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/mm.h>
  18 #include <linux/proc_fs.h>
  19 #include <linux/smp_lock.h>
  20 #include <linux/quotaops.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/dcache.h>
  23
  24 #include <asm/uaccess.h>
  25 #include <asm/unaligned.h>
  26 #include <asm/semaphore.h>
  27 #include <asm/page.h>
  28 #include <asm/pgtable.h>
  29
  30 #include <asm/namei.h>
  31
  32 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
  33
  34 /* [Feb-1997 T. Schoebel-Theuer]
  35  * Fundamental changes in the pathname lookup mechanisms (namei)
  36  * were necessary because of omirr.  The reason is that omirr needs
  37  * to know the _real_ pathname, not the user-supplied one, in case
  38  * of symlinks (and also when transname replacements occur).
  39  *
  40  * The new code replaces the old recursive symlink resolution with
  41  * an iterative one (in case of non-nested symlink chains).  It does
  42  * this with calls to <fs>_follow_link().
  43  * As a side effect, dir_namei(), _namei() and follow_link() are now
  44  * replaced with a single function lookup_dentry() that can handle all
  45  * the special cases of the former code.
  46  *
  47  * With the new dcache, the pathname is stored at each inode, at least as
  48  * long as the refcount of the inode is positive.  As a side effect, the
  49  * size of the dcache depends on the inode cache and thus is dynamic.
  50  *
  51  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  52  * resolution to correspond with current state of the code.
  53  *
  54  * Note that the symlink resolution is not *completely* iterative.
  55  * There is still a significant amount of tail- and mid- recursion in
  56  * the algorithm.  Also, note that <fs>_readlink() is not used in
  57  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  58  * may return different results than <fs>_follow_link().  Many virtual
  59  * filesystems (including /proc) exhibit this behavior.
  60  */
  61
  62 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  63  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  64  * and the name already exists in form of a symlink, try to create the new
  65  * name indicated by the symlink. The old code always complained that the
  66  * name already exists, due to not following the symlink even if its target
  67  * is nonexistent.  The new semantics affects also mknod() and link() when
  68  * the name is a symlink pointing to a non-existant name.
  69  *
  70  * I don't know which semantics is the right one, since I have no access
  71  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  72  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  73  * "old" one. Personally, I think the new semantics is much more logical.
  74  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  75  * file does succeed in both HP-UX and SunOs, but not in Solaris
  76  * and in the old Linux semantics.
  77  */
  78
  79 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  80  * semantics.  See the comments in "open_namei" and "do_link" below.
  81  *
  82  * [10-Sep-98 Alan Modra] Another symlink change.
  83  */
  84
  85 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  86  *      inside the path - always follow.
  87  *      in the last component in creation/removal/renaming - never follow.
  88  *      if LOOKUP_FOLLOW passed - follow.
  89  *      if the pathname has trailing slashes - follow.
  90  *      otherwise - don't follow.
  91  * (applied in that order).
  92  *
  93  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
  94  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
  95  * During the 2.4 we need to fix the userland stuff depending on it -
  96  * hopefully we will be able to get rid of that wart in 2.5. So far only
  97  * XEmacs seems to be relying on it...
  98  */
  99
 100 /* In order to reduce some races, while at the same time doing additional
 101  * checking and hopefully speeding things up, we copy filenames to the
 102  * kernel data space before using them..
 103  *
 104  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 105  */
 106 static inline int do_getname(const char *filename, char *page)
 107 {
 108         int retval;
 109         unsigned long len = PAGE_SIZE;
 110
 111         if ((unsigned long) filename >= TASK_SIZE) {
 112                 if (!segment_eq(get_fs(), KERNEL_DS))
 113                         return -EFAULT;
 114         } else if (TASK_SIZE - (unsigned long) filename < PAGE_SIZE)
 115                 len = TASK_SIZE - (unsigned long) filename;
 116
 117         retval = strncpy_from_user((char *)page, filename, len);
 118         if (retval > 0) {
 119                 if (retval < len)
 120                         return 0;
 121                 return -ENAMETOOLONG;
 122         } else if (!retval)
 123                 retval = -ENOENT;
 124         return retval;
 125 }
 126
 127 char * getname(const char * filename)
 128 {
 129         char *tmp, *result;
 130
 131         result = ERR_PTR(-ENOMEM);
 132         tmp = __getname();
 133         if (tmp)  {
 134                 int retval = do_getname(filename, tmp);
 135
 136                 result = tmp;
 137                 if (retval < 0) {
 138                         putname(tmp);
 139                         result = ERR_PTR(retval);
 140                 }
 141         }
 142         return result;
 143 }
 144
 145 /*
 146  *      permission()
 147  *
 148  * is used to check for read/write/execute permissions on a file.
 149  * We use "fsuid" for this, letting us set arbitrary permissions
 150  * for filesystem access without changing the "normal" uids which
 151  * are used for other things..
 152  */
 153 int permission(struct inode * inode,int mask)
 154 {
 155         int mode = inode->i_mode;
 156
 157         if (inode->i_op && inode->i_op->permission) {
 158                 int retval;
 159                 lock_kernel();
 160                 retval = inode->i_op->permission(inode, mask);
 161                 unlock_kernel();
 162                 return retval;
 163         }
 164
 165         if ((mask & S_IWOTH) && IS_RDONLY(inode) &&
 166                  (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 167                 return -EROFS; /* Nobody gets write access to a read-only fs */
 168
 169         if ((mask & S_IWOTH) && IS_IMMUTABLE(inode))
 170                 return -EACCES; /* Nobody gets write access to an immutable file */
 171
 172         if (current->fsuid == inode->i_uid)
 173                 mode >>= 6;
 174         else if (in_group_p(inode->i_gid))
 175                 mode >>= 3;
 176
 177         if (((mode & mask & S_IRWXO) == mask) || capable(CAP_DAC_OVERRIDE))
 178                 return 0;
 179
 180         /* read and search access */
 181         if ((mask == S_IROTH) ||
 182             (S_ISDIR(mode)  && !(mask & ~(S_IROTH | S_IXOTH))))
 183                 if (capable(CAP_DAC_READ_SEARCH))
 184                         return 0;
 185
 186         return -EACCES;
 187 }
 188
 189 /*
 190  * get_write_access() gets write permission for a file.
 191  * put_write_access() releases this write permission.
 192  * This is used for regular files.
 193  * We cannot support write (and maybe mmap read-write shared) accesses and
 194  * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
 195  * can have the following values:
 196  * 0: no writers, no VM_DENYWRITE mappings
 197  * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
 198  * > 0: (i_writecount) users are writing to the file.
 199  *
 200  * Normally we operate on that counter with atomic_{inc,dec} and it's safe
 201  * except for the cases where we don't hold i_writecount yet. Then we need to
 202  * use {get,deny}_write_access() - these functions check the sign and refuse
 203  * to do the change if sign is wrong. Exclusion between them is provided by
 204  * spinlock (arbitration_lock) and I'll rip the second arsehole to the first
 205  * who will try to move it in struct inode - just leave it here.
 206  */
 207 static spinlock_t arbitration_lock = SPIN_LOCK_UNLOCKED;
 208 int get_write_access(struct inode * inode)
 209 {
 210         spin_lock(&arbitration_lock);
 211         if (atomic_read(&inode->i_writecount) < 0) {
 212                 spin_unlock(&arbitration_lock);
 213                 return -ETXTBSY;
 214         }
 215         atomic_inc(&inode->i_writecount);
 216         spin_unlock(&arbitration_lock);
 217         return 0;
 218 }
 219 int deny_write_access(struct file * file)
 220 {
 221         spin_lock(&arbitration_lock);
 222         if (atomic_read(&file->f_dentry->d_inode->i_writecount) > 0) {
 223                 spin_unlock(&arbitration_lock);
 224                 return -ETXTBSY;
 225         }
 226         atomic_dec(&file->f_dentry->d_inode->i_writecount);
 227         spin_unlock(&arbitration_lock);
 228         return 0;
 229 }
 230
 231 void path_release(struct nameidata *nd)
 232 {
 233         dput(nd->dentry);
 234         mntput(nd->mnt);
 235 }
 236
 237 /*
 238  * Internal lookup() using the new generic dcache.
