fs/namei.c

   1 /*
   2  *  linux/fs/namei.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * Some corrections by tytso.
   9  */
  10
  11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12  * lookup logic.
  13  */
  14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15  */
  16
  17 #include <linux/init.h>
  18 #include <linux/module.h>
  19 #include <linux/slab.h>
  20 #include <linux/fs.h>
  21 #include <linux/namei.h>
  22 #include <linux/quotaops.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/fsnotify.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/personality.h>
  27 #include <linux/security.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/mount.h>
  30 #include <linux/audit.h>
  31 #include <asm/namei.h>
  32 #include <asm/uaccess.h>
  33
  34 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
  35
  36 /* [Feb-1997 T. Schoebel-Theuer]
  37  * Fundamental changes in the pathname lookup mechanisms (namei)
  38  * were necessary because of omirr.  The reason is that omirr needs
  39  * to know the _real_ pathname, not the user-supplied one, in case
  40  * of symlinks (and also when transname replacements occur).
  41  *
  42  * The new code replaces the old recursive symlink resolution with
  43  * an iterative one (in case of non-nested symlink chains).  It does
  44  * this with calls to <fs>_follow_link().
  45  * As a side effect, dir_namei(), _namei() and follow_link() are now
  46  * replaced with a single function lookup_dentry() that can handle all
  47  * the special cases of the former code.
  48  *
  49  * With the new dcache, the pathname is stored at each inode, at least as
  50  * long as the refcount of the inode is positive.  As a side effect, the
  51  * size of the dcache depends on the inode cache and thus is dynamic.
  52  *
  53  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  54  * resolution to correspond with current state of the code.
  55  *
  56  * Note that the symlink resolution is not *completely* iterative.
  57  * There is still a significant amount of tail- and mid- recursion in
  58  * the algorithm.  Also, note that <fs>_readlink() is not used in
  59  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  60  * may return different results than <fs>_follow_link().  Many virtual
  61  * filesystems (including /proc) exhibit this behavior.
  62  */
  63
  64 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  65  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  66  * and the name already exists in form of a symlink, try to create the new
  67  * name indicated by the symlink. The old code always complained that the
  68  * name already exists, due to not following the symlink even if its target
  69  * is nonexistent.  The new semantics affects also mknod() and link() when
  70  * the name is a symlink pointing to a non-existant name.
  71  *
  72  * I don't know which semantics is the right one, since I have no access
  73  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  74  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  75  * "old" one. Personally, I think the new semantics is much more logical.
  76  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  77  * file does succeed in both HP-UX and SunOs, but not in Solaris
  78  * and in the old Linux semantics.
  79  */
  80
  81 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  82  * semantics.  See the comments in "open_namei" and "do_link" below.
  83  *
  84  * [10-Sep-98 Alan Modra] Another symlink change.
  85  */
  86
  87 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  88  *      inside the path - always follow.
  89  *      in the last component in creation/removal/renaming - never follow.
  90  *      if LOOKUP_FOLLOW passed - follow.
  91  *      if the pathname has trailing slashes - follow.
  92  *      otherwise - don't follow.
  93  * (applied in that order).
  94  *
  95  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
  96  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
  97  * During the 2.4 we need to fix the userland stuff depending on it -
  98  * hopefully we will be able to get rid of that wart in 2.5. So far only
  99  * XEmacs seems to be relying on it...
 100  */
 101 /*
 102  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
 103  * implemented.  Let's see if raised priority of ->s_vfs_rename_sem gives
 104  * any extra contention...
 105  */
 106
 107 /* In order to reduce some races, while at the same time doing additional
 108  * checking and hopefully speeding things up, we copy filenames to the
 109  * kernel data space before using them..
 110  *
 111  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 112  * PATH_MAX includes the nul terminator --RR.
 113  */
 114 static inline int do_getname(const char __user *filename, char *page)
 115 {
 116         int retval;
 117         unsigned long len = PATH_MAX;
 118
 119         if (!segment_eq(get_fs(), KERNEL_DS)) {
 120                 if ((unsigned long) filename >= TASK_SIZE)
 121                         return -EFAULT;
 122                 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
 123                         len = TASK_SIZE - (unsigned long) filename;
 124         }
 125
 126         retval = strncpy_from_user(page, filename, len);
 127         if (retval > 0) {
 128                 if (retval < len)
 129                         return 0;
 130                 return -ENAMETOOLONG;
 131         } else if (!retval)
 132                 retval = -ENOENT;
 133         return retval;
 134 }
 135
 136 char * getname(const char __user * filename)
 137 {
 138         char *tmp, *result;
 139
 140         result = ERR_PTR(-ENOMEM);
 141         tmp = __getname();
 142         if (tmp)  {
 143                 int retval = do_getname(filename, tmp);
 144
 145                 result = tmp;
 146                 if (retval < 0) {
 147                         __putname(tmp);
 148                         result = ERR_PTR(retval);
 149                 }
 150         }
 151         audit_getname(result);
 152         return result;
 153 }
 154
 155 #ifdef CONFIG_AUDITSYSCALL
 156 void putname(const char *name)
 157 {
 158         if (unlikely(current->audit_context))
 159                 audit_putname(name);
 160         else
 161                 __putname(name);
 162 }
 163 EXPORT_SYMBOL(putname);
 164 #endif
 165
 166
 167 /**
 168  * generic_permission  -  check for access rights on a Posix-like filesystem
 169  * @inode:      inode to check access rights for
 170  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
 171  * @check_acl:  optional callback to check for Posix ACLs
 172  *
 173  * Used to check for read/write/execute permissions on a file.
 174  * We use "fsuid" for this, letting us set arbitrary permissions
 175  * for filesystem access without changing the "normal" uids which
 176  * are used for other things..
 177  */
 178 int generic_permission(struct inode *inode, int mask,
 179                 int (*check_acl)(struct inode *inode, int mask))
 180 {
 181         umode_t                 mode = inode->i_mode;
 182
 183         if (current->fsuid == inode->i_uid)
 184                 mode >>= 6;
 185         else {
 186                 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
 187                         int error = check_acl(inode, mask);
 188                         if (error == -EACCES)
 189                                 goto check_capabilities;
 190                         else if (error != -EAGAIN)
 191                                 return error;
 192                 }
 193
 194                 if (in_group_p(inode->i_gid))
 195                         mode >>= 3;
 196         }
 197
 198         /*
 199          * If the DACs are ok we don't need any capability check.
 200          */
 201         if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
 202                 return 0;
 203
 204  check_capabilities:
 205         /*
 206          * Read/write DACs are always overridable.
 207          * Executable DACs are overridable if at least one exec bit is set.
 208          */
 209         if (!(mask & MAY_EXEC) ||
 210             (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
 211                 if (capable(CAP_DAC_OVERRIDE))
 212                         return 0;
 213
 214         /*
 215          * Searching includes executable on directories, else just read.
 216          */
 217         if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
 218                 if (capable(CAP_DAC_READ_SEARCH))
 219                         return 0;
 220
 221         return -EACCES;
 222 }
 223
 224 int permission(struct inode *inode, int mask, struct nameidata *nd)
 225 {
 226         int retval, submask;
 227
 228         if (mask & MAY_WRITE) {
 229                 umode_t mode = inode->i_mode;
 230
 231                 /*
 232                  * Nobody gets write access to a read-only fs.
 233                  */
 234                 if (IS_RDONLY(inode) &&
 235                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 236                         return -EROFS;
 237
 238                 /*
 239                  * Nobody gets write access to an immutable file.
 240                  */
 241                 if (IS_IMMUTABLE(inode))
 242                         return -EACCES;
 243         }
 244
 245
 246         /* Ordinary permission routines do not understand MAY_APPEND. */
 247         submask = mask & ~MAY_APPEND;
 248         if (inode->i_op && inode->i_op->permission)
 249                 retval = inode->i_op->permission(inode, submask, nd);
 250         else
 251                 retval = generic_permission(inode, submask, NULL);
 252         if (retval)
 253                 return retval;
 254
 255         return security_inode_permission(inode, mask, nd);
 256 }
 257
 258 /*
 259  * get_write_access() gets write permission for a file.
 260  * put_write_access() releases this write permission.
 261  * This is used for regular files.
 262  * We cannot support write (and maybe mmap read-write shared) accesses and
 263  * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
 264  * can have the following values:
 265  * 0: no writers, no VM_DENYWRITE mappings
 266  * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
 267  * > 0: (i_writecount) users are writing to the file.
 268  *
 269  * Normally we operate on that counter with atomic_{inc,dec} and it's safe
 270  * except for the cases where we don't hold i_writecount yet. Then we need to
 271  * use {get,deny}_write_access() - these functions check the sign and refuse
 272  * to do the change if sign is wrong. Exclusion between them is provided by
 273  * the inode->i_lock spinlock.
 274  */
 275
 276 int get_write_access(struct inode * inode)
 277 {
 278         spin_lock(&inode->i_lock);
 279         if (atomic_read(&inode->i_writecount) < 0) {
 280                 spin_unlock(&inode->i_lock);
 281                 return -ETXTBSY;
 282         }
 283         atomic_inc(&inode->i_writecount);
 284         spin_unlock(&inode->i_lock);
 285
 286         return 0;
 287 }
 288
 289 int deny_write_access(struct file * file)
 290 {
 291         struct inode *inode = file->f_dentry->d_inode;
 292
 293         spin_lock(&inode->i_lock);
 294         if (atomic_read(&inode->i_writecount) > 0) {
 295                 spin_unlock(&inode->i_lock);
 296                 return -ETXTBSY;
 297         }
 298         atomic_dec(&inode->i_writecount);
 299         spin_unlock(&inode->i_lock);
 300
 301         return 0;
 302 }
 303
 304 void path_release(struct nameidata *nd)
 305 {
 306         dput(nd->dentry);
 307         mntput(nd->mnt);
 308 }
 309
 310 /*
 311  * umount() mustn't call path_release()/mntput() as that would clear
 312  * mnt_expiry_mark
 313  */
 314 void path_release_on_umount(struct nameidata *nd)
 315 {
 316         dput(nd->dentry);
 317         mntput_no_expire(nd->mnt);
 318 }
 319
 320 /*
 321  * Internal lookup() using the new generic dcache.
