, while
1362  * limiting consecutive symlinks to 40.
1363  *
1364  * Without that kind of total limit, nasty chains of consecutive
1365  * symlinks can cause almost arbitrarily long lookups.
1366  */
1367 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1368 {
1369         int res;
1370
1371         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1372                 path_put_conditional(path, nd);
1373                 path_put(&nd->path);
1374                 return -ELOOP;
1375         }
1376         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1377
1378         nd->depth++;
1379         current->link_count++;
1380
1381         do {
1382                 struct path link = *path;
1383                 void *cookie;
1384
1385                 res = follow_link(&link, nd, &cookie);
1386                 if (!res)
1387                         res = walk_component(nd, path, &nd->last,
1388                                              nd->last_type, LOOKUP_FOLLOW);
1389                 put_link(nd, &link, cookie);
1390         } while (res > 0);
1391
1392         current->link_count--;
1393         nd->depth--;
1394         return res;
1395 }
1396
1397 /*
1398  * We really don't want to look at inode->i_op->lookup
1399  * when we don't have to. So we keep a cache bit in
1400  * the inode ->i_opflags field that says "yes, we can
1401  * do lookup on this inode".
1402  */
1403 static inline int can_lookup(struct inode *inode)
1404 {
1405         if (likely(inode->i_opflags & IOP_LOOKUP))
1406                 return 1;
1407         if (likely(!inode->i_op->lookup))
1408                 return 0;
1409
1410         /* We do this once for the lifetime of the inode */
1411         spin_lock(&inode->i_lock);
1412         inode->i_opflags |= IOP_LOOKUP;
1413         spin_unlock(&inode->i_lock);
1414         return 1;
1415 }
1416
1417 /*
1418  * We can do the critical dentry name comparison and hashing
1419  * operations one word at a time, but we are limited to:
1420  *
1421  * - Architectures with fast unaligned word accesses. We could
1422  *   do a "get_unaligned()" if this helps and is sufficiently
1423  *   fast.
1424  *
1425  * - Little-endian machines (so that we can generate the mask
1426  *   of low bytes efficiently). Again, we *could* do a byte
1427  *   swapping load on big-endian architectures if that is not
1428  *   expensive enough to make the optimization worthless.
1429  *
1430  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1431  *   do not trap on the (extremely unlikely) case of a page
1432  *   crossing operation.
1433  *
1434  * - Furthermore, we need an efficient 64-bit compile for the
1435  *   64-bit case in order to generate the "number of bytes in
1436  *   the final mask". Again, that could be replaced with a
1437  *   efficient population count instruction or similar.
1438  */
1439 #ifdef CONFIG_DCACHE_WORD_ACCESS
1440
1441 #include <asm/word-at-a-time.h>
1442
1443 #ifdef CONFIG_64BIT
1444
1445 static inline unsigned int fold_hash(unsigned long hash)
1446 {
1447         hash += hash >> (8*sizeof(int));
1448         return hash;
1449 }
1450
1451 #else   /* 32-bit case */
1452
1453 #define fold_hash(x) (x)
1454
1455 #endif
1456
1457 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1458 {
1459         unsigned long a, mask;
1460         unsigned long hash = 0;
1461
1462         for (;;) {
1463                 a = load_unaligned_zeropad(name);
1464                 if (len < sizeof(unsigned long))
1465                         break;
1466                 hash += a;
1467                 hash *= 9;
1468                 name += sizeof(unsigned long);
1469                 len -= sizeof(unsigned long);
1470                 if (!len)
1471                         goto done;
1472         }
1473         mask = ~(~0ul << len*8);
1474         hash += mask & a;
1475 done:
1476         return fold_hash(hash);
1477 }
1478 EXPORT_SYMBOL(full_name_hash);
1479
1480 /*
1481  * Calculate the length and hash of the path component, and
1482  * return the length of the component;
1483  */
1484 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1485 {
1486         unsigned long a, b, adata, bdata, mask, hash, len;
1487         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1488
1489         hash = a = 0;
1490         len = -sizeof(unsigned long);
1491         do {
1492                 hash = (hash + a) * 9;
1493                 len += sizeof(unsigned long);
1494                 a = load_unaligned_zeropad(name+len);
1495                 b = a ^ REPEAT_BYTE('/');
1496         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1497
1498         adata = prep_zero_mask(a, adata, &constants);
1499         bdata = prep_zero_mask(b, bdata, &constants);
1500
1501         mask = create_zero_mask(adata | bdata);
1502
1503         hash += a & zero_bytemask(mask);
1504         *hashp = fold_hash(hash);
1505
1506         return len + find_zero(mask);
1507 }
1508
1509 #else
1510
1511 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1512 {
1513         unsigned long hash = init_name_hash();
1514         while (len--)
1515                 hash = partial_name_hash(*name++, hash);
1516         return end_name_hash(hash);
1517 }
1518 EXPORT_SYMBOL(full_name_hash);
1519
1520 /*
1521  * We know there's a real path component here of at least
1522  * one character.
1523  */
1524 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1525 {
1526         unsigned long hash = init_name_hash();
1527         unsigned long len = 0, c;
1528
1529         c = (unsigned char)*name;
1530         do {
1531                 len++;
1532                 hash = partial_name_hash(c, hash);
1533                 c = (unsigned char)name[len];
1534         } while (c && c != '/');
1535         *hashp = end_name_hash(hash);
1536         return len;
1537 }
1538
1539 #endif
1540
1541 /*
1542  * Name resolution.
1543  * This is the basic name resolution function, turning a pathname into
1544  * the final dentry. We expect 'base' to be positive and a directory.
1545  *
1546  * Returns 0 and nd will have valid dentry and mnt on success.
1547  * Returns error and drops reference to input namei data on failure.
1548  */
1549 static int link_path_walk(const char *name, struct nameidata *nd)
1550 {
1551         struct path next;
1552         int err;
1553
1554         while (*name=='/')
1555                 name++;
1556         if (!*name)
1557                 return 0;
1558
1559         /* At this point we know we have a real path component. */
1560         for(;;) {
1561                 struct qstr this;
1562                 long len;
1563                 int type;
1564
1565                 err = may_lookup(nd);
1566                 if (err)
1567                         break;
1568
1569                 len = hash_name(name, &this.hash);
1570                 this.name = name;
1571                 this.len = len;
1572
1573                 type = LAST_NORM;
1574                 if (name[0] == '.') switch (len) {
1575                         case 2:
1576                                 if (name[1] == '.') {
1577                                         type = LAST_DOTDOT;
1578                                         nd->flags |= LOOKUP_JUMPED;
1579                                 }
1580                                 break;
1581                         case 1:
1582                                 type = LAST_DOT;
1583                 }
1584                 if (likely(type == LAST_NORM)) {
1585                         struct dentry *parent = nd->path.dentry;
1586                         nd->flags &= ~LOOKUP_JUMPED;
1587                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1588                                 err = parent->d_op->d_hash(parent, nd->inode,
1589                                                            &this);
1590                                 if (err < 0)
1591                                         break;
1592                         }
1593                 }
1594
1595                 if (!name[len])
1596                         goto last_component;
1597                 /*
1598                  * If it wasn't NUL, we know it was '/'. Skip that
1599                  * slash, and continue until no more slashes.