 239  * SMP-safe
 240  */
 241 static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, int flags)
 242 {
 243         struct dentry * dentry = d_lookup(parent, name);
 244
 245         if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
 246                 if (!dentry->d_op->d_revalidate(dentry, flags) && !d_invalidate(dentry)) {
 247                         dput(dentry);
 248                         dentry = NULL;
 249                 }
 250         }
 251         return dentry;
 252 }
 253
 254 /*
 255  * This is called when everything else fails, and we actually have
 256  * to go to the low-level filesystem to find out what we should do..
 257  *
 258  * We get the directory semaphore, and after getting that we also
 259  * make sure that nobody added the entry to the dcache in the meantime..
 260  * SMP-safe
 261  */
 262 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, int flags)
 263 {
 264         struct dentry * result;
 265         struct inode *dir = parent->d_inode;
 266
 267         down(&dir->i_sem);
 268         /*
 269          * First re-do the cached lookup just in case it was created
 270          * while we waited for the directory semaphore..
 271          *
 272          * FIXME! This could use version numbering or similar to
 273          * avoid unnecessary cache lookups.
 274          */
 275         result = d_lookup(parent, name);
 276         if (!result) {
 277                 struct dentry * dentry = d_alloc(parent, name);
 278                 result = ERR_PTR(-ENOMEM);
 279                 if (dentry) {
 280                         lock_kernel();
 281                         result = dir->i_op->lookup(dir, dentry);
 282                         unlock_kernel();
 283                         if (result)
 284                                 dput(dentry);
 285                         else
 286                                 result = dentry;
 287                 }
 288                 up(&dir->i_sem);
 289                 return result;
 290         }
 291
 292         /*
 293          * Uhhuh! Nasty case: the cache was re-populated while
 294          * we waited on the semaphore. Need to revalidate, but
 295          * we're going to return this entry regardless (same
 296          * as if it was busy).
 297          */
 298         up(&dir->i_sem);
 299         if (result->d_op && result->d_op->d_revalidate)
 300                 result->d_op->d_revalidate(result, flags);
 301         return result;
 302 }
 303
 304 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
 305 {
 306         int err;
 307         if (current->link_count >= 32)
 308                 goto loop;
 309         current->link_count++;
 310         UPDATE_ATIME(dentry->d_inode);
 311         err = dentry->d_inode->i_op->follow_link(dentry, nd);
 312         current->link_count--;
 313         return err;
 314 loop:
 315         path_release(nd);
 316         return -ELOOP;
 317 }
 318
 319 static inline int __follow_up(struct vfsmount **mnt, struct dentry **base)
 320 {
 321         struct vfsmount *parent;
 322         struct dentry *dentry;
 323         spin_lock(&dcache_lock);
 324         parent=(*mnt)->mnt_parent;
 325         if (parent == *mnt) {
 326                 spin_unlock(&dcache_lock);
 327                 return 0;
 328         }
 329         mntget(parent);
 330         dentry=dget((*mnt)->mnt_mountpoint);
 331         spin_unlock(&dcache_lock);
 332         dput(*base);
 333         *base = dentry;
 334         mntput(*mnt);
 335         *mnt = parent;
 336         return 1;
 337 }
 338
 339 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 340 {
 341         return __follow_up(mnt, dentry);
 342 }
 343
 344 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
 345 {
 346         struct list_head *p;
 347         spin_lock(&dcache_lock);
 348         p = (*dentry)->d_vfsmnt.next;
 349         while (p != &(*dentry)->d_vfsmnt) {
 350                 struct vfsmount *tmp;
 351                 tmp = list_entry(p, struct vfsmount, mnt_clash);
 352                 if (tmp->mnt_parent == *mnt) {
 353                         *mnt = mntget(tmp);
 354                         spin_unlock(&dcache_lock);
 355                         mntput(tmp->mnt_parent);
 356                         /* tmp holds the mountpoint, so... */
 357                         dput(*dentry);
 358                         *dentry = dget(tmp->mnt_root);
 359                         return 1;
 360                 }
 361                 p = p->next;
 362         }
 363         spin_unlock(&dcache_lock);
 364         return 0;
 365 }
 366
 367 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 368 {
 369         return __follow_down(mnt,dentry);
 370 }
 371
 372 static inline void follow_dotdot(struct nameidata *nd)
 373 {
 374         while(1) {
 375                 struct vfsmount *parent;
 376                 struct dentry *dentry;
 377                 read_lock(&current->fs->lock);
 378                 if (nd->dentry == current->fs->root &&
 379                     nd->mnt == current->fs->rootmnt)  {
 380                         read_unlock(&current->fs->lock);
 381                         break;
 382                 }
 383                 read_unlock(&current->fs->lock);
 384                 spin_lock(&dcache_lock);
 385                 if (nd->dentry != nd->mnt->mnt_root) {
 386                         dentry = dget(nd->dentry->d_parent);
 387                         spin_unlock(&dcache_lock);
 388                         dput(nd->dentry);
 389                         nd->dentry = dentry;
 390                         break;
 391                 }
 392                 parent=nd->mnt->mnt_parent;
 393                 if (parent == nd->mnt) {
 394                         spin_unlock(&dcache_lock);
 395                         break;
 396                 }
 397                 mntget(parent);
 398                 dentry=dget(nd->mnt->mnt_mountpoint);
 399                 spin_unlock(&dcache_lock);
 400                 dput(nd->dentry);
 401                 nd->dentry = dentry;
 402                 mntput(nd->mnt);
 403                 nd->mnt = parent;
 404         }
 405 }
 406 /*
 407  * Name resolution.
 408  *
 409  * This is the basic name resolution function, turning a pathname
 410  * into the final dentry.
 411  *
 412  * We expect 'base' to be positive and a directory.
 413  */
 414 int path_walk(const char * name, struct nameidata *nd)
 415 {
 416         struct dentry *dentry;
 417         struct inode *inode;
 418         int err;
 419         unsigned int lookup_flags = nd->flags;
 420
 421         while (*name=='/')
 422                 name++;
 423         if (!*name)
 424                 goto return_base;
 425
 426         inode = nd->dentry->d_inode;
 427         if (current->link_count)
 428                 lookup_flags = LOOKUP_FOLLOW;
 429
 430         /* At this point we know we have a real path component. */
 431         for(;;) {
 432                 unsigned long hash;
 433                 struct qstr this;
 434                 unsigned int c;
 435
 436                 err = permission(inode, MAY_EXEC);
 437                 dentry = ERR_PTR(err);
 438                 if (err)
 439                         break;
 440
 441                 this.name = name;
 442                 c = *(const unsigned char *)name;
 443
 444                 hash = init_name_hash();
 445                 do {
 446                         name++;
 447                         hash = partial_name_hash(c, hash);
 448                         c = *(const unsigned char *)name;
 449                 } while (c && (c != '/'));
 450                 this.len = name - (const char *) this.name;
 451                 this.hash = end_name_hash(hash);
 452
 453                 /* remove trailing slashes? */
 454                 if (!c)
 455                         goto last_component;
 456                 while (*++name == '/');
 457                 if (!*name)
 458                         goto last_with_slashes;
 459
 460                 /*
 461                  * "." and ".." are special - ".." especially so because it has
 462                  * to be able to know about the current root directory and
 463                  * parent relationships.
 464                  */
 465                 if (this.name[0] == '.') switch (this.len) {
 466                         default:
 467                                 break;
 468                         case 2:
 469                                 if (this.name[1] != '.')
 470                                         break;
 471                                 follow_dotdot(nd);
 472                                 inode = nd->dentry->d_inode;
 473                                 /* fallthrough */
 474                         case 1:
 475                                 continue;
 476                 }
 477                 /*
 478                  * See if the low-level filesystem might want
 479                  * to use its own hash..