 322  * SMP-safe
 323  */
 324 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
 325 {
 326         struct dentry * dentry = __d_lookup(parent, name);
 327
 328         /* lockess __d_lookup may fail due to concurrent d_move()
 329          * in some unrelated directory, so try with d_lookup
 330          */
 331         if (!dentry)
 332                 dentry = d_lookup(parent, name);
 333
 334         if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
 335                 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
 336                         dput(dentry);
 337                         dentry = NULL;
 338                 }
 339         }
 340         return dentry;
 341 }
 342
 343 /*
 344  * Short-cut version of permission(), for calling by
 345  * path_walk(), when dcache lock is held.  Combines parts
 346  * of permission() and generic_permission(), and tests ONLY for
 347  * MAY_EXEC permission.
 348  *
 349  * If appropriate, check DAC only.  If not appropriate, or
 350  * short-cut DAC fails, then call permission() to do more
 351  * complete permission check.
 352  */
 353 static inline int exec_permission_lite(struct inode *inode,
 354                                        struct nameidata *nd)
 355 {
 356         umode_t mode = inode->i_mode;
 357
 358         if (inode->i_op && inode->i_op->permission)
 359                 return -EAGAIN;
 360
 361         if (current->fsuid == inode->i_uid)
 362                 mode >>= 6;
 363         else if (in_group_p(inode->i_gid))
 364                 mode >>= 3;
 365
 366         if (mode & MAY_EXEC)
 367                 goto ok;
 368
 369         if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
 370                 goto ok;
 371
 372         if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
 373                 goto ok;
 374
 375         if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
 376                 goto ok;
 377
 378         return -EACCES;
 379 ok:
 380         return security_inode_permission(inode, MAY_EXEC, nd);
 381 }
 382
 383 /*
 384  * This is called when everything else fails, and we actually have
 385  * to go to the low-level filesystem to find out what we should do..
 386  *
 387  * We get the directory semaphore, and after getting that we also
 388  * make sure that nobody added the entry to the dcache in the meantime..
 389  * SMP-safe
 390  */
 391 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
 392 {
 393         struct dentry * result;
 394         struct inode *dir = parent->d_inode;
 395
 396         down(&dir->i_sem);
 397         /*
 398          * First re-do the cached lookup just in case it was created
 399          * while we waited for the directory semaphore..
 400          *
 401          * FIXME! This could use version numbering or similar to
 402          * avoid unnecessary cache lookups.
 403          *
 404          * The "dcache_lock" is purely to protect the RCU list walker
 405          * from concurrent renames at this point (we mustn't get false
 406          * negatives from the RCU list walk here, unlike the optimistic
 407          * fast walk).
 408          *
 409          * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
 410          */
 411         result = d_lookup(parent, name);
 412         if (!result) {
 413                 struct dentry * dentry = d_alloc(parent, name);
 414                 result = ERR_PTR(-ENOMEM);
 415                 if (dentry) {
 416                         result = dir->i_op->lookup(dir, dentry, nd);
 417                         if (result)
 418                                 dput(dentry);
 419                         else
 420                                 result = dentry;
 421                 }
 422                 up(&dir->i_sem);
 423                 return result;
 424         }
 425
 426         /*
 427          * Uhhuh! Nasty case: the cache was re-populated while
 428          * we waited on the semaphore. Need to revalidate.
 429          */
 430         up(&dir->i_sem);
 431         if (result->d_op && result->d_op->d_revalidate) {
 432                 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
 433                         dput(result);
 434                         result = ERR_PTR(-ENOENT);
 435                 }
 436         }
 437         return result;
 438 }
 439
 440 static int __emul_lookup_dentry(const char *, struct nameidata *);
 441
 442 /* SMP-safe */
 443 static inline int
 444 walk_init_root(const char *name, struct nameidata *nd)
 445 {
 446         read_lock(&current->fs->lock);
 447         if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
 448                 nd->mnt = mntget(current->fs->altrootmnt);
 449                 nd->dentry = dget(current->fs->altroot);
 450                 read_unlock(&current->fs->lock);
 451                 if (__emul_lookup_dentry(name,nd))
 452                         return 0;
 453                 read_lock(&current->fs->lock);
 454         }
 455         nd->mnt = mntget(current->fs->rootmnt);
 456         nd->dentry = dget(current->fs->root);
 457         read_unlock(&current->fs->lock);
 458         return 1;
 459 }
 460
 461 static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
 462 {
 463         int res = 0;
 464         char *name;
 465         if (IS_ERR(link))
 466                 goto fail;
 467
 468         if (*link == '/') {
 469                 path_release(nd);
 470                 if (!walk_init_root(link, nd))
 471                         /* weird __emul_prefix() stuff did it */
 472                         goto out;
 473         }
 474         res = link_path_walk(link, nd);
 475 out:
 476         if (nd->depth || res || nd->last_type!=LAST_NORM)
 477                 return res;
 478         /*
 479          * If it is an iterative symlinks resolution in open_namei() we
 480          * have to copy the last component. And all that crap because of
 481          * bloody create() on broken symlinks. Furrfu...
 482          */
 483         name = __getname();
 484         if (unlikely(!name)) {
 485                 path_release(nd);
 486                 return -ENOMEM;
 487         }
 488         strcpy(name, nd->last.name);
 489         nd->last.name = name;
 490         return 0;
 491 fail:
 492         path_release(nd);
 493         return PTR_ERR(link);
 494 }
 495
 496 struct path {
 497         struct vfsmount *mnt;
 498         struct dentry *dentry;
 499 };
 500
 501 static inline int __do_follow_link(struct path *path, struct nameidata *nd)
 502 {
 503         int error;
 504         void *cookie;
 505         struct dentry *dentry = path->dentry;
 506
 507         touch_atime(path->mnt, dentry);
 508         nd_set_link(nd, NULL);
 509
 510         if (path->mnt == nd->mnt)
 511                 mntget(path->mnt);
 512         cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
 513         error = PTR_ERR(cookie);
 514         if (!IS_ERR(cookie)) {
 515                 char *s = nd_get_link(nd);
 516                 error = 0;
 517                 if (s)
 518                         error = __vfs_follow_link(nd, s);
 519                 if (dentry->d_inode->i_op->put_link)
 520                         dentry->d_inode->i_op->put_link(dentry, nd, cookie);
 521         }
 522         dput(dentry);
 523         mntput(path->mnt);
 524
 525         return error;
 526 }
 527
 528 static inline void dput_path(struct path *path, struct nameidata *nd)
 529 {
 530         dput(path->dentry);
 531         if (path->mnt != nd->mnt)
 532                 mntput(path->mnt);
 533 }
 534
 535 static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
 536 {
 537         dput(nd->dentry);
 538         if (nd->mnt != path->mnt)
 539                 mntput(nd->mnt);
 540         nd->mnt = path->mnt;
 541         nd->dentry = path->dentry;
 542 }
 543
 544 /*
 545  * This limits recursive symlink follows to 8, while
 546  * limiting consecutive symlinks to 40.
 547  *
 548  * Without that kind of total limit, nasty chains of consecutive
 549  * symlinks can cause almost arbitrarily long lookups.
 550  */
 551 static inline int do_follow_link(struct path *path, struct nameidata *nd)
 552 {
 553         int err = -ELOOP;
 554         if (current->link_count >= MAX_NESTED_LINKS)
 555                 goto loop;
 556         if (current->total_link_count >= 40)
 557                 goto loop;
 558         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
 559         cond_resched();
 560         err = security_inode_follow_link(path->dentry, nd);
 561         if (err)
 562                 goto loop;
 563         current->link_count++;
 564         current->total_link_count++;
 565         nd->depth++;
 566         err = __do_follow_link(path, nd);
 567         current->link_count--;
 568         nd->depth--;
 569         return err;
 570 loop:
 571         dput_path(path, nd);
 572         path_release(nd);
 573         return err;
 574 }
 575
 576 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 577 {
 578         struct vfsmount *parent;
 579         struct dentry *mountpoint;
 580         spin_lock(&vfsmount_lock);
 581         parent=(*mnt)->mnt_parent;
 582         if (parent == *mnt) {
 583                 spin_unlock(&vfsmount_lock);
 584                 return 0;
 585         }
 586         mntget(parent);
 587         mountpoint=dget((*mnt)->mnt_mountpoint);
 588         spin_unlock(&vfsmount_lock);
 589         dput(*dentry);
 590         *dentry = mountpoint;
 591         mntput(*mnt);
 592         *mnt = parent;
 593         return 1;
 594 }
 595
 596 /* no need for dcache_lock, as serialization is taken care in
 597  * namespace.c
 598  */
 599 static int __follow_mount(struct path *path)
 600 {
 601         int res = 0;
 602         while (d_mountpoint(path->dentry)) {
 603                 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
 604                 if (!mounted)
 605                         break;
 606                 dput(path->dentry);
 607                 if (res)
 608                         mntput(path->mnt);
 609                 path->mnt = mounted;
 610                 path->dentry = dget(mounted->mnt_root);
 611                 res = 1;
 612         }
 613         return res;
 614 }
 615
 616 static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
 617 {
 618         while (d_mountpoint(*dentry)) {
 619                 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
 620                 if (!mounted)
 621                         break;
 622                 dput(*dentry);
 623                 mntput(*mnt);
 624                 *mnt = mounted;
 625                 *dentry = dget(mounted->mnt_root);
 626         }
 627 }
 628
 629 /* no need for dcache_lock, as serialization is taken care in
 630  * namespace.c
 631  */
 632 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 633 {
 634         struct vfsmount *mounted;
 635
 636         mounted = lookup_mnt(*mnt, *dentry);
 637         if (mounted) {
 638                 dput(*dentry);
 639                 mntput(*mnt);
 640                 *mnt = mounted;
 641                 *dentry = dget(mounted->mnt_root);
 642                 return 1;
 643         }
 644         return 0;
 645 }
 646
 647 static inline void follow_dotdot(struct nameidata *nd)
 648 {
 649         while(1) {
 650                 struct vfsmount *parent;
 651                 struct dentry *old = nd->dentry;
 652
 653                 read_lock(&current->fs->lock);
 654                 if (nd->dentry == current->fs->root &&
 655                     nd->mnt == current->fs->rootmnt) {
 656                         read_unlock(&current->fs->lock);
 657                         break;
 658                 }
 659                 read_unlock(&current->fs->lock);
 660                 spin_lock(&dcache_lock);
 661                 if (nd->dentry != nd->mnt->mnt_root) {
 662                         nd->dentry = dget(nd->dentry->d_parent);
 663                         spin_unlock(&dcache_lock);
 664                         dput(old);
 665                         break;
 666                 }
 667                 spin_unlock(&dcache_lock);
 668                 spin_lock(&vfsmount_lock);
 669                 parent = nd->mnt->mnt_parent;
 670                 if (parent == nd->mnt) {
 671                         spin_unlock(&vfsmount_lock);
 672                         break;
 673                 }
 674                 mntget(parent);
 675                 nd->dentry = dget(nd->mnt->mnt_mountpoint);
 676                 spin_unlock(&vfsmount_lock);
 677                 dput(old);
 678                 mntput(nd->mnt);
 679                 nd->mnt = parent;
 680         }
 681         follow_mount(&nd->mnt, &nd->dentry);
 682 }
 683
 684 /*
 685  *  It's more convoluted than I'd like it to be, but... it's still fairly
 686  *  small and for now I'd prefer to have fast path as straight as possible.