1600                  */
1601                 do {
1602                         len++;
1603                 } while (unlikely(name[len] == '/'));
1604                 if (!name[len])
1605                         goto last_component;
1606                 name += len;
1607
1608                 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1609                 if (err < 0)
1610                         return err;
1611
1612                 if (err) {
1613                         err = nested_symlink(&next, nd);
1614                         if (err)
1615                                 return err;
1616                 }
1617                 if (can_lookup(nd->inode))
1618                         continue;
1619                 err = -ENOTDIR;
1620                 break;
1621                 /* here ends the main loop */
1622
1623 last_component:
1624                 nd->last = this;
1625                 nd->last_type = type;
1626                 return 0;
1627         }
1628         terminate_walk(nd);
1629         return err;
1630 }
1631
1632 static int path_init(int dfd, const char *name, unsigned int flags,
1633                      struct nameidata *nd, struct file **fp)
1634 {
1635         int retval = 0;
1636         int fput_needed;
1637         struct file *file;
1638
1639         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1640         nd->flags = flags | LOOKUP_JUMPED;
1641         nd->depth = 0;
1642         if (flags & LOOKUP_ROOT) {
1643                 struct inode *inode = nd->root.dentry->d_inode;
1644                 if (*name) {
1645                         if (!inode->i_op->lookup)
1646                                 return -ENOTDIR;
1647                         retval = inode_permission(inode, MAY_EXEC);
1648                         if (retval)
1649                                 return retval;
1650                 }
1651                 nd->path = nd->root;
1652                 nd->inode = inode;
1653                 if (flags & LOOKUP_RCU) {
1654                         br_read_lock(&vfsmount_lock);
1655                         rcu_read_lock();
1656                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1657                 } else {
1658                         path_get(&nd->path);
1659                 }
1660                 return 0;
1661         }
1662
1663         nd->root.mnt = NULL;
1664
1665         if (*name=='/') {
1666                 if (flags & LOOKUP_RCU) {
1667                         br_read_lock(&vfsmount_lock);
1668                         rcu_read_lock();
1669                         set_root_rcu(nd);
1670                 } else {
1671                         set_root(nd);
1672                         path_get(&nd->root);
1673                 }
1674                 nd->path = nd->root;
1675         } else if (dfd == AT_FDCWD) {
1676                 if (flags & LOOKUP_RCU) {
1677                         struct fs_struct *fs = current->fs;
1678                         unsigned seq;
1679
1680                         br_read_lock(&vfsmount_lock);
1681                         rcu_read_lock();
1682
1683                         do {
1684                                 seq = read_seqcount_begin(&fs->seq);
1685                                 nd->path = fs->pwd;
1686                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1687                         } while (read_seqcount_retry(&fs->seq, seq));
1688                 } else {
1689                         get_fs_pwd(current->fs, &nd->path);
1690                 }
1691         } else {
1692                 struct dentry *dentry;
1693
1694                 file = fget_raw_light(dfd, &fput_needed);
1695                 retval = -EBADF;
1696                 if (!file)
1697                         goto out_fail;
1698
1699                 dentry = file->f_path.dentry;
1700
1701                 if (*name) {
1702                         retval = -ENOTDIR;
1703                         if (!S_ISDIR(dentry->d_inode->i_mode))
1704                                 goto fput_fail;
1705
1706                         retval = inode_permission(dentry->d_inode, MAY_EXEC);
1707                         if (retval)
1708                                 goto fput_fail;
1709                 }
1710
1711                 nd->path = file->f_path;
1712                 if (flags & LOOKUP_RCU) {
1713                         if (fput_needed)
1714                                 *fp = file;
1715                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1716                         br_read_lock(&vfsmount_lock);
1717                         rcu_read_lock();
1718                 } else {
1719                         path_get(&file->f_path);
1720                         fput_light(file, fput_needed);
1721                 }
1722         }
1723
1724         nd->inode = nd->path.dentry->d_inode;
1725         return 0;
1726
1727 fput_fail:
1728         fput_light(file, fput_needed);
1729 out_fail:
1730         return retval;
1731 }
1732
1733 static inline int lookup_last(struct nameidata *nd, struct path *path)
1734 {
1735         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1736                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1737
1738         nd->flags &= ~LOOKUP_PARENT;
1739         return walk_component(nd, path, &nd->last, nd->last_type,
1740                                         nd->flags & LOOKUP_FOLLOW);
1741 }
1742
1743 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1744 static int path_lookupat(int dfd, const char *name,
1745                                 unsigned int flags, struct nameidata *nd)
1746 {
1747         struct file *base = NULL;
1748         struct path path;
1749         int err;
1750
1751         /*
1752          * Path walking is largely split up into 2 different synchronisation
1753          * schemes, rcu-walk and ref-walk (explained in
1754          * Documentation/filesystems/path-lookup.txt). These share much of the
1755          * path walk code, but some things particularly setup, cleanup, and
1756          * following mounts are sufficiently divergent that functions are
1757          * duplicated. Typically there is a function foo(), and its RCU
1758          * analogue, foo_rcu().
1759          *
1760          * -ECHILD is the error number of choice (just to avoid clashes) that
1761          * is returned if some aspect of an rcu-walk fails. Such an error must
1762          * be handled by restarting a traditional ref-walk (which will always
1763          * be able to complete).
1764          */
1765         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1766
1767         if (unlikely(err))
1768                 return err;
1769
1770         current->total_link_count = 0;
1771         err = link_path_walk(name, nd);
1772
1773         if (!err && !(flags & LOOKUP_PARENT)) {
1774                 err = lookup_last(nd, &path);
1775                 while (err > 0) {
1776                         void *cookie;
1777                         struct path link = path;
1778                         nd->flags |= LOOKUP_PARENT;
1779                         err = follow_link(&link, nd, &cookie);
1780                         if (!err)
1781                                 err = lookup_last(nd, &path);
1782                         put_link(nd, &link, cookie);
1783                 }
1784         }
1785
1786         if (!err)
1787                 err = complete_walk(nd);
1788
1789         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1790                 if (!nd->inode->i_op->lookup) {
1791                         path_put(&nd->path);
1792                         err = -ENOTDIR;
1793                 }
1794         }
1795
1796         if (base)
1797                 fput(base);
1798
1799         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1800                 path_put(&nd->root);
1801                 nd->root.mnt = NULL;
1802         }
1803         return err;
1804 }
1805
1806 static int do_path_lookup(int dfd, const char *name,
1807                                 unsigned int flags, struct nameidata *nd)
1808 {
1809         int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1810         if (unlikely(retval == -ECHILD))
1811                 retval = path_lookupat(dfd, name, flags, nd);
1812         if (unlikely(retval == -ESTALE))
1813                 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1814
1815         if (likely(!retval)) {
1816                 if (unlikely(!audit_dummy_context())) {
1817                         if (nd->path.dentry && nd->inode)
1818                                 audit_inode(name, nd->path.dentry);
1819                 }
1820         }
1821         return retval;
1822 }
1823
1824 int kern_path_parent(const char *name, struct nameidata *nd)
1825 {
1826         return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1827 }
1828
1829 int kern_path(const char *name, unsigned int flags, struct path *path)
1830 {
1831         struct nameidata nd;
1832         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1833         if (!res)
1834                 *path = nd.path;
1835         return res;
1836 }
1837
1838 /**
1839  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1840  * @dentry:  pointer to dentry of the base directory
1841  * @mnt: pointer to vfs mount of the base directory
1842  * @name: pointer to file name
1843  * @flags: lookup flags
1844  * @path: pointer to struct path to fill
1845  */
1846 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1847                     const char *name, unsigned int flags,
1848                     struct path *path)
1849 {
1850         struct nameidata nd;
1851         int err;
1852         nd.root.dentry = dentry;
1853         nd.root.mnt = mnt;
1854         BUG_ON(flags & LOOKUP_PARENT);
1855         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1856         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1857         if (!err)
1858                 *path = nd.path;
1859         return err;
1860 }
1861
1862 /*
1863  * Restricted form of lookup. Doesn't follow links, single-component only,
1864  * needs parent already locked. Doesn't follow mounts.
1865  * SMP-safe.
1866  */
1867 static struct dentry *lookup_hash(struct nameidata *nd)
1868 {
1869         return __lookup_hash(&nd->last, nd->path.dentry, nd);
1870 }
1871
1872 /**
1873  * lookup_one_len - filesystem helper to lookup single pathname component
1874  * @name:       pathname component to lookup
1875  * @base:       base directory to lookup from
1876  * @len:        maximum length @len should be interpreted to
1877  *
1878  * Note that this routine is purely a helper for filesystem usage and should
1879  * not be called by generic code.  Also note that by using this function the
1880  * nameidata argument is passed to the filesystem methods and a filesystem
1881  * using this helper needs to be prepared for that.
1882  */
1883 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1884 {
1885         struct qstr this;
1886         unsigned int c;
1887         int err;
1888
1889         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1890
1891         this.name = name;
1892         this.len = len;
1893         this.hash = full_name_hash(name, len);
1894         if (!len)
1895                 return ERR_PTR(-EACCES);
1896
1897         while (len--) {
1898                 c = *(const unsigned char *)name++;
1899                 if (c == '/' || c == '\0')
1900                         return ERR_PTR(-EACCES);
1901         }
1902         /*
1903          * See if the low-level filesystem might want
1904          * to use its own hash..
