fs/namei.c

   1 /*
   2  *  linux/fs/namei.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * Some corrections by tytso.
   9  */
  10
  11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12  * lookup logic.
  13  */
  14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15  */
  16
  17 #include <linux/init.h>
  18 #include <linux/module.h>
  19 #include <linux/slab.h>
  20 #include <linux/fs.h>
  21 #include <linux/namei.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/fsnotify.h>
  24 #include <linux/personality.h>
  25 #include <linux/security.h>
  26 #include <linux/ima.h>
  27 #include <linux/syscalls.h>
  28 #include <linux/mount.h>
  29 #include <linux/audit.h>
  30 #include <linux/capability.h>
  31 #include <linux/file.h>
  32 #include <linux/fcntl.h>
  33 #include <linux/device_cgroup.h>
  34 #include <linux/fs_struct.h>
  35 #include <linux/posix_acl.h>
  36 #include <asm/uaccess.h>
  37
  38 #include "internal.h"
  39
  40 /* [Feb-1997 T. Schoebel-Theuer]
  41  * Fundamental changes in the pathname lookup mechanisms (namei)
  42  * were necessary because of omirr.  The reason is that omirr needs
  43  * to know the _real_ pathname, not the user-supplied one, in case
  44  * of symlinks (and also when transname replacements occur).
  45  *
  46  * The new code replaces the old recursive symlink resolution with
  47  * an iterative one (in case of non-nested symlink chains).  It does
  48  * this with calls to <fs>_follow_link().
  49  * As a side effect, dir_namei(), _namei() and follow_link() are now
  50  * replaced with a single function lookup_dentry() that can handle all
  51  * the special cases of the former code.
  52  *
  53  * With the new dcache, the pathname is stored at each inode, at least as
  54  * long as the refcount of the inode is positive.  As a side effect, the
  55  * size of the dcache depends on the inode cache and thus is dynamic.
  56  *
  57  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  58  * resolution to correspond with current state of the code.
  59  *
  60  * Note that the symlink resolution is not *completely* iterative.
  61  * There is still a significant amount of tail- and mid- recursion in
  62  * the algorithm.  Also, note that <fs>_readlink() is not used in
  63  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  64  * may return different results than <fs>_follow_link().  Many virtual
  65  * filesystems (including /proc) exhibit this behavior.
  66  */
  67
  68 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  69  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  70  * and the name already exists in form of a symlink, try to create the new
  71  * name indicated by the symlink. The old code always complained that the
  72  * name already exists, due to not following the symlink even if its target
  73  * is nonexistent.  The new semantics affects also mknod() and link() when
  74  * the name is a symlink pointing to a non-existent name.
  75  *
  76  * I don't know which semantics is the right one, since I have no access
  77  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  78  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  79  * "old" one. Personally, I think the new semantics is much more logical.
  80  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  81  * file does succeed in both HP-UX and SunOs, but not in Solaris
  82  * and in the old Linux semantics.
  83  */
  84
  85 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  86  * semantics.  See the comments in "open_namei" and "do_link" below.
  87  *
  88  * [10-Sep-98 Alan Modra] Another symlink change.
  89  */
  90
  91 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  92  *      inside the path - always follow.
  93  *      in the last component in creation/removal/renaming - never follow.
  94  *      if LOOKUP_FOLLOW passed - follow.
  95  *      if the pathname has trailing slashes - follow.
  96  *      otherwise - don't follow.
  97  * (applied in that order).
  98  *
  99  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
 100  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
 101  * During the 2.4 we need to fix the userland stuff depending on it -
 102  * hopefully we will be able to get rid of that wart in 2.5. So far only
 103  * XEmacs seems to be relying on it...
 104  */
 105 /*
 106  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 107  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
 108  * any extra contention...
 109  */
 110
 111 /* In order to reduce some races, while at the same time doing additional
 112  * checking and hopefully speeding things up, we copy filenames to the
 113  * kernel data space before using them..
 114  *
 115  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 116  * PATH_MAX includes the nul terminator --RR.
 117  */
 118 static int do_getname(const char __user *filename, char *page)
 119 {
 120         int retval;
 121         unsigned long len = PATH_MAX;
 122
 123         if (!segment_eq(get_fs(), KERNEL_DS)) {
 124                 if ((unsigned long) filename >= TASK_SIZE)
 125                         return -EFAULT;
 126                 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
 127                         len = TASK_SIZE - (unsigned long) filename;
 128         }
 129
 130         retval = strncpy_from_user(page, filename, len);
 131         if (retval > 0) {
 132                 if (retval < len)
 133                         return 0;
 134                 return -ENAMETOOLONG;
 135         } else if (!retval)
 136                 retval = -ENOENT;
 137         return retval;
 138 }
 139
 140 static char *getname_flags(const char __user * filename, int flags)
 141 {
 142         char *tmp, *result;
 143
 144         result = ERR_PTR(-ENOMEM);
 145         tmp = __getname();
 146         if (tmp)  {
 147                 int retval = do_getname(filename, tmp);
 148
 149                 result = tmp;
 150                 if (retval < 0) {
 151                         if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
 152                                 __putname(tmp);
 153                                 result = ERR_PTR(retval);
 154                         }
 155                 }
 156         }
 157         audit_getname(result);
 158         return result;
 159 }
 160
 161 char *getname(const char __user * filename)
 162 {
 163         return getname_flags(filename, 0);
 164 }
 165
 166 #ifdef CONFIG_AUDITSYSCALL
 167 void putname(const char *name)
 168 {
 169         if (unlikely(!audit_dummy_context()))
 170                 audit_putname(name);
 171         else
 172                 __putname(name);
 173 }
 174 EXPORT_SYMBOL(putname);
 175 #endif
 176
 177 static int check_acl(struct inode *inode, int mask)
 178 {
 179         struct posix_acl *acl;
 180
 181         /*
 182          * Under RCU walk, we cannot even do a "get_cached_acl()",
 183          * because that involves locking and getting a refcount on
 184          * a cached ACL.
 185          *
 186          * So the only case we handle during RCU walking is the
 187          * case of a cached "no ACL at all", which needs no locks
 188          * or refcounts.
 189          */
 190         if (mask & MAY_NOT_BLOCK) {
 191                 if (negative_cached_acl(inode, ACL_TYPE_ACCESS))
 192                         return -EAGAIN;
 193                 return -ECHILD;
 194         }
 195
 196         acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
 197
 198         /*
 199          * A filesystem can force a ACL callback by just never
 200          * filling the ACL cache. But normally you'd fill the
 201          * cache either at inode instantiation time, or on the
 202          * first ->check_acl call.
 203          *
 204          * If the filesystem doesn't have a check_acl() function
 205          * at all, we'll just create the negative cache entry.
 206          */
 207         if (acl == ACL_NOT_CACHED) {
 208                 if (inode->i_op->check_acl)
 209                         return inode->i_op->check_acl(inode, mask);
 210
 211                 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
 212                 return -EAGAIN;
 213         }
 214
 215         if (acl) {
 216                 int error = posix_acl_permission(inode, acl, mask);
 217                 posix_acl_release(acl);
 218                 return error;
 219         }
 220
 221         return -EAGAIN;
 222 }
 223
 224 /*
 225  * This does basic POSIX ACL permission checking
 226  */
 227 static int acl_permission_check(struct inode *inode, int mask)
 228 {
 229         unsigned int mode = inode->i_mode;
 230
 231         mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
 232
 233         if (current_user_ns() != inode_userns(inode))
 234                 goto other_perms;
 235
 236         if (current_fsuid() == inode->i_uid)
 237                 mode >>= 6;
 238         else {
 239                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
 240                         int error = check_acl(inode, mask);
 241                         if (error != -EAGAIN)
 242                                 return error;
 243                 }
 244
 245                 if (in_group_p(inode->i_gid))
 246                         mode >>= 3;
 247         }
 248
 249 other_perms:
 250         /*
 251          * If the DACs are ok we don't need any capability check.
 252          */
 253         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
 254                 return 0;
 255         return -EACCES;
 256 }
 257
 258 /**
 259  * generic_permission -  check for access rights on a Posix-like filesystem
 260  * @inode:      inode to check access rights for
 261  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 262  *
 263  * Used to check for read/write/execute permissions on a file.
 264  * We use "fsuid" for this, letting us set arbitrary permissions
 265  * for filesystem access without changing the "normal" uids which
 266  * are used for other things.
 267  *
 268  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
 269  * request cannot be satisfied (eg. requires blocking or too much complexity).
 270  * It would then be called again in ref-walk mode.
 271  */
 272 int generic_permission(struct inode *inode, int mask)
 273 {
 274         int ret;
 275
 276         /*
 277          * Do the basic POSIX ACL permission checks.
 278          */
 279         ret = acl_permission_check(inode, mask);
 280         if (ret != -EACCES)
 281                 return ret;
 282
 283         if (S_ISDIR(inode->i_mode)) {
 284                 /* DACs are overridable for directories */
 285                 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
 286                         return 0;
 287                 if (!(mask & MAY_WRITE))
 288                         if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
 289                                 return 0;
 290                 return -EACCES;
 291         }
 292         /*
 293          * Read/write DACs are always overridable.
 294          * Executable DACs are overridable when there is
 295          * at least one exec bit set.
 296          */
 297         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
 298                 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
 299                         return 0;
 300
 301         /*
 302          * Searching includes executable on directories, else just read.
 303          */
 304         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
 305         if (mask == MAY_READ)
 306                 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
 307                         return 0;
 308
 309         return -EACCES;
 310 }
 311
 312 /**
 313  * inode_permission  -  check for access rights to a given inode
 314  * @inode:      inode to check permission on
 315  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 316  *
 317  * Used to check for read/write/execute permissions on an inode.
 318  * We use "fsuid" for this, letting us set arbitrary permissions
 319  * for filesystem access without changing the "normal" uids which
 320  * are used for other things.
 321  */
 322 int inode_permission(struct inode *inode, int mask)
 323 {
 324         int retval;
 325
 326         if (mask & MAY_WRITE) {
 327                 umode_t mode = inode->i_mode;
 328
 329                 /*
 330                  * Nobody gets write access to a read-only fs.
 331                  */
 332                 if (IS_RDONLY(inode) &&
 333                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 334                         return -EROFS;
 335
 336                 /*
 337                  * Nobody gets write access to an immutable file.
 338                  */
 339                 if (IS_IMMUTABLE(inode))
 340                         return -EACCES;
 341         }
 342
 343         if (inode->i_op->permission)
 344                 retval = inode->i_op->permission(inode, mask);
 345         else
 346                 retval = generic_permission(inode, mask);
 347
 348         if (retval)
 349                 return retval;
 350
 351         retval = devcgroup_inode_permission(inode, mask);
 352         if (retval)
 353                 return retval;
 354
 355         return security_inode_permission(inode, mask);
 356 }
 357
 358 /**
 359  * path_get - get a reference to a path
 360  * @path: path to get the reference to
 361  *
 362  * Given a path increment the reference count to the dentry and the vfsmount.
 363  */
 364 void path_get(struct path *path)
 365 {
 366         mntget(path->mnt);
 367         dget(path->dentry);
 368 }
 369 EXPORT_SYMBOL(path_get);
 370
 371 /**
 372  * path_put - put a reference to a path
 373  * @path: path to put the reference to
 374  *
 375  * Given a path decrement the reference count to the dentry and the vfsmount.
 376  */
 377 void path_put(struct path *path)
 378 {
 379         dput(path->dentry);
 380         mntput(path->mnt);
 381 }
 382 EXPORT_SYMBOL(path_put);
 383
 384 /*
 385  * Path walking has 2 modes, rcu-walk and ref-walk (see
 386  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
 387  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
 388  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
 389  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
 390  * got stuck, so ref-walk may continue from there. If this is not successful
 391  * (eg. a seqcount has changed), then failure is returned and it's up to caller
 392  * to restart the path walk from the beginning in ref-walk mode.
 393  */
 394
 395 /**
 396  * unlazy_walk - try to switch to ref-walk mode.
 397  * @nd: nameidata pathwalk data
 398  * @dentry: child of nd->path.dentry or NULL
 399  * Returns: 0 on success, -ECHILD on failure
 400  *
 401  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
 402  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
 403  * @nd or NULL.  Must be called from rcu-walk context.
 404  */
 405 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
 406 {
 407         struct fs_struct *fs = current->fs;
 408         struct dentry *parent = nd->path.dentry;
 409         int want_root = 0;
 410
 411         BUG_ON(!(nd->flags & LOOKUP_RCU));
 412         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
 413                 want_root = 1;
 414                 spin_lock(&fs->lock);
 415                 if (nd->root.mnt != fs->root.mnt ||
 416                                 nd->root.dentry != fs->root.dentry)
 417                         goto err_root;
 418         }
 419         spin_lock(&parent->d_lock);
 420         if (!dentry) {
 421                 if (!__d_rcu_to_refcount(parent, nd->seq))
 422                         goto err_parent;
 423                 BUG_ON(nd->inode != parent->d_inode);
 424         } else {
 425                 if (dentry->d_parent != parent)
 426                         goto err_parent;
 427                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
 428                 if (!__d_rcu_to_refcount(dentry, nd->seq))
 429                         goto err_child;
 430                 /*
 431                  * If the sequence check on the child dentry passed, then
 432                  * the child has not been removed from its parent. This
 433                  * means the parent dentry must be valid and able to take
 434                  * a reference at this point.