 687  *  It _is_ time-critical.
 688  */
 689 static int do_lookup(struct nameidata *nd, struct qstr *name,
 690                      struct path *path)
 691 {
 692         struct vfsmount *mnt = nd->mnt;
 693         struct dentry *dentry = __d_lookup(nd->dentry, name);
 694
 695         if (!dentry)
 696                 goto need_lookup;
 697         if (dentry->d_op && dentry->d_op->d_revalidate)
 698                 goto need_revalidate;
 699 done:
 700         path->mnt = mnt;
 701         path->dentry = dentry;
 702         __follow_mount(path);
 703         return 0;
 704
 705 need_lookup:
 706         dentry = real_lookup(nd->dentry, name, nd);
 707         if (IS_ERR(dentry))
 708                 goto fail;
 709         goto done;
 710
 711 need_revalidate:
 712         if (dentry->d_op->d_revalidate(dentry, nd))
 713                 goto done;
 714         if (d_invalidate(dentry))
 715                 goto done;
 716         dput(dentry);
 717         goto need_lookup;
 718
 719 fail:
 720         return PTR_ERR(dentry);
 721 }
 722
 723 /*
 724  * Name resolution.
 725  * This is the basic name resolution function, turning a pathname into
 726  * the final dentry. We expect 'base' to be positive and a directory.
 727  *
 728  * Returns 0 and nd will have valid dentry and mnt on success.
 729  * Returns error and drops reference to input namei data on failure.
 730  */
 731 static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
 732 {
 733         struct path next;
 734         struct inode *inode;
 735         int err;
 736         unsigned int lookup_flags = nd->flags;
 737
 738         while (*name=='/')
 739                 name++;
 740         if (!*name)
 741                 goto return_reval;
 742
 743         inode = nd->dentry->d_inode;
 744         if (nd->depth)
 745                 lookup_flags = LOOKUP_FOLLOW;
 746
 747         /* At this point we know we have a real path component. */
 748         for(;;) {
 749                 unsigned long hash;
 750                 struct qstr this;
 751                 unsigned int c;
 752
 753                 err = exec_permission_lite(inode, nd);
 754                 if (err == -EAGAIN) {
 755                         err = permission(inode, MAY_EXEC, nd);
 756                 }
 757                 if (err)
 758                         break;
 759
 760                 this.name = name;
 761                 c = *(const unsigned char *)name;
 762
 763                 hash = init_name_hash();
 764                 do {
 765                         name++;
 766                         hash = partial_name_hash(c, hash);
 767                         c = *(const unsigned char *)name;
 768                 } while (c && (c != '/'));
 769                 this.len = name - (const char *) this.name;
 770                 this.hash = end_name_hash(hash);
 771
 772                 /* remove trailing slashes? */
 773                 if (!c)
 774                         goto last_component;
 775                 while (*++name == '/');
 776                 if (!*name)
 777                         goto last_with_slashes;
 778
 779                 /*
 780                  * "." and ".." are special - ".." especially so because it has
 781                  * to be able to know about the current root directory and
 782                  * parent relationships.
 783                  */
 784                 if (this.name[0] == '.') switch (this.len) {
 785                         default:
 786                                 break;
 787                         case 2:
 788                                 if (this.name[1] != '.')
 789                                         break;
 790                                 follow_dotdot(nd);
 791                                 inode = nd->dentry->d_inode;
 792                                 /* fallthrough */
 793                         case 1:
 794                                 continue;
 795                 }
 796                 /*
 797                  * See if the low-level filesystem might want
 798                  * to use its own hash..
 799                  */
 800                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 801                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 802                         if (err < 0)
 803                                 break;
 804                 }
 805                 nd->flags |= LOOKUP_CONTINUE;
 806                 /* This does the actual lookups.. */
 807                 err = do_lookup(nd, &this, &next);
 808                 if (err)
 809                         break;
 810
 811                 err = -ENOENT;
 812                 inode = next.dentry->d_inode;
 813                 if (!inode)
 814                         goto out_dput;
 815                 err = -ENOTDIR;
 816                 if (!inode->i_op)
 817                         goto out_dput;
 818
 819                 if (inode->i_op->follow_link) {
 820                         err = do_follow_link(&next, nd);
 821                         if (err)
 822                                 goto return_err;
 823                         err = -ENOENT;
 824                         inode = nd->dentry->d_inode;
 825                         if (!inode)
 826                                 break;
 827                         err = -ENOTDIR;
 828                         if (!inode->i_op)
 829                                 break;
 830                 } else
 831                         path_to_nameidata(&next, nd);
 832                 err = -ENOTDIR;
 833                 if (!inode->i_op->lookup)
 834                         break;
 835                 continue;
 836                 /* here ends the main loop */
 837
 838 last_with_slashes:
 839                 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 840 last_component:
 841                 nd->flags &= ~LOOKUP_CONTINUE;
 842                 if (lookup_flags & LOOKUP_PARENT)
 843                         goto lookup_parent;
 844                 if (this.name[0] == '.') switch (this.len) {
 845                         default:
 846                                 break;
 847                         case 2:
 848                                 if (this.name[1] != '.')
 849                                         break;
 850                                 follow_dotdot(nd);
 851                                 inode = nd->dentry->d_inode;
 852                                 /* fallthrough */
 853                         case 1:
 854                                 goto return_reval;
 855                 }
 856                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 857                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 858                         if (err < 0)
 859                                 break;
 860                 }
 861                 err = do_lookup(nd, &this, &next);
 862                 if (err)
 863                         break;
 864                 inode = next.dentry->d_inode;
 865                 if ((lookup_flags & LOOKUP_FOLLOW)
 866                     && inode && inode->i_op && inode->i_op->follow_link) {
 867                         err = do_follow_link(&next, nd);
 868                         if (err)
 869                                 goto return_err;
 870                         inode = nd->dentry->d_inode;
 871                 } else
 872                         path_to_nameidata(&next, nd);
 873                 err = -ENOENT;
 874                 if (!inode)
 875                         break;
 876                 if (lookup_flags & LOOKUP_DIRECTORY) {
 877                         err = -ENOTDIR;
 878                         if (!inode->i_op || !inode->i_op->lookup)
 879                                 break;
 880                 }
 881                 goto return_base;
 882 lookup_parent:
 883                 nd->last = this;
 884                 nd->last_type = LAST_NORM;
 885                 if (this.name[0] != '.')
 886                         goto return_base;
 887                 if (this.len == 1)
 888                         nd->last_type = LAST_DOT;
 889                 else if (this.len == 2 && this.name[1] == '.')
 890                         nd->last_type = LAST_DOTDOT;
 891                 else
 892                         goto return_base;
 893 return_reval:
 894                 /*
 895                  * We bypassed the ordinary revalidation routines.
 896                  * We may need to check the cached dentry for staleness.
 897                  */
 898                 if (nd->dentry && nd->dentry->d_sb &&
 899                     (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
 900                         err = -ESTALE;
 901                         /* Note: we do not d_invalidate() */
 902                         if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
 903                                 break;
 904                 }
 905 return_base:
 906                 return 0;
 907 out_dput:
 908                 dput_path(&next, nd);
 909                 break;
 910         }
 911         path_release(nd);
 912 return_err:
 913         return err;
 914 }
 915
 916 /*
 917  * Wrapper to retry pathname resolution whenever the underlying
 918  * file system returns an ESTALE.
 919  *
 920  * Retry the whole path once, forcing real lookup requests
 921  * instead of relying on the dcache.