 480                  */
 481                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 482                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 483                         if (err < 0)
 484                                 break;
 485                 }
 486                 /* This does the actual lookups.. */
 487                 dentry = cached_lookup(nd->dentry, &this, LOOKUP_CONTINUE);
 488                 if (!dentry) {
 489                         dentry = real_lookup(nd->dentry, &this, LOOKUP_CONTINUE);
 490                         err = PTR_ERR(dentry);
 491                         if (IS_ERR(dentry))
 492                                 break;
 493                 }
 494                 /* Check mountpoints.. */
 495                 while (d_mountpoint(dentry) && __follow_down(&nd->mnt, &dentry))
 496                         ;
 497
 498                 err = -ENOENT;
 499                 inode = dentry->d_inode;
 500                 if (!inode)
 501                         goto out_dput;
 502                 err = -ENOTDIR;
 503                 if (!inode->i_op)
 504                         goto out_dput;
 505
 506                 if (inode->i_op->follow_link) {
 507                         err = do_follow_link(dentry, nd);
 508                         dput(dentry);
 509                         if (err)
 510                                 goto return_err;
 511                         err = -ENOENT;
 512                         inode = nd->dentry->d_inode;
 513                         if (!inode)
 514                                 break;
 515                         err = -ENOTDIR;
 516                         if (!inode->i_op)
 517                                 break;
 518                 } else {
 519                         dput(nd->dentry);
 520                         nd->dentry = dentry;
 521                 }
 522                 err = -ENOTDIR;
 523                 if (!inode->i_op->lookup)
 524                         break;
 525                 continue;
 526                 /* here ends the main loop */
 527
 528 last_with_slashes:
 529                 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 530 last_component:
 531                 if (lookup_flags & LOOKUP_PARENT)
 532                         goto lookup_parent;
 533                 if (this.name[0] == '.') switch (this.len) {
 534                         default:
 535                                 break;
 536                         case 2:
 537                                 if (this.name[1] != '.')
 538                                         break;
 539                                 follow_dotdot(nd);
 540                                 inode = nd->dentry->d_inode;
 541                                 /* fallthrough */
 542                         case 1:
 543                                 goto return_base;
 544                 }
 545                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 546                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 547                         if (err < 0)
 548                                 break;
 549                 }
 550                 dentry = cached_lookup(nd->dentry, &this, 0);
 551                 if (!dentry) {
 552                         dentry = real_lookup(nd->dentry, &this, 0);
 553                         err = PTR_ERR(dentry);
 554                         if (IS_ERR(dentry))
 555                                 break;
 556                 }
 557                 while (d_mountpoint(dentry) && __follow_down(&nd->mnt, &dentry))
 558                         ;
 559                 inode = dentry->d_inode;
 560                 if ((lookup_flags & LOOKUP_FOLLOW)
 561                     && inode && inode->i_op && inode->i_op->follow_link) {
 562                         err = do_follow_link(dentry, nd);
 563                         dput(dentry);
 564                         if (err)
 565                                 goto return_err;
 566                         inode = nd->dentry->d_inode;
 567                 } else {
 568                         dput(nd->dentry);
 569                         nd->dentry = dentry;
 570                 }
 571                 err = -ENOENT;
 572                 if (!inode)
 573                         goto no_inode;
 574                 if (lookup_flags & LOOKUP_DIRECTORY) {
 575                         err = -ENOTDIR;
 576                         if (!inode->i_op || !inode->i_op->lookup)
 577                                 break;
 578                 }
 579                 goto return_base;
 580 no_inode:
 581                 err = -ENOENT;
 582                 if (lookup_flags & (LOOKUP_POSITIVE|LOOKUP_DIRECTORY))
 583                         break;
 584                 goto return_base;
 585 lookup_parent:
 586                 nd->last = this;
 587                 nd->last_type = LAST_NORM;
 588                 if (this.name[0] != '.')
 589                         goto return_base;
 590                 if (this.len == 1)
 591                         nd->last_type = LAST_DOT;
 592                 else if (this.len == 2 && this.name[1] == '.')
 593                         nd->last_type = LAST_DOTDOT;
 594 return_base:
 595                 return 0;
 596 out_dput:
 597                 dput(dentry);
 598                 break;
 599         }
 600         path_release(nd);
 601 return_err:
 602         return err;
 603 }
 604
 605 /* SMP-safe */
 606 /* returns 1 if everything is done */
 607 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
 608 {
 609         if (path_walk(name, nd))
 610                 return 0;
 611
 612         if (!nd->dentry->d_inode) {
 613                 struct nameidata nd_root;
 614                 nd_root.last_type = LAST_ROOT;
 615                 nd_root.flags = nd->flags;
 616                 read_lock(&current->fs->lock);
 617                 nd_root.mnt = mntget(current->fs->rootmnt);
 618                 nd_root.dentry = dget(current->fs->root);
 619                 read_unlock(&current->fs->lock);
 620                 if (path_walk(name, &nd_root))
 621                         return 1;
 622                 if (nd_root.dentry->d_inode) {
 623                         path_release(nd);
 624                         nd->dentry = nd_root.dentry;
 625                         nd->mnt = nd_root.mnt;
 626                         nd->last = nd_root.last;
 627                         return 1;
 628                 }
 629                 path_release(&nd_root);
 630         }
 631         return 1;
 632 }
 633
 634 void set_fs_altroot(void)
 635 {
 636         char *emul = __emul_prefix();
 637         struct nameidata nd;
 638         struct vfsmount *mnt = NULL, *oldmnt;
 639         struct dentry *dentry = NULL, *olddentry;
 640         if (emul) {
 641                 read_lock(&current->fs->lock);
 642                 nd.mnt = mntget(current->fs->rootmnt);
 643                 nd.dentry = dget(current->fs->root);
 644                 read_unlock(&current->fs->lock);
 645                 nd.flags = LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_POSITIVE;
 646                 if (path_walk(emul,&nd) == 0) {
 647                         mnt = nd.mnt;
 648                         dentry = nd.dentry;
 649                 }
 650         }
 651         write_lock(&current->fs->lock);
 652         oldmnt = current->fs->altrootmnt;
 653         olddentry = current->fs->altroot;
 654         current->fs->altrootmnt = mnt;
 655         current->fs->altroot = dentry;
 656         write_unlock(&current->fs->lock);
 657         if (olddentry) {
 658                 dput(olddentry);
 659                 mntput(oldmnt);
 660         }
 661 }
 662
 663 /* SMP-safe */
 664 static inline int
 665 walk_init_root(const char *name, struct nameidata *nd)
 666 {
 667         read_lock(&current->fs->lock);
 668         if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
 669                 nd->mnt = mntget(current->fs->altrootmnt);
 670                 nd->dentry = dget(current->fs->altroot);
 671                 read_unlock(&current->fs->lock);
 672                 if (__emul_lookup_dentry(name,nd))
 673                         return 0;
 674                 read_lock(&current->fs->lock);
 675         }
 676         nd->mnt = mntget(current->fs->rootmnt);
 677         nd->dentry = dget(current->fs->root);
 678         read_unlock(&current->fs->lock);
 679         return 1;
 680 }
 681
 682 /* SMP-safe */
 683 int path_init(const char *name,unsigned int flags,struct nameidata *nd)
 684 {
 685         nd->last_type = LAST_ROOT; /* if there are only slashes... */
 686         nd->flags = flags;
 687         if (*name=='/')
 688                 return walk_init_root(name,nd);
 689         read_lock(&current->fs->lock);
 690         nd->mnt = mntget(current->fs->pwdmnt);
 691         nd->dentry = dget(current->fs->pwd);
 692         read_unlock(&current->fs->lock);
 693         return 1;
 694 }
 695
 696 /*
 697  * Restricted form of lookup. Doesn't follow links, single-component only,
 698  * needs parent already locked. Doesn't follow mounts.
 699  * SMP-safe.