 435                  */
 436                 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
 437                 BUG_ON(!parent->d_count);
 438                 parent->d_count++;
 439                 spin_unlock(&dentry->d_lock);
 440         }
 441         spin_unlock(&parent->d_lock);
 442         if (want_root) {
 443                 path_get(&nd->root);
 444                 spin_unlock(&fs->lock);
 445         }
 446         mntget(nd->path.mnt);
 447
 448         rcu_read_unlock();
 449         br_read_unlock(vfsmount_lock);
 450         nd->flags &= ~LOOKUP_RCU;
 451         return 0;
 452
 453 err_child:
 454         spin_unlock(&dentry->d_lock);
 455 err_parent:
 456         spin_unlock(&parent->d_lock);
 457 err_root:
 458         if (want_root)
 459                 spin_unlock(&fs->lock);
 460         return -ECHILD;
 461 }
 462
 463 /**
 464  * release_open_intent - free up open intent resources
 465  * @nd: pointer to nameidata
 466  */
 467 void release_open_intent(struct nameidata *nd)
 468 {
 469         struct file *file = nd->intent.open.file;
 470
 471         if (file && !IS_ERR(file)) {
 472                 if (file->f_path.dentry == NULL)
 473                         put_filp(file);
 474                 else
 475                         fput(file);
 476         }
 477 }
 478
 479 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
 480 {
 481         return dentry->d_op->d_revalidate(dentry, nd);
 482 }
 483
 484 /**
 485  * complete_walk - successful completion of path walk
 486  * @nd:  pointer nameidata
 487  *
 488  * If we had been in RCU mode, drop out of it and legitimize nd->path.
 489  * Revalidate the final result, unless we'd already done that during
 490  * the path walk or the filesystem doesn't ask for it.  Return 0 on
 491  * success, -error on failure.  In case of failure caller does not
 492  * need to drop nd->path.
 493  */
 494 static int complete_walk(struct nameidata *nd)
 495 {
 496         struct dentry *dentry = nd->path.dentry;
 497         int status;
 498
 499         if (nd->flags & LOOKUP_RCU) {
 500                 nd->flags &= ~LOOKUP_RCU;
 501                 if (!(nd->flags & LOOKUP_ROOT))
 502                         nd->root.mnt = NULL;
 503                 spin_lock(&dentry->d_lock);
 504                 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
 505                         spin_unlock(&dentry->d_lock);
 506                         rcu_read_unlock();
 507                         br_read_unlock(vfsmount_lock);
 508                         return -ECHILD;
 509                 }
 510                 BUG_ON(nd->inode != dentry->d_inode);
 511                 spin_unlock(&dentry->d_lock);
 512                 mntget(nd->path.mnt);
 513                 rcu_read_unlock();
 514                 br_read_unlock(vfsmount_lock);
 515         }
 516
 517         if (likely(!(nd->flags & LOOKUP_JUMPED)))
 518                 return 0;
 519
 520         if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
 521                 return 0;
 522
 523         if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
 524                 return 0;
 525
 526         /* Note: we do not d_invalidate() */
 527         status = d_revalidate(dentry, nd);
 528         if (status > 0)
 529                 return 0;
 530
 531         if (!status)
 532                 status = -ESTALE;
 533
 534         path_put(&nd->path);
 535         return status;
 536 }
 537
 538 static __always_inline void set_root(struct nameidata *nd)
 539 {
 540         if (!nd->root.mnt)
 541                 get_fs_root(current->fs, &nd->root);
 542 }
 543
 544 static int link_path_walk(const char *, struct nameidata *);
 545
 546 static __always_inline void set_root_rcu(struct nameidata *nd)
 547 {
 548         if (!nd->root.mnt) {
 549                 struct fs_struct *fs = current->fs;
 550                 unsigned seq;
 551
 552                 do {
 553                         seq = read_seqcount_begin(&fs->seq);
 554                         nd->root = fs->root;
 555                         nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
 556                 } while (read_seqcount_retry(&fs->seq, seq));
 557         }
 558 }
 559
 560 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 561 {
 562         int ret;
 563
 564         if (IS_ERR(link))
 565                 goto fail;
 566
 567         if (*link == '/') {
 568                 set_root(nd);
 569                 path_put(&nd->path);
 570                 nd->path = nd->root;
 571                 path_get(&nd->root);
 572                 nd->flags |= LOOKUP_JUMPED;
 573         }
 574         nd->inode = nd->path.dentry->d_inode;
 575
 576         ret = link_path_walk(link, nd);
 577         return ret;
 578 fail:
 579         path_put(&nd->path);
 580         return PTR_ERR(link);
 581 }
 582
 583 static void path_put_conditional(struct path *path, struct nameidata *nd)
 584 {
 585         dput(path->dentry);
 586         if (path->mnt != nd->path.mnt)
 587                 mntput(path->mnt);
 588 }
 589
 590 static inline void path_to_nameidata(const struct path *path,
 591                                         struct nameidata *nd)
 592 {
 593         if (!(nd->flags & LOOKUP_RCU)) {
 594                 dput(nd->path.dentry);
 595                 if (nd->path.mnt != path->mnt)
 596                         mntput(nd->path.mnt);
 597         }
 598         nd->path.mnt = path->mnt;
 599         nd->path.dentry = path->dentry;
 600 }
 601
 602 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
 603 {
 604         struct inode *inode = link->dentry->d_inode;
 605         if (!IS_ERR(cookie) && inode->i_op->put_link)
 606                 inode->i_op->put_link(link->dentry, nd, cookie);
 607         path_put(link);
 608 }
 609
 610 static __always_inline int
 611 follow_link(struct path *link, struct nameidata *nd, void **p)
 612 {
 613         int error;
 614         struct dentry *dentry = link->dentry;
 615
 616         BUG_ON(nd->flags & LOOKUP_RCU);
 617
 618         if (link->mnt == nd->path.mnt)
 619                 mntget(link->mnt);
 620
 621         if (unlikely(current->total_link_count >= 40)) {
 622                 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
 623                 path_put(&nd->path);
 624                 return -ELOOP;
 625         }
 626         cond_resched();
 627         current->total_link_count++;
 628
 629         touch_atime(link->mnt, dentry);
 630         nd_set_link(nd, NULL);
 631
 632         error = security_inode_follow_link(link->dentry, nd);
 633         if (error) {
 634                 *p = ERR_PTR(error); /* no ->put_link(), please */
 635                 path_put(&nd->path);
 636                 return error;
 637         }
 638
 639         nd->last_type = LAST_BIND;
 640         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
 641         error = PTR_ERR(*p);
 642         if (!IS_ERR(*p)) {
 643                 char *s = nd_get_link(nd);
 644                 error = 0;
 645                 if (s)
 646                         error = __vfs_follow_link(nd, s);
 647                 else if (nd->last_type == LAST_BIND) {
 648                         nd->flags |= LOOKUP_JUMPED;
 649                         nd->inode = nd->path.dentry->d_inode;
 650                         if (nd->inode->i_op->follow_link) {
 651                                 /* stepped on a _really_ weird one */
 652                                 path_put(&nd->path);
 653                                 error = -ELOOP;
 654                         }
 655                 }
 656         }
 657         return error;
 658 }
 659
 660 static int follow_up_rcu(struct path *path)
 661 {
 662         struct vfsmount *parent;
 663         struct dentry *mountpoint;
 664
 665         parent = path->mnt->mnt_parent;
 666         if (parent == path->mnt)
 667                 return 0;
 668         mountpoint = path->mnt->mnt_mountpoint;
 669         path->dentry = mountpoint;
 670         path->mnt = parent;
 671         return 1;
 672 }
 673
 674 int follow_up(struct path *path)
 675 {
 676         struct vfsmount *parent;
 677         struct dentry *mountpoint;
 678
 679         br_read_lock(vfsmount_lock);
 680         parent = path->mnt->mnt_parent;
 681         if (parent == path->mnt) {
 682                 br_read_unlock(vfsmount_lock);
 683                 return 0;
 684         }
 685         mntget(parent);
 686         mountpoint = dget(path->mnt->mnt_mountpoint);
 687         br_read_unlock(vfsmount_lock);
 688         dput(path->dentry);
 689         path->dentry = mountpoint;
 690         mntput(path->mnt);
 691         path->mnt = parent;
 692         return 1;
 693 }
 694
 695 /*
 696  * Perform an automount
 697  * - return -EISDIR to tell follow_managed() to stop and return the path we
 698  *   were called with.
 699  */
 700 static int follow_automount(struct path *path, unsigned flags,
 701                             bool *need_mntput)
 702 {
 703         struct vfsmount *mnt;
 704         int err;
 705
 706         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
 707                 return -EREMOTE;
 708
 709         /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
 710          * and this is the terminal part of the path.
 711          */
 712         if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
 713                 return -EISDIR; /* we actually want to stop here */
 714
 715         /* We want to mount if someone is trying to open/create a file of any
 716          * type under the mountpoint, wants to traverse through the mountpoint
 717          * or wants to open the mounted directory.
 718          *
 719          * We don't want to mount if someone's just doing a stat and they've
 720          * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
 721          * appended a '/' to the name.
 722          */
 723         if (!(flags & LOOKUP_FOLLOW) &&
 724             !(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
 725                        LOOKUP_OPEN | LOOKUP_CREATE)))
 726                 return -EISDIR;
 727
 728         current->total_link_count++;
 729         if (current->total_link_count >= 40)
 730                 return -ELOOP;
 731
 732         mnt = path->dentry->d_op->d_automount(path);
 733         if (IS_ERR(mnt)) {
 734                 /*
 735                  * The filesystem is allowed to return -EISDIR here to indicate
 736                  * it doesn't want to automount.  For instance, autofs would do
 737                  * this so that its userspace daemon can mount on this dentry.
 738                  *
 739                  * However, we can only permit this if it's a terminal point in
 740                  * the path being looked up; if it wasn't then the remainder of
 741                  * the path is inaccessible and we should say so.
 742                  */
 743                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
 744                         return -EREMOTE;
 745                 return PTR_ERR(mnt);
 746         }
 747
 748         if (!mnt) /* mount collision */
 749                 return 0;
 750
 751         if (!*need_mntput) {
 752                 /* lock_mount() may release path->mnt on error */
 753                 mntget(path->mnt);
 754                 *need_mntput = true;
 755         }
 756         err = finish_automount(mnt, path);
 757
 758         switch (err) {
 759         case -EBUSY:
 760                 /* Someone else made a mount here whilst we were busy */
 761                 return 0;
 762         case 0:
 763                 path_put(path);
 764                 path->mnt = mnt;
 765                 path->dentry = dget(mnt->mnt_root);
 766                 return 0;
 767         default:
 768                 return err;
 769         }
 770
 771 }
 772
 773 /*
 774  * Handle a dentry that is managed in some way.
 775  * - Flagged for transit management (autofs)
 776  * - Flagged as mountpoint
 777  * - Flagged as automount point
 778  *
 779  * This may only be called in refwalk mode.