 922  */
 923 int fastcall link_path_walk(const char *name, struct nameidata *nd)
 924 {
 925         struct nameidata save = *nd;
 926         int result;
 927
 928         /* make sure the stuff we saved doesn't go away */
 929         dget(save.dentry);
 930         mntget(save.mnt);
 931
 932         result = __link_path_walk(name, nd);
 933         if (result == -ESTALE) {
 934                 *nd = save;
 935                 dget(nd->dentry);
 936                 mntget(nd->mnt);
 937                 nd->flags |= LOOKUP_REVAL;
 938                 result = __link_path_walk(name, nd);
 939         }
 940
 941         dput(save.dentry);
 942         mntput(save.mnt);
 943
 944         return result;
 945 }
 946
 947 int fastcall path_walk(const char * name, struct nameidata *nd)
 948 {
 949         current->total_link_count = 0;
 950         return link_path_walk(name, nd);
 951 }
 952
 953 /*
 954  * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
 955  * everything is done. Returns 0 and drops input nd, if lookup failed;
 956  */
 957 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
 958 {
 959         if (path_walk(name, nd))
 960                 return 0;               /* something went wrong... */
 961
 962         if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
 963                 struct dentry *old_dentry = nd->dentry;
 964                 struct vfsmount *old_mnt = nd->mnt;
 965                 struct qstr last = nd->last;
 966                 int last_type = nd->last_type;
 967                 /*
 968                  * NAME was not found in alternate root or it's a directory.  Try to find
 969                  * it in the normal root:
 970                  */
 971                 nd->last_type = LAST_ROOT;
 972                 read_lock(&current->fs->lock);
 973                 nd->mnt = mntget(current->fs->rootmnt);
 974                 nd->dentry = dget(current->fs->root);
 975                 read_unlock(&current->fs->lock);
 976                 if (path_walk(name, nd) == 0) {
 977                         if (nd->dentry->d_inode) {
 978                                 dput(old_dentry);
 979                                 mntput(old_mnt);
 980                                 return 1;
 981                         }
 982                         path_release(nd);
 983                 }
 984                 nd->dentry = old_dentry;
 985                 nd->mnt = old_mnt;
 986                 nd->last = last;
 987                 nd->last_type = last_type;
 988         }
 989         return 1;
 990 }
 991
 992 void set_fs_altroot(void)
 993 {
 994         char *emul = __emul_prefix();
 995         struct nameidata nd;
 996         struct vfsmount *mnt = NULL, *oldmnt;
 997         struct dentry *dentry = NULL, *olddentry;
 998         int err;
 999
1000         if (!emul)
1001                 goto set_it;
1002         err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1003         if (!err) {
1004                 mnt = nd.mnt;
1005                 dentry = nd.dentry;
1006         }
1007 set_it:
1008         write_lock(&current->fs->lock);
1009         oldmnt = current->fs->altrootmnt;
1010         olddentry = current->fs->altroot;
1011         current->fs->altrootmnt = mnt;
1012         current->fs->altroot = dentry;
1013         write_unlock(&current->fs->lock);
1014         if (olddentry) {
1015                 dput(olddentry);
1016                 mntput(oldmnt);
1017         }
1018 }
1019
1020 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1021 int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1022 {
1023         int retval = 0;
1024
1025         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1026         nd->flags = flags;
1027         nd->depth = 0;
1028
1029         read_lock(&current->fs->lock);
1030         if (*name=='/') {
1031                 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1032                         nd->mnt = mntget(current->fs->altrootmnt);
1033                         nd->dentry = dget(current->fs->altroot);
1034                         read_unlock(&current->fs->lock);
1035                         if (__emul_lookup_dentry(name,nd))
1036                                 goto out; /* found in altroot */
1037                         read_lock(&current->fs->lock);
1038                 }
1039                 nd->mnt = mntget(current->fs->rootmnt);
1040                 nd->dentry = dget(current->fs->root);
1041         } else {
1042                 nd->mnt = mntget(current->fs->pwdmnt);
1043                 nd->dentry = dget(current->fs->pwd);
1044         }
1045         read_unlock(&current->fs->lock);
1046         current->total_link_count = 0;
1047         retval = link_path_walk(name, nd);
1048 out:
1049         if (unlikely(current->audit_context
1050                      && nd && nd->dentry && nd->dentry->d_inode))
1051                 audit_inode(name, nd->dentry->d_inode);
1052         return retval;
1053 }
1054
1055 /*
1056  * Restricted form of lookup. Doesn't follow links, single-component only,
1057  * needs parent already locked. Doesn't follow mounts.
1058  * SMP-safe.
1059  */
1060 static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1061 {
1062         struct dentry * dentry;
1063         struct inode *inode;
1064         int err;
1065
1066         inode = base->d_inode;
1067         err = permission(inode, MAY_EXEC, nd);
1068         dentry = ERR_PTR(err);
1069         if (err)
1070                 goto out;
1071
1072         /*
1073          * See if the low-level filesystem might want
1074          * to use its own hash..
1075          */
1076         if (base->d_op && base->d_op->d_hash) {
1077                 err = base->d_op->d_hash(base, name);
1078                 dentry = ERR_PTR(err);
1079                 if (err < 0)
1080                         goto out;
1081         }
1082
1083         dentry = cached_lookup(base, name, nd);
1084         if (!dentry) {
1085                 struct dentry *new = d_alloc(base, name);
1086                 dentry = ERR_PTR(-ENOMEM);
1087                 if (!new)
1088                         goto out;
1089                 dentry = inode->i_op->lookup(inode, new, nd);
1090                 if (!dentry)
1091                         dentry = new;
1092                 else
1093                         dput(new);
1094         }
1095 out:
1096         return dentry;
1097 }
1098
1099 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
1100 {
1101         return __lookup_hash(name, base, NULL);
1102 }
1103
1104 /* SMP-safe */
1105 struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1106 {
1107         unsigned long hash;
1108         struct qstr this;
1109         unsigned int c;
1110
1111         this.name = name;
1112         this.len = len;
1113         if (!len)
1114                 goto access;
1115
1116         hash = init_name_hash();
1117         while (len--) {
1118                 c = *(const unsigned char *)name++;
1119                 if (c == '/' || c == '\0')
1120                         goto access;
1121                 hash = partial_name_hash(c, hash);
1122         }
1123         this.hash = end_name_hash(hash);
1124
1125         return lookup_hash(&this, base);
1126 access:
1127         return ERR_PTR(-EACCES);
1128 }
1129
1130 /*
1131  *      namei()
1132  *
1133  * is used by most simple commands to get the inode of a specified name.
1134  * Open, link etc use their own routines, but this is enough for things
1135  * like 'chmod' etc.
1136  *
1137  * namei exists in two versions: namei/lnamei. The only difference is
1138  * that namei follows links, while lnamei does not.
1139  * SMP-safe
1140  */
1141 int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1142 {
1143         char *tmp = getname(name);
1144         int err = PTR_ERR(tmp);
1145
1146         if (!IS_ERR(tmp)) {
1147                 err = path_lookup(tmp, flags, nd);
1148                 putname(tmp);
1149         }
1150         return err;
1151 }
1152
1153 /*
1154  * It's inline, so penalty for filesystems that don't use sticky bit is
1155  * minimal.
1156  */
1157 static inline int check_sticky(struct inode *dir, struct inode *inode)
1158 {
1159         if (!(dir->i_mode & S_ISVTX))
1160                 return 0;
1161         if (inode->i_uid == current->fsuid)
1162                 return 0;
1163         if (dir->i_uid == current->fsuid)
1164                 return 0;
1165         return !capable(CAP_FOWNER);
1166 }
1167
1168 /*
1169  *      Check whether we can remove a link victim from directory dir, check
1170  *  whether the type of victim is right.
1171  *  1. We can't do it if dir is read-only (done in permission())
1172  *  2. We should have write and exec permissions on dir
1173  *  3. We can't remove anything from append-only dir
1174  *  4. We can't do anything with immutable dir (done in permission())
1175  *  5. If the sticky bit on dir is set we should either
1176  *      a. be owner of dir, or
1177  *      b. be owner of victim, or
1178  *      c. have CAP_FOWNER capability
1179  *  6. If the victim is append-only or immutable we can't do antyhing with
1180  *     links pointing to it.
1181  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1182  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1183  *  9. We can't remove a root or mountpoint.
1184  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1185  *     nfs_async_unlink().
1186  */
1187 static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1188 {
1189         int error;
1190
1191         if (!victim->d_inode)
1192                 return -ENOENT;
1193
1194         BUG_ON(victim->d_parent->d_inode != dir);
1195
1196         error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1197         if (error)
1198                 return error;
1199         if (IS_APPEND(dir))
1200                 return -EPERM;
1201         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1202             IS_IMMUTABLE(victim->d_inode))
1203                 return -EPERM;
1204         if (isdir) {
1205                 if (!S_ISDIR(victim->d_inode->i_mode))
1206                         return -ENOTDIR;
1207                 if (IS_ROOT(victim))
1208                         return -EBUSY;
1209         } else if (S_ISDIR(victim->d_inode->i_mode))
1210                 return -EISDIR;
1211         if (IS_DEADDIR(dir))
1212                 return -ENOENT;
1213         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1214                 return -EBUSY;
1215         return 0;
1216 }
1217
1218 /*      Check whether we can create an object with dentry child in directory
1219  *  dir.
1220  *  1. We can't do it if child already exists (open has special treatment for
1221  *     this case, but since we are inlined it's OK)
1222  *  2. We can't do it if dir is read-only (done in permission())
1223  *  3. We should have write and exec permissions on dir
1224  *  4. We can't do it if dir is immutable (done in permission())
1225  */
1226 static inline int may_create(struct inode *dir, struct dentry *child,
1227                              struct nameidata *nd)
1228 {
1229         if (child->d_inode)
1230                 return -EEXIST;
1231         if (IS_DEADDIR(dir))
1232                 return -ENOENT;
1233         return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1234 }
1235
1236 /*
1237  * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
1238  * reasons.
1239  *
1240  * O_DIRECTORY translates into forcing a directory lookup.