1905          */
1906         if (base->d_flags & DCACHE_OP_HASH) {
1907                 int err = base->d_op->d_hash(base, base->d_inode, &this);
1908                 if (err < 0)
1909                         return ERR_PTR(err);
1910         }
1911
1912         err = inode_permission(base->d_inode, MAY_EXEC);
1913         if (err)
1914                 return ERR_PTR(err);
1915
1916         return __lookup_hash(&this, base, NULL);
1917 }
1918
1919 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
1920                  struct path *path, int *empty)
1921 {
1922         struct nameidata nd;
1923         char *tmp = getname_flags(name, flags, empty);
1924         int err = PTR_ERR(tmp);
1925         if (!IS_ERR(tmp)) {
1926
1927                 BUG_ON(flags & LOOKUP_PARENT);
1928
1929                 err = do_path_lookup(dfd, tmp, flags, &nd);
1930                 putname(tmp);
1931                 if (!err)
1932                         *path = nd.path;
1933         }
1934         return err;
1935 }
1936
1937 int user_path_at(int dfd, const char __user *name, unsigned flags,
1938                  struct path *path)
1939 {
1940         return user_path_at_empty(dfd, name, flags, path, NULL);
1941 }
1942
1943 static int user_path_parent(int dfd, const char __user *path,
1944                         struct nameidata *nd, char **name)
1945 {
1946         char *s = getname(path);
1947         int error;
1948
1949         if (IS_ERR(s))
1950                 return PTR_ERR(s);
1951
1952         error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1953         if (error)
1954                 putname(s);
1955         else
1956                 *name = s;
1957
1958         return error;
1959 }
1960
1961 /*
1962  * It's inline, so penalty for filesystems that don't use sticky bit is
1963  * minimal.
1964  */
1965 static inline int check_sticky(struct inode *dir, struct inode *inode)
1966 {
1967         kuid_t fsuid = current_fsuid();
1968
1969         if (!(dir->i_mode & S_ISVTX))
1970                 return 0;
1971         if (uid_eq(inode->i_uid, fsuid))
1972                 return 0;
1973         if (uid_eq(dir->i_uid, fsuid))
1974                 return 0;
1975         return !inode_capable(inode, CAP_FOWNER);
1976 }
1977
1978 /*
1979  *      Check whether we can remove a link victim from directory dir, check
1980  *  whether the type of victim is right.
1981  *  1. We can't do it if dir is read-only (done in permission())
1982  *  2. We should have write and exec permissions on dir
1983  *  3. We can't remove anything from append-only dir
1984  *  4. We can't do anything with immutable dir (done in permission())
1985  *  5. If the sticky bit on dir is set we should either
1986  *      a. be owner of dir, or
1987  *      b. be owner of victim, or
1988  *      c. have CAP_FOWNER capability
1989  *  6. If the victim is append-only or immutable we can't do antyhing with
1990  *     links pointing to it.
1991  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1992  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1993  *  9. We can't remove a root or mountpoint.
1994  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1995  *     nfs_async_unlink().
1996  */
1997 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1998 {
1999         int error;
2000
2001         if (!victim->d_inode)
2002                 return -ENOENT;
2003
2004         BUG_ON(victim->d_parent->d_inode != dir);
2005         audit_inode_child(victim, dir);
2006
2007         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2008         if (error)
2009                 return error;
2010         if (IS_APPEND(dir))
2011                 return -EPERM;
2012         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2013             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2014                 return -EPERM;
2015         if (isdir) {
2016                 if (!S_ISDIR(victim->d_inode->i_mode))
2017                         return -ENOTDIR;
2018                 if (IS_ROOT(victim))
2019                         return -EBUSY;
2020         } else if (S_ISDIR(victim->d_inode->i_mode))
2021                 return -EISDIR;
2022         if (IS_DEADDIR(dir))
2023                 return -ENOENT;
2024         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2025                 return -EBUSY;
2026         return 0;
2027 }
2028
2029 /*      Check whether we can create an object with dentry child in directory
2030  *  dir.
2031  *  1. We can't do it if child already exists (open has special treatment for
2032  *     this case, but since we are inlined it's OK)
2033  *  2. We can't do it if dir is read-only (done in permission())
2034  *  3. We should have write and exec permissions on dir
2035  *  4. We can't do it if dir is immutable (done in permission())
2036  */
2037 static inline int may_create(struct inode *dir, struct dentry *child)
2038 {
2039         if (child->d_inode)
2040                 return -EEXIST;
2041         if (IS_DEADDIR(dir))
2042                 return -ENOENT;
2043         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2044 }
2045
2046 /*
2047  * p1 and p2 should be directories on the same fs.
2048  */
2049 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2050 {
2051         struct dentry *p;
2052
2053         if (p1 == p2) {
2054                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2055                 return NULL;
2056         }
2057
2058         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2059
2060         p = d_ancestor(p2, p1);
2061         if (p) {
2062                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2063                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2064                 return p;
2065         }
2066
2067         p = d_ancestor(p1, p2);
2068         if (p) {
2069                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2070                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2071                 return p;
2072         }
2073
2074         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2075         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2076         return NULL;
2077 }
2078
2079 void unlock_rename(struct dentry *p1, struct dentry *p2)
2080 {
2081         mutex_unlock(&p1->d_inode->i_mutex);
2082         if (p1 != p2) {
2083                 mutex_unlock(&p2->d_inode->i_mutex);
2084                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2085         }
2086 }
2087
2088 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2089                 struct nameidata *nd)
2090 {
2091         int error = may_create(dir, dentry);
2092
2093         if (error)
2094                 return error;
2095
2096         if (!dir->i_op->create)
2097                 return -EACCES; /* shouldn't it be ENOSYS? */
2098         mode &= S_IALLUGO;
2099         mode |= S_IFREG;
2100         error = security_inode_create(dir, dentry, mode);
2101         if (error)
2102                 return error;
2103         error = dir->i_op->create(dir, dentry, mode, nd);
2104         if (!error)
2105                 fsnotify_create(dir, dentry);
2106         return error;
2107 }
2108
2109 static int may_open(struct path *path, int acc_mode, int flag)
2110 {
2111         struct dentry *dentry = path->dentry;
2112         struct inode *inode = dentry->d_inode;
2113         int error;
2114
2115         /* O_PATH? */
2116         if (!acc_mode)
2117                 return 0;
2118
2119         if (!inode)
2120                 return -ENOENT;
2121
2122         switch (inode->i_mode & S_IFMT) {
2123         case S_IFLNK:
2124                 return -ELOOP;
2125         case S_IFDIR:
2126                 if (acc_mode & MAY_WRITE)
2127                         return -EISDIR;
2128                 break;
2129         case S_IFBLK:
2130         case S_IFCHR:
2131                 if (path->mnt->mnt_flags & MNT_NODEV)
2132                         return -EACCES;
2133                 /*FALLTHRU*/
2134         case S_IFIFO:
2135         case S_IFSOCK:
2136                 flag &= ~O_TRUNC;
2137                 break;
2138         }
2139
2140         error = inode_permission(inode, acc_mode);
2141         if (error)
2142                 return error;
2143
2144         /*
2145          * An append-only file must be opened in append mode for writing.
2146          */
2147         if (IS_APPEND(inode)) {
2148                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2149                         return -EPERM;
2150                 if (flag & O_TRUNC)
2151                         return -EPERM;
2152         }
2153
2154         /* O_NOATIME can only be set by the owner or superuser */
2155         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2156                 return -EPERM;
2157
2158         return 0;
2159 }
2160
2161 static int handle_truncate(struct file *filp)
2162 {
2163         struct path *path = &filp->f_path;
2164         struct inode *inode = path->dentry->d_inode;
2165         int error = get_write_access(inode);
2166         if (error)
2167                 return error;
2168         /*
2169          * Refuse to truncate files with mandatory locks held on them.