 780  *
 781  * Serialization is taken care of in namespace.c
 782  */
 783 static int follow_managed(struct path *path, unsigned flags)
 784 {
 785         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
 786         unsigned managed;
 787         bool need_mntput = false;
 788         int ret = 0;
 789
 790         /* Given that we're not holding a lock here, we retain the value in a
 791          * local variable for each dentry as we look at it so that we don't see
 792          * the components of that value change under us */
 793         while (managed = ACCESS_ONCE(path->dentry->d_flags),
 794                managed &= DCACHE_MANAGED_DENTRY,
 795                unlikely(managed != 0)) {
 796                 /* Allow the filesystem to manage the transit without i_mutex
 797                  * being held. */
 798                 if (managed & DCACHE_MANAGE_TRANSIT) {
 799                         BUG_ON(!path->dentry->d_op);
 800                         BUG_ON(!path->dentry->d_op->d_manage);
 801                         ret = path->dentry->d_op->d_manage(path->dentry, false);
 802                         if (ret < 0)
 803                                 break;
 804                 }
 805
 806                 /* Transit to a mounted filesystem. */
 807                 if (managed & DCACHE_MOUNTED) {
 808                         struct vfsmount *mounted = lookup_mnt(path);
 809                         if (mounted) {
 810                                 dput(path->dentry);
 811                                 if (need_mntput)
 812                                         mntput(path->mnt);
 813                                 path->mnt = mounted;
 814                                 path->dentry = dget(mounted->mnt_root);
 815                                 need_mntput = true;
 816                                 continue;
 817                         }
 818
 819                         /* Something is mounted on this dentry in another
 820                          * namespace and/or whatever was mounted there in this
 821                          * namespace got unmounted before we managed to get the
 822                          * vfsmount_lock */
 823                 }
 824
 825                 /* Handle an automount point */
 826                 if (managed & DCACHE_NEED_AUTOMOUNT) {
 827                         ret = follow_automount(path, flags, &need_mntput);
 828                         if (ret < 0)
 829                                 break;
 830                         continue;
 831                 }
 832
 833                 /* We didn't change the current path point */
 834                 break;
 835         }
 836
 837         if (need_mntput && path->mnt == mnt)
 838                 mntput(path->mnt);
 839         if (ret == -EISDIR)
 840                 ret = 0;
 841         return ret;
 842 }
 843
 844 int follow_down_one(struct path *path)
 845 {
 846         struct vfsmount *mounted;
 847
 848         mounted = lookup_mnt(path);
 849         if (mounted) {
 850                 dput(path->dentry);
 851                 mntput(path->mnt);
 852                 path->mnt = mounted;
 853                 path->dentry = dget(mounted->mnt_root);
 854                 return 1;
 855         }
 856         return 0;
 857 }
 858
 859 static inline bool managed_dentry_might_block(struct dentry *dentry)
 860 {
 861         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
 862                 dentry->d_op->d_manage(dentry, true) < 0);
 863 }
 864
 865 /*
 866  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
 867  * we meet a managed dentry that would need blocking.
 868  */
 869 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
 870                                struct inode **inode)
 871 {
 872         for (;;) {
 873                 struct vfsmount *mounted;
 874                 /*
 875                  * Don't forget we might have a non-mountpoint managed dentry
 876                  * that wants to block transit.
 877                  */
 878                 if (unlikely(managed_dentry_might_block(path->dentry)))
 879                         return false;
 880
 881                 if (!d_mountpoint(path->dentry))
 882                         break;
 883
 884                 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
 885                 if (!mounted)
 886                         break;
 887                 path->mnt = mounted;
 888                 path->dentry = mounted->mnt_root;
 889                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
 890                 /*
 891                  * Update the inode too. We don't need to re-check the
 892                  * dentry sequence number here after this d_inode read,
 893                  * because a mount-point is always pinned.
 894                  */
 895                 *inode = path->dentry->d_inode;
 896         }
 897         return true;
 898 }
 899
 900 static void follow_mount_rcu(struct nameidata *nd)
 901 {
 902         while (d_mountpoint(nd->path.dentry)) {
 903                 struct vfsmount *mounted;
 904                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
 905                 if (!mounted)
 906                         break;
 907                 nd->path.mnt = mounted;
 908                 nd->path.dentry = mounted->mnt_root;
 909                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
 910         }
 911 }
 912
 913 static int follow_dotdot_rcu(struct nameidata *nd)
 914 {
 915         set_root_rcu(nd);
 916
 917         while (1) {
 918                 if (nd->path.dentry == nd->root.dentry &&
 919                     nd->path.mnt == nd->root.mnt) {
 920                         break;
 921                 }
 922                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
 923                         struct dentry *old = nd->path.dentry;
 924                         struct dentry *parent = old->d_parent;
 925                         unsigned seq;
 926
 927                         seq = read_seqcount_begin(&parent->d_seq);
 928                         if (read_seqcount_retry(&old->d_seq, nd->seq))
 929                                 goto failed;
 930                         nd->path.dentry = parent;
 931                         nd->seq = seq;
 932                         break;
 933                 }
 934                 if (!follow_up_rcu(&nd->path))
 935                         break;
 936                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
 937         }
 938         follow_mount_rcu(nd);
 939         nd->inode = nd->path.dentry->d_inode;
 940         return 0;
 941
 942 failed:
 943         nd->flags &= ~LOOKUP_RCU;
 944         if (!(nd->flags & LOOKUP_ROOT))
 945                 nd->root.mnt = NULL;
 946         rcu_read_unlock();
 947         br_read_unlock(vfsmount_lock);
 948         return -ECHILD;
 949 }
 950
 951 /*
 952  * Follow down to the covering mount currently visible to userspace.  At each
 953  * point, the filesystem owning that dentry may be queried as to whether the
 954  * caller is permitted to proceed or not.
 955  */
 956 int follow_down(struct path *path)
 957 {
 958         unsigned managed;
 959         int ret;
 960
 961         while (managed = ACCESS_ONCE(path->dentry->d_flags),
 962                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
 963                 /* Allow the filesystem to manage the transit without i_mutex
 964                  * being held.
 965                  *
 966                  * We indicate to the filesystem if someone is trying to mount
 967                  * something here.  This gives autofs the chance to deny anyone
 968                  * other than its daemon the right to mount on its
 969                  * superstructure.
 970                  *
 971                  * The filesystem may sleep at this point.
 972                  */
 973                 if (managed & DCACHE_MANAGE_TRANSIT) {
 974                         BUG_ON(!path->dentry->d_op);
 975                         BUG_ON(!path->dentry->d_op->d_manage);
 976                         ret = path->dentry->d_op->d_manage(
 977                                 path->dentry, false);
 978                         if (ret < 0)
 979                                 return ret == -EISDIR ? 0 : ret;
 980                 }
 981
 982                 /* Transit to a mounted filesystem. */
 983                 if (managed & DCACHE_MOUNTED) {
 984                         struct vfsmount *mounted = lookup_mnt(path);
 985                         if (!mounted)
 986                                 break;
 987                         dput(path->dentry);
 988                         mntput(path->mnt);
 989                         path->mnt = mounted;
 990                         path->dentry = dget(mounted->mnt_root);
 991                         continue;
 992                 }
 993
 994                 /* Don't handle automount points here */
 995                 break;
 996         }
 997         return 0;
 998 }
 999
1000 /*
1001  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1002  */
1003 static void follow_mount(struct path *path)
1004 {
1005         while (d_mountpoint(path->dentry)) {
1006                 struct vfsmount *mounted = lookup_mnt(path);
1007                 if (!mounted)
1008                         break;
1009                 dput(path->dentry);
1010                 mntput(path->mnt);
1011                 path->mnt = mounted;
1012                 path->dentry = dget(mounted->mnt_root);
1013         }
1014 }
1015
1016 static void follow_dotdot(struct nameidata *nd)
1017 {
1018         set_root(nd);
1019
1020         while(1) {
1021                 struct dentry *old = nd->path.dentry;
1022
1023                 if (nd->path.dentry == nd->root.dentry &&
1024                     nd->path.mnt == nd->root.mnt) {
1025                         break;
1026                 }
1027                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1028                         /* rare case of legitimate dget_parent()... */
1029                         nd->path.dentry = dget_parent(nd->path.dentry);
1030                         dput(old);
1031                         break;
1032                 }
1033                 if (!follow_up(&nd->path))
1034                         break;
1035         }
1036         follow_mount(&nd->path);
1037         nd->inode = nd->path.dentry->d_inode;
1038 }
1039
1040 /*
1041  * Allocate a dentry with name and parent, and perform a parent
1042  * directory ->lookup on it. Returns the new dentry, or ERR_PTR
1043  * on error. parent->d_inode->i_mutex must be held. d_lookup must
1044  * have verified that no child exists while under i_mutex.
1045  */
1046 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1047                                 struct qstr *name, struct nameidata *nd)
1048 {
1049         struct inode *inode = parent->d_inode;
1050         struct dentry *dentry;
1051         struct dentry *old;
1052
1053         /* Don't create child dentry for a dead directory. */
1054         if (unlikely(IS_DEADDIR(inode)))
1055                 return ERR_PTR(-ENOENT);
1056
1057         dentry = d_alloc(parent, name);
1058         if (unlikely(!dentry))
1059                 return ERR_PTR(-ENOMEM);
1060
1061         old = inode->i_op->lookup(inode, dentry, nd);
1062         if (unlikely(old)) {
1063                 dput(dentry);
1064                 dentry = old;
1065         }
1066         return dentry;
1067 }
1068
1069 /*
1070  * We already have a dentry, but require a lookup to be performed on the parent
1071  * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1072  * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1073  * child exists while under i_mutex.
1074  */
1075 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1076                                      struct nameidata *nd)
1077 {
1078         struct inode *inode = parent->d_inode;
1079         struct dentry *old;
1080
1081         /* Don't create child dentry for a dead directory. */
1082         if (unlikely(IS_DEADDIR(inode)))
1083                 return ERR_PTR(-ENOENT);
1084
1085         old = inode->i_op->lookup(inode, dentry, nd);
1086         if (unlikely(old)) {
1087                 dput(dentry);
1088                 dentry = old;
1089         }
1090         return dentry;
1091 }
1092
1093 /*
1094  *  It's more convoluted than I'd like it to be, but... it's still fairly
1095  *  small and for now I'd prefer to have fast path as straight as possible.
1096  *  It _is_ time-critical.
1097  */
1098 static int do_lookup(struct nameidata *nd, struct qstr *name,
1099                         struct path *path, struct inode **inode)
1100 {
1101         struct vfsmount *mnt = nd->path.mnt;
1102         struct dentry *dentry, *parent = nd->path.dentry;
1103         int need_reval = 1;
1104         int status = 1;
1105         int err;
1106
1107         /*
1108          * Rename seqlock is not required here because in the off chance
1109          * of a false negative due to a concurrent rename, we're going to
1110          * do the non-racy lookup, below.
1111          */
1112         if (nd->flags & LOOKUP_RCU) {
1113                 unsigned seq;
1114                 *inode = nd->inode;
1115                 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1116                 if (!dentry)
1117                         goto unlazy;
1118
1119                 /* Memory barrier in read_seqcount_begin of child is enough */
1120                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1121                         return -ECHILD;
1122                 nd->seq = seq;
1123
1124                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1125                         status = d_revalidate(dentry, nd);
1126                         if (unlikely(status <= 0)) {
1127                                 if (status != -ECHILD)
1128                                         need_reval = 0;
1129                                 goto unlazy;
1130                         }
1131                 }
1132                 if (unlikely(d_need_lookup(dentry)))
1133                         goto unlazy;
1134                 path->mnt = mnt;
1135                 path->dentry = dentry;
1136                 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1137                         goto unlazy;
1138                 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1139                         goto unlazy;
1140                 return 0;
1141 unlazy:
1142                 if (unlazy_walk(nd, dentry))
1143                         return -ECHILD;
1144         } else {
1145                 dentry = __d_lookup(parent, name);
1146         }
1147
1148         if (dentry && unlikely(d_need_lookup(dentry))) {
1149                 dput(dentry);
1150                 dentry = NULL;
1151         }
1152 retry:
1153         if (unlikely(!dentry)) {
1154                 struct inode *dir = parent->d_inode;
1155                 BUG_ON(nd->inode != dir);
1156
1157                 mutex_lock(&dir->i_mutex);
1158                 dentry = d_lookup(parent, name);
1159                 if (likely(!dentry)) {
1160                         dentry = d_alloc_and_lookup(parent, name, nd);
1161                         if (IS_ERR(dentry)) {
1162                                 mutex_unlock(&dir->i_mutex);
1163                                 return PTR_ERR(dentry);
1164                         }
1165                         /* known good */
1166                         need_reval = 0;
1167                         status = 1;
1168                 } else if (unlikely(d_need_lookup(dentry))) {
1169                         dentry = d_inode_lookup(parent, dentry, nd);
1170                         if (IS_ERR(dentry)) {
1171                                 mutex_unlock(&dir->i_mutex);
1172                                 return PTR_ERR(dentry);
1173                         }
1174                         /* known good */
1175                         need_reval = 0;
1176                         status = 1;
1177                 }
1178                 mutex_unlock(&dir->i_mutex);
1179         }
1180         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1181                 status = d_revalidate(dentry, nd);
1182         if (unlikely(status <= 0)) {
1183                 if (status < 0) {
1184                         dput(dentry);
1185                         return status;
1186                 }
1187                 if (!d_invalidate(dentry)) {
1188                         dput(dentry);
1189                         dentry = NULL;
1190                         need_reval = 1;
1191                         goto retry;
1192                 }
1193         }
1194
1195         path->mnt = mnt;
1196         path->dentry = dentry;
1197         err = follow_managed(path, nd->flags);
1198         if (unlikely(err < 0)) {
1199                 path_put_conditional(path, nd);
1200                 return err;
1201         }
1202         *inode = path->dentry->d_inode;
1203         return 0;
1204 }
1205
1206 static inline int may_lookup(struct nameidata *nd)
1207 {
1208         if (nd->flags & LOOKUP_RCU) {
1209                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1210                 if (err != -ECHILD)
1211                         return err;
1212                 if (unlazy_walk(nd, NULL))
1213                         return -ECHILD;
1214         }
1215         return inode_permission(nd->inode, MAY_EXEC);
1216 }
1217
1218 static inline int handle_dots(struct nameidata *nd, int type)
1219 {
1220         if (type == LAST_DOTDOT) {
1221                 if (nd->flags & LOOKUP_RCU) {
1222                         if (follow_dotdot_rcu(nd))
1223                                 return -ECHILD;
1224                 } else
1225                         follow_dotdot(nd);
1226         }
1227         return 0;
1228 }
1229
1230 static void terminate_walk(struct nameidata *nd)
1231 {
1232         if (!(nd->flags & LOOKUP_RCU)) {
1233                 path_put(&nd->path);
1234         } else {
1235                 nd->flags &= ~LOOKUP_RCU;
1236                 if (!(nd->flags & LOOKUP_ROOT))
1237                         nd->root.mnt = NULL;
1238                 rcu_read_unlock();
1239                 br_read_unlock(vfsmount_lock);
1240         }
1241 }
1242
1243 static inline int walk_component(struct nameidata *nd, struct path *path,
1244                 struct qstr *name, int type, int follow)
1245 {
1246         struct inode *inode;
1247         int err;
1248         /*
1249          * "." and ".." are special - ".." especially so because it has
1250          * to be able to know about the current root directory and
1251          * parent relationships.