1241  */
1242 static inline int lookup_flags(unsigned int f)
1243 {
1244         unsigned long retval = LOOKUP_FOLLOW;
1245
1246         if (f & O_NOFOLLOW)
1247                 retval &= ~LOOKUP_FOLLOW;
1248
1249         if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1250                 retval &= ~LOOKUP_FOLLOW;
1251
1252         if (f & O_DIRECTORY)
1253                 retval |= LOOKUP_DIRECTORY;
1254
1255         return retval;
1256 }
1257
1258 /*
1259  * p1 and p2 should be directories on the same fs.
1260  */
1261 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1262 {
1263         struct dentry *p;
1264
1265         if (p1 == p2) {
1266                 down(&p1->d_inode->i_sem);
1267                 return NULL;
1268         }
1269
1270         down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1271
1272         for (p = p1; p->d_parent != p; p = p->d_parent) {
1273                 if (p->d_parent == p2) {
1274                         down(&p2->d_inode->i_sem);
1275                         down(&p1->d_inode->i_sem);
1276                         return p;
1277                 }
1278         }
1279
1280         for (p = p2; p->d_parent != p; p = p->d_parent) {
1281                 if (p->d_parent == p1) {
1282                         down(&p1->d_inode->i_sem);
1283                         down(&p2->d_inode->i_sem);
1284                         return p;
1285                 }
1286         }
1287
1288         down(&p1->d_inode->i_sem);
1289         down(&p2->d_inode->i_sem);
1290         return NULL;
1291 }
1292
1293 void unlock_rename(struct dentry *p1, struct dentry *p2)
1294 {
1295         up(&p1->d_inode->i_sem);
1296         if (p1 != p2) {
1297                 up(&p2->d_inode->i_sem);
1298                 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1299         }
1300 }
1301
1302 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1303                 struct nameidata *nd)
1304 {
1305         int error = may_create(dir, dentry, nd);
1306
1307         if (error)
1308                 return error;
1309
1310         if (!dir->i_op || !dir->i_op->create)
1311                 return -EACCES; /* shouldn't it be ENOSYS? */
1312         mode &= S_IALLUGO;
1313         mode |= S_IFREG;
1314         error = security_inode_create(dir, dentry, mode);
1315         if (error)
1316                 return error;
1317         DQUOT_INIT(dir);
1318         error = dir->i_op->create(dir, dentry, mode, nd);
1319         if (!error) {
1320                 fsnotify_create(dir, dentry->d_name.name);
1321                 security_inode_post_create(dir, dentry, mode);
1322         }
1323         return error;
1324 }
1325
1326 int may_open(struct nameidata *nd, int acc_mode, int flag)
1327 {
1328         struct dentry *dentry = nd->dentry;
1329         struct inode *inode = dentry->d_inode;
1330         int error;
1331
1332         if (!inode)
1333                 return -ENOENT;
1334
1335         if (S_ISLNK(inode->i_mode))
1336                 return -ELOOP;
1337
1338         if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1339                 return -EISDIR;
1340
1341         error = permission(inode, acc_mode, nd);
1342         if (error)
1343                 return error;
1344
1345         /*
1346          * FIFO's, sockets and device files are special: they don't
1347          * actually live on the filesystem itself, and as such you
1348          * can write to them even if the filesystem is read-only.
1349          */
1350         if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1351                 flag &= ~O_TRUNC;
1352         } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1353                 if (nd->mnt->mnt_flags & MNT_NODEV)
1354                         return -EACCES;
1355
1356                 flag &= ~O_TRUNC;
1357         } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1358                 return -EROFS;
1359         /*
1360          * An append-only file must be opened in append mode for writing.
1361          */
1362         if (IS_APPEND(inode)) {
1363                 if  ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1364                         return -EPERM;
1365                 if (flag & O_TRUNC)
1366                         return -EPERM;
1367         }
1368
1369         /* O_NOATIME can only be set by the owner or superuser */
1370         if (flag & O_NOATIME)
1371                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1372                         return -EPERM;
1373
1374         /*
1375          * Ensure there are no outstanding leases on the file.
1376          */
1377         error = break_lease(inode, flag);
1378         if (error)
1379                 return error;
1380
1381         if (flag & O_TRUNC) {
1382                 error = get_write_access(inode);
1383                 if (error)
1384                         return error;
1385
1386                 /*
1387                  * Refuse to truncate files with mandatory locks held on them.
1388                  */
1389                 error = locks_verify_locked(inode);
1390                 if (!error) {
1391                         DQUOT_INIT(inode);
1392
1393                         error = do_truncate(dentry, 0);
1394                 }
1395                 put_write_access(inode);
1396                 if (error)
1397                         return error;
1398         } else
1399                 if (flag & FMODE_WRITE)
1400                         DQUOT_INIT(inode);
1401
1402         return 0;
1403 }
1404
1405 /*
1406  *      open_namei()
1407  *
1408  * namei for open - this is in fact almost the whole open-routine.
1409  *
1410  * Note that the low bits of "flag" aren't the same as in the open
1411  * system call - they are 00 - no permissions needed
1412  *                        01 - read permission needed
1413  *                        10 - write permission needed
1414  *                        11 - read/write permissions needed
1415  * which is a lot more logical, and also allows the "no perm" needed
1416  * for symlinks (where the permissions are checked later).
1417  * SMP-safe
1418  */
1419 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1420 {
1421         int acc_mode, error = 0;
1422         struct path path;
1423         struct dentry *dir;
1424         int count = 0;
1425
1426         acc_mode = ACC_MODE(flag);
1427
1428         /* Allow the LSM permission hook to distinguish append
1429            access from general write access. */
1430         if (flag & O_APPEND)
1431                 acc_mode |= MAY_APPEND;
1432
1433         /* Fill in the open() intent data */
1434         nd->intent.open.flags = flag;
1435         nd->intent.open.create_mode = mode;
1436
1437         /*
1438          * The simplest case - just a plain lookup.
1439          */
1440         if (!(flag & O_CREAT)) {
1441                 error = path_lookup(pathname, lookup_flags(flag)|LOOKUP_OPEN, nd);
1442                 if (error)
1443                         return error;
1444                 goto ok;
1445         }
1446
1447         /*
1448          * Create - we need to know the parent.
1449          */
1450         error = path_lookup(pathname, LOOKUP_PARENT|LOOKUP_OPEN|LOOKUP_CREATE, nd);
1451         if (error)
1452                 return error;
1453
1454         /*
1455          * We have the parent and last component. First of all, check
1456          * that we are not asked to creat(2) an obvious directory - that
1457          * will not do.
1458          */
1459         error = -EISDIR;
1460         if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1461                 goto exit;
1462
1463         dir = nd->dentry;
1464         nd->flags &= ~LOOKUP_PARENT;
1465         down(&dir->d_inode->i_sem);
1466         path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1467         path.mnt = nd->mnt;
1468
1469 do_last:
1470         error = PTR_ERR(path.dentry);
1471         if (IS_ERR(path.dentry)) {
1472                 up(&dir->d_inode->i_sem);
1473                 goto exit;
1474         }
1475
1476         /* Negative dentry, just create the file */
1477         if (!path.dentry->d_inode) {
1478                 if (!IS_POSIXACL(dir->d_inode))
1479                         mode &= ~current->fs->umask;
1480                 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
1481                 up(&dir->d_inode->i_sem);
1482                 dput(nd->dentry);
1483                 nd->dentry = path.dentry;
1484                 if (error)
1485                         goto exit;
1486                 /* Don't check for write permission, don't truncate */
1487                 acc_mode = 0;
1488                 flag &= ~O_TRUNC;
1489                 goto ok;
1490         }
1491
1492         /*
1493          * It already exists.
1494          */
1495         up(&dir->d_inode->i_sem);
1496
1497         error = -EEXIST;
1498         if (flag & O_EXCL)
1499                 goto exit_dput;
1500
1501         if (__follow_mount(&path)) {
1502                 error = -ELOOP;
1503                 if (flag & O_NOFOLLOW)
1504                         goto exit_dput;
1505         }
1506         error = -ENOENT;
1507         if (!path.dentry->d_inode)
1508                 goto exit_dput;
1509         if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
1510                 goto do_link;
1511
1512         path_to_nameidata(&path, nd);
1513         error = -EISDIR;
1514         if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
1515                 goto exit;
1516 ok:
1517         error = may_open(nd, acc_mode, flag);
1518         if (error)
1519                 goto exit;
1520         return 0;
1521
1522 exit_dput:
1523         dput_path(&path, nd);
1524 exit:
1525         path_release(nd);
1526         return error;
1527
1528 do_link:
1529         error = -ELOOP;
1530         if (flag & O_NOFOLLOW)
1531                 goto exit_dput;
1532         /*
1533          * This is subtle. Instead of calling do_follow_link() we do the
1534          * thing by hands. The reason is that this way we have zero link_count
1535          * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1536          * After that we have the parent and last component, i.e.
1537          * we are in the same situation as after the first path_walk().
1538          * Well, almost - if the last component is normal we get its copy
1539          * stored in nd->last.name and we will have to putname() it when we
1540          * are done. Procfs-like symlinks just set LAST_BIND.