 700  */
 701 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
 702 {
 703         struct dentry * dentry;
 704         struct inode *inode;
 705         int err;
 706
 707         inode = base->d_inode;
 708         err = permission(inode, MAY_EXEC);
 709         dentry = ERR_PTR(err);
 710         if (err)
 711                 goto out;
 712
 713         /*
 714          * See if the low-level filesystem might want
 715          * to use its own hash..
 716          */
 717         if (base->d_op && base->d_op->d_hash) {
 718                 err = base->d_op->d_hash(base, name);
 719                 dentry = ERR_PTR(err);
 720                 if (err < 0)
 721                         goto out;
 722         }
 723
 724         dentry = cached_lookup(base, name, 0);
 725         if (!dentry) {
 726                 struct dentry *new = d_alloc(base, name);
 727                 dentry = ERR_PTR(-ENOMEM);
 728                 if (!new)
 729                         goto out;
 730                 lock_kernel();
 731                 dentry = inode->i_op->lookup(inode, new);
 732                 unlock_kernel();
 733                 if (!dentry)
 734                         dentry = new;
 735                 else
 736                         dput(new);
 737         }
 738 out:
 739         return dentry;
 740 }
 741
 742 /* SMP-safe */
 743 struct dentry * lookup_one(const char * name, struct dentry * base)
 744 {
 745         unsigned long hash;
 746         struct qstr this;
 747         unsigned int c;
 748
 749         this.name = name;
 750         c = *(const unsigned char *)name;
 751         if (!c)
 752                 goto access;
 753
 754         hash = init_name_hash();
 755         do {
 756                 name++;
 757                 if (c == '/')
 758                         goto access;
 759                 hash = partial_name_hash(c, hash);
 760                 c = *(const unsigned char *)name;
 761         } while (c);
 762         this.len = name - (const char *) this.name;
 763         this.hash = end_name_hash(hash);
 764
 765         return lookup_hash(&this, base);
 766 access:
 767         return ERR_PTR(-EACCES);
 768 }
 769
 770 /*
 771  *      namei()
 772  *
 773  * is used by most simple commands to get the inode of a specified name.
 774  * Open, link etc use their own routines, but this is enough for things
 775  * like 'chmod' etc.
 776  *
 777  * namei exists in two versions: namei/lnamei. The only difference is
 778  * that namei follows links, while lnamei does not.
 779  * SMP-safe
 780  */
 781 int __user_walk(const char *name, unsigned flags, struct nameidata *nd)
 782 {
 783         char *tmp;
 784         int err;
 785
 786         tmp = getname(name);
 787         err = PTR_ERR(tmp);
 788         if (!IS_ERR(tmp)) {
 789                 err = 0;
 790                 if (path_init(tmp, flags, nd))
 791                         err = path_walk(tmp, nd);
 792                 putname(tmp);
 793         }
 794         return err;
 795 }
 796
 797 /*
 798  * It's inline, so penalty for filesystems that don't use sticky bit is
 799  * minimal.
 800  */
 801 static inline int check_sticky(struct inode *dir, struct inode *inode)
 802 {
 803         if (!(dir->i_mode & S_ISVTX))
 804                 return 0;
 805         if (inode->i_uid == current->fsuid)
 806                 return 0;
 807         if (dir->i_uid == current->fsuid)
 808                 return 0;
 809         return !capable(CAP_FOWNER);
 810 }
 811
 812 /*
 813  *      Check whether we can remove a link victim from directory dir, check
 814  *  whether the type of victim is right.
 815  *  1. We can't do it if dir is read-only (done in permission())
 816  *  2. We should have write and exec permissions on dir
 817  *  3. We can't remove anything from append-only dir
 818  *  4. We can't do anything with immutable dir (done in permission())
 819  *  5. If the sticky bit on dir is set we should either
 820  *      a. be owner of dir, or
 821  *      b. be owner of victim, or
 822  *      c. have CAP_FOWNER capability
 823  *  6. If the victim is append-only or immutable we can't do antyhing with
 824  *     links pointing to it.
 825  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 826  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 827  *  9. We can't remove a root or mountpoint.
 828  */
 829 static inline int may_delete(struct inode *dir,struct dentry *victim, int isdir)
 830 {
 831         int error;
 832         if (!victim->d_inode || victim->d_parent->d_inode != dir)
 833                 return -ENOENT;
 834         error = permission(dir,MAY_WRITE | MAY_EXEC);
 835         if (error)
 836                 return error;
 837         if (IS_APPEND(dir))
 838                 return -EPERM;
 839         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
 840             IS_IMMUTABLE(victim->d_inode))
 841                 return -EPERM;
 842         if (isdir) {
 843                 if (!S_ISDIR(victim->d_inode->i_mode))
 844                         return -ENOTDIR;
 845                 if (IS_ROOT(victim))
 846                         return -EBUSY;
 847         } else if (S_ISDIR(victim->d_inode->i_mode))
 848                 return -EISDIR;
 849         return 0;
 850 }
 851
 852 /*      Check whether we can create an object with dentry child in directory
 853  *  dir.
 854  *  1. We can't do it if child already exists (open has special treatment for
 855  *     this case, but since we are inlined it's OK)
 856  *  2. We can't do it if dir is read-only (done in permission())
 857  *  3. We should have write and exec permissions on dir
 858  *  4. We can't do it if dir is immutable (done in permission())
 859  */
 860 static inline int may_create(struct inode *dir, struct dentry *child) {
 861         if (child->d_inode)
 862                 return -EEXIST;
 863         if (IS_DEADDIR(dir))
 864                 return -ENOENT;
 865         return permission(dir,MAY_WRITE | MAY_EXEC);
 866 }
 867
 868 /*
 869  * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
 870  * reasons.
 871  *
 872  * O_DIRECTORY translates into forcing a directory lookup.
 873  */
 874 static inline int lookup_flags(unsigned int f)
 875 {
 876         unsigned long retval = LOOKUP_FOLLOW;
 877
 878         if (f & O_NOFOLLOW)
 879                 retval &= ~LOOKUP_FOLLOW;
 880
 881         if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 882                 retval &= ~LOOKUP_FOLLOW;
 883
 884         if (f & O_DIRECTORY)
 885                 retval |= LOOKUP_DIRECTORY;
 886
 887         return retval;
 888 }
 889
 890 int vfs_create(struct inode *dir, struct dentry *dentry, int mode)
 891 {
 892         int error;
 893
 894         mode &= S_IALLUGO & ~current->fs->umask;
 895         mode |= S_IFREG;
 896
 897         down(&dir->i_zombie);
 898         error = may_create(dir, dentry);
 899         if (error)
 900                 goto exit_lock;
 901
 902         error = -EACCES;        /* shouldn't it be ENOSYS? */
 903         if (!dir->i_op || !dir->i_op->create)
 904                 goto exit_lock;
 905
 906         DQUOT_INIT(dir);
 907         lock_kernel();
 908         error = dir->i_op->create(dir, dentry, mode);
 909         unlock_kernel();
 910 exit_lock:
 911         up(&dir->i_zombie);
 912         return error;
 913 }
 914
 915 /*
 916  *      open_namei()
 917  *
 918  * namei for open - this is in fact almost the whole open-routine.
 919  *
 920  * Note that the low bits of "flag" aren't the same as in the open
 921  * system call - they are 00 - no permissions needed
 922  *                        01 - read permission needed
 923  *                        10 - write permission needed
 924  *                        11 - read/write permissions needed
 925  * which is a lot more logical, and also allows the "no perm" needed
 926  * for symlinks (where the permissions are checked later).
 927  * SMP-safe
 928  */
 929 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
 930 {
 931         int acc_mode, error = 0;
 932         struct inode *inode;
 933         struct dentry *dentry;
 934         struct dentry *dir;
 935         int count = 0;
 936
 937         acc_mode = ACC_MODE(flag);
 938
 939         /*
 940          * The simplest case - just a plain lookup.
 941          */
 942         if (!(flag & O_CREAT)) {
 943                 if (path_init(pathname, lookup_flags(flag), nd))
 944                         error = path_walk(pathname, nd);
 945                 if (error)
 946                         return error;
 947                 dentry = nd->dentry;
 948                 goto ok;
 949         }
 950
 951         /*
 952          * Create - we need to know the parent.