1252          */
1253         if (unlikely(type != LAST_NORM))
1254                 return handle_dots(nd, type);
1255         err = do_lookup(nd, name, path, &inode);
1256         if (unlikely(err)) {
1257                 terminate_walk(nd);
1258                 return err;
1259         }
1260         if (!inode) {
1261                 path_to_nameidata(path, nd);
1262                 terminate_walk(nd);
1263                 return -ENOENT;
1264         }
1265         if (unlikely(inode->i_op->follow_link) && follow) {
1266                 if (nd->flags & LOOKUP_RCU) {
1267                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1268                                 terminate_walk(nd);
1269                                 return -ECHILD;
1270                         }
1271                 }
1272                 BUG_ON(inode != path->dentry->d_inode);
1273                 return 1;
1274         }
1275         path_to_nameidata(path, nd);
1276         nd->inode = inode;
1277         return 0;
1278 }
1279
1280 /*
1281  * This limits recursive symlink follows to 8, while
1282  * limiting consecutive symlinks to 40.
1283  *
1284  * Without that kind of total limit, nasty chains of consecutive
1285  * symlinks can cause almost arbitrarily long lookups.
1286  */
1287 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1288 {
1289         int res;
1290
1291         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1292                 path_put_conditional(path, nd);
1293                 path_put(&nd->path);
1294                 return -ELOOP;
1295         }
1296         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1297
1298         nd->depth++;
1299         current->link_count++;
1300
1301         do {
1302                 struct path link = *path;
1303                 void *cookie;
1304
1305                 res = follow_link(&link, nd, &cookie);
1306                 if (!res)
1307                         res = walk_component(nd, path, &nd->last,
1308                                              nd->last_type, LOOKUP_FOLLOW);
1309                 put_link(nd, &link, cookie);
1310         } while (res > 0);
1311
1312         current->link_count--;
1313         nd->depth--;
1314         return res;
1315 }
1316
1317 /*
1318  * Name resolution.
1319  * This is the basic name resolution function, turning a pathname into
1320  * the final dentry. We expect 'base' to be positive and a directory.
1321  *
1322  * Returns 0 and nd will have valid dentry and mnt on success.
1323  * Returns error and drops reference to input namei data on failure.
1324  */
1325 static int link_path_walk(const char *name, struct nameidata *nd)
1326 {
1327         struct path next;
1328         int err;
1329
1330         while (*name=='/')
1331                 name++;
1332         if (!*name)
1333                 return 0;
1334
1335         /* At this point we know we have a real path component. */
1336         for(;;) {
1337                 unsigned long hash;
1338                 struct qstr this;
1339                 unsigned int c;
1340                 int type;
1341
1342                 err = may_lookup(nd);
1343                 if (err)
1344                         break;
1345
1346                 this.name = name;
1347                 c = *(const unsigned char *)name;
1348
1349                 hash = init_name_hash();
1350                 do {
1351                         name++;
1352                         hash = partial_name_hash(c, hash);
1353                         c = *(const unsigned char *)name;
1354                 } while (c && (c != '/'));
1355                 this.len = name - (const char *) this.name;
1356                 this.hash = end_name_hash(hash);
1357
1358                 type = LAST_NORM;
1359                 if (this.name[0] == '.') switch (this.len) {
1360                         case 2:
1361                                 if (this.name[1] == '.') {
1362                                         type = LAST_DOTDOT;
1363                                         nd->flags |= LOOKUP_JUMPED;
1364                                 }
1365                                 break;
1366                         case 1:
1367                                 type = LAST_DOT;
1368                 }
1369                 if (likely(type == LAST_NORM)) {
1370                         struct dentry *parent = nd->path.dentry;
1371                         nd->flags &= ~LOOKUP_JUMPED;
1372                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1373                                 err = parent->d_op->d_hash(parent, nd->inode,
1374                                                            &this);
1375                                 if (err < 0)
1376                                         break;
1377                         }
1378                 }
1379
1380                 /* remove trailing slashes? */
1381                 if (!c)
1382                         goto last_component;
1383                 while (*++name == '/');
1384                 if (!*name)
1385                         goto last_component;
1386
1387                 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1388                 if (err < 0)
1389                         return err;
1390
1391                 if (err) {
1392                         err = nested_symlink(&next, nd);
1393                         if (err)
1394                                 return err;
1395                 }
1396                 err = -ENOTDIR;
1397                 if (!nd->inode->i_op->lookup)
1398                         break;
1399                 continue;
1400                 /* here ends the main loop */
1401
1402 last_component:
1403                 nd->last = this;
1404                 nd->last_type = type;
1405                 return 0;
1406         }
1407         terminate_walk(nd);
1408         return err;
1409 }
1410
1411 static int path_init(int dfd, const char *name, unsigned int flags,
1412                      struct nameidata *nd, struct file **fp)
1413 {
1414         int retval = 0;
1415         int fput_needed;
1416         struct file *file;
1417
1418         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1419         nd->flags = flags | LOOKUP_JUMPED;
1420         nd->depth = 0;
1421         if (flags & LOOKUP_ROOT) {
1422                 struct inode *inode = nd->root.dentry->d_inode;
1423                 if (*name) {
1424                         if (!inode->i_op->lookup)
1425                                 return -ENOTDIR;
1426                         retval = inode_permission(inode, MAY_EXEC);
1427                         if (retval)
1428                                 return retval;
1429                 }
1430                 nd->path = nd->root;
1431                 nd->inode = inode;
1432                 if (flags & LOOKUP_RCU) {
1433                         br_read_lock(vfsmount_lock);
1434                         rcu_read_lock();
1435                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1436                 } else {
1437                         path_get(&nd->path);
1438                 }
1439                 return 0;
1440         }
1441
1442         nd->root.mnt = NULL;
1443
1444         if (*name=='/') {
1445                 if (flags & LOOKUP_RCU) {
1446                         br_read_lock(vfsmount_lock);
1447                         rcu_read_lock();
1448                         set_root_rcu(nd);
1449                 } else {
1450                         set_root(nd);
1451                         path_get(&nd->root);
1452                 }
1453                 nd->path = nd->root;
1454         } else if (dfd == AT_FDCWD) {
1455                 if (flags & LOOKUP_RCU) {
1456                         struct fs_struct *fs = current->fs;
1457                         unsigned seq;
1458
1459                         br_read_lock(vfsmount_lock);
1460                         rcu_read_lock();
1461
1462                         do {
1463                                 seq = read_seqcount_begin(&fs->seq);
1464                                 nd->path = fs->pwd;
1465                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1466                         } while (read_seqcount_retry(&fs->seq, seq));
1467                 } else {
1468                         get_fs_pwd(current->fs, &nd->path);
1469                 }
1470         } else {
1471                 struct dentry *dentry;
1472
1473                 file = fget_raw_light(dfd, &fput_needed);
1474                 retval = -EBADF;
1475                 if (!file)
1476                         goto out_fail;
1477
1478                 dentry = file->f_path.dentry;
1479
1480                 if (*name) {
1481                         retval = -ENOTDIR;
1482                         if (!S_ISDIR(dentry->d_inode->i_mode))
1483                                 goto fput_fail;
1484
1485                         retval = inode_permission(dentry->d_inode, MAY_EXEC);
1486                         if (retval)
1487                                 goto fput_fail;
1488                 }
1489
1490                 nd->path = file->f_path;
1491                 if (flags & LOOKUP_RCU) {
1492                         if (fput_needed)
1493                                 *fp = file;
1494                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1495                         br_read_lock(vfsmount_lock);
1496                         rcu_read_lock();
1497                 } else {
1498                         path_get(&file->f_path);
1499                         fput_light(file, fput_needed);
1500                 }
1501         }
1502
1503         nd->inode = nd->path.dentry->d_inode;
1504         return 0;
1505
1506 fput_fail:
1507         fput_light(file, fput_needed);
1508 out_fail:
1509         return retval;
1510 }
1511
1512 static inline int lookup_last(struct nameidata *nd, struct path *path)
1513 {
1514         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1515                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1516
1517         nd->flags &= ~LOOKUP_PARENT;
1518         return walk_component(nd, path, &nd->last, nd->last_type,
1519                                         nd->flags & LOOKUP_FOLLOW);
1520 }
1521
1522 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1523 static int path_lookupat(int dfd, const char *name,
1524                                 unsigned int flags, struct nameidata *nd)
1525 {
1526         struct file *base = NULL;
1527         struct path path;
1528         int err;
1529
1530         /*
1531          * Path walking is largely split up into 2 different synchronisation
1532          * schemes, rcu-walk and ref-walk (explained in
1533          * Documentation/filesystems/path-lookup.txt). These share much of the
1534          * path walk code, but some things particularly setup, cleanup, and
1535          * following mounts are sufficiently divergent that functions are
1536          * duplicated. Typically there is a function foo(), and its RCU
1537          * analogue, foo_rcu().
1538          *
1539          * -ECHILD is the error number of choice (just to avoid clashes) that
1540          * is returned if some aspect of an rcu-walk fails. Such an error must
1541          * be handled by restarting a traditional ref-walk (which will always
1542          * be able to complete).
1543          */
1544         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1545
1546         if (unlikely(err))
1547                 return err;
1548
1549         current->total_link_count = 0;
1550         err = link_path_walk(name, nd);
1551
1552         if (!err && !(flags & LOOKUP_PARENT)) {
1553                 err = lookup_last(nd, &path);
1554                 while (err > 0) {
1555                         void *cookie;
1556                         struct path link = path;
1557                         nd->flags |= LOOKUP_PARENT;
1558                         err = follow_link(&link, nd, &cookie);
1559                         if (!err)
1560                                 err = lookup_last(nd, &path);
1561                         put_link(nd, &link, cookie);
1562                 }
1563         }
1564
1565         if (!err)
1566                 err = complete_walk(nd);
1567
1568         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1569                 if (!nd->inode->i_op->lookup) {
1570                         path_put(&nd->path);
1571                         err = -ENOTDIR;
1572                 }
1573         }
1574
1575         if (base)
1576                 fput(base);
1577
1578         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1579                 path_put(&nd->root);
1580                 nd->root.mnt = NULL;
1581         }
1582         return err;
1583 }
1584
1585 static int do_path_lookup(int dfd, const char *name,
1586                                 unsigned int flags, struct nameidata *nd)
1587 {
1588         int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1589         if (unlikely(retval == -ECHILD))
1590                 retval = path_lookupat(dfd, name, flags, nd);
1591         if (unlikely(retval == -ESTALE))
1592                 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1593
1594         if (likely(!retval)) {
1595                 if (unlikely(!audit_dummy_context())) {
1596                         if (nd->path.dentry && nd->inode)
1597                                 audit_inode(name, nd->path.dentry);
1598                 }
1599         }
1600         return retval;
1601 }
1602
1603 int kern_path_parent(const char *name, struct nameidata *nd)
1604 {
1605         return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1606 }
1607
1608 int kern_path(const char *name, unsigned int flags, struct path *path)
1609 {
1610         struct nameidata nd;
1611         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1612         if (!res)
1613                 *path = nd.path;
1614         return res;
1615 }
1616
1617 /**
1618  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1619  * @dentry:  pointer to dentry of the base directory
1620  * @mnt: pointer to vfs mount of the base directory
1621  * @name: pointer to file name
1622  * @flags: lookup flags
1623  * @path: pointer to struct path to fill
1624  */
1625 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1626                     const char *name, unsigned int flags,
1627                     struct path *path)
1628 {
1629         struct nameidata nd;
1630         int err;
1631         nd.root.dentry = dentry;
1632         nd.root.mnt = mnt;
1633         BUG_ON(flags & LOOKUP_PARENT);
1634         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1635         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
1636         if (!err)
1637                 *path = nd.path;
1638         return err;
1639 }
1640
1641 static struct dentry *__lookup_hash(struct qstr *name,
1642                 struct dentry *base, struct nameidata *nd)
1643 {
1644         struct inode *inode = base->d_inode;
1645         struct dentry *dentry;
1646         int err;
1647
1648         err = inode_permission(inode, MAY_EXEC);
1649         if (err)
1650                 return ERR_PTR(err);
1651
1652         /*
1653          * Don't bother with __d_lookup: callers are for creat as
1654          * well as unlink, so a lot of the time it would cost
1655          * a double lookup.