2170          */
2171         error = locks_verify_locked(inode);
2172         if (!error)
2173                 error = security_path_truncate(path);
2174         if (!error) {
2175                 error = do_truncate(path->dentry, 0,
2176                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2177                                     filp);
2178         }
2179         put_write_access(inode);
2180         return error;
2181 }
2182
2183 static inline int open_to_namei_flags(int flag)
2184 {
2185         if ((flag & O_ACCMODE) == 3)
2186                 flag--;
2187         return flag;
2188 }
2189
2190 /*
2191  * Handle the last step of open()
2192  */
2193 static struct file *do_last(struct nameidata *nd, struct path *path,
2194                             const struct open_flags *op, const char *pathname)
2195 {
2196         struct dentry *dir = nd->path.dentry;
2197         struct dentry *dentry;
2198         int open_flag = op->open_flag;
2199         int will_truncate = open_flag & O_TRUNC;
2200         int want_write = 0;
2201         int acc_mode = op->acc_mode;
2202         struct file *filp;
2203         struct inode *inode;
2204         int symlink_ok = 0;
2205         struct path save_parent = { .dentry = NULL, .mnt = NULL };
2206         bool retried = false;
2207         int error;
2208
2209         nd->flags &= ~LOOKUP_PARENT;
2210         nd->flags |= op->intent;
2211
2212         switch (nd->last_type) {
2213         case LAST_DOTDOT:
2214         case LAST_DOT:
2215                 error = handle_dots(nd, nd->last_type);
2216                 if (error)
2217                         return ERR_PTR(error);
2218                 /* fallthrough */
2219         case LAST_ROOT:
2220                 error = complete_walk(nd);
2221                 if (error)
2222                         return ERR_PTR(error);
2223                 audit_inode(pathname, nd->path.dentry);
2224                 if (open_flag & O_CREAT) {
2225                         error = -EISDIR;
2226                         goto exit;
2227                 }
2228                 goto ok;
2229         case LAST_BIND:
2230                 error = complete_walk(nd);
2231                 if (error)
2232                         return ERR_PTR(error);
2233                 audit_inode(pathname, dir);
2234                 goto ok;
2235         }
2236
2237         if (!(open_flag & O_CREAT)) {
2238                 if (nd->last.name[nd->last.len])
2239                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2240                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2241                         symlink_ok = 1;
2242                 /* we _can_ be in RCU mode here */
2243                 error = lookup_fast(nd, &nd->last, path, &inode);
2244                 if (unlikely(error)) {
2245                         if (error < 0)
2246                                 goto exit;
2247
2248                         error = lookup_slow(nd, &nd->last, path);
2249                         if (error < 0)
2250                                 goto exit;
2251
2252                         inode = path->dentry->d_inode;
2253                 }
2254                 goto finish_lookup;
2255         }
2256
2257         /* create side of things */
2258         /*
2259          * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been
2260          * cleared when we got to the last component we are about to look up
2261          */
2262         error = complete_walk(nd);
2263         if (error)
2264                 return ERR_PTR(error);
2265
2266         audit_inode(pathname, dir);
2267         error = -EISDIR;
2268         /* trailing slashes? */
2269         if (nd->last.name[nd->last.len])
2270                 goto exit;
2271
2272 retry_lookup:
2273         mutex_lock(&dir->d_inode->i_mutex);
2274
2275         dentry = lookup_hash(nd);
2276         error = PTR_ERR(dentry);
2277         if (IS_ERR(dentry)) {
2278                 mutex_unlock(&dir->d_inode->i_mutex);
2279                 goto exit;
2280         }
2281
2282         path->dentry = dentry;
2283         path->mnt = nd->path.mnt;
2284
2285         /* Negative dentry, just create the file */
2286         if (!dentry->d_inode) {
2287                 umode_t mode = op->mode;
2288                 if (!IS_POSIXACL(dir->d_inode))
2289                         mode &= ~current_umask();
2290                 /*
2291                  * This write is needed to ensure that a
2292                  * rw->ro transition does not occur between
2293                  * the time when the file is created and when
2294                  * a permanent write count is taken through
2295                  * the 'struct file' in nameidata_to_filp().
2296                  */
2297                 error = mnt_want_write(nd->path.mnt);
2298                 if (error)
2299                         goto exit_mutex_unlock;
2300                 want_write = 1;
2301                 /* Don't check for write permission, don't truncate */
2302                 open_flag &= ~O_TRUNC;
2303                 will_truncate = 0;
2304                 acc_mode = MAY_OPEN;
2305                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2306                 if (error)
2307                         goto exit_mutex_unlock;
2308                 error = vfs_create(dir->d_inode, dentry, mode, nd);
2309                 if (error)
2310                         goto exit_mutex_unlock;
2311                 mutex_unlock(&dir->d_inode->i_mutex);
2312                 dput(nd->path.dentry);
2313                 nd->path.dentry = dentry;
2314                 goto common;
2315         }
2316
2317         /*
2318          * It already exists.
2319          */
2320         mutex_unlock(&dir->d_inode->i_mutex);
2321         audit_inode(pathname, path->dentry);
2322
2323         error = -EEXIST;
2324         if (open_flag & O_EXCL)
2325                 goto exit_dput;
2326
2327         error = follow_managed(path, nd->flags);
2328         if (error < 0)
2329                 goto exit_dput;
2330
2331         if (error)
2332                 nd->flags |= LOOKUP_JUMPED;
2333
2334         BUG_ON(nd->flags & LOOKUP_RCU);
2335         inode = path->dentry->d_inode;
2336 finish_lookup:
2337         /* we _can_ be in RCU mode here */
2338         error = -ENOENT;
2339         if (!inode) {
2340                 path_to_nameidata(path, nd);
2341                 goto exit;
2342         }
2343
2344         if (should_follow_link(inode, !symlink_ok)) {
2345                 if (nd->flags & LOOKUP_RCU) {
2346                         if (unlikely(unlazy_walk(nd, path->dentry))) {
2347                                 error = -ECHILD;
2348                                 goto exit;
2349                         }
2350                 }
2351                 BUG_ON(inode != path->dentry->d_inode);
2352                 return NULL;
2353         }
2354
2355         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2356                 path_to_nameidata(path, nd);
2357         } else {
2358                 save_parent.dentry = nd->path.dentry;
2359                 save_parent.mnt = mntget(path->mnt);
2360                 nd->path.dentry = path->dentry;
2361
2362         }
2363         nd->inode = inode;
2364         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
2365         error = complete_walk(nd);
2366         if (error) {
2367                 path_put(&save_parent);
2368                 return ERR_PTR(error);
2369         }
2370         error = -EISDIR;
2371         if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
2372                 goto exit;
2373         error = -ENOTDIR;
2374         if ((nd->flags & LOOKUP_DIRECTORY) && !nd->inode->i_op->lookup)
2375                 goto exit;
2376         audit_inode(pathname, nd->path.dentry);
2377 ok:
2378         if (!S_ISREG(nd->inode->i_mode))
2379                 will_truncate = 0;
2380
2381         if (will_truncate) {
2382                 error = mnt_want_write(nd->path.mnt);
2383                 if (error)
2384                         goto exit;
2385                 want_write = 1;
2386         }
2387 common:
2388         error = may_open(&nd->path, acc_mode, open_flag);
2389         if (error)
2390                 goto exit;
2391         filp = nameidata_to_filp(nd);
2392         if (filp == ERR_PTR(-EOPENSTALE) && save_parent.dentry && !retried) {
2393                 BUG_ON(save_parent.dentry != dir);
2394                 path_put(&nd->path);
2395                 nd->path = save_parent;
2396                 nd->inode = dir->d_inode;
2397                 save_parent.mnt = NULL;
2398                 save_parent.dentry = NULL;
2399                 if (want_write) {
2400                         mnt_drop_write(nd->path.mnt);
2401                         want_write = 0;
2402                 }
2403                 retried = true;
2404                 goto retry_lookup;
2405         }
2406         if (!IS_ERR(filp)) {
2407                 error = ima_file_check(filp, op->acc_mode);
2408                 if (error) {
2409                         fput(filp);
2410                         filp = ERR_PTR(error);
2411                 }
2412         }
2413         if (!IS_ERR(filp)) {
2414                 if (will_truncate) {
2415                         error = handle_truncate(filp);
2416                         if (error) {
2417                                 fput(filp);
2418                                 filp = ERR_PTR(error);
2419                         }
2420                 }
2421         }
2422 out:
2423         if (want_write)
2424                 mnt_drop_write(nd->path.mnt);
2425         path_put(&save_parent);
2426         terminate_walk(nd);
2427         return filp;
2428
2429 exit_mutex_unlock:
2430         mutex_unlock(&dir->d_inode->i_mutex);
2431 exit_dput:
2432         path_put_conditional(path, nd);
2433 exit:
2434         filp = ERR_PTR(error);
2435         goto out;
2436 }
2437
2438 static struct file *path_openat(int dfd, const char *pathname,
2439                 struct nameidata *nd, const struct open_flags *op, int flags)
2440 {
2441         struct file *base = NULL;
2442         struct file *filp;
2443         struct path path;
2444         int error;
2445
2446         filp = get_empty_filp();
2447         if (!filp)
2448                 return ERR_PTR(-ENFILE);
2449
2450         filp->f_flags = op->open_flag;
2451         nd->intent.open.file = filp;
2452         nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2453         nd->intent.open.create_mode = op->mode;