 953          */
 954         if (path_init(pathname, LOOKUP_PARENT, nd))
 955                 error = path_walk(pathname, nd);
 956         if (error)
 957                 return error;
 958
 959         /*
 960          * We have the parent and last component. First of all, check
 961          * that we are not asked to creat(2) an obvious directory - that
 962          * will not do.
 963          */
 964         error = -EISDIR;
 965         if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
 966                 goto exit;
 967
 968         dir = nd->dentry;
 969         down(&dir->d_inode->i_sem);
 970         dentry = lookup_hash(&nd->last, nd->dentry);
 971
 972 do_last:
 973         error = PTR_ERR(dentry);
 974         if (IS_ERR(dentry)) {
 975                 up(&dir->d_inode->i_sem);
 976                 goto exit;
 977         }
 978
 979         /* Negative dentry, just create the file */
 980         if (!dentry->d_inode) {
 981                 error = vfs_create(dir->d_inode, dentry, mode);
 982                 up(&dir->d_inode->i_sem);
 983                 dput(nd->dentry);
 984                 nd->dentry = dentry;
 985                 if (error)
 986                         goto exit;
 987                 /* Don't check for write permission, don't truncate */
 988                 acc_mode = 0;
 989                 flag &= ~O_TRUNC;
 990                 goto ok;
 991         }
 992
 993         /*
 994          * It already exists.
 995          */
 996         up(&dir->d_inode->i_sem);
 997
 998         error = -EEXIST;
 999         if (flag & O_EXCL)
1000                 goto exit_dput;
1001
1002         if (d_mountpoint(dentry)) {
1003                 error = -ELOOP;
1004                 if (flag & O_NOFOLLOW)
1005                         goto exit_dput;
1006                 do __follow_down(&nd->mnt,&dentry); while(d_mountpoint(dentry));
1007         }
1008         error = -ENOENT;
1009         if (!dentry->d_inode)
1010                 goto exit_dput;
1011         if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1012                 goto do_link;
1013
1014         dput(nd->dentry);
1015         nd->dentry = dentry;
1016         error = -EISDIR;
1017         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1018                 goto exit;
1019 ok:
1020         error = -ENOENT;
1021         inode = dentry->d_inode;
1022         if (!inode)
1023                 goto exit;
1024
1025         error = -ELOOP;
1026         if (S_ISLNK(inode->i_mode))
1027                 goto exit;
1028
1029         error = -EISDIR;
1030         if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1031                 goto exit;
1032
1033         error = permission(inode,acc_mode);
1034         if (error)
1035                 goto exit;
1036
1037         /*
1038          * FIFO's, sockets and device files are special: they don't
1039          * actually live on the filesystem itself, and as such you
1040          * can write to them even if the filesystem is read-only.
1041          */
1042         if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1043                 flag &= ~O_TRUNC;
1044         } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1045                 error = -EACCES;
1046                 if (IS_NODEV(inode))
1047                         goto exit;
1048
1049                 flag &= ~O_TRUNC;
1050         } else {
1051                 error = -EROFS;
1052                 if (IS_RDONLY(inode) && (flag & 2))
1053                         goto exit;
1054         }
1055         /*
1056          * An append-only file must be opened in append mode for writing.
1057          */
1058         error = -EPERM;
1059         if (IS_APPEND(inode)) {
1060                 if  ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1061                         goto exit;
1062                 if (flag & O_TRUNC)
1063                         goto exit;
1064         }
1065
1066         if (flag & O_TRUNC) {
1067                 error = get_write_access(inode);
1068                 if (error)
1069                         goto exit;
1070
1071                 /*
1072                  * Refuse to truncate files with mandatory locks held on them.
1073                  */
1074                 error = locks_verify_locked(inode);
1075                 if (!error) {
1076                         DQUOT_INIT(inode);
1077
1078                         error = do_truncate(dentry, 0);
1079                 }
1080                 put_write_access(inode);
1081                 if (error)
1082                         goto exit;
1083         } else
1084                 if (flag & FMODE_WRITE)
1085                         DQUOT_INIT(inode);
1086
1087         return 0;
1088
1089 exit_dput:
1090         dput(dentry);
1091 exit:
1092         path_release(nd);
1093         return error;
1094
1095 do_link:
1096         error = -ELOOP;
1097         if (flag & O_NOFOLLOW)
1098                 goto exit_dput;
1099         /*
1100          * This is subtle. Instead of calling do_follow_link() we do the
1101          * thing by hands. The reason is that this way we have zero link_count
1102          * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1103          * After that we have the parent and last component, i.e.
1104          * we are in the same situation as after the first path_walk().
1105          * Well, almost - if the last component is normal we get its copy
1106          * stored in nd->last.name and we will have to putname() it when we
1107          * are done. Procfs-like symlinks just set LAST_BIND.
1108          */
1109         UPDATE_ATIME(dentry->d_inode);
1110         error = dentry->d_inode->i_op->follow_link(dentry, nd);
1111         dput(dentry);
1112         if (error)
1113                 return error;
1114         if (nd->last_type == LAST_BIND) {
1115                 dentry = nd->dentry;
1116                 goto ok;
1117         }
1118         error = -EISDIR;
1119         if (nd->last_type != LAST_NORM)
1120                 goto exit;
1121         if (nd->last.name[nd->last.len]) {
1122                 putname(nd->last.name);
1123                 goto exit;
1124         }
1125         if (count++==32) {
1126                 dentry = nd->dentry;
1127                 putname(nd->last.name);
1128                 goto ok;
1129         }
1130         dir = nd->dentry;
1131         down(&dir->d_inode->i_sem);
1132         dentry = lookup_hash(&nd->last, nd->dentry);
1133         putname(nd->last.name);
1134         goto do_last;
1135 }
1136
1137 /* SMP-safe */
1138 static struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1139 {
1140         struct dentry *dentry;
1141
1142         down(&nd->dentry->d_inode->i_sem);
1143         dentry = ERR_PTR(-EEXIST);
1144         if (nd->last_type != LAST_NORM)
1145                 goto fail;
1146         dentry = lookup_hash(&nd->last, nd->dentry);
1147         if (IS_ERR(dentry))
1148                 goto fail;
1149         if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1150                 goto enoent;
1151         return dentry;
1152 enoent:
1153         dput(dentry);
1154         dentry = ERR_PTR(-ENOENT);
1155 fail:
1156         return dentry;
1157 }
1158
1159 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1160 {
1161         int error = -EPERM;
1162
1163         mode &= ~current->fs->umask;
1164
1165         down(&dir->i_zombie);
1166         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1167                 goto exit_lock;
1168
1169         error = may_create(dir, dentry);
1170         if (error)
1171                 goto exit_lock;
1172
1173         error = -EPERM;
1174         if (!dir->i_op || !dir->i_op->mknod)
1175                 goto exit_lock;
1176
1177         DQUOT_INIT(dir);
1178         lock_kernel();
1179         error = dir->i_op->mknod(dir, dentry, mode, dev);
1180         unlock_kernel();
1181 exit_lock:
1182         up(&dir->i_zombie);
1183         return error;
1184 }
1185
1186 asmlinkage long sys_mknod(const char * filename, int mode, dev_t dev)
1187 {
1188         int error = 0;
1189         char * tmp;
1190         struct dentry * dentry;
1191         struct nameidata nd;
1192
1193         if (S_ISDIR(mode))
1194                 return -EPERM;
1195         tmp = getname(filename);
1196         if (IS_ERR(tmp))
1197                 return PTR_ERR(tmp);
1198
1199         if (path_init(tmp, LOOKUP_PARENT, &nd))
1200                 error = path_walk(tmp, &nd);
1201         if (error)
1202                 goto out;
1203         dentry = lookup_create(&nd, 0);
1204         error = PTR_ERR(dentry);
1205         if (!IS_ERR(dentry)) {
1206                 switch (mode & S_IFMT) {
1207                 case 0: case S_IFREG:
1208                         error = vfs_create(nd.dentry->d_inode,dentry,mode);
1209                         break;
1210                 case S_IFCHR: case S_IFBLK: case S_IFIFO: case S_IFSOCK:
1211                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,dev);
1212                         break;
1213                 case S_IFDIR:
1214                         error = -EPERM;
1215                         break;
1216                 default:
1217                         error = -EINVAL;
1218                 }
1219                 dput(dentry);
1220         }
1221         up(&nd.dentry->d_inode->i_sem);
1222         path_release(&nd);
1223 out:
1224         putname(tmp);
1225
1226         return error;
1227 }
1228
1229 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1230 {