1541          */
1542         nd->flags |= LOOKUP_PARENT;
1543         error = security_inode_follow_link(path.dentry, nd);
1544         if (error)
1545                 goto exit_dput;
1546         error = __do_follow_link(&path, nd);
1547         if (error)
1548                 return error;
1549         nd->flags &= ~LOOKUP_PARENT;
1550         if (nd->last_type == LAST_BIND)
1551                 goto ok;
1552         error = -EISDIR;
1553         if (nd->last_type != LAST_NORM)
1554                 goto exit;
1555         if (nd->last.name[nd->last.len]) {
1556                 putname(nd->last.name);
1557                 goto exit;
1558         }
1559         error = -ELOOP;
1560         if (count++==32) {
1561                 putname(nd->last.name);
1562                 goto exit;
1563         }
1564         dir = nd->dentry;
1565         down(&dir->d_inode->i_sem);
1566         path.dentry = __lookup_hash(&nd->last, nd->dentry, nd);
1567         path.mnt = nd->mnt;
1568         putname(nd->last.name);
1569         goto do_last;
1570 }
1571
1572 /**
1573  * lookup_create - lookup a dentry, creating it if it doesn't exist
1574  * @nd: nameidata info
1575  * @is_dir: directory flag
1576  *
1577  * Simple function to lookup and return a dentry and create it
1578  * if it doesn't exist.  Is SMP-safe.
1579  *
1580  * Returns with nd->dentry->d_inode->i_sem locked.
1581  */
1582 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1583 {
1584         struct dentry *dentry = ERR_PTR(-EEXIST);
1585
1586         down(&nd->dentry->d_inode->i_sem);
1587         /*
1588          * Yucky last component or no last component at all?
1589          * (foo/., foo/.., /////)
1590          */
1591         if (nd->last_type != LAST_NORM)
1592                 goto fail;
1593         nd->flags &= ~LOOKUP_PARENT;
1594
1595         /*
1596          * Do the final lookup.
1597          */
1598         dentry = lookup_hash(&nd->last, nd->dentry);
1599         if (IS_ERR(dentry))
1600                 goto fail;
1601
1602         /*
1603          * Special case - lookup gave negative, but... we had foo/bar/
1604          * From the vfs_mknod() POV we just have a negative dentry -
1605          * all is fine. Let's be bastards - you had / on the end, you've
1606          * been asking for (non-existent) directory. -ENOENT for you.
1607          */
1608         if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1609                 goto enoent;
1610         return dentry;
1611 enoent:
1612         dput(dentry);
1613         dentry = ERR_PTR(-ENOENT);
1614 fail:
1615         return dentry;
1616 }
1617 EXPORT_SYMBOL_GPL(lookup_create);
1618
1619 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1620 {
1621         int error = may_create(dir, dentry, NULL);
1622
1623         if (error)
1624                 return error;
1625
1626         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1627                 return -EPERM;
1628
1629         if (!dir->i_op || !dir->i_op->mknod)
1630                 return -EPERM;
1631
1632         error = security_inode_mknod(dir, dentry, mode, dev);
1633         if (error)
1634                 return error;
1635
1636         DQUOT_INIT(dir);
1637         error = dir->i_op->mknod(dir, dentry, mode, dev);
1638         if (!error) {
1639                 fsnotify_create(dir, dentry->d_name.name);
1640                 security_inode_post_mknod(dir, dentry, mode, dev);
1641         }
1642         return error;
1643 }
1644
1645 asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1646 {
1647         int error = 0;
1648         char * tmp;
1649         struct dentry * dentry;
1650         struct nameidata nd;
1651
1652         if (S_ISDIR(mode))
1653                 return -EPERM;
1654         tmp = getname(filename);
1655         if (IS_ERR(tmp))
1656                 return PTR_ERR(tmp);
1657
1658         error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1659         if (error)
1660                 goto out;
1661         dentry = lookup_create(&nd, 0);
1662         error = PTR_ERR(dentry);
1663
1664         if (!IS_POSIXACL(nd.dentry->d_inode))
1665                 mode &= ~current->fs->umask;
1666         if (!IS_ERR(dentry)) {
1667                 switch (mode & S_IFMT) {
1668                 case 0: case S_IFREG:
1669                         error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1670                         break;
1671                 case S_IFCHR: case S_IFBLK:
1672                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1673                                         new_decode_dev(dev));
1674                         break;
1675                 case S_IFIFO: case S_IFSOCK:
1676                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1677                         break;
1678                 case S_IFDIR:
1679                         error = -EPERM;
1680                         break;
1681                 default:
1682                         error = -EINVAL;
1683                 }
1684                 dput(dentry);
1685         }
1686         up(&nd.dentry->d_inode->i_sem);
1687         path_release(&nd);
1688 out:
1689         putname(tmp);
1690
1691         return error;
1692 }
1693
1694 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1695 {
1696         int error = may_create(dir, dentry, NULL);
1697
1698         if (error)
1699                 return error;
1700
1701         if (!dir->i_op || !dir->i_op->mkdir)
1702                 return -EPERM;
1703
1704         mode &= (S_IRWXUGO|S_ISVTX);
1705         error = security_inode_mkdir(dir, dentry, mode);
1706         if (error)
1707                 return error;
1708
1709         DQUOT_INIT(dir);
1710         error = dir->i_op->mkdir(dir, dentry, mode);
1711         if (!error) {
1712                 fsnotify_mkdir(dir, dentry->d_name.name);
1713                 security_inode_post_mkdir(dir,dentry, mode);
1714         }
1715         return error;
1716 }
1717
1718 asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1719 {
1720         int error = 0;
1721         char * tmp;
1722
1723         tmp = getname(pathname);
1724         error = PTR_ERR(tmp);
1725         if (!IS_ERR(tmp)) {
1726                 struct dentry *dentry;
1727                 struct nameidata nd;
1728
1729                 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1730                 if (error)
1731                         goto out;
1732                 dentry = lookup_create(&nd, 1);
1733                 error = PTR_ERR(dentry);
1734                 if (!IS_ERR(dentry)) {
1735                         if (!IS_POSIXACL(nd.dentry->d_inode))
1736                                 mode &= ~current->fs->umask;
1737                         error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1738                         dput(dentry);
1739                 }
1740                 up(&nd.dentry->d_inode->i_sem);
1741                 path_release(&nd);
1742 out:
1743                 putname(tmp);
1744         }
1745
1746         return error;
1747 }
1748
1749 /*
1750  * We try to drop the dentry early: we should have
1751  * a usage count of 2 if we're the only user of this
1752  * dentry, and if that is true (possibly after pruning
1753  * the dcache), then we drop the dentry now.
1754  *
1755  * A low-level filesystem can, if it choses, legally
1756  * do a
1757  *
1758  *      if (!d_unhashed(dentry))
1759  *              return -EBUSY;
1760  *
1761  * if it cannot handle the case of removing a directory
1762  * that is still in use by something else..
1763  */
1764 void dentry_unhash(struct dentry *dentry)
1765 {
1766         dget(dentry);
1767         if (atomic_read(&dentry->d_count))
1768                 shrink_dcache_parent(dentry);
1769         spin_lock(&dcache_lock);
1770         spin_lock(&dentry->d_lock);
1771         if (atomic_read(&dentry->d_count) == 2)
1772                 __d_drop(dentry);
1773         spin_unlock(&dentry->d_lock);
1774         spin_unlock(&dcache_lock);
1775 }
1776
1777 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1778 {
1779         int error = may_delete(dir, dentry, 1);
1780
1781         if (error)
1782                 return error;
1783
1784         if (!dir->i_op || !dir->i_op->rmdir)
1785                 return -EPERM;
1786
1787         DQUOT_INIT(dir);
1788
1789         down(&dentry->d_inode->i_sem);
1790         dentry_unhash(dentry);
1791         if (d_mountpoint(dentry))
1792                 error = -EBUSY;
1793         else {
1794                 error = security_inode_rmdir(dir, dentry);
1795                 if (!error) {
1796                         error = dir->i_op->rmdir(dir, dentry);
1797                         if (!error)
1798                                 dentry->d_inode->i_flags |= S_DEAD;
1799                 }
1800         }
1801         up(&dentry->d_inode->i_sem);
1802         if (!error) {
1803                 d_delete(dentry);
1804         }
1805         dput(dentry);
1806
1807         return error;
1808 }
1809
1810 asmlinkage long sys_rmdir(const char __user * pathname)
1811 {
1812         int error = 0;
1813         char * name;
1814         struct dentry *dentry;
1815         struct nameidata nd;
1816
1817         name = getname(pathname);
1818         if(IS_ERR(name))
1819                 return PTR_ERR(name);
1820
1821         error = path_lookup(name, LOOKUP_PARENT, &nd);
1822         if (error)
1823                 goto exit;
1824
1825         switch(nd.last_type) {
1826                 case LAST_DOTDOT:
1827                         error = -ENOTEMPTY;
1828                         goto exit1;
1829                 case LAST_DOT:
1830                         error = -EINVAL;
1831                         goto exit1;
1832                 case LAST_ROOT:
1833                         error = -EBUSY;
1834                         goto exit1;
1835         }
1836         down(&nd.dentry->d_inode->i_sem);
1837         dentry = lookup_hash(&nd.last, nd.dentry);
1838         error = PTR_ERR(dentry);
1839         if (!IS_ERR(dentry)) {
1840                 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1841                 dput(dentry);
1842         }
1843         up(&nd.dentry->d_inode->i_sem);
1844 exit1:
1845         path_release(&nd);
1846 exit:
1847         putname(name);
1848         return error;
1849 }
1850
1851 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1852 {
1853         int error = may_delete(dir, dentry, 0);
1854
1855         if (error)
1856                 return error;
1857
1858         if (!dir->i_op || !dir->i_op->unlink)
1859                 return -EPERM;
1860
1861         DQUOT_INIT(dir);
1862
1863         down(&dentry->d_inode->i_sem);
1864         if (d_mountpoint(dentry))
1865                 error = -EBUSY;
1866         else {
1867                 error = security_inode_unlink(dir, dentry);
1868                 if (!error)
1869                         error = dir->i_op->unlink(dir, dentry);
1870         }
1871         up(&dentry->d_inode->i_sem);
1872
1873         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1874         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1875                 d_delete(dentry);
1876         }
1877
1878         return error;
1879 }
1880
1881 /*
1882  * Make sure that the actual truncation of the file will occur outside its
1883  * directory's i_sem.  Truncate can take a long time if there is a lot of
1884  * writeout happening, and we don't want to prevent access to the directory
1885  * while waiting on the I/O.