2454
2455         error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2456         if (unlikely(error))
2457                 goto out_filp;
2458
2459         current->total_link_count = 0;
2460         error = link_path_walk(pathname, nd);
2461         if (unlikely(error))
2462                 goto out_filp;
2463
2464         filp = do_last(nd, &path, op, pathname);
2465         while (unlikely(!filp)) { /* trailing symlink */
2466                 struct path link = path;
2467                 void *cookie;
2468                 if (!(nd->flags & LOOKUP_FOLLOW)) {
2469                         path_put_conditional(&path, nd);
2470                         path_put(&nd->path);
2471                         filp = ERR_PTR(-ELOOP);
2472                         break;
2473                 }
2474                 nd->flags |= LOOKUP_PARENT;
2475                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2476                 error = follow_link(&link, nd, &cookie);
2477                 if (unlikely(error))
2478                         filp = ERR_PTR(error);
2479                 else
2480                         filp = do_last(nd, &path, op, pathname);
2481                 put_link(nd, &link, cookie);
2482         }
2483 out:
2484         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2485                 path_put(&nd->root);
2486         if (base)
2487                 fput(base);
2488         release_open_intent(nd);
2489         if (filp == ERR_PTR(-EOPENSTALE)) {
2490                 if (flags & LOOKUP_RCU)
2491                         filp = ERR_PTR(-ECHILD);
2492                 else
2493                         filp = ERR_PTR(-ESTALE);
2494         }
2495         return filp;
2496
2497 out_filp:
2498         filp = ERR_PTR(error);
2499         goto out;
2500 }
2501
2502 struct file *do_filp_open(int dfd, const char *pathname,
2503                 const struct open_flags *op, int flags)
2504 {
2505         struct nameidata nd;
2506         struct file *filp;
2507
2508         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2509         if (unlikely(filp == ERR_PTR(-ECHILD)))
2510                 filp = path_openat(dfd, pathname, &nd, op, flags);
2511         if (unlikely(filp == ERR_PTR(-ESTALE)))
2512                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2513         return filp;
2514 }
2515
2516 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2517                 const char *name, const struct open_flags *op, int flags)
2518 {
2519         struct nameidata nd;
2520         struct file *file;
2521
2522         nd.root.mnt = mnt;
2523         nd.root.dentry = dentry;
2524
2525         flags |= LOOKUP_ROOT;
2526
2527         if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2528                 return ERR_PTR(-ELOOP);
2529
2530         file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2531         if (unlikely(file == ERR_PTR(-ECHILD)))
2532                 file = path_openat(-1, name, &nd, op, flags);
2533         if (unlikely(file == ERR_PTR(-ESTALE)))
2534                 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2535         return file;
2536 }
2537
2538 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2539 {
2540         struct dentry *dentry = ERR_PTR(-EEXIST);
2541         struct nameidata nd;
2542         int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2543         if (error)
2544                 return ERR_PTR(error);
2545
2546         /*
2547          * Yucky last component or no last component at all?
2548          * (foo/., foo/.., /////)
2549          */
2550         if (nd.last_type != LAST_NORM)
2551                 goto out;
2552         nd.flags &= ~LOOKUP_PARENT;
2553         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2554         nd.intent.open.flags = O_EXCL;
2555
2556         /*
2557          * Do the final lookup.
2558          */
2559         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2560         dentry = lookup_hash(&nd);
2561         if (IS_ERR(dentry))
2562                 goto fail;
2563
2564         if (dentry->d_inode)
2565                 goto eexist;
2566         /*
2567          * Special case - lookup gave negative, but... we had foo/bar/
2568          * From the vfs_mknod() POV we just have a negative dentry -
2569          * all is fine. Let's be bastards - you had / on the end, you've
2570          * been asking for (non-existent) directory. -ENOENT for you.
2571          */
2572         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2573                 dput(dentry);
2574                 dentry = ERR_PTR(-ENOENT);
2575                 goto fail;
2576         }
2577         *path = nd.path;
2578         return dentry;
2579 eexist:
2580         dput(dentry);
2581         dentry = ERR_PTR(-EEXIST);
2582 fail:
2583         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2584 out:
2585         path_put(&nd.path);
2586         return dentry;
2587 }
2588 EXPORT_SYMBOL(kern_path_create);
2589
2590 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2591 {
2592         char *tmp = getname(pathname);
2593         struct dentry *res;
2594         if (IS_ERR(tmp))
2595                 return ERR_CAST(tmp);
2596         res = kern_path_create(dfd, tmp, path, is_dir);
2597         putname(tmp);
2598         return res;
2599 }
2600 EXPORT_SYMBOL(user_path_create);
2601
2602 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2603 {
2604         int error = may_create(dir, dentry);
2605
2606         if (error)
2607                 return error;
2608
2609         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
2610                 return -EPERM;
2611
2612         if (!dir->i_op->mknod)
2613                 return -EPERM;
2614
2615         error = devcgroup_inode_mknod(mode, dev);
2616         if (error)
2617                 return error;
2618
2619         error = security_inode_mknod(dir, dentry, mode, dev);
2620         if (error)
2621                 return error;
2622
2623         error = dir->i_op->mknod(dir, dentry, mode, dev);
2624         if (!error)
2625                 fsnotify_create(dir, dentry);
2626         return error;
2627 }
2628
2629 static int may_mknod(umode_t mode)
2630 {
2631         switch (mode & S_IFMT) {
2632         case S_IFREG:
2633         case S_IFCHR:
2634         case S_IFBLK:
2635         case S_IFIFO:
2636         case S_IFSOCK:
2637         case 0: /* zero mode translates to S_IFREG */
2638                 return 0;
2639         case S_IFDIR:
2640                 return -EPERM;
2641         default:
2642                 return -EINVAL;
2643         }
2644 }
2645
2646 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
2647                 unsigned, dev)
2648 {
2649         struct dentry *dentry;
2650         struct path path;
2651         int error;
2652
2653         if (S_ISDIR(mode))
2654                 return -EPERM;
2655
2656         dentry = user_path_create(dfd, filename, &path, 0);
2657         if (IS_ERR(dentry))
2658                 return PTR_ERR(dentry);
2659
2660         if (!IS_POSIXACL(path.dentry->d_inode))
2661                 mode &= ~current_umask();
2662         error = may_mknod(mode);
2663         if (error)
2664                 goto out_dput;
2665         error = mnt_want_write(path.mnt);
2666         if (error)
2667                 goto out_dput;
2668         error = security_path_mknod(&path, dentry, mode, dev);
2669         if (error)
2670                 goto out_drop_write;
2671         switch (mode & S_IFMT) {
2672                 case 0: case S_IFREG:
2673                         error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2674                         break;
2675                 case S_IFCHR: case S_IFBLK:
2676                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2677                                         new_decode_dev(dev));
2678                         break;
2679                 case S_IFIFO: case S_IFSOCK:
2680                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2681                         break;
2682         }
2683 out_drop_write:
2684         mnt_drop_write(path.mnt);
2685 out_dput:
2686         dput(dentry);
2687         mutex_unlock(&path.dentry->d_inode->i_mutex);
2688         path_put(&path);
2689
2690         return error;
2691 }
2692
2693 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
2694 {
2695         return sys_mknodat(AT_FDCWD, filename, mode, dev);
2696 }
2697
2698 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2699 {
2700         int error = may_create(dir, dentry);
2701         unsigned max_links = dir->i_sb->s_max_links;
2702
2703         if (error)
2704                 return error;
2705
2706         if (!dir->i_op->mkdir)
2707                 return -EPERM;
2708
2709         mode &= (S_IRWXUGO|S_ISVTX);
2710         error = security_inode_mkdir(dir, dentry, mode);
2711         if (error)
2712                 return error;
2713
2714         if (max_links && dir->i_nlink >= max_links)
2715                 return -EMLINK;
2716
2717         error = dir->i_op->mkdir(dir, dentry, mode);
2718         if (!error)
2719                 fsnotify_mkdir(dir, dentry);
2720         return error;
2721 }
2722
2723 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
2724 {
2725         struct dentry *dentry;
2726         struct path path;
2727         int error;
2728
2729         dentry = user_path_create(dfd, pathname, &path, 1);
2730         if (IS_ERR(dentry))
2731                 return PTR_ERR(dentry);
2732
2733         if (!IS_POSIXACL(path.dentry->d_inode))
2734                 mode &= ~current_umask();
2735         error = mnt_want_write(path.mnt);
2736         if (error)
2737                 goto out_dput;
2738         error = security_path_mkdir(&path, dentry, mode);
2739         if (error)
2740                 goto out_drop_write;
2741         error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2742 out_drop_write:
2743         mnt_drop_write(path.mnt);
2744 out_dput:
2745         dput(dentry);
2746         mutex_unlock(&path.dentry->d_inode->i_mutex);
2747         path_put(&path);
2748         return error;
2749 }
2750
2751 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
2752 {
2753         return sys_mkdirat(AT_FDCWD, pathname, mode);
2754 }
2755
2756 /*
2757  * The dentry_unhash() helper will try to drop the dentry early: we
2758  * should have a usage count of 1 if we're the only user of this
2759  * dentry, and if that is true (possibly after pruning the dcache),
2760  * then we drop the dentry now.