1231         int error;
1232
1233         down(&dir->i_zombie);
1234         error = may_create(dir, dentry);
1235         if (error)
1236                 goto exit_lock;
1237
1238         error = -EPERM;
1239         if (!dir->i_op || !dir->i_op->mkdir)
1240                 goto exit_lock;
1241
1242         DQUOT_INIT(dir);
1243         mode &= (S_IRWXUGO|S_ISVTX) & ~current->fs->umask;
1244         lock_kernel();
1245         error = dir->i_op->mkdir(dir, dentry, mode);
1246         unlock_kernel();
1247
1248 exit_lock:
1249         up(&dir->i_zombie);
1250         return error;
1251 }
1252
1253 asmlinkage long sys_mkdir(const char * pathname, int mode)
1254 {
1255         int error = 0;
1256         char * tmp;
1257
1258         tmp = getname(pathname);
1259         error = PTR_ERR(tmp);
1260         if (!IS_ERR(tmp)) {
1261                 struct dentry *dentry;
1262                 struct nameidata nd;
1263
1264                 if (path_init(tmp, LOOKUP_PARENT, &nd))
1265                         error = path_walk(tmp, &nd);
1266                 if (error)
1267                         goto out;
1268                 dentry = lookup_create(&nd, 1);
1269                 error = PTR_ERR(dentry);
1270                 if (!IS_ERR(dentry)) {
1271                         error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1272                         dput(dentry);
1273                 }
1274                 up(&nd.dentry->d_inode->i_sem);
1275                 path_release(&nd);
1276 out:
1277                 putname(tmp);
1278         }
1279
1280         return error;
1281 }
1282
1283 /*
1284  * We try to drop the dentry early: we should have
1285  * a usage count of 2 if we're the only user of this
1286  * dentry, and if that is true (possibly after pruning
1287  * the dcache), then we drop the dentry now.
1288  *
1289  * A low-level filesystem can, if it choses, legally
1290  * do a
1291  *
1292  *      if (!d_unhashed(dentry))
1293  *              return -EBUSY;
1294  *
1295  * if it cannot handle the case of removing a directory
1296  * that is still in use by something else..
1297  */
1298 static void d_unhash(struct dentry *dentry)
1299 {
1300         dget(dentry);
1301         switch (atomic_read(&dentry->d_count)) {
1302         default:
1303                 shrink_dcache_parent(dentry);
1304                 if (atomic_read(&dentry->d_count) != 2)
1305                         break;
1306         case 2:
1307                 d_drop(dentry);
1308         }
1309 }
1310
1311 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1312 {
1313         int error;
1314
1315         error = may_delete(dir, dentry, 1);
1316         if (error)
1317                 return error;
1318
1319         if (!dir->i_op || !dir->i_op->rmdir)
1320                 return -EPERM;
1321
1322         DQUOT_INIT(dir);
1323
1324         double_down(&dir->i_zombie, &dentry->d_inode->i_zombie);
1325         d_unhash(dentry);
1326         if (IS_DEADDIR(dir))
1327                 error = -ENOENT;
1328         else if (d_mountpoint(dentry))
1329                 error = -EBUSY;
1330         else {
1331                 lock_kernel();
1332                 error = dir->i_op->rmdir(dir, dentry);
1333                 unlock_kernel();
1334                 if (!error)
1335                         dentry->d_inode->i_flags |= S_DEAD;
1336         }
1337         double_up(&dir->i_zombie, &dentry->d_inode->i_zombie);
1338         if (!error)
1339                 d_delete(dentry);
1340         dput(dentry);
1341
1342         return error;
1343 }
1344
1345 asmlinkage long sys_rmdir(const char * pathname)
1346 {
1347         int error = 0;
1348         char * name;
1349         struct dentry *dentry;
1350         struct nameidata nd;
1351
1352         name = getname(pathname);
1353         if(IS_ERR(name))
1354                 return PTR_ERR(name);
1355
1356         if (path_init(name, LOOKUP_PARENT, &nd))
1357                 error = path_walk(name, &nd);
1358         if (error)
1359                 goto exit;
1360
1361         switch(nd.last_type) {
1362                 case LAST_DOTDOT:
1363                         error = -ENOTEMPTY;
1364                         goto exit1;
1365                 case LAST_ROOT: case LAST_DOT:
1366                         error = -EBUSY;
1367                         goto exit1;
1368         }
1369         down(&nd.dentry->d_inode->i_sem);
1370         dentry = lookup_hash(&nd.last, nd.dentry);
1371         error = PTR_ERR(dentry);
1372         if (!IS_ERR(dentry)) {
1373                 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1374                 dput(dentry);
1375         }
1376         up(&nd.dentry->d_inode->i_sem);
1377 exit1:
1378         path_release(&nd);
1379 exit:
1380         putname(name);
1381         return error;
1382 }
1383
1384 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1385 {
1386         int error;
1387
1388         down(&dir->i_zombie);
1389         error = may_delete(dir, dentry, 0);
1390         if (!error) {
1391                 error = -EPERM;
1392                 if (dir->i_op && dir->i_op->unlink) {
1393                         DQUOT_INIT(dir);
1394                         if (d_mountpoint(dentry))
1395                                 error = -EBUSY;
1396                         else {
1397                                 lock_kernel();
1398                                 error = dir->i_op->unlink(dir, dentry);
1399                                 unlock_kernel();
1400                                 if (!error)
1401                                         d_delete(dentry);
1402                         }
1403                 }
1404         }
1405         up(&dir->i_zombie);
1406         return error;
1407 }
1408
1409 asmlinkage long sys_unlink(const char * pathname)
1410 {
1411         int error = 0;
1412         char * name;
1413         struct dentry *dentry;
1414         struct nameidata nd;
1415
1416         name = getname(pathname);
1417         if(IS_ERR(name))
1418                 return PTR_ERR(name);
1419
1420         if (path_init(name, LOOKUP_PARENT, &nd))
1421                 error = path_walk(name, &nd);
1422         if (error)
1423                 goto exit;
1424         error = -EISDIR;
1425         if (nd.last_type != LAST_NORM)
1426                 goto exit1;
1427         down(&nd.dentry->d_inode->i_sem);
1428         dentry = lookup_hash(&nd.last, nd.dentry);
1429         error = PTR_ERR(dentry);
1430         if (!IS_ERR(dentry)) {
1431                 /* Why not before? Because we want correct error value */
1432                 if (nd.last.name[nd.last.len])
1433                         goto slashes;
1434                 error = vfs_unlink(nd.dentry->d_inode, dentry);
1435         exit2:
1436                 dput(dentry);
1437         }
1438         up(&nd.dentry->d_inode->i_sem);
1439 exit1:
1440         path_release(&nd);
1441 exit:
1442         putname(name);
1443
1444         return error;
1445
1446 slashes:
1447         error = !dentry->d_inode ? -ENOENT :
1448                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1449         goto exit2;
1450 }
1451
1452 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1453 {
1454         int error;
1455
1456         down(&dir->i_zombie);
1457         error = may_create(dir, dentry);
1458         if (error)
1459                 goto exit_lock;
1460
1461         error = -EPERM;
1462         if (!dir->i_op || !dir->i_op->symlink)
1463                 goto exit_lock;
1464
1465         DQUOT_INIT(dir);
1466         lock_kernel();
1467         error = dir->i_op->symlink(dir, dentry, oldname);
1468         unlock_kernel();
1469
1470 exit_lock:
1471         up(&dir->i_zombie);
1472         return error;
1473 }
1474
1475 asmlinkage long sys_symlink(const char * oldname, const char * newname)
1476 {
1477         int error = 0;
1478         char * from;
1479         char * to;
1480
1481         from = getname(oldname);
1482         if(IS_ERR(from))
1483                 return PTR_ERR(from);
1484         to = getname(newname);
1485         error = PTR_ERR(to);
1486         if (!IS_ERR(to)) {
1487                 struct dentry *dentry;
1488                 struct nameidata nd;
1489
1490                 if (path_init(to, LOOKUP_PARENT, &nd))
1491                         error = path_walk(to, &nd);
1492                 if (error)
1493                         goto out;
1494                 dentry = lookup_create(&nd, 0);
1495                 error = PTR_ERR(dentry);
1496                 if (!IS_ERR(dentry)) {
1497                         error = vfs_symlink(nd.dentry->d_inode, dentry, from);
1498                         dput(dentry);
1499                 }
1500                 up(&nd.dentry->d_inode->i_sem);
1501                 path_release(&nd);
1502 out:
1503                 putname(to);
1504         }
1505         putname(from);
1506         return error;
1507 }
1508
1509 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1510 {
1511         struct inode *inode;
1512         int error;
1513
1514         down(&dir->i_zombie);
1515         error = -ENOENT;
1516         inode = old_dentry->d_inode;
1517         if (!inode)
1518                 goto exit_lock;
1519
1520         error = may_create(dir, new_dentry);
1521         if (error)
1522                 goto exit_lock;
1523
1524         error = -EXDEV;
1525         if (dir->i_dev != inode->i_dev)
1526                 goto exit_lock;
1527
1528         /*
1529          * A link to an append-only or immutable file cannot be created.
