fs/namei.c

   1 /*
   2  *  linux/fs/namei.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  */
   6
   7 /*
   8  * Some corrections by tytso.
   9  */
  10
  11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
  12  * lookup logic.
  13  */
  14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
  15  */
  16
  17 #include <linux/init.h>
  18 #include <linux/mm.h>
  19 #include <linux/proc_fs.h>
  20 #include <linux/smp_lock.h>
  21 #include <linux/quotaops.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/dcache.h>
  24
  25 #include <asm/uaccess.h>
  26 #include <asm/unaligned.h>
  27 #include <asm/semaphore.h>
  28 #include <asm/page.h>
  29 #include <asm/pgtable.h>
  30
  31 #include <asm/namei.h>
  32
  33 #define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
  34
  35 /* [Feb-1997 T. Schoebel-Theuer]
  36  * Fundamental changes in the pathname lookup mechanisms (namei)
  37  * were necessary because of omirr.  The reason is that omirr needs
  38  * to know the _real_ pathname, not the user-supplied one, in case
  39  * of symlinks (and also when transname replacements occur).
  40  *
  41  * The new code replaces the old recursive symlink resolution with
  42  * an iterative one (in case of non-nested symlink chains).  It does
  43  * this with calls to <fs>_follow_link().
  44  * As a side effect, dir_namei(), _namei() and follow_link() are now
  45  * replaced with a single function lookup_dentry() that can handle all
  46  * the special cases of the former code.
  47  *
  48  * With the new dcache, the pathname is stored at each inode, at least as
  49  * long as the refcount of the inode is positive.  As a side effect, the
  50  * size of the dcache depends on the inode cache and thus is dynamic.
  51  *
  52  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
  53  * resolution to correspond with current state of the code.
  54  *
  55  * Note that the symlink resolution is not *completely* iterative.
  56  * There is still a significant amount of tail- and mid- recursion in
  57  * the algorithm.  Also, note that <fs>_readlink() is not used in
  58  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
  59  * may return different results than <fs>_follow_link().  Many virtual
  60  * filesystems (including /proc) exhibit this behavior.
  61  */
  62
  63 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
  64  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
  65  * and the name already exists in form of a symlink, try to create the new
  66  * name indicated by the symlink. The old code always complained that the
  67  * name already exists, due to not following the symlink even if its target
  68  * is nonexistent.  The new semantics affects also mknod() and link() when
  69  * the name is a symlink pointing to a non-existant name.
  70  *
  71  * I don't know which semantics is the right one, since I have no access
  72  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
  73  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
  74  * "old" one. Personally, I think the new semantics is much more logical.
  75  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
  76  * file does succeed in both HP-UX and SunOs, but not in Solaris
  77  * and in the old Linux semantics.
  78  */
  79
  80 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
  81  * semantics.  See the comments in "open_namei" and "do_link" below.
  82  *
  83  * [10-Sep-98 Alan Modra] Another symlink change.
  84  */
  85
  86 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
  87  *      inside the path - always follow.
  88  *      in the last component in creation/removal/renaming - never follow.
  89  *      if LOOKUP_FOLLOW passed - follow.
  90  *      if the pathname has trailing slashes - follow.
  91  *      otherwise - don't follow.
  92  * (applied in that order).
  93  *
  94  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
  95  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
  96  * During the 2.4 we need to fix the userland stuff depending on it -
  97  * hopefully we will be able to get rid of that wart in 2.5. So far only
  98  * XEmacs seems to be relying on it...
  99  */
 100
 101 /* In order to reduce some races, while at the same time doing additional
 102  * checking and hopefully speeding things up, we copy filenames to the
 103  * kernel data space before using them..
 104  *
 105  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
 106  */
 107 static inline int do_getname(const char *filename, char *page)
 108 {
 109         int retval;
 110         unsigned long len = PAGE_SIZE;
 111
 112         if ((unsigned long) filename >= TASK_SIZE) {
 113                 if (!segment_eq(get_fs(), KERNEL_DS))
 114                         return -EFAULT;
 115         } else if (TASK_SIZE - (unsigned long) filename < PAGE_SIZE)
 116                 len = TASK_SIZE - (unsigned long) filename;
 117
 118         retval = strncpy_from_user((char *)page, filename, len);
 119         if (retval > 0) {
 120                 if (retval < len)
 121                         return 0;
 122                 return -ENAMETOOLONG;
 123         } else if (!retval)
 124                 retval = -ENOENT;
 125         return retval;
 126 }
 127
 128 char * getname(const char * filename)
 129 {
 130         char *tmp, *result;
 131
 132         result = ERR_PTR(-ENOMEM);
 133         tmp = __getname();
 134         if (tmp)  {
 135                 int retval = do_getname(filename, tmp);
 136
 137                 result = tmp;
 138                 if (retval < 0) {
 139                         putname(tmp);
 140                         result = ERR_PTR(retval);
 141                 }
 142         }
 143         return result;
 144 }
 145
 146 /*
 147  *      permission()
 148  *
 149  * is used to check for read/write/execute permissions on a file.
 150  * We use "fsuid" for this, letting us set arbitrary permissions
 151  * for filesystem access without changing the "normal" uids which
 152  * are used for other things..