1886  */
1887 asmlinkage long sys_unlink(const char __user * pathname)
1888 {
1889         int error = 0;
1890         char * name;
1891         struct dentry *dentry;
1892         struct nameidata nd;
1893         struct inode *inode = NULL;
1894
1895         name = getname(pathname);
1896         if(IS_ERR(name))
1897                 return PTR_ERR(name);
1898
1899         error = path_lookup(name, LOOKUP_PARENT, &nd);
1900         if (error)
1901                 goto exit;
1902         error = -EISDIR;
1903         if (nd.last_type != LAST_NORM)
1904                 goto exit1;
1905         down(&nd.dentry->d_inode->i_sem);
1906         dentry = lookup_hash(&nd.last, nd.dentry);
1907         error = PTR_ERR(dentry);
1908         if (!IS_ERR(dentry)) {
1909                 /* Why not before? Because we want correct error value */
1910                 if (nd.last.name[nd.last.len])
1911                         goto slashes;
1912                 inode = dentry->d_inode;
1913                 if (inode)
1914                         atomic_inc(&inode->i_count);
1915                 error = vfs_unlink(nd.dentry->d_inode, dentry);
1916         exit2:
1917                 dput(dentry);
1918         }
1919         up(&nd.dentry->d_inode->i_sem);
1920         if (inode)
1921                 iput(inode);    /* truncate the inode here */
1922 exit1:
1923         path_release(&nd);
1924 exit:
1925         putname(name);
1926         return error;
1927
1928 slashes:
1929         error = !dentry->d_inode ? -ENOENT :
1930                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1931         goto exit2;
1932 }
1933
1934 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
1935 {
1936         int error = may_create(dir, dentry, NULL);
1937
1938         if (error)
1939                 return error;
1940
1941         if (!dir->i_op || !dir->i_op->symlink)
1942                 return -EPERM;
1943
1944         error = security_inode_symlink(dir, dentry, oldname);
1945         if (error)
1946                 return error;
1947
1948         DQUOT_INIT(dir);
1949         error = dir->i_op->symlink(dir, dentry, oldname);
1950         if (!error) {
1951                 fsnotify_create(dir, dentry->d_name.name);
1952                 security_inode_post_symlink(dir, dentry, oldname);
1953         }
1954         return error;
1955 }
1956
1957 asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
1958 {
1959         int error = 0;
1960         char * from;
1961         char * to;
1962
1963         from = getname(oldname);
1964         if(IS_ERR(from))
1965                 return PTR_ERR(from);
1966         to = getname(newname);
1967         error = PTR_ERR(to);
1968         if (!IS_ERR(to)) {
1969                 struct dentry *dentry;
1970                 struct nameidata nd;
1971
1972                 error = path_lookup(to, LOOKUP_PARENT, &nd);
1973                 if (error)
1974                         goto out;
1975                 dentry = lookup_create(&nd, 0);
1976                 error = PTR_ERR(dentry);
1977                 if (!IS_ERR(dentry)) {
1978                         error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
1979                         dput(dentry);
1980                 }
1981                 up(&nd.dentry->d_inode->i_sem);
1982                 path_release(&nd);
1983 out:
1984                 putname(to);
1985         }
1986         putname(from);
1987         return error;
1988 }
1989
1990 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1991 {
1992         struct inode *inode = old_dentry->d_inode;
1993         int error;
1994
1995         if (!inode)
1996                 return -ENOENT;
1997
1998         error = may_create(dir, new_dentry, NULL);
1999         if (error)
2000                 return error;
2001
2002         if (dir->i_sb != inode->i_sb)
2003                 return -EXDEV;
2004
2005         /*
2006          * A link to an append-only or immutable file cannot be created.
2007          */
2008         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2009                 return -EPERM;
2010         if (!dir->i_op || !dir->i_op->link)
2011                 return -EPERM;
2012         if (S_ISDIR(old_dentry->d_inode->i_mode))
2013                 return -EPERM;
2014
2015         error = security_inode_link(old_dentry, dir, new_dentry);
2016         if (error)
2017                 return error;
2018
2019         down(&old_dentry->d_inode->i_sem);
2020         DQUOT_INIT(dir);
2021         error = dir->i_op->link(old_dentry, dir, new_dentry);
2022         up(&old_dentry->d_inode->i_sem);
2023         if (!error) {
2024                 fsnotify_create(dir, new_dentry->d_name.name);
2025                 security_inode_post_link(old_dentry, dir, new_dentry);
2026         }
2027         return error;
2028 }
2029
2030 /*
2031  * Hardlinks are often used in delicate situations.  We avoid
2032  * security-related surprises by not following symlinks on the
2033  * newname.  --KAB
2034  *
2035  * We don't follow them on the oldname either to be compatible
2036  * with linux 2.0, and to avoid hard-linking to directories
2037  * and other special files.  --ADM
2038  */
2039 asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2040 {
2041         struct dentry *new_dentry;
2042         struct nameidata nd, old_nd;
2043         int error;
2044         char * to;
2045
2046         to = getname(newname);
2047         if (IS_ERR(to))
2048                 return PTR_ERR(to);
2049
2050         error = __user_walk(oldname, 0, &old_nd);
2051         if (error)
2052                 goto exit;
2053         error = path_lookup(to, LOOKUP_PARENT, &nd);
2054         if (error)
2055                 goto out;
2056         error = -EXDEV;
2057         if (old_nd.mnt != nd.mnt)
2058                 goto out_release;
2059         new_dentry = lookup_create(&nd, 0);
2060         error = PTR_ERR(new_dentry);
2061         if (!IS_ERR(new_dentry)) {
2062                 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2063                 dput(new_dentry);
2064         }
2065         up(&nd.dentry->d_inode->i_sem);
2066 out_release:
2067         path_release(&nd);
2068 out:
2069         path_release(&old_nd);
2070 exit:
2071         putname(to);
2072
2073         return error;
2074 }
2075
2076 /*
2077  * The worst of all namespace operations - renaming directory. "Perverted"
2078  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2079  * Problems:
2080  *      a) we can get into loop creation. Check is done in is_subdir().
2081  *      b) race potential - two innocent renames can create a loop together.
2082  *         That's where 4.4 screws up. Current fix: serialization on
2083  *         sb->s_vfs_rename_sem. We might be more accurate, but that's another
2084  *         story.
2085  *      c) we have to lock _three_ objects - parents and victim (if it exists).
2086  *         And that - after we got ->i_sem on parents (until then we don't know
2087  *         whether the target exists).  Solution: try to be smart with locking
2088  *         order for inodes.  We rely on the fact that tree topology may change
2089  *         only under ->s_vfs_rename_sem _and_ that parent of the object we
2090  *         move will be locked.  Thus we can rank directories by the tree
2091  *         (ancestors first) and rank all non-directories after them.
2092  *         That works since everybody except rename does "lock parent, lookup,
2093  *         lock child" and rename is under ->s_vfs_rename_sem.
2094  *         HOWEVER, it relies on the assumption that any object with ->lookup()
2095  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
2096  *         we'd better make sure that there's no link(2) for them.
2097  *      d) some filesystems don't support opened-but-unlinked directories,
2098  *         either because of layout or because they are not ready to deal with
2099  *         all cases correctly. The latter will be fixed (taking this sort of
2100  *         stuff into VFS), but the former is not going away. Solution: the same
2101  *         trick as in rmdir().
2102  *      e) conversion from fhandle to dentry may come in the wrong moment - when
2103  *         we are removing the target. Solution: we will have to grab ->i_sem
2104  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2105  *         ->i_sem on parents, which works but leads to some truely excessive
2106  *         locking].
2107  */
2108 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2109                           struct inode *new_dir, struct dentry *new_dentry)
2110 {
2111         int error = 0;
2112         struct inode *target;
2113
2114         /*
2115          * If we are going to change the parent - check write permissions,
2116          * we'll need to flip '..'.