1530          */
1531         error = -EPERM;
1532         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1533                 goto exit_lock;
1534         if (!dir->i_op || !dir->i_op->link)
1535                 goto exit_lock;
1536
1537         DQUOT_INIT(dir);
1538         lock_kernel();
1539         error = dir->i_op->link(old_dentry, dir, new_dentry);
1540         unlock_kernel();
1541
1542 exit_lock:
1543         up(&dir->i_zombie);
1544         return error;
1545 }
1546
1547 /*
1548  * Hardlinks are often used in delicate situations.  We avoid
1549  * security-related surprises by not following symlinks on the
1550  * newname.  --KAB
1551  *
1552  * We don't follow them on the oldname either to be compatible
1553  * with linux 2.0, and to avoid hard-linking to directories
1554  * and other special files.  --ADM
1555  */
1556 asmlinkage long sys_link(const char * oldname, const char * newname)
1557 {
1558         int error;
1559         char * from;
1560         char * to;
1561
1562         from = getname(oldname);
1563         if(IS_ERR(from))
1564                 return PTR_ERR(from);
1565         to = getname(newname);
1566         error = PTR_ERR(to);
1567         if (!IS_ERR(to)) {
1568                 struct dentry *new_dentry;
1569                 struct nameidata nd, old_nd;
1570
1571                 error = 0;
1572                 if (path_init(from, LOOKUP_POSITIVE, &old_nd))
1573                         error = path_walk(from, &old_nd);
1574                 if (error)
1575                         goto exit;
1576                 if (path_init(to, LOOKUP_PARENT, &nd))
1577                         error = path_walk(to, &nd);
1578                 if (error)
1579                         goto out;
1580                 error = -EXDEV;
1581                 if (old_nd.mnt != nd.mnt)
1582                         goto out;
1583                 new_dentry = lookup_create(&nd, 0);
1584                 error = PTR_ERR(new_dentry);
1585                 if (!IS_ERR(new_dentry)) {
1586                         error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1587                         dput(new_dentry);
1588                 }
1589                 up(&nd.dentry->d_inode->i_sem);
1590                 path_release(&nd);
1591 out:
1592                 path_release(&old_nd);
1593 exit:
1594                 putname(to);
1595         }
1596         putname(from);
1597
1598         return error;
1599 }
1600
1601 /*
1602  * The worst of all namespace operations - renaming directory. "Perverted"
1603  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
1604  * Problems:
1605  *      a) we can get into loop creation. Check is done in is_subdir().
1606  *      b) race potential - two innocent renames can create a loop together.
1607  *         That's where 4.4 screws up. Current fix: serialization on
1608  *         sb->s_vfs_rename_sem. We might be more accurate, but that's another
1609  *         story.
1610  *      c) we have to lock _three_ objects - parents and victim (if it exists).
1611  *         And that - after we got ->i_sem on parents (until then we don't know
1612  *         whether the target exists at all, let alone whether it is a directory
1613  *         or not). Solution: ->i_zombie. Taken only after ->i_sem. Always taken
1614  *         on link creation/removal of any kind. And taken (without ->i_sem) on
1615  *         directory that will be removed (both in rmdir() and here).
1616  *      d) some filesystems don't support opened-but-unlinked directories,
1617  *         either because of layout or because they are not ready to deal with
1618  *         all cases correctly. The latter will be fixed (taking this sort of
1619  *         stuff into VFS), but the former is not going away. Solution: the same
1620  *         trick as in rmdir().
1621  *      e) conversion from fhandle to dentry may come in the wrong moment - when
1622  *         we are removing the target. Solution: we will have to grab ->i_zombie
1623  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1624  *         ->i_sem on parents, which works but leads to some truely excessive
1625  *         locking].
1626  */
1627 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
1628                struct inode *new_dir, struct dentry *new_dentry)
1629 {
1630         int error;
1631         struct inode *target;
1632
1633         if (old_dentry->d_inode == new_dentry->d_inode)
1634                 return 0;
1635
1636         error = may_delete(old_dir, old_dentry, 1);
1637         if (error)
1638                 return error;
1639
1640         if (new_dir->i_dev != old_dir->i_dev)
1641                 return -EXDEV;
1642
1643         if (!new_dentry->d_inode)
1644                 error = may_create(new_dir, new_dentry);
1645         else
1646                 error = may_delete(new_dir, new_dentry, 1);
1647         if (error)
1648                 return error;
1649
1650         if (!old_dir->i_op || !old_dir->i_op->rename)
1651                 return -EPERM;
1652
1653         /*
1654          * If we are going to change the parent - check write permissions,
1655          * we'll need to flip '..'.