2761  *
2762  * A low-level filesystem can, if it choses, legally
2763  * do a
2764  *
2765  *      if (!d_unhashed(dentry))
2766  *              return -EBUSY;
2767  *
2768  * if it cannot handle the case of removing a directory
2769  * that is still in use by something else..
2770  */
2771 void dentry_unhash(struct dentry *dentry)
2772 {
2773         shrink_dcache_parent(dentry);
2774         spin_lock(&dentry->d_lock);
2775         if (dentry->d_count == 1)
2776                 __d_drop(dentry);
2777         spin_unlock(&dentry->d_lock);
2778 }
2779
2780 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2781 {
2782         int error = may_delete(dir, dentry, 1);
2783
2784         if (error)
2785                 return error;
2786
2787         if (!dir->i_op->rmdir)
2788                 return -EPERM;
2789
2790         dget(dentry);
2791         mutex_lock(&dentry->d_inode->i_mutex);
2792
2793         error = -EBUSY;
2794         if (d_mountpoint(dentry))
2795                 goto out;
2796
2797         error = security_inode_rmdir(dir, dentry);
2798         if (error)
2799                 goto out;
2800
2801         shrink_dcache_parent(dentry);
2802         error = dir->i_op->rmdir(dir, dentry);
2803         if (error)
2804                 goto out;
2805
2806         dentry->d_inode->i_flags |= S_DEAD;
2807         dont_mount(dentry);
2808
2809 out:
2810         mutex_unlock(&dentry->d_inode->i_mutex);
2811         dput(dentry);
2812         if (!error)
2813                 d_delete(dentry);
2814         return error;
2815 }
2816
2817 static long do_rmdir(int dfd, const char __user *pathname)
2818 {
2819         int error = 0;
2820         char * name;
2821         struct dentry *dentry;
2822         struct nameidata nd;
2823
2824         error = user_path_parent(dfd, pathname, &nd, &name);
2825         if (error)
2826                 return error;
2827
2828         switch(nd.last_type) {
2829         case LAST_DOTDOT:
2830                 error = -ENOTEMPTY;
2831                 goto exit1;
2832         case LAST_DOT:
2833                 error = -EINVAL;
2834                 goto exit1;
2835         case LAST_ROOT:
2836                 error = -EBUSY;
2837                 goto exit1;
2838         }
2839
2840         nd.flags &= ~LOOKUP_PARENT;
2841
2842         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2843         dentry = lookup_hash(&nd);
2844         error = PTR_ERR(dentry);
2845         if (IS_ERR(dentry))
2846                 goto exit2;
2847         if (!dentry->d_inode) {
2848                 error = -ENOENT;
2849                 goto exit3;
2850         }
2851         error = mnt_want_write(nd.path.mnt);
2852         if (error)
2853                 goto exit3;
2854         error = security_path_rmdir(&nd.path, dentry);
2855         if (error)
2856                 goto exit4;
2857         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2858 exit4:
2859         mnt_drop_write(nd.path.mnt);
2860 exit3:
2861         dput(dentry);
2862 exit2:
2863         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2864 exit1:
2865         path_put(&nd.path);
2866         putname(name);
2867         return error;
2868 }
2869
2870 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2871 {
2872         return do_rmdir(AT_FDCWD, pathname);
2873 }
2874
2875 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2876 {
2877         int error = may_delete(dir, dentry, 0);
2878
2879         if (error)
2880                 return error;
2881
2882         if (!dir->i_op->unlink)
2883                 return -EPERM;
2884
2885         mutex_lock(&dentry->d_inode->i_mutex);
2886         if (d_mountpoint(dentry))
2887                 error = -EBUSY;
2888         else {
2889                 error = security_inode_unlink(dir, dentry);
2890                 if (!error) {
2891                         error = dir->i_op->unlink(dir, dentry);
2892                         if (!error)
2893                                 dont_mount(dentry);
2894                 }
2895         }
2896         mutex_unlock(&dentry->d_inode->i_mutex);
2897
2898         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2899         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2900                 fsnotify_link_count(dentry->d_inode);
2901                 d_delete(dentry);
2902         }
2903
2904         return error;
2905 }
2906
2907 /*
2908  * Make sure that the actual truncation of the file will occur outside its
2909  * directory's i_mutex.  Truncate can take a long time if there is a lot of
2910  * writeout happening, and we don't want to prevent access to the directory
2911  * while waiting on the I/O.
2912  */
2913 static long do_unlinkat(int dfd, const char __user *pathname)
2914 {
2915         int error;
2916         char *name;
2917         struct dentry *dentry;
2918         struct nameidata nd;
2919         struct inode *inode = NULL;
2920
2921         error = user_path_parent(dfd, pathname, &nd, &name);
2922         if (error)
2923                 return error;
2924
2925         error = -EISDIR;
2926         if (nd.last_type != LAST_NORM)
2927                 goto exit1;
2928
2929         nd.flags &= ~LOOKUP_PARENT;
2930
2931         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2932         dentry = lookup_hash(&nd);
2933         error = PTR_ERR(dentry);
2934         if (!IS_ERR(dentry)) {
2935                 /* Why not before? Because we want correct error value */
2936                 if (nd.last.name[nd.last.len])
2937                         goto slashes;
2938                 inode = dentry->d_inode;
2939                 if (!inode)
2940                         goto slashes;
2941                 ihold(inode);
2942                 error = mnt_want_write(nd.path.mnt);
2943                 if (error)
2944                         goto exit2;
2945                 error = security_path_unlink(&nd.path, dentry);
2946                 if (error)
2947                         goto exit3;
2948                 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2949 exit3:
2950                 mnt_drop_write(nd.path.mnt);
2951         exit2:
2952                 dput(dentry);
2953         }
2954         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2955         if (inode)
2956                 iput(inode);    /* truncate the inode here */
2957 exit1:
2958         path_put(&nd.path);
2959         putname(name);
2960         return error;
2961
2962 slashes:
2963         error = !dentry->d_inode ? -ENOENT :
2964                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2965         goto exit2;
2966 }
2967
2968 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2969 {
2970         if ((flag & ~AT_REMOVEDIR) != 0)
2971                 return -EINVAL;
2972
2973         if (flag & AT_REMOVEDIR)
2974                 return do_rmdir(dfd, pathname);
2975
2976         return do_unlinkat(dfd, pathname);
2977 }
2978
2979 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2980 {
2981         return do_unlinkat(AT_FDCWD, pathname);
2982 }
2983
2984 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2985 {
2986         int error = may_create(dir, dentry);
2987
2988         if (error)
2989                 return error;
2990
2991         if (!dir->i_op->symlink)
2992                 return -EPERM;
2993
2994         error = security_inode_symlink(dir, dentry, oldname);
2995         if (error)
2996                 return error;
2997
2998         error = dir->i_op->symlink(dir, dentry, oldname);
2999         if (!error)
3000                 fsnotify_create(dir, dentry);
3001         return error;
3002 }
3003
3004 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3005                 int, newdfd, const char __user *, newname)
3006 {
3007         int error;
3008         char *from;
3009         struct dentry *dentry;
3010         struct path path;
3011
3012         from = getname(oldname);
3013         if (IS_ERR(from))
3014                 return PTR_ERR(from);
3015
3016         dentry = user_path_create(newdfd, newname, &path, 0);
3017         error = PTR_ERR(dentry);
3018         if (IS_ERR(dentry))
3019                 goto out_putname;
3020
3021         error = mnt_want_write(path.mnt);
3022         if (error)
3023                 goto out_dput;
3024         error = security_path_symlink(&path, dentry, from);
3025         if (error)
3026                 goto out_drop_write;
3027         error = vfs_symlink(path.dentry->d_inode, dentry, from);
3028 out_drop_write:
3029         mnt_drop_write(path.mnt);
3030 out_dput:
3031         dput(dentry);
3032         mutex_unlock(&path.dentry->d_inode->i_mutex);
3033         path_put(&path);
3034 out_putname:
3035         putname(from);
3036         return error;
3037 }
3038
3039 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3040 {
3041         return sys_symlinkat(oldname, AT_FDCWD, newname);
3042 }
3043
3044 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3045 {
3046         struct inode *inode = old_dentry->d_inode;
3047         unsigned max_links = dir->i_sb->s_max_links;
3048         int error;
3049
3050         if (!inode)
3051                 return -ENOENT;
3052
3053         error = may_create(dir, new_dentry);
3054         if (error)
3055                 return error;
3056
3057         if (dir->i_sb != inode->i_sb)
3058                 return -EXDEV;
3059
3060         /*
3061          * A link to an append-only or immutable file cannot be created.