1656          */
1657         dentry = d_lookup(base, name);
1658
1659         if (dentry && d_need_lookup(dentry)) {
1660                 /*
1661                  * __lookup_hash is called with the parent dir's i_mutex already
1662                  * held, so we are good to go here.
1663                  */
1664                 dentry = d_inode_lookup(base, dentry, nd);
1665                 if (IS_ERR(dentry))
1666                         return dentry;
1667         }
1668
1669         if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1670                 int status = d_revalidate(dentry, nd);
1671                 if (unlikely(status <= 0)) {
1672                         /*
1673                          * The dentry failed validation.
1674                          * If d_revalidate returned 0 attempt to invalidate
1675                          * the dentry otherwise d_revalidate is asking us
1676                          * to return a fail status.
1677                          */
1678                         if (status < 0) {
1679                                 dput(dentry);
1680                                 return ERR_PTR(status);
1681                         } else if (!d_invalidate(dentry)) {
1682                                 dput(dentry);
1683                                 dentry = NULL;
1684                         }
1685                 }
1686         }
1687
1688         if (!dentry)
1689                 dentry = d_alloc_and_lookup(base, name, nd);
1690
1691         return dentry;
1692 }
1693
1694 /*
1695  * Restricted form of lookup. Doesn't follow links, single-component only,
1696  * needs parent already locked. Doesn't follow mounts.
1697  * SMP-safe.
1698  */
1699 static struct dentry *lookup_hash(struct nameidata *nd)
1700 {
1701         return __lookup_hash(&nd->last, nd->path.dentry, nd);
1702 }
1703
1704 /**
1705  * lookup_one_len - filesystem helper to lookup single pathname component
1706  * @name:       pathname component to lookup
1707  * @base:       base directory to lookup from
1708  * @len:        maximum length @len should be interpreted to
1709  *
1710  * Note that this routine is purely a helper for filesystem usage and should
1711  * not be called by generic code.  Also note that by using this function the
1712  * nameidata argument is passed to the filesystem methods and a filesystem
1713  * using this helper needs to be prepared for that.
1714  */
1715 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1716 {
1717         struct qstr this;
1718         unsigned long hash;
1719         unsigned int c;
1720
1721         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1722
1723         this.name = name;
1724         this.len = len;
1725         if (!len)
1726                 return ERR_PTR(-EACCES);
1727
1728         hash = init_name_hash();
1729         while (len--) {
1730                 c = *(const unsigned char *)name++;
1731                 if (c == '/' || c == '\0')
1732                         return ERR_PTR(-EACCES);
1733                 hash = partial_name_hash(c, hash);
1734         }
1735         this.hash = end_name_hash(hash);
1736         /*
1737          * See if the low-level filesystem might want
1738          * to use its own hash..
1739          */
1740         if (base->d_flags & DCACHE_OP_HASH) {
1741                 int err = base->d_op->d_hash(base, base->d_inode, &this);
1742                 if (err < 0)
1743                         return ERR_PTR(err);
1744         }
1745
1746         return __lookup_hash(&this, base, NULL);
1747 }
1748
1749 int user_path_at(int dfd, const char __user *name, unsigned flags,
1750                  struct path *path)
1751 {
1752         struct nameidata nd;
1753         char *tmp = getname_flags(name, flags);
1754         int err = PTR_ERR(tmp);
1755         if (!IS_ERR(tmp)) {
1756
1757                 BUG_ON(flags & LOOKUP_PARENT);
1758
1759                 err = do_path_lookup(dfd, tmp, flags, &nd);
1760                 putname(tmp);
1761                 if (!err)
1762                         *path = nd.path;
1763         }
1764         return err;
1765 }
1766
1767 static int user_path_parent(int dfd, const char __user *path,
1768                         struct nameidata *nd, char **name)
1769 {
1770         char *s = getname(path);
1771         int error;
1772
1773         if (IS_ERR(s))
1774                 return PTR_ERR(s);
1775
1776         error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1777         if (error)
1778                 putname(s);
1779         else
1780                 *name = s;
1781
1782         return error;
1783 }
1784
1785 /*
1786  * It's inline, so penalty for filesystems that don't use sticky bit is
1787  * minimal.
1788  */
1789 static inline int check_sticky(struct inode *dir, struct inode *inode)
1790 {
1791         uid_t fsuid = current_fsuid();
1792
1793         if (!(dir->i_mode & S_ISVTX))
1794                 return 0;
1795         if (current_user_ns() != inode_userns(inode))
1796                 goto other_userns;
1797         if (inode->i_uid == fsuid)
1798                 return 0;
1799         if (dir->i_uid == fsuid)
1800                 return 0;
1801
1802 other_userns:
1803         return !ns_capable(inode_userns(inode), CAP_FOWNER);
1804 }
1805
1806 /*
1807  *      Check whether we can remove a link victim from directory dir, check
1808  *  whether the type of victim is right.
1809  *  1. We can't do it if dir is read-only (done in permission())
1810  *  2. We should have write and exec permissions on dir
1811  *  3. We can't remove anything from append-only dir
1812  *  4. We can't do anything with immutable dir (done in permission())
1813  *  5. If the sticky bit on dir is set we should either
1814  *      a. be owner of dir, or
1815  *      b. be owner of victim, or
1816  *      c. have CAP_FOWNER capability
1817  *  6. If the victim is append-only or immutable we can't do antyhing with
1818  *     links pointing to it.
1819  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1820  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1821  *  9. We can't remove a root or mountpoint.
1822  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1823  *     nfs_async_unlink().
1824  */
1825 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1826 {
1827         int error;
1828
1829         if (!victim->d_inode)
1830                 return -ENOENT;
1831
1832         BUG_ON(victim->d_parent->d_inode != dir);
1833         audit_inode_child(victim, dir);
1834
1835         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1836         if (error)
1837                 return error;
1838         if (IS_APPEND(dir))
1839                 return -EPERM;
1840         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1841             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1842                 return -EPERM;
1843         if (isdir) {
1844                 if (!S_ISDIR(victim->d_inode->i_mode))
1845                         return -ENOTDIR;
1846                 if (IS_ROOT(victim))
1847                         return -EBUSY;
1848         } else if (S_ISDIR(victim->d_inode->i_mode))
1849                 return -EISDIR;
1850         if (IS_DEADDIR(dir))
1851                 return -ENOENT;
1852         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1853                 return -EBUSY;
1854         return 0;
1855 }
1856
1857 /*      Check whether we can create an object with dentry child in directory
1858  *  dir.
1859  *  1. We can't do it if child already exists (open has special treatment for
1860  *     this case, but since we are inlined it's OK)
1861  *  2. We can't do it if dir is read-only (done in permission())
1862  *  3. We should have write and exec permissions on dir
1863  *  4. We can't do it if dir is immutable (done in permission())
1864  */
1865 static inline int may_create(struct inode *dir, struct dentry *child)
1866 {
1867         if (child->d_inode)
1868                 return -EEXIST;
1869         if (IS_DEADDIR(dir))
1870                 return -ENOENT;
1871         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1872 }
1873
1874 /*
1875  * p1 and p2 should be directories on the same fs.
1876  */
1877 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1878 {
1879         struct dentry *p;
1880
1881         if (p1 == p2) {
1882                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1883                 return NULL;
1884         }
1885
1886         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1887
1888         p = d_ancestor(p2, p1);
1889         if (p) {
1890                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1891                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1892                 return p;
1893         }
1894
1895         p = d_ancestor(p1, p2);
1896         if (p) {
1897                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1898                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1899                 return p;
1900         }
1901
1902         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1903         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1904         return NULL;
1905 }
1906
1907 void unlock_rename(struct dentry *p1, struct dentry *p2)
1908 {
1909         mutex_unlock(&p1->d_inode->i_mutex);
1910         if (p1 != p2) {
1911                 mutex_unlock(&p2->d_inode->i_mutex);
1912                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1913         }
1914 }
1915
1916 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1917                 struct nameidata *nd)
1918 {
1919         int error = may_create(dir, dentry);
1920
1921         if (error)
1922                 return error;
1923
1924         if (!dir->i_op->create)
1925                 return -EACCES; /* shouldn't it be ENOSYS? */
1926         mode &= S_IALLUGO;
1927         mode |= S_IFREG;
1928         error = security_inode_create(dir, dentry, mode);
1929         if (error)
1930                 return error;
1931         error = dir->i_op->create(dir, dentry, mode, nd);
1932         if (!error)
1933                 fsnotify_create(dir, dentry);
1934         return error;
1935 }
1936
1937 static int may_open(struct path *path, int acc_mode, int flag)
1938 {
1939         struct dentry *dentry = path->dentry;
1940         struct inode *inode = dentry->d_inode;
1941         int error;
1942
1943         /* O_PATH? */
1944         if (!acc_mode)
1945                 return 0;
1946
1947         if (!inode)
1948                 return -ENOENT;
1949
1950         switch (inode->i_mode & S_IFMT) {
1951         case S_IFLNK:
1952                 return -ELOOP;
1953         case S_IFDIR:
1954                 if (acc_mode & MAY_WRITE)
1955                         return -EISDIR;
1956                 break;
1957         case S_IFBLK:
1958         case S_IFCHR:
1959                 if (path->mnt->mnt_flags & MNT_NODEV)
1960                         return -EACCES;
1961                 /*FALLTHRU*/
1962         case S_IFIFO:
1963         case S_IFSOCK:
1964                 flag &= ~O_TRUNC;
1965                 break;
1966         }
1967
1968         error = inode_permission(inode, acc_mode);
1969         if (error)
1970                 return error;
1971
1972         /*
1973          * An append-only file must be opened in append mode for writing.
1974          */
1975         if (IS_APPEND(inode)) {
1976                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1977                         return -EPERM;
1978                 if (flag & O_TRUNC)
1979                         return -EPERM;
1980         }
1981
1982         /* O_NOATIME can only be set by the owner or superuser */
1983         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
1984                 return -EPERM;
1985
1986         /*
1987          * Ensure there are no outstanding leases on the file.
1988          */
1989         return break_lease(inode, flag);
1990 }
1991
1992 static int handle_truncate(struct file *filp)
1993 {
1994         struct path *path = &filp->f_path;
1995         struct inode *inode = path->dentry->d_inode;
1996         int error = get_write_access(inode);
1997         if (error)
1998                 return error;
1999         /*
2000          * Refuse to truncate files with mandatory locks held on them.
