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