3062          */
3063         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3064                 return -EPERM;
3065         if (!dir->i_op->link)
3066                 return -EPERM;
3067         if (S_ISDIR(inode->i_mode))
3068                 return -EPERM;
3069
3070         error = security_inode_link(old_dentry, dir, new_dentry);
3071         if (error)
3072                 return error;
3073
3074         mutex_lock(&inode->i_mutex);
3075         /* Make sure we don't allow creating hardlink to an unlinked file */
3076         if (inode->i_nlink == 0)
3077                 error =  -ENOENT;
3078         else if (max_links && inode->i_nlink >= max_links)
3079                 error = -EMLINK;
3080         else
3081                 error = dir->i_op->link(old_dentry, dir, new_dentry);
3082         mutex_unlock(&inode->i_mutex);
3083         if (!error)
3084                 fsnotify_link(dir, inode, new_dentry);
3085         return error;
3086 }
3087
3088 /*
3089  * Hardlinks are often used in delicate situations.  We avoid
3090  * security-related surprises by not following symlinks on the
3091  * newname.  --KAB
3092  *
3093  * We don't follow them on the oldname either to be compatible
3094  * with linux 2.0, and to avoid hard-linking to directories
3095  * and other special files.  --ADM
3096  */
3097 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3098                 int, newdfd, const char __user *, newname, int, flags)
3099 {
3100         struct dentry *new_dentry;
3101         struct path old_path, new_path;
3102         int how = 0;
3103         int error;
3104
3105         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3106                 return -EINVAL;
3107         /*
3108          * To use null names we require CAP_DAC_READ_SEARCH
3109          * This ensures that not everyone will be able to create
3110          * handlink using the passed filedescriptor.
3111          */
3112         if (flags & AT_EMPTY_PATH) {
3113                 if (!capable(CAP_DAC_READ_SEARCH))
3114                         return -ENOENT;
3115                 how = LOOKUP_EMPTY;
3116         }
3117
3118         if (flags & AT_SYMLINK_FOLLOW)
3119                 how |= LOOKUP_FOLLOW;
3120
3121         error = user_path_at(olddfd, oldname, how, &old_path);
3122         if (error)
3123                 return error;
3124
3125         new_dentry = user_path_create(newdfd, newname, &new_path, 0);
3126         error = PTR_ERR(new_dentry);
3127         if (IS_ERR(new_dentry))
3128                 goto out;
3129
3130         error = -EXDEV;
3131         if (old_path.mnt != new_path.mnt)
3132                 goto out_dput;
3133         error = mnt_want_write(new_path.mnt);
3134         if (error)
3135                 goto out_dput;
3136         error = security_path_link(old_path.dentry, &new_path, new_dentry);
3137         if (error)
3138                 goto out_drop_write;
3139         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3140 out_drop_write:
3141         mnt_drop_write(new_path.mnt);
3142 out_dput:
3143         dput(new_dentry);
3144         mutex_unlock(&new_path.dentry->d_inode->i_mutex);
3145         path_put(&new_path);
3146 out:
3147         path_put(&old_path);
3148
3149         return error;
3150 }
3151
3152 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3153 {
3154         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3155 }
3156
3157 /*
3158  * The worst of all namespace operations - renaming directory. "Perverted"
3159  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3160  * Problems:
3161  *      a) we can get into loop creation. Check is done in is_subdir().
3162  *      b) race potential - two innocent renames can create a loop together.
3163  *         That's where 4.4 screws up. Current fix: serialization on
3164  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3165  *         story.
3166  *      c) we have to lock _three_ objects - parents and victim (if it exists).
3167  *         And that - after we got ->i_mutex on parents (until then we don't know
3168  *         whether the target exists).  Solution: try to be smart with locking
3169  *         order for inodes.  We rely on the fact that tree topology may change
3170  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
3171  *         move will be locked.  Thus we can rank directories by the tree
3172  *         (ancestors first) and rank all non-directories after them.
3173  *         That works since everybody except rename does "lock parent, lookup,
3174  *         lock child" and rename is under ->s_vfs_rename_mutex.
3175  *         HOWEVER, it relies on the assumption that any object with ->lookup()
3176  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
3177  *         we'd better make sure that there's no link(2) for them.
3178  *      d) conversion from fhandle to dentry may come in the wrong moment - when
3179  *         we are removing the target. Solution: we will have to grab ->i_mutex
3180  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3181  *         ->i_mutex on parents, which works but leads to some truly excessive
3182  *         locking].
3183  */
3184 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3185                           struct inode *new_dir, struct dentry *new_dentry)
3186 {
3187         int error = 0;
3188         struct inode *target = new_dentry->d_inode;
3189         unsigned max_links = new_dir->i_sb->s_max_links;
3190
3191         /*
3192          * If we are going to change the parent - check write permissions,
3193          * we'll need to flip '..'.
3194          */
3195         if (new_dir != old_dir) {
3196                 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3197                 if (error)
3198                         return error;
3199         }
3200
3201         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3202         if (error)
3203                 return error;
3204
3205         dget(new_dentry);
3206         if (target)
3207                 mutex_lock(&target->i_mutex);
3208
3209         error = -EBUSY;
3210         if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3211                 goto out;
3212
3213         error = -EMLINK;
3214         if (max_links && !target && new_dir != old_dir &&
3215             new_dir->i_nlink >= max_links)
3216                 goto out;
3217
3218         if (target)
3219                 shrink_dcache_parent(new_dentry);
3220         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3221         if (error)
3222                 goto out;
3223
3224         if (target) {
3225                 target->i_flags |= S_DEAD;
3226                 dont_mount(new_dentry);
3227         }
3228 out:
3229         if (target)
3230                 mutex_unlock(&target->i_mutex);
3231         dput(new_dentry);
3232         if (!error)
3233                 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3234                         d_move(old_dentry,new_dentry);
3235         return error;
3236 }
3237
3238 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
3239                             struct inode *new_dir, struct dentry *new_dentry)
3240 {
3241         struct inode *target = new_dentry->d_inode;
3242         int error;
3243
3244         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3245         if (error)
3246                 return error;
3247
3248         dget(new_dentry);
3249         if (target)
3250                 mutex_lock(&target->i_mutex);
3251
3252         error = -EBUSY;
3253         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3254                 goto out;
3255
3256         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3257         if (error)
3258                 goto out;
3259
3260         if (target)
3261                 dont_mount(new_dentry);
3262         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3263                 d_move(old_dentry, new_dentry);
3264 out:
3265         if (target)
3266                 mutex_unlock(&target->i_mutex);
3267         dput(new_dentry);
3268         return error;
3269 }
3270
3271 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3272                struct inode *new_dir, struct dentry *new_dentry)
3273 {
3274         int error;
3275         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3276         const unsigned char *old_name;
3277
3278         if (old_dentry->d_inode == new_dentry->d_inode)
3279                 return 0;
3280
3281         error = may_delete(old_dir, old_dentry, is_dir);
3282         if (error)
3283                 return error;
3284
3285         if (!new_dentry->d_inode)
3286                 error = may_create(new_dir, new_dentry);
3287         else
3288                 error = may_delete(new_dir, new_dentry, is_dir);
3289         if (error)
3290                 return error;
3291
3292         if (!old_dir->i_op->rename)
3293                 return -EPERM;
3294
3295         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3296
3297         if (is_dir)
3298                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3299         else