2001          */
2002         error = locks_verify_locked(inode);
2003         if (!error)
2004                 error = security_path_truncate(path);
2005         if (!error) {
2006                 error = do_truncate(path->dentry, 0,
2007                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2008                                     filp);
2009         }
2010         put_write_access(inode);
2011         return error;
2012 }
2013
2014 static inline int open_to_namei_flags(int flag)
2015 {
2016         if ((flag & O_ACCMODE) == 3)
2017                 flag--;
2018         return flag;
2019 }
2020
2021 /*
2022  * Handle the last step of open()
2023  */
2024 static struct file *do_last(struct nameidata *nd, struct path *path,
2025                             const struct open_flags *op, const char *pathname)
2026 {
2027         struct dentry *dir = nd->path.dentry;
2028         struct dentry *dentry;
2029         int open_flag = op->open_flag;
2030         int will_truncate = open_flag & O_TRUNC;
2031         int want_write = 0;
2032         int acc_mode = op->acc_mode;
2033         struct file *filp;
2034         int error;
2035
2036         nd->flags &= ~LOOKUP_PARENT;
2037         nd->flags |= op->intent;
2038
2039         switch (nd->last_type) {
2040         case LAST_DOTDOT:
2041         case LAST_DOT:
2042                 error = handle_dots(nd, nd->last_type);
2043                 if (error)
2044                         return ERR_PTR(error);
2045                 /* fallthrough */
2046         case LAST_ROOT:
2047                 error = complete_walk(nd);
2048                 if (error)
2049                         return ERR_PTR(error);
2050                 audit_inode(pathname, nd->path.dentry);
2051                 if (open_flag & O_CREAT) {
2052                         error = -EISDIR;
2053                         goto exit;
2054                 }
2055                 goto ok;
2056         case LAST_BIND:
2057                 error = complete_walk(nd);
2058                 if (error)
2059                         return ERR_PTR(error);
2060                 audit_inode(pathname, dir);
2061                 goto ok;
2062         }
2063
2064         if (!(open_flag & O_CREAT)) {
2065                 int symlink_ok = 0;
2066                 if (nd->last.name[nd->last.len])
2067                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2068                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2069                         symlink_ok = 1;
2070                 /* we _can_ be in RCU mode here */
2071                 error = walk_component(nd, path, &nd->last, LAST_NORM,
2072                                         !symlink_ok);
2073                 if (error < 0)
2074                         return ERR_PTR(error);
2075                 if (error) /* symlink */
2076                         return NULL;
2077                 /* sayonara */
2078                 error = complete_walk(nd);
2079                 if (error)
2080                         return ERR_PTR(-ECHILD);
2081
2082                 error = -ENOTDIR;
2083                 if (nd->flags & LOOKUP_DIRECTORY) {
2084                         if (!nd->inode->i_op->lookup)
2085                                 goto exit;
2086                 }
2087                 audit_inode(pathname, nd->path.dentry);
2088                 goto ok;
2089         }
2090
2091         /* create side of things */
2092         error = complete_walk(nd);
2093         if (error)
2094                 return ERR_PTR(error);
2095
2096         audit_inode(pathname, dir);
2097         error = -EISDIR;
2098         /* trailing slashes? */
2099         if (nd->last.name[nd->last.len])
2100                 goto exit;
2101
2102         mutex_lock(&dir->d_inode->i_mutex);
2103
2104         dentry = lookup_hash(nd);
2105         error = PTR_ERR(dentry);
2106         if (IS_ERR(dentry)) {
2107                 mutex_unlock(&dir->d_inode->i_mutex);
2108                 goto exit;
2109         }
2110
2111         path->dentry = dentry;
2112         path->mnt = nd->path.mnt;
2113
2114         /* Negative dentry, just create the file */
2115         if (!dentry->d_inode) {
2116                 int mode = op->mode;
2117                 if (!IS_POSIXACL(dir->d_inode))
2118                         mode &= ~current_umask();
2119                 /*
2120                  * This write is needed to ensure that a
2121                  * rw->ro transition does not occur between
2122                  * the time when the file is created and when
2123                  * a permanent write count is taken through
2124                  * the 'struct file' in nameidata_to_filp().
2125                  */
2126                 error = mnt_want_write(nd->path.mnt);
2127                 if (error)
2128                         goto exit_mutex_unlock;
2129                 want_write = 1;
2130                 /* Don't check for write permission, don't truncate */
2131                 open_flag &= ~O_TRUNC;
2132                 will_truncate = 0;
2133                 acc_mode = MAY_OPEN;
2134                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2135                 if (error)
2136                         goto exit_mutex_unlock;
2137                 error = vfs_create(dir->d_inode, dentry, mode, nd);
2138                 if (error)
2139                         goto exit_mutex_unlock;
2140                 mutex_unlock(&dir->d_inode->i_mutex);
2141                 dput(nd->path.dentry);
2142                 nd->path.dentry = dentry;
2143                 goto common;
2144         }
2145
2146         /*
2147          * It already exists.
2148          */
2149         mutex_unlock(&dir->d_inode->i_mutex);
2150         audit_inode(pathname, path->dentry);
2151
2152         error = -EEXIST;
2153         if (open_flag & O_EXCL)
2154                 goto exit_dput;
2155
2156         error = follow_managed(path, nd->flags);
2157         if (error < 0)
2158                 goto exit_dput;
2159
2160         error = -ENOENT;
2161         if (!path->dentry->d_inode)
2162                 goto exit_dput;
2163
2164         if (path->dentry->d_inode->i_op->follow_link)
2165                 return NULL;
2166
2167         path_to_nameidata(path, nd);
2168         nd->inode = path->dentry->d_inode;
2169         error = -EISDIR;
2170         if (S_ISDIR(nd->inode->i_mode))
2171                 goto exit;
2172 ok:
2173         if (!S_ISREG(nd->inode->i_mode))
2174                 will_truncate = 0;
2175
2176         if (will_truncate) {
2177                 error = mnt_want_write(nd->path.mnt);
2178                 if (error)
2179                         goto exit;
2180                 want_write = 1;
2181         }
2182 common:
2183         error = may_open(&nd->path, acc_mode, open_flag);
2184         if (error)
2185                 goto exit;
2186         filp = nameidata_to_filp(nd);
2187         if (!IS_ERR(filp)) {
2188                 error = ima_file_check(filp, op->acc_mode);
2189                 if (error) {
2190                         fput(filp);
2191                         filp = ERR_PTR(error);
2192                 }
2193         }
2194         if (!IS_ERR(filp)) {
2195                 if (will_truncate) {
2196                         error = handle_truncate(filp);
2197                         if (error) {
2198                                 fput(filp);
2199                                 filp = ERR_PTR(error);
2200                         }
2201                 }
2202         }
2203 out:
2204         if (want_write)
2205                 mnt_drop_write(nd->path.mnt);
2206         path_put(&nd->path);
2207         return filp;
2208
2209 exit_mutex_unlock:
2210         mutex_unlock(&dir->d_inode->i_mutex);
2211 exit_dput:
2212         path_put_conditional(path, nd);
2213 exit:
2214         filp = ERR_PTR(error);
2215         goto out;
2216 }
2217
2218 static struct file *path_openat(int dfd, const char *pathname,
2219                 struct nameidata *nd, const struct open_flags *op, int flags)
2220 {
2221         struct file *base = NULL;
2222         struct file *filp;
2223         struct path path;
2224         int error;
2225
2226         filp = get_empty_filp();
2227         if (!filp)
2228                 return ERR_PTR(-ENFILE);
2229
2230         filp->f_flags = op->open_flag;
2231         nd->intent.open.file = filp;
2232         nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2233         nd->intent.open.create_mode = op->mode;
2234
2235         error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2236         if (unlikely(error))
2237                 goto out_filp;
2238
2239         current->total_link_count = 0;
2240         error = link_path_walk(pathname, nd);
2241         if (unlikely(error))
2242                 goto out_filp;
2243
2244         filp = do_last(nd, &path, op, pathname);
2245         while (unlikely(!filp)) { /* trailing symlink */
2246                 struct path link = path;
2247                 void *cookie;
2248                 if (!(nd->flags & LOOKUP_FOLLOW)) {
2249                         path_put_conditional(&path, nd);
2250                         path_put(&nd->path);
2251                         filp = ERR_PTR(-ELOOP);
2252                         break;
2253                 }
2254                 nd->flags |= LOOKUP_PARENT;
2255                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2256                 error = follow_link(&link, nd, &cookie);
2257                 if (unlikely(error))
2258                         filp = ERR_PTR(error);
2259                 else
2260                         filp = do_last(nd, &path, op, pathname);
2261                 put_link(nd, &link, cookie);
2262         }
2263 out:
2264         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2265                 path_put(&nd->root);
2266         if (base)
2267                 fput(base);
2268         release_open_intent(nd);
2269         return filp;
2270
2271 out_filp:
2272         filp = ERR_PTR(error);
2273         goto out;
2274 }
2275
2276 struct file *do_filp_open(int dfd, const char *pathname,
2277                 const struct open_flags *op, int flags)
2278 {
2279         struct nameidata nd;
2280         struct file *filp;
2281
2282         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2283         if (unlikely(filp == ERR_PTR(-ECHILD)))
2284                 filp = path_openat(dfd, pathname, &nd, op, flags);
2285         if (unlikely(filp == ERR_PTR(-ESTALE)))
2286                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2287         return filp;
2288 }
2289
2290 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2291                 const char *name, const struct open_flags *op, int flags)
2292 {
2293         struct nameidata nd;
2294         struct file *file;
2295
2296         nd.root.mnt = mnt;
2297         nd.root.dentry = dentry;
2298
2299         flags |= LOOKUP_ROOT;
2300
2301         if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2302                 return ERR_PTR(-ELOOP);
2303
2304         file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2305         if (unlikely(file == ERR_PTR(-ECHILD)))
2306                 file = path_openat(-1, name, &nd, op, flags);
2307         if (unlikely(file == ERR_PTR(-ESTALE)))
2308                 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2309         return file;
2310 }
2311
2312 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2313 {
2314         struct dentry *dentry = ERR_PTR(-EEXIST);
2315         struct nameidata nd;
2316         int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2317         if (error)
2318                 return ERR_PTR(error);
2319
2320         /*
2321          * Yucky last component or no last component at all?
2322          * (foo/., foo/.., /////)
2323          */
2324         if (nd.last_type != LAST_NORM)
2325                 goto out;
2326         nd.flags &= ~LOOKUP_PARENT;
2327         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2328         nd.intent.open.flags = O_EXCL;
2329
2330         /*
2331          * Do the final lookup.
2332          */
2333         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2334         dentry = lookup_hash(&nd);
2335         if (IS_ERR(dentry))
2336                 goto fail;
2337
2338         if (dentry->d_inode)
2339                 goto eexist;
2340         /*
2341          * Special case - lookup gave negative, but... we had foo/bar/
2342          * From the vfs_mknod() POV we just have a negative dentry -
2343          * all is fine. Let's be bastards - you had / on the end, you've
2344          * been asking for (non-existent) directory. -ENOENT for you.
2345          */
2346         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
2347                 dput(dentry);
2348                 dentry = ERR_PTR(-ENOENT);
2349                 goto fail;
2350         }
2351         *path = nd.path;
2352         return dentry;
2353 eexist:
2354         dput(dentry);
2355         dentry = ERR_PTR(-EEXIST);
2356 fail:
2357         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2358 out:
2359         path_put(&nd.path);
2360         return dentry;
2361 }
2362 EXPORT_SYMBOL(kern_path_create);
2363
2364 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2365 {
2366         char *tmp = getname(pathname);
2367         struct dentry *res;
2368         if (IS_ERR(tmp))
2369                 return ERR_CAST(tmp);
2370         res = kern_path_create(dfd, tmp, path, is_dir);
2371         putname(tmp);
2372         return res;
2373 }
2374 EXPORT_SYMBOL(user_path_create);
2375
2376 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2377 {
2378         int error = may_create(dir, dentry);
2379
2380         if (error)
2381                 return error;
2382
2383         if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2384             !ns_capable(inode_userns(dir), CAP_MKNOD))
2385                 return -EPERM;
2386
2387         if (!dir->i_op->mknod)
2388                 return -EPERM;
2389
2390         error = devcgroup_inode_mknod(mode, dev);
2391         if (error)
2392                 return error;
2393
2394         error = security_inode_mknod(dir, dentry, mode, dev);
2395         if (error)
2396                 return error;
2397
2398         error = dir->i_op->mknod(dir, dentry, mode, dev);
2399         if (!error)
2400                 fsnotify_create(dir, dentry);
2401         return error;
2402 }
2403
2404 static int may_mknod(mode_t mode)
2405 {
2406         switch (mode & S_IFMT) {
2407         case S_IFREG:
2408         case S_IFCHR:
2409         case S_IFBLK:
2410         case S_IFIFO:
2411         case S_IFSOCK:
2412         case 0: /* zero mode translates to S_IFREG */
2413                 return 0;
2414         case S_IFDIR:
2415                 return -EPERM;
2416         default:
2417                 return -EINVAL;
2418         }
2419 }
2420
2421 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2422                 unsigned, dev)
2423 {
2424         struct dentry *dentry;
2425         struct path path;
2426         int error;
2427
2428         if (S_ISDIR(mode))
2429                 return -EPERM;
2430
2431         dentry = user_path_create(dfd, filename, &path, 0);
2432         if (IS_ERR(dentry))
2433                 return PTR_ERR(dentry);
2434
2435         if (!IS_POSIXACL(path.dentry->d_inode))
2436                 mode &= ~current_umask();
2437         error = may_mknod(mode);
2438         if (error)
2439                 goto out_dput;
2440         error = mnt_want_write(path.mnt);
2441         if (error)
2442                 goto out_dput;
2443         error = security_path_mknod(&path, dentry, mode, dev);
2444         if (error)
2445                 goto out_drop_write;
2446         switch (mode & S_IFMT) {
2447                 case 0: case S_IFREG:
2448                         error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2449                         break;
2450                 case S_IFCHR: case S_IFBLK:
2451                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2452                                         new_decode_dev(dev));
2453                         break;
2454                 case S_IFIFO: case S_IFSOCK:
2455                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2456                         break;
2457         }
2458 out_drop_write:
2459         mnt_drop_write(path.mnt);
2460 out_dput:
2461         dput(dentry);
2462         mutex_unlock(&path.dentry->d_inode->i_mutex);
2463         path_put(&path);
2464
2465         return error;
2466 }
2467
2468 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2469 {
2470         return sys_mknodat(AT_FDCWD, filename, mode, dev);
2471 }
2472
2473 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2474 {
2475         int error = may_create(dir, dentry);
2476
2477         if (error)
2478                 return error;
2479
2480         if (!dir->i_op->mkdir)
2481                 return -EPERM;
2482
2483         mode &= (S_IRWXUGO|S_ISVTX);
2484         error = security_inode_mkdir(dir, dentry, mode);
2485         if (error)
2486                 return error;
2487
2488         error = dir->i_op->mkdir(dir, dentry, mode);
2489         if (!error)
2490                 fsnotify_mkdir(dir, dentry);
2491         return error;
2492 }
2493
2494 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2495 {
2496         struct dentry *dentry;
2497         struct path path;
2498         int error;
2499
2500         dentry = user_path_create(dfd, pathname, &path, 1);
2501         if (IS_ERR(dentry))
2502                 return PTR_ERR(dentry);
2503
2504         if (!IS_POSIXACL(path.dentry->d_inode))
2505                 mode &= ~current_umask();
2506         error = mnt_want_write(path.mnt);
2507         if (error)
2508                 goto out_dput;
2509         error = security_path_mkdir(&path, dentry, mode);
2510         if (error)
2511                 goto out_drop_write;
2512         error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2513 out_drop_write:
2514         mnt_drop_write(path.mnt);
2515 out_dput:
2516         dput(dentry);
2517         mutex_unlock(&path.dentry->d_inode->i_mutex);
2518         path_put(&path);
2519         return error;
2520 }
2521
2522 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2523 {
2524         return sys_mkdirat(AT_FDCWD, pathname, mode);
2525 }
2526
2527 /*
2528  * The dentry_unhash() helper will try to drop the dentry early: we
2529  * should have a usage count of 2 if we're the only user of this
2530  * dentry, and if that is true (possibly after pruning the dcache),
2531  * then we drop the dentry now.