1656          */
1657         if (new_dir != old_dir) {
1658                 error = permission(old_dentry->d_inode, MAY_WRITE);
1659         }
1660         if (error)
1661                 return error;
1662
1663         DQUOT_INIT(old_dir);
1664         DQUOT_INIT(new_dir);
1665         down(&old_dir->i_sb->s_vfs_rename_sem);
1666         error = -EINVAL;
1667         if (is_subdir(new_dentry, old_dentry))
1668                 goto out_unlock;
1669         target = new_dentry->d_inode;
1670         if (target) { /* Hastur! Hastur! Hastur! */
1671                 triple_down(&old_dir->i_zombie,
1672                             &new_dir->i_zombie,
1673                             &target->i_zombie);
1674                 d_unhash(new_dentry);
1675         } else
1676                 double_down(&old_dir->i_zombie,
1677                             &new_dir->i_zombie);
1678         if (IS_DEADDIR(old_dir)||IS_DEADDIR(new_dir))
1679                 error = -ENOENT;
1680         else if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1681                 error = -EBUSY;
1682         else
1683                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1684         if (target) {
1685                 if (!error)
1686                         target->i_flags |= S_DEAD;
1687                 triple_up(&old_dir->i_zombie,
1688                           &new_dir->i_zombie,
1689                           &target->i_zombie);
1690                 d_rehash(new_dentry);
1691                 dput(new_dentry);
1692         } else
1693                 double_up(&old_dir->i_zombie,
1694                           &new_dir->i_zombie);
1695
1696         if (!error)
1697                 d_move(old_dentry,new_dentry);
1698 out_unlock:
1699         up(&old_dir->i_sb->s_vfs_rename_sem);
1700         return error;
1701 }
1702
1703 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
1704                struct inode *new_dir, struct dentry *new_dentry)
1705 {
1706         int error;
1707
1708         if (old_dentry->d_inode == new_dentry->d_inode)
1709                 return 0;
1710
1711         error = may_delete(old_dir, old_dentry, 0);
1712         if (error)
1713                 return error;
1714
1715         if (new_dir->i_dev != old_dir->i_dev)
1716                 return -EXDEV;
1717
1718         if (!new_dentry->d_inode)
1719                 error = may_create(new_dir, new_dentry);
1720         else
1721                 error = may_delete(new_dir, new_dentry, 0);
1722         if (error)
1723                 return error;
1724
1725         if (!old_dir->i_op || !old_dir->i_op->rename)
1726                 return -EPERM;
1727
1728         DQUOT_INIT(old_dir);
1729         DQUOT_INIT(new_dir);
1730         double_down(&old_dir->i_zombie, &new_dir->i_zombie);
1731         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1732                 error = -EBUSY;
1733         else
1734                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1735         double_up(&old_dir->i_zombie, &new_dir->i_zombie);
1736         if (error)
1737                 return error;
1738         /* The following d_move() should become unconditional */
1739         if (!(old_dir->i_sb->s_flags & MS_ODD_RENAME)) {
1740                 d_move(old_dentry, new_dentry);
1741         }
1742         return 0;
1743 }
1744
1745 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1746                struct inode *new_dir, struct dentry *new_dentry)
1747 {
1748         if (S_ISDIR(old_dentry->d_inode->i_mode))
1749                 return vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
1750         else
1751                 return vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
1752 }
1753
1754 static inline int do_rename(const char * oldname, const char * newname)
1755 {
1756         int error = 0;
1757         struct dentry * old_dir, * new_dir;
1758         struct dentry * old_dentry, *new_dentry;
1759         struct nameidata oldnd, newnd;
1760
1761         if (path_init(oldname, LOOKUP_PARENT, &oldnd))
1762                 error = path_walk(oldname, &oldnd);
1763
1764         if (error)
1765                 goto exit;
1766
1767         if (path_init(newname, LOOKUP_PARENT, &newnd))
1768                 error = path_walk(newname, &newnd);
1769         if (error)
1770                 goto exit1;
1771
1772         error = -EXDEV;
1773         if (oldnd.mnt != newnd.mnt)
1774                 goto exit2;
1775
1776         old_dir = oldnd.dentry;
1777         error = -EBUSY;
1778         if (oldnd.last_type != LAST_NORM)
1779                 goto exit2;
1780
1781         new_dir = newnd.dentry;
1782         if (newnd.last_type != LAST_NORM)
1783                 goto exit2;
1784
1785         double_lock(new_dir, old_dir);
1786
1787         old_dentry = lookup_hash(&oldnd.last, old_dir);
1788         error = PTR_ERR(old_dentry);
1789         if (IS_ERR(old_dentry))
1790                 goto exit3;
1791         /* source must exist */
1792         error = -ENOENT;
1793         if (!old_dentry->d_inode)
1794                 goto exit4;
1795         /* unless the source is a directory trailing slashes give -ENOTDIR */
1796         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
1797                 error = -ENOTDIR;
1798                 if (oldnd.last.name[oldnd.last.len])
1799                         goto exit4;
1800                 if (newnd.last.name[newnd.last.len])
1801                         goto exit4;
1802         }
1803         new_dentry = lookup_hash(&newnd.last, new_dir);
1804         error = PTR_ERR(new_dentry);
1805         if (IS_ERR(new_dentry))
1806                 goto exit4;
1807
1808         lock_kernel();
1809         error = vfs_rename(old_dir->d_inode, old_dentry,
1810                                    new_dir->d_inode, new_dentry);
1811         unlock_kernel();
1812
1813         dput(new_dentry);
1814 exit4:
1815         dput(old_dentry);
1816 exit3:
1817         double_up(&new_dir->d_inode->i_sem, &old_dir->d_inode->i_sem);
1818 exit2:
1819         path_release(&newnd);
1820 exit1:
1821         path_release(&oldnd);
1822 exit:
1823         return error;
1824 }
1825
1826 asmlinkage long sys_rename(const char * oldname, const char * newname)
1827 {
1828         int error;
1829         char * from;
1830         char * to;
1831
1832         from = getname(oldname);
1833         if(IS_ERR(from))
1834                 return PTR_ERR(from);
1835         to = getname(newname);
1836         error = PTR_ERR(to);
1837         if (!IS_ERR(to)) {
1838                 error = do_rename(from,to);
1839                 putname(to);
1840         }
1841         putname(from);
1842         return error;
1843 }
1844
1845 int vfs_readlink(struct dentry *dentry, char *buffer, int buflen, const char *link)
1846 {
1847         int len;
1848
1849         len = PTR_ERR(link);
1850         if (IS_ERR(link))
1851                 goto out;
1852
1853         len = strlen(link);
1854         if (len > (unsigned) buflen)
1855                 len = buflen;
1856         if (copy_to_user(buffer, link, len))
1857                 len = -EFAULT;
1858 out:
1859         return len;
1860 }
1861
1862 static inline int
1863 __vfs_follow_link(struct nameidata *nd, const char *link)
1864 {
1865         int res = 0;
1866         char *name;
1867         if (IS_ERR(link))
1868                 goto fail;
1869
1870         if (*link == '/') {
1871                 path_release(nd);
1872                 if (!walk_init_root(link, nd))
1873                         /* weird __emul_prefix() stuff did it */
1874                         goto out;
1875         }
1876         res = path_walk(link, nd);
1877 out:
1878         if (current->link_count || res || nd->last_type!=LAST_NORM)
1879                 return res;
1880         /*
1881          * If it is an iterative symlinks resolution in open_namei() we
1882          * have to copy the last component. And all that crap because of
1883          * bloody create() on broken symlinks. Furrfu...
1884          */
1885         name = __getname();
1886         if (IS_ERR(name))
1887                 goto fail_name;
1888         strcpy(name, nd->last.name);
1889         nd->last.name = name;
1890         return 0;
1891 fail_name:
1892         link = name;
1893 fail:
1894         path_release(nd);
1895         return PTR_ERR(link);
1896 }
1897
1898 int vfs_follow_link(struct nameidata *nd, const char *link)
1899 {
1900         return __vfs_follow_link(nd, link);
1901 }
1902
1903 /* get the link contents into pagecache */
1904 static char *page_getlink(struct dentry * dentry, struct page **ppage)
1905 {
1906         struct page * page;
1907         struct address_space *mapping = dentry->d_inode->i_mapping;
1908         page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
1909                                 NULL);
1910         if (IS_ERR(page))
1911                 goto sync_fail;
1912         wait_on_page(page);
1913         if (!Page_Uptodate(page))
1914                 goto async_fail;
1915         *ppage = page;
1916         return (char*) kmap(page);
1917
1918 async_fail:
1919         page_cache_release(page);
1920         return ERR_PTR(-EIO);
1921
1922 sync_fail:
1923         return (char*)page;
1924 }
1925
1926 int page_readlink(struct dentry *dentry, char *buffer, int buflen)
1927 {
1928         struct page *page = NULL;
1929         char *s = page_getlink(dentry, &page);
1930         int res = vfs_readlink(dentry,buffer,buflen,s);
1931         if (page) {
1932                 kunmap(page);
1933                 page_cache_release(page);
1934         }
1935         return res;
1936 }
1937
1938 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
1939 {
1940         struct page *page = NULL;
1941         char *s = page_getlink(dentry, &page);
1942         int res = __vfs_follow_link(nd, s);
1943         if (page) {
1944                 kunmap(page);
1945                 page_cache_release(page);
1946         }
1947         return res;
1948 }
1949
1950 struct inode_operations page_symlink_inode_operations = {
1951         readlink:       page_readlink,
1952         follow_link:    page_follow_link,
1953 };