3300                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3301         if (!error)
3302                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3303                               new_dentry->d_inode, old_dentry);
3304         fsnotify_oldname_free(old_name);
3305
3306         return error;
3307 }
3308
3309 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3310                 int, newdfd, const char __user *, newname)
3311 {
3312         struct dentry *old_dir, *new_dir;
3313         struct dentry *old_dentry, *new_dentry;
3314         struct dentry *trap;
3315         struct nameidata oldnd, newnd;
3316         char *from;
3317         char *to;
3318         int error;
3319
3320         error = user_path_parent(olddfd, oldname, &oldnd, &from);
3321         if (error)
3322                 goto exit;
3323
3324         error = user_path_parent(newdfd, newname, &newnd, &to);
3325         if (error)
3326                 goto exit1;
3327
3328         error = -EXDEV;
3329         if (oldnd.path.mnt != newnd.path.mnt)
3330                 goto exit2;
3331
3332         old_dir = oldnd.path.dentry;
3333         error = -EBUSY;
3334         if (oldnd.last_type != LAST_NORM)
3335                 goto exit2;
3336
3337         new_dir = newnd.path.dentry;
3338         if (newnd.last_type != LAST_NORM)
3339                 goto exit2;
3340
3341         oldnd.flags &= ~LOOKUP_PARENT;
3342         newnd.flags &= ~LOOKUP_PARENT;
3343         newnd.flags |= LOOKUP_RENAME_TARGET;
3344
3345         trap = lock_rename(new_dir, old_dir);
3346
3347         old_dentry = lookup_hash(&oldnd);
3348         error = PTR_ERR(old_dentry);
3349         if (IS_ERR(old_dentry))
3350                 goto exit3;
3351         /* source must exist */
3352         error = -ENOENT;
3353         if (!old_dentry->d_inode)
3354                 goto exit4;
3355         /* unless the source is a directory trailing slashes give -ENOTDIR */
3356         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3357                 error = -ENOTDIR;
3358                 if (oldnd.last.name[oldnd.last.len])
3359                         goto exit4;
3360                 if (newnd.last.name[newnd.last.len])
3361                         goto exit4;
3362         }
3363         /* source should not be ancestor of target */
3364         error = -EINVAL;
3365         if (old_dentry == trap)
3366                 goto exit4;
3367         new_dentry = lookup_hash(&newnd);
3368         error = PTR_ERR(new_dentry);
3369         if (IS_ERR(new_dentry))
3370                 goto exit4;
3371         /* target should not be an ancestor of source */
3372         error = -ENOTEMPTY;
3373         if (new_dentry == trap)
3374                 goto exit5;
3375
3376         error = mnt_want_write(oldnd.path.mnt);
3377         if (error)
3378                 goto exit5;
3379         error = security_path_rename(&oldnd.path, old_dentry,
3380                                      &newnd.path, new_dentry);
3381         if (error)
3382                 goto exit6;
3383         error = vfs_rename(old_dir->d_inode, old_dentry,
3384                                    new_dir->d_inode, new_dentry);
3385 exit6:
3386         mnt_drop_write(oldnd.path.mnt);
3387 exit5:
3388         dput(new_dentry);
3389 exit4:
3390         dput(old_dentry);
3391 exit3:
3392         unlock_rename(new_dir, old_dir);
3393 exit2:
3394         path_put(&newnd.path);
3395         putname(to);
3396 exit1:
3397         path_put(&oldnd.path);
3398         putname(from);
3399 exit:
3400         return error;
3401 }
3402
3403 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3404 {
3405         return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3406 }
3407
3408 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3409 {
3410         int len;
3411
3412         len = PTR_ERR(link);
3413         if (IS_ERR(link))
3414                 goto out;
3415
3416         len = strlen(link);
3417         if (len > (unsigned) buflen)
3418                 len = buflen;
3419         if (copy_to_user(buffer, link, len))
3420                 len = -EFAULT;
3421 out:
3422         return len;
3423 }
3424
3425 /*
3426  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
3427  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
3428  * using) it for any given inode is up to filesystem.
3429  */
3430 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3431 {
3432         struct nameidata nd;
3433         void *cookie;
3434         int res;
3435
3436         nd.depth = 0;
3437         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3438         if (IS_ERR(cookie))
3439                 return PTR_ERR(cookie);
3440
3441         res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3442         if (dentry->d_inode->i_op->put_link)
3443                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3444         return res;
3445 }
3446
3447 int vfs_follow_link(struct nameidata *nd, const char *link)
3448 {
3449         return __vfs_follow_link(nd, link);
3450 }
3451
3452 /* get the link contents into pagecache */
3453 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3454 {
3455         char *kaddr;
3456         struct page *page;
3457         struct address_space *mapping = dentry->d_inode->i_mapping;
3458         page = read_mapping_page(mapping, 0, NULL);
3459         if (IS_ERR(page))
3460                 return (char*)page;
3461         *ppage = page;
3462         kaddr = kmap(page);
3463         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3464         return kaddr;
3465 }
3466
3467 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3468 {
3469         struct page *page = NULL;
3470         char *s = page_getlink(dentry, &page);
3471         int res = vfs_readlink(dentry,buffer,buflen,s);
3472         if (page) {
3473                 kunmap(page);
3474                 page_cache_release(page);
3475         }
3476         return res;
3477 }
3478
3479 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3480 {
3481         struct page *page = NULL;
3482         nd_set_link(nd, page_getlink(dentry, &page));
3483         return page;
3484 }
3485
3486 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3487 {
3488         struct page *page = cookie;
3489
3490         if (page) {
3491                 kunmap(page);
3492                 page_cache_release(page);
3493         }
3494 }
3495
3496 /*
3497  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3498  */
3499 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3500 {
3501         struct address_space *mapping = inode->i_mapping;
3502         struct page *page;
3503         void *fsdata;
3504         int err;
3505         char *kaddr;
3506         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3507         if (nofs)
3508                 flags |= AOP_FLAG_NOFS;
3509
3510 retry:
3511         err = pagecache_write_begin(NULL, mapping, 0, len-1,
3512                                 flags, &page, &fsdata);
3513         if (err)
3514                 goto fail;
3515
3516         kaddr = kmap_atomic(page);
3517         memcpy(kaddr, symname, len-1);
3518         kunmap_atomic(kaddr);
3519
3520         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3521                                                         page, fsdata);
3522         if (err < 0)
3523                 goto fail;
3524         if (err < len-1)
3525                 goto retry;
3526
3527         mark_inode_dirty(inode);
3528         return 0;
3529 fail:
3530         return err;
3531 }
3532
3533 int page_symlink(struct inode *inode, const char *symname, int len)
3534 {
3535         return __page_symlink(inode, symname, len,
3536                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3537 }
3538
3539 const struct inode_operations page_symlink_inode_operations = {
3540         .readlink       = generic_readlink,
3541         .follow_link    = page_follow_link_light,
3542         .put_link       = page_put_link,
3543 };
3544
3545 EXPORT_SYMBOL(user_path_at);
3546 EXPORT_SYMBOL(follow_down_one);
3547 EXPORT_SYMBOL(follow_down);
3548 EXPORT_SYMBOL(follow_up);
3549 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3550 EXPORT_SYMBOL(getname);
3551 EXPORT_SYMBOL(lock_rename);
3552 EXPORT_SYMBOL(lookup_one_len);
3553 EXPORT_SYMBOL(page_follow_link_light);
3554 EXPORT_SYMBOL(page_put_link);
3555 EXPORT_SYMBOL(page_readlink);
3556 EXPORT_SYMBOL(__page_symlink);
3557 EXPORT_SYMBOL(page_symlink);
3558 EXPORT_SYMBOL(page_symlink_inode_operations);
3559 EXPORT_SYMBOL(kern_path);
3560 EXPORT_SYMBOL(vfs_path_lookup);
3561 EXPORT_SYMBOL(inode_permission);
3562 EXPORT_SYMBOL(unlock_rename);
3563 EXPORT_SYMBOL(vfs_create);
3564 EXPORT_SYMBOL(vfs_follow_link);
3565 EXPORT_SYMBOL(vfs_link);
3566 EXPORT_SYMBOL(vfs_mkdir);
3567 EXPORT_SYMBOL(vfs_mknod);
3568 EXPORT_SYMBOL(generic_permission);
3569 EXPORT_SYMBOL(vfs_readlink);
3570 EXPORT_SYMBOL(vfs_rename);
3571 EXPORT_SYMBOL(vfs_rmdir);
3572 EXPORT_SYMBOL(vfs_symlink);
3573 EXPORT_SYMBOL(vfs_unlink);
3574 EXPORT_SYMBOL(dentry_unhash);
3575 EXPORT_SYMBOL(generic_readlink);