2532  *
2533  * A low-level filesystem can, if it choses, legally
2534  * do a
2535  *
2536  *      if (!d_unhashed(dentry))
2537  *              return -EBUSY;
2538  *
2539  * if it cannot handle the case of removing a directory
2540  * that is still in use by something else..
2541  */
2542 void dentry_unhash(struct dentry *dentry)
2543 {
2544         shrink_dcache_parent(dentry);
2545         spin_lock(&dentry->d_lock);
2546         if (dentry->d_count == 1)
2547                 __d_drop(dentry);
2548         spin_unlock(&dentry->d_lock);
2549 }
2550
2551 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2552 {
2553         int error = may_delete(dir, dentry, 1);
2554
2555         if (error)
2556                 return error;
2557
2558         if (!dir->i_op->rmdir)
2559                 return -EPERM;
2560
2561         mutex_lock(&dentry->d_inode->i_mutex);
2562
2563         error = -EBUSY;
2564         if (d_mountpoint(dentry))
2565                 goto out;
2566
2567         error = security_inode_rmdir(dir, dentry);
2568         if (error)
2569                 goto out;
2570
2571         shrink_dcache_parent(dentry);
2572         error = dir->i_op->rmdir(dir, dentry);
2573         if (error)
2574                 goto out;
2575
2576         dentry->d_inode->i_flags |= S_DEAD;
2577         dont_mount(dentry);
2578
2579 out:
2580         mutex_unlock(&dentry->d_inode->i_mutex);
2581         if (!error)
2582                 d_delete(dentry);
2583         return error;
2584 }
2585
2586 static long do_rmdir(int dfd, const char __user *pathname)
2587 {
2588         int error = 0;
2589         char * name;
2590         struct dentry *dentry;
2591         struct nameidata nd;
2592
2593         error = user_path_parent(dfd, pathname, &nd, &name);
2594         if (error)
2595                 return error;
2596
2597         switch(nd.last_type) {
2598         case LAST_DOTDOT:
2599                 error = -ENOTEMPTY;
2600                 goto exit1;
2601         case LAST_DOT:
2602                 error = -EINVAL;
2603                 goto exit1;
2604         case LAST_ROOT:
2605                 error = -EBUSY;
2606                 goto exit1;
2607         }
2608
2609         nd.flags &= ~LOOKUP_PARENT;
2610
2611         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2612         dentry = lookup_hash(&nd);
2613         error = PTR_ERR(dentry);
2614         if (IS_ERR(dentry))
2615                 goto exit2;
2616         if (!dentry->d_inode) {
2617                 error = -ENOENT;
2618                 goto exit3;
2619         }
2620         error = mnt_want_write(nd.path.mnt);
2621         if (error)
2622                 goto exit3;
2623         error = security_path_rmdir(&nd.path, dentry);
2624         if (error)
2625                 goto exit4;
2626         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2627 exit4:
2628         mnt_drop_write(nd.path.mnt);
2629 exit3:
2630         dput(dentry);
2631 exit2:
2632         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2633 exit1:
2634         path_put(&nd.path);
2635         putname(name);
2636         return error;
2637 }
2638
2639 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2640 {
2641         return do_rmdir(AT_FDCWD, pathname);
2642 }
2643
2644 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2645 {
2646         int error = may_delete(dir, dentry, 0);
2647
2648         if (error)
2649                 return error;
2650
2651         if (!dir->i_op->unlink)
2652                 return -EPERM;
2653
2654         mutex_lock(&dentry->d_inode->i_mutex);
2655         if (d_mountpoint(dentry))
2656                 error = -EBUSY;
2657         else {
2658                 error = security_inode_unlink(dir, dentry);
2659                 if (!error) {
2660                         error = dir->i_op->unlink(dir, dentry);
2661                         if (!error)
2662                                 dont_mount(dentry);
2663                 }
2664         }
2665         mutex_unlock(&dentry->d_inode->i_mutex);
2666
2667         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2668         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2669                 fsnotify_link_count(dentry->d_inode);
2670                 d_delete(dentry);
2671         }
2672
2673         return error;
2674 }
2675
2676 /*
2677  * Make sure that the actual truncation of the file will occur outside its
2678  * directory's i_mutex.  Truncate can take a long time if there is a lot of
2679  * writeout happening, and we don't want to prevent access to the directory
2680  * while waiting on the I/O.
2681  */
2682 static long do_unlinkat(int dfd, const char __user *pathname)
2683 {
2684         int error;
2685         char *name;
2686         struct dentry *dentry;
2687         struct nameidata nd;
2688         struct inode *inode = NULL;
2689
2690         error = user_path_parent(dfd, pathname, &nd, &name);
2691         if (error)
2692                 return error;
2693
2694         error = -EISDIR;
2695         if (nd.last_type != LAST_NORM)
2696                 goto exit1;
2697
2698         nd.flags &= ~LOOKUP_PARENT;
2699
2700         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2701         dentry = lookup_hash(&nd);
2702         error = PTR_ERR(dentry);
2703         if (!IS_ERR(dentry)) {
2704                 /* Why not before? Because we want correct error value */
2705                 if (nd.last.name[nd.last.len])
2706                         goto slashes;
2707                 inode = dentry->d_inode;
2708                 if (!inode)
2709                         goto slashes;
2710                 ihold(inode);
2711                 error = mnt_want_write(nd.path.mnt);
2712                 if (error)
2713                         goto exit2;
2714                 error = security_path_unlink(&nd.path, dentry);
2715                 if (error)
2716                         goto exit3;
2717                 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2718 exit3:
2719                 mnt_drop_write(nd.path.mnt);
2720         exit2:
2721                 dput(dentry);
2722         }
2723         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2724         if (inode)
2725                 iput(inode);    /* truncate the inode here */
2726 exit1:
2727         path_put(&nd.path);
2728         putname(name);
2729         return error;
2730
2731 slashes:
2732         error = !dentry->d_inode ? -ENOENT :
2733                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2734         goto exit2;
2735 }
2736
2737 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2738 {
2739         if ((flag & ~AT_REMOVEDIR) != 0)
2740                 return -EINVAL;
2741
2742         if (flag & AT_REMOVEDIR)
2743                 return do_rmdir(dfd, pathname);
2744
2745         return do_unlinkat(dfd, pathname);
2746 }
2747
2748 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2749 {
2750         return do_unlinkat(AT_FDCWD, pathname);
2751 }
2752
2753 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2754 {
2755         int error = may_create(dir, dentry);
2756
2757         if (error)
2758                 return error;
2759
2760         if (!dir->i_op->symlink)
2761                 return -EPERM;
2762
2763         error = security_inode_symlink(dir, dentry, oldname);
2764         if (error)
2765                 return error;
2766
2767         error = dir->i_op->symlink(dir, dentry, oldname);
2768         if (!error)
2769                 fsnotify_create(dir, dentry);
2770         return error;
2771 }
2772
2773 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2774                 int, newdfd, const char __user *, newname)
2775 {
2776         int error;
2777         char *from;
2778         struct dentry *dentry;
2779         struct path path;
2780
2781         from = getname(oldname);
2782         if (IS_ERR(from))
2783                 return PTR_ERR(from);
2784
2785         dentry = user_path_create(newdfd, newname, &path, 0);
2786         error = PTR_ERR(dentry);
2787         if (IS_ERR(dentry))
2788                 goto out_putname;
2789
2790         error = mnt_want_write(path.mnt);
2791         if (error)
2792                 goto out_dput;
2793         error = security_path_symlink(&path, dentry, from);
2794         if (error)
2795                 goto out_drop_write;
2796         error = vfs_symlink(path.dentry->d_inode, dentry, from);
2797 out_drop_write:
2798         mnt_drop_write(path.mnt);
2799 out_dput:
2800         dput(dentry);
2801         mutex_unlock(&path.dentry->d_inode->i_mutex);
2802         path_put(&path);
2803 out_putname:
2804         putname(from);
2805         return error;
2806 }
2807
2808 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2809 {
2810         return sys_symlinkat(oldname, AT_FDCWD, newname);
2811 }
2812
2813 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2814 {
2815         struct inode *inode = old_dentry->d_inode;
2816         int error;
2817
2818         if (!inode)
2819                 return -ENOENT;
2820
2821         error = may_create(dir, new_dentry);
2822         if (error)
2823                 return error;
2824
2825         if (dir->i_sb != inode->i_sb)
2826                 return -EXDEV;
2827
2828         /*
2829          * A link to an append-only or immutable file cannot be created.
2830          */
2831         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2832                 return -EPERM;
2833         if (!dir->i_op->link)
2834                 return -EPERM;
2835         if (S_ISDIR(inode->i_mode))
2836                 return -EPERM;
2837
2838         error = security_inode_link(old_dentry, dir, new_dentry);
2839         if (error)
2840                 return error;
2841
2842         mutex_lock(&inode->i_mutex);
2843         /* Make sure we don't allow creating hardlink to an unlinked file */
2844         if (inode->i_nlink == 0)
2845                 error =  -ENOENT;
2846         else
2847                 error = dir->i_op->link(old_dentry, dir, new_dentry);
2848         mutex_unlock(&inode->i_mutex);
2849         if (!error)
2850                 fsnotify_link(dir, inode, new_dentry);
2851         return error;
2852 }
2853
2854 /*
2855  * Hardlinks are often used in delicate situations.  We avoid
2856  * security-related surprises by not following symlinks on the
2857  * newname.  --KAB
2858  *
2859  * We don't follow them on the oldname either to be compatible
2860  * with linux 2.0, and to avoid hard-linking to directories
2861  * and other special files.  --ADM
2862  */
2863 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2864                 int, newdfd, const char __user *, newname, int, flags)
2865 {
2866         struct dentry *new_dentry;
2867         struct path old_path, new_path;
2868         int how = 0;
2869         int error;
2870
2871         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2872                 return -EINVAL;
2873         /*
2874          * To use null names we require CAP_DAC_READ_SEARCH
2875          * This ensures that not everyone will be able to create
2876          * handlink using the passed filedescriptor.
2877          */
2878         if (flags & AT_EMPTY_PATH) {
2879                 if (!capable(CAP_DAC_READ_SEARCH))
2880                         return -ENOENT;
2881                 how = LOOKUP_EMPTY;
2882         }
2883
2884         if (flags & AT_SYMLINK_FOLLOW)
2885                 how |= LOOKUP_FOLLOW;
2886
2887         error = user_path_at(olddfd, oldname, how, &old_path);
2888         if (error)
2889                 return error;
2890
2891         new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2892         error = PTR_ERR(new_dentry);
2893         if (IS_ERR(new_dentry))
2894                 goto out;
2895
2896         error = -EXDEV;
2897         if (old_path.mnt != new_path.mnt)
2898                 goto out_dput;
2899         error = mnt_want_write(new_path.mnt);
2900         if (error)
2901                 goto out_dput;
2902         error = security_path_link(old_path.dentry, &new_path, new_dentry);
2903         if (error)
2904                 goto out_drop_write;
2905         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2906 out_drop_write:
2907         mnt_drop_write(new_path.mnt);
2908 out_dput:
2909         dput(new_dentry);
2910         mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2911         path_put(&new_path);
2912 out:
2913         path_put(&old_path);
2914
2915         return error;
2916 }
2917
2918 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2919 {
2920         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2921 }
2922
2923 /*
2924  * The worst of all namespace operations - renaming directory. "Perverted"
2925  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2926  * Problems:
2927  *      a) we can get into loop creation. Check is done in is_subdir().