2117          */
2118         if (new_dir != old_dir) {
2119                 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2120                 if (error)
2121                         return error;
2122         }
2123
2124         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2125         if (error)
2126                 return error;
2127
2128         target = new_dentry->d_inode;
2129         if (target) {
2130                 down(&target->i_sem);
2131                 dentry_unhash(new_dentry);
2132         }
2133         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2134                 error = -EBUSY;
2135         else
2136                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2137         if (target) {
2138                 if (!error)
2139                         target->i_flags |= S_DEAD;
2140                 up(&target->i_sem);
2141                 if (d_unhashed(new_dentry))
2142                         d_rehash(new_dentry);
2143                 dput(new_dentry);
2144         }
2145         if (!error) {
2146                 d_move(old_dentry,new_dentry);
2147                 security_inode_post_rename(old_dir, old_dentry,
2148                                            new_dir, new_dentry);
2149         }
2150         return error;
2151 }
2152
2153 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2154                             struct inode *new_dir, struct dentry *new_dentry)
2155 {
2156         struct inode *target;
2157         int error;
2158
2159         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2160         if (error)
2161                 return error;
2162
2163         dget(new_dentry);
2164         target = new_dentry->d_inode;
2165         if (target)
2166                 down(&target->i_sem);
2167         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2168                 error = -EBUSY;
2169         else
2170                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2171         if (!error) {
2172                 /* The following d_move() should become unconditional */
2173                 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2174                         d_move(old_dentry, new_dentry);
2175                 security_inode_post_rename(old_dir, old_dentry, new_dir, new_dentry);
2176         }
2177         if (target)
2178                 up(&target->i_sem);
2179         dput(new_dentry);
2180         return error;
2181 }
2182
2183 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2184                struct inode *new_dir, struct dentry *new_dentry)
2185 {
2186         int error;
2187         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
2188         const char *old_name;
2189
2190         if (old_dentry->d_inode == new_dentry->d_inode)
2191                 return 0;
2192
2193         error = may_delete(old_dir, old_dentry, is_dir);
2194         if (error)
2195                 return error;
2196
2197         if (!new_dentry->d_inode)
2198                 error = may_create(new_dir, new_dentry, NULL);
2199         else
2200                 error = may_delete(new_dir, new_dentry, is_dir);
2201         if (error)
2202                 return error;
2203
2204         if (!old_dir->i_op || !old_dir->i_op->rename)
2205                 return -EPERM;
2206
2207         DQUOT_INIT(old_dir);
2208         DQUOT_INIT(new_dir);
2209
2210         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2211
2212         if (is_dir)
2213                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2214         else
2215                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2216         if (!error) {
2217                 const char *new_name = old_dentry->d_name.name;
2218                 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2219                               new_dentry->d_inode, old_dentry->d_inode);
2220         }
2221         fsnotify_oldname_free(old_name);
2222
2223         return error;
2224 }
2225
2226 static inline int do_rename(const char * oldname, const char * newname)
2227 {
2228         int error = 0;
2229         struct dentry * old_dir, * new_dir;
2230         struct dentry * old_dentry, *new_dentry;
2231         struct dentry * trap;
2232         struct nameidata oldnd, newnd;
2233
2234         error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2235         if (error)
2236                 goto exit;
2237
2238         error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2239         if (error)
2240                 goto exit1;
2241
2242         error = -EXDEV;
2243         if (oldnd.mnt != newnd.mnt)
2244                 goto exit2;
2245
2246         old_dir = oldnd.dentry;
2247         error = -EBUSY;
2248         if (oldnd.last_type != LAST_NORM)
2249                 goto exit2;
2250
2251         new_dir = newnd.dentry;
2252         if (newnd.last_type != LAST_NORM)
2253                 goto exit2;
2254
2255         trap = lock_rename(new_dir, old_dir);
2256
2257         old_dentry = lookup_hash(&oldnd.last, old_dir);
2258         error = PTR_ERR(old_dentry);
2259         if (IS_ERR(old_dentry))
2260                 goto exit3;
2261         /* source must exist */
2262         error = -ENOENT;
2263         if (!old_dentry->d_inode)
2264                 goto exit4;
2265         /* unless the source is a directory trailing slashes give -ENOTDIR */
2266         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2267                 error = -ENOTDIR;
2268                 if (oldnd.last.name[oldnd.last.len])
2269                         goto exit4;
2270                 if (newnd.last.name[newnd.last.len])
2271                         goto exit4;
2272         }
2273         /* source should not be ancestor of target */
2274         error = -EINVAL;
2275         if (old_dentry == trap)
2276                 goto exit4;
2277         new_dentry = lookup_hash(&newnd.last, new_dir);
2278         error = PTR_ERR(new_dentry);
2279         if (IS_ERR(new_dentry))
2280                 goto exit4;
2281         /* target should not be an ancestor of source */
2282         error = -ENOTEMPTY;
2283         if (new_dentry == trap)
2284                 goto exit5;
2285
2286         error = vfs_rename(old_dir->d_inode, old_dentry,
2287                                    new_dir->d_inode, new_dentry);
2288 exit5:
2289         dput(new_dentry);
2290 exit4:
2291         dput(old_dentry);
2292 exit3:
2293         unlock_rename(new_dir, old_dir);
2294 exit2:
2295         path_release(&newnd);
2296 exit1:
2297         path_release(&oldnd);
2298 exit:
2299         return error;
2300 }
2301
2302 asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2303 {
2304         int error;
2305         char * from;
2306         char * to;
2307
2308         from = getname(oldname);
2309         if(IS_ERR(from))
2310                 return PTR_ERR(from);
2311         to = getname(newname);
2312         error = PTR_ERR(to);
2313         if (!IS_ERR(to)) {
2314                 error = do_rename(from,to);
2315                 putname(to);
2316         }
2317         putname(from);
2318         return error;
2319 }
2320
2321 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2322 {
2323         int len;
2324
2325         len = PTR_ERR(link);
2326         if (IS_ERR(link))
2327                 goto out;
2328
2329         len = strlen(link);
2330         if (len > (unsigned) buflen)
2331                 len = buflen;
2332         if (copy_to_user(buffer, link, len))
2333                 len = -EFAULT;
2334 out:
2335         return len;
2336 }
2337
2338 /*
2339  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
2340  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
2341  * using) it for any given inode is up to filesystem.
2342  */
2343 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2344 {
2345         struct nameidata nd;
2346         void *cookie;
2347
2348         nd.depth = 0;
2349         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2350         if (!IS_ERR(cookie)) {
2351                 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
2352                 if (dentry->d_inode->i_op->put_link)
2353                         dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2354                 cookie = ERR_PTR(res);
2355         }
2356         return PTR_ERR(cookie);
2357 }
2358
2359 int vfs_follow_link(struct nameidata *nd, const char *link)
2360 {
2361         return __vfs_follow_link(nd, link);
2362 }
2363
2364 /* get the link contents into pagecache */
2365 static char *page_getlink(struct dentry * dentry, struct page **ppage)
2366 {
2367         struct page * page;
2368         struct address_space *mapping = dentry->d_inode->i_mapping;
2369         page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2370                                 NULL);
2371         if (IS_ERR(page))
2372                 goto sync_fail;
2373         wait_on_page_locked(page);
2374         if (!PageUptodate(page))
2375                 goto async_fail;
2376         *ppage = page;
2377         return kmap(page);
2378
2379 async_fail:
2380         page_cache_release(page);
2381         return ERR_PTR(-EIO);
2382
2383 sync_fail:
2384         return (char*)page;
2385 }
2386
2387 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2388 {
2389         struct page *page = NULL;
2390         char *s = page_getlink(dentry, &page);
2391         int res = vfs_readlink(dentry,buffer,buflen,s);
2392         if (page) {
2393                 kunmap(page);
2394                 page_cache_release(page);
2395         }
2396         return res;
2397 }
2398
2399 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
2400 {
2401         struct page *page = NULL;
2402         nd_set_link(nd, page_getlink(dentry, &page));
2403         return page;
2404 }
2405
2406 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2407 {
2408         struct page *page = cookie;
2409
2410         if (page) {
2411                 kunmap(page);
2412                 page_cache_release(page);
2413         }
2414 }
2415
2416 int page_symlink(struct inode *inode, const char *symname, int len)
2417 {
2418         struct address_space *mapping = inode->i_mapping;
2419         struct page *page = grab_cache_page(mapping, 0);
2420         int err = -ENOMEM;
2421         char *kaddr;
2422
2423         if (!page)
2424                 goto fail;
2425         err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2426         if (err)
2427                 goto fail_map;
2428         kaddr = kmap_atomic(page, KM_USER0);
2429         memcpy(kaddr, symname, len-1);
2430         kunmap_atomic(kaddr, KM_USER0);
2431         mapping->a_ops->commit_write(NULL, page, 0, len-1);
2432         /*
2433          * Notice that we are _not_ going to block here - end of page is
2434          * unmapped, so this will only try to map the rest of page, see
2435          * that it is unmapped (typically even will not look into inode -
2436          * ->i_size will be enough for everything) and zero it out.
2437          * OTOH it's obviously correct and should make the page up-to-date.
2438          */
2439         if (!PageUptodate(page)) {
2440                 err = mapping->a_ops->readpage(NULL, page);
2441                 wait_on_page_locked(page);
2442         } else {
2443                 unlock_page(page);
2444         }
2445         page_cache_release(page);
2446         if (err < 0)
2447                 goto fail;
2448         mark_inode_dirty(inode);
2449         return 0;
2450 fail_map:
2451         unlock_page(page);
2452         page_cache_release(page);
2453 fail:
2454         return err;
2455 }
2456
2457 struct inode_operations page_symlink_inode_operations = {
2458         .readlink       = generic_readlink,
2459         .follow_link    = page_follow_link_light,
2460         .put_link       = page_put_link,
2461 };
2462
2463 EXPORT_SYMBOL(__user_walk);
2464 EXPORT_SYMBOL(follow_down);
2465 EXPORT_SYMBOL(follow_up);
2466 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2467 EXPORT_SYMBOL(getname);
2468 EXPORT_SYMBOL(lock_rename);
2469 EXPORT_SYMBOL(lookup_hash);
2470 EXPORT_SYMBOL(lookup_one_len);
2471 EXPORT_SYMBOL(page_follow_link_light);
2472 EXPORT_SYMBOL(page_put_link);
2473 EXPORT_SYMBOL(page_readlink);
2474 EXPORT_SYMBOL(page_symlink);
2475 EXPORT_SYMBOL(page_symlink_inode_operations);
2476 EXPORT_SYMBOL(path_lookup);
2477 EXPORT_SYMBOL(path_release);
2478 EXPORT_SYMBOL(path_walk);
2479 EXPORT_SYMBOL(permission);
2480 EXPORT_SYMBOL(unlock_rename);
2481 EXPORT_SYMBOL(vfs_create);
2482 EXPORT_SYMBOL(vfs_follow_link);
2483 EXPORT_SYMBOL(vfs_link);
2484 EXPORT_SYMBOL(vfs_mkdir);
2485 EXPORT_SYMBOL(vfs_mknod);
2486 EXPORT_SYMBOL(generic_permission);
2487 EXPORT_SYMBOL(vfs_readlink);
2488 EXPORT_SYMBOL(vfs_rename);
2489 EXPORT_SYMBOL(vfs_rmdir);
2490 EXPORT_SYMBOL(vfs_symlink);
2491 EXPORT_SYMBOL(vfs_unlink);
2492 EXPORT_SYMBOL(dentry_unhash);
2493 EXPORT_SYMBOL(generic_readlink);