2928  *      b) race potential - two innocent renames can create a loop together.
2929  *         That's where 4.4 screws up. Current fix: serialization on
2930  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2931  *         story.
2932  *      c) we have to lock _three_ objects - parents and victim (if it exists).
2933  *         And that - after we got ->i_mutex on parents (until then we don't know
2934  *         whether the target exists).  Solution: try to be smart with locking
2935  *         order for inodes.  We rely on the fact that tree topology may change
2936  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
2937  *         move will be locked.  Thus we can rank directories by the tree
2938  *         (ancestors first) and rank all non-directories after them.
2939  *         That works since everybody except rename does "lock parent, lookup,
2940  *         lock child" and rename is under ->s_vfs_rename_mutex.
2941  *         HOWEVER, it relies on the assumption that any object with ->lookup()
2942  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
2943  *         we'd better make sure that there's no link(2) for them.
2944  *      d) conversion from fhandle to dentry may come in the wrong moment - when
2945  *         we are removing the target. Solution: we will have to grab ->i_mutex
2946  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2947  *         ->i_mutex on parents, which works but leads to some truly excessive
2948  *         locking].
2949  */
2950 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2951                           struct inode *new_dir, struct dentry *new_dentry)
2952 {
2953         int error = 0;
2954         struct inode *target = new_dentry->d_inode;
2955
2956         /*
2957          * If we are going to change the parent - check write permissions,
2958          * we'll need to flip '..'.
2959          */
2960         if (new_dir != old_dir) {
2961                 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2962                 if (error)
2963                         return error;
2964         }
2965
2966         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2967         if (error)
2968                 return error;
2969
2970         if (target)
2971                 mutex_lock(&target->i_mutex);
2972
2973         error = -EBUSY;
2974         if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2975                 goto out;
2976
2977         if (target)
2978                 shrink_dcache_parent(new_dentry);
2979         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2980         if (error)
2981                 goto out;
2982
2983         if (target) {
2984                 target->i_flags |= S_DEAD;
2985                 dont_mount(new_dentry);
2986         }
2987 out:
2988         if (target)
2989                 mutex_unlock(&target->i_mutex);
2990         if (!error)
2991                 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2992                         d_move(old_dentry,new_dentry);
2993         return error;
2994 }
2995
2996 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2997                             struct inode *new_dir, struct dentry *new_dentry)
2998 {
2999         struct inode *target = new_dentry->d_inode;
3000         int error;
3001
3002         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3003         if (error)
3004                 return error;
3005
3006         dget(new_dentry);
3007         if (target)
3008                 mutex_lock(&target->i_mutex);
3009
3010         error = -EBUSY;
3011         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
3012                 goto out;
3013
3014         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
3015         if (error)
3016                 goto out;
3017
3018         if (target)
3019                 dont_mount(new_dentry);
3020         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
3021                 d_move(old_dentry, new_dentry);
3022 out:
3023         if (target)
3024                 mutex_unlock(&target->i_mutex);
3025         dput(new_dentry);
3026         return error;
3027 }
3028
3029 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
3030                struct inode *new_dir, struct dentry *new_dentry)
3031 {
3032         int error;
3033         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3034         const unsigned char *old_name;
3035
3036         if (old_dentry->d_inode == new_dentry->d_inode)
3037                 return 0;
3038
3039         error = may_delete(old_dir, old_dentry, is_dir);
3040         if (error)
3041                 return error;
3042
3043         if (!new_dentry->d_inode)
3044                 error = may_create(new_dir, new_dentry);
3045         else
3046                 error = may_delete(new_dir, new_dentry, is_dir);
3047         if (error)
3048                 return error;
3049
3050         if (!old_dir->i_op->rename)
3051                 return -EPERM;
3052
3053         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3054
3055         if (is_dir)
3056                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3057         else
3058                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3059         if (!error)
3060                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3061                               new_dentry->d_inode, old_dentry);
3062         fsnotify_oldname_free(old_name);
3063
3064         return error;
3065 }
3066
3067 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3068                 int, newdfd, const char __user *, newname)
3069 {
3070         struct dentry *old_dir, *new_dir;
3071         struct dentry *old_dentry, *new_dentry;
3072         struct dentry *trap;
3073         struct nameidata oldnd, newnd;
3074         char *from;
3075         char *to;
3076         int error;
3077
3078         error = user_path_parent(olddfd, oldname, &oldnd, &from);
3079         if (error)
3080                 goto exit;
3081
3082         error = user_path_parent(newdfd, newname, &newnd, &to);
3083         if (error)
3084                 goto exit1;
3085
3086         error = -EXDEV;
3087         if (oldnd.path.mnt != newnd.path.mnt)
3088                 goto exit2;
3089
3090         old_dir = oldnd.path.dentry;
3091         error = -EBUSY;
3092         if (oldnd.last_type != LAST_NORM)
3093                 goto exit2;
3094
3095         new_dir = newnd.path.dentry;
3096         if (newnd.last_type != LAST_NORM)
3097                 goto exit2;
3098
3099         oldnd.flags &= ~LOOKUP_PARENT;
3100         newnd.flags &= ~LOOKUP_PARENT;
3101         newnd.flags |= LOOKUP_RENAME_TARGET;
3102
3103         trap = lock_rename(new_dir, old_dir);
3104
3105         old_dentry = lookup_hash(&oldnd);
3106         error = PTR_ERR(old_dentry);
3107         if (IS_ERR(old_dentry))
3108                 goto exit3;
3109         /* source must exist */
3110         error = -ENOENT;
3111         if (!old_dentry->d_inode)
3112                 goto exit4;
3113         /* unless the source is a directory trailing slashes give -ENOTDIR */
3114         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3115                 error = -ENOTDIR;
3116                 if (oldnd.last.name[oldnd.last.len])
3117                         goto exit4;
3118                 if (newnd.last.name[newnd.last.len])
3119                         goto exit4;
3120         }
3121         /* source should not be ancestor of target */
3122         error = -EINVAL;
3123         if (old_dentry == trap)
3124                 goto exit4;
3125         new_dentry = lookup_hash(&newnd);
3126         error = PTR_ERR(new_dentry);
3127         if (IS_ERR(new_dentry))
3128                 goto exit4;
3129         /* target should not be an ancestor of source */
3130         error = -ENOTEMPTY;
3131         if (new_dentry == trap)
3132                 goto exit5;
3133
3134         error = mnt_want_write(oldnd.path.mnt);
3135         if (error)
3136                 goto exit5;
3137         error = security_path_rename(&oldnd.path, old_dentry,
3138                                      &newnd.path, new_dentry);
3139         if (error)
3140                 goto exit6;
3141         error = vfs_rename(old_dir->d_inode, old_dentry,
3142                                    new_dir->d_inode, new_dentry);
3143 exit6:
3144         mnt_drop_write(oldnd.path.mnt);
3145 exit5:
3146         dput(new_dentry);
3147 exit4:
3148         dput(old_dentry);
3149 exit3:
3150         unlock_rename(new_dir, old_dir);
3151 exit2:
3152         path_put(&newnd.path);
3153         putname(to);
3154 exit1:
3155         path_put(&oldnd.path);
3156         putname(from);
3157 exit:
3158         return error;
3159 }
3160
3161 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3162 {
3163         return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3164 }
3165
3166 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3167 {
3168         int len;
3169
3170         len = PTR_ERR(link);
3171         if (IS_ERR(link))
3172                 goto out;
3173
3174         len = strlen(link);
3175         if (len > (unsigned) buflen)
3176                 len = buflen;
3177         if (copy_to_user(buffer, link, len))
3178                 len = -EFAULT;
3179 out:
3180         return len;
3181 }
3182
3183 /*
3184  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
3185  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
3186  * using) it for any given inode is up to filesystem.
3187  */
3188 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3189 {
3190         struct nameidata nd;
3191         void *cookie;
3192         int res;
3193
3194         nd.depth = 0;
3195         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3196         if (IS_ERR(cookie))
3197                 return PTR_ERR(cookie);
3198
3199         res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3200         if (dentry->d_inode->i_op->put_link)
3201                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3202         return res;
3203 }
3204
3205 int vfs_follow_link(struct nameidata *nd, const char *link)
3206 {
3207         return __vfs_follow_link(nd, link);
3208 }
3209
3210 /* get the link contents into pagecache */
3211 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3212 {
3213         char *kaddr;
3214         struct page *page;
3215         struct address_space *mapping = dentry->d_inode->i_mapping;
3216         page = read_mapping_page(mapping, 0, NULL);
3217         if (IS_ERR(page))
3218                 return (char*)page;
3219         *ppage = page;
3220         kaddr = kmap(page);
3221         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3222         return kaddr;
3223 }
3224
3225 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3226 {
3227         struct page *page = NULL;
3228         char *s = page_getlink(dentry, &page);
3229         int res = vfs_readlink(dentry,buffer,buflen,s);
3230         if (page) {
3231                 kunmap(page);
3232                 page_cache_release(page);
3233         }
3234         return res;
3235 }
3236
3237 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3238 {
3239         struct page *page = NULL;
3240         nd_set_link(nd, page_getlink(dentry, &page));
3241         return page;
3242 }
3243
3244 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3245 {
3246         struct page *page = cookie;
3247
3248         if (page) {
3249                 kunmap(page);
3250                 page_cache_release(page);
3251         }
3252 }
3253
3254 /*
3255  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3256  */
3257 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3258 {
3259         struct address_space *mapping = inode->i_mapping;
3260         struct page *page;
3261         void *fsdata;
3262         int err;
3263         char *kaddr;
3264         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3265         if (nofs)
3266                 flags |= AOP_FLAG_NOFS;
3267
3268 retry:
3269         err = pagecache_write_begin(NULL, mapping, 0, len-1,
3270                                 flags, &page, &fsdata);
3271         if (err)
3272                 goto fail;
3273
3274         kaddr = kmap_atomic(page, KM_USER0);
3275         memcpy(kaddr, symname, len-1);
3276         kunmap_atomic(kaddr, KM_USER0);
3277
3278         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3279                                                         page, fsdata);
3280         if (err < 0)
3281                 goto fail;
3282         if (err < len-1)
3283                 goto retry;
3284
3285         mark_inode_dirty(inode);
3286         return 0;
3287 fail:
3288         return err;
3289 }
3290
3291 int page_symlink(struct inode *inode, const char *symname, int len)
3292 {
3293         return __page_symlink(inode, symname, len,
3294                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3295 }
3296
3297 const struct inode_operations page_symlink_inode_operations = {
3298         .readlink       = generic_readlink,
3299         .follow_link    = page_follow_link_light,
3300         .put_link       = page_put_link,
3301 };
3302
3303 EXPORT_SYMBOL(user_path_at);
3304 EXPORT_SYMBOL(follow_down_one);
3305 EXPORT_SYMBOL(follow_down);
3306 EXPORT_SYMBOL(follow_up);
3307 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3308 EXPORT_SYMBOL(getname);
3309 EXPORT_SYMBOL(lock_rename);
3310 EXPORT_SYMBOL(lookup_one_len);
3311 EXPORT_SYMBOL(page_follow_link_light);
3312 EXPORT_SYMBOL(page_put_link);
3313 EXPORT_SYMBOL(page_readlink);
3314 EXPORT_SYMBOL(__page_symlink);
3315 EXPORT_SYMBOL(page_symlink);
3316 EXPORT_SYMBOL(page_symlink_inode_operations);
3317 EXPORT_SYMBOL(kern_path);
3318 EXPORT_SYMBOL(vfs_path_lookup);
3319 EXPORT_SYMBOL(inode_permission);
3320 EXPORT_SYMBOL(unlock_rename);
3321 EXPORT_SYMBOL(vfs_create);
3322 EXPORT_SYMBOL(vfs_follow_link);
3323 EXPORT_SYMBOL(vfs_link);
3324 EXPORT_SYMBOL(vfs_mkdir);
3325 EXPORT_SYMBOL(vfs_mknod);
3326 EXPORT_SYMBOL(generic_permission);
3327 EXPORT_SYMBOL(vfs_readlink);
3328 EXPORT_SYMBOL(vfs_rename);
3329 EXPORT_SYMBOL(vfs_rmdir);
3330 EXPORT_SYMBOL(vfs_symlink);
3331 EXPORT_SYMBOL(vfs_unlink);
3332 EXPORT_SYMBOL(dentry_unhash);
3333 EXPORT_SYMBOL(generic_readlink);