 153  */
 154 int permission(struct inode * inode,int mask)
 155 {
 156         int mode = inode->i_mode;
 157
 158         if (inode->i_op && inode->i_op->permission) {
 159                 int retval;
 160                 lock_kernel();
 161                 retval = inode->i_op->permission(inode, mask);
 162                 unlock_kernel();
 163                 return retval;
 164         }
 165
 166         if ((mask & S_IWOTH) && IS_RDONLY(inode) &&
 167                  (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 168                 return -EROFS; /* Nobody gets write access to a read-only fs */
 169
 170         if ((mask & S_IWOTH) && IS_IMMUTABLE(inode))
 171                 return -EACCES; /* Nobody gets write access to an immutable file */
 172
 173         if (current->fsuid == inode->i_uid)
 174                 mode >>= 6;
 175         else if (in_group_p(inode->i_gid))
 176                 mode >>= 3;
 177
 178         if (((mode & mask & S_IRWXO) == mask) || capable(CAP_DAC_OVERRIDE))
 179                 return 0;
 180
 181         /* read and search access */
 182         if ((mask == S_IROTH) ||
 183             (S_ISDIR(mode)  && !(mask & ~(S_IROTH | S_IXOTH))))
 184                 if (capable(CAP_DAC_READ_SEARCH))
 185                         return 0;
 186
 187         return -EACCES;
 188 }
 189
 190 /*
 191  * get_write_access() gets write permission for a file.
 192  * put_write_access() releases this write permission.
 193  * This is used for regular files.
 194  * We cannot support write (and maybe mmap read-write shared) accesses and
 195  * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
 196  * can have the following values:
 197  * 0: no writers, no VM_DENYWRITE mappings
 198  * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
 199  * > 0: (i_writecount) users are writing to the file.
 200  *
 201  * Normally we operate on that counter with atomic_{inc,dec} and it's safe
 202  * except for the cases where we don't hold i_writecount yet. Then we need to
 203  * use {get,deny}_write_access() - these functions check the sign and refuse
 204  * to do the change if sign is wrong. Exclusion between them is provided by
 205  * spinlock (arbitration_lock) and I'll rip the second arsehole to the first
 206  * who will try to move it in struct inode - just leave it here.
 207  */
 208 static spinlock_t arbitration_lock = SPIN_LOCK_UNLOCKED;
 209 int get_write_access(struct inode * inode)
 210 {
 211         spin_lock(&arbitration_lock);
 212         if (atomic_read(&inode->i_writecount) < 0) {
 213                 spin_unlock(&arbitration_lock);
 214                 return -ETXTBSY;
 215         }
 216         atomic_inc(&inode->i_writecount);
 217         spin_unlock(&arbitration_lock);
 218         return 0;
 219 }
 220 int deny_write_access(struct file * file)
 221 {
 222         spin_lock(&arbitration_lock);
 223         if (atomic_read(&file->f_dentry->d_inode->i_writecount) > 0) {
 224                 spin_unlock(&arbitration_lock);
 225                 return -ETXTBSY;
 226         }
 227         atomic_dec(&file->f_dentry->d_inode->i_writecount);
 228         spin_unlock(&arbitration_lock);
 229         return 0;
 230 }
 231
 232 void path_release(struct nameidata *nd)
 233 {
 234         dput(nd->dentry);
 235         mntput(nd->mnt);
 236 }
 237
 238 /*
 239  * Internal lookup() using the new generic dcache.
 240  * SMP-safe
 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  * SMP-safe
 262  */
 263 static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, int flags)
 264 {
 265         struct dentry * result;
 266         struct inode *dir = parent->d_inode;
 267
 268         down(&dir->i_sem);
 269         /*
 270          * First re-do the cached lookup just in case it was created
 271          * while we waited for the directory semaphore..
 272          *
 273          * FIXME! This could use version numbering or similar to
 274          * avoid unnecessary cache lookups.
 275          */
 276         result = d_lookup(parent, name);
 277         if (!result) {
 278                 struct dentry * dentry = d_alloc(parent, name);
 279                 result = ERR_PTR(-ENOMEM);
 280                 if (dentry) {
 281                         lock_kernel();
 282                         result = dir->i_op->lookup(dir, dentry);
 283                         unlock_kernel();
 284                         if (result)
 285                                 dput(dentry);
 286                         else
 287                                 result = dentry;
 288                 }
 289                 up(&dir->i_sem);
 290                 return result;
 291         }
 292
 293         /*
 294          * Uhhuh! Nasty case: the cache was re-populated while
 295          * we waited on the semaphore. Need to revalidate.
 296          */
 297         up(&dir->i_sem);
 298         if (result->d_op && result->d_op->d_revalidate) {
 299                 if (!result->d_op->d_revalidate(result, flags) && !d_invalidate(result)) {
 300                         dput(result);
 301                         result = ERR_PTR(-ENOENT);
 302                 }
 303         }
 304         return result;
 305 }
 306
 307 static inline int do_follow_link(struct dentry *dentry, struct nameidata *nd)
 308 {
 309         int err;
 310         if (current->link_count >= 8)
 311                 goto loop;
 312         current->link_count++;
 313         UPDATE_ATIME(dentry->d_inode);
 314         err = dentry->d_inode->i_op->follow_link(dentry, nd);
 315         current->link_count--;
 316         return err;
 317 loop:
 318         path_release(nd);
 319         return -ELOOP;
 320 }
 321
 322 static inline int __follow_up(struct vfsmount **mnt, struct dentry **base)
 323 {
 324         struct vfsmount *parent;
 325         struct dentry *dentry;
 326         spin_lock(&dcache_lock);
 327         parent=(*mnt)->mnt_parent;
 328         if (parent == *mnt) {
 329                 spin_unlock(&dcache_lock);
 330                 return 0;
 331         }
 332         mntget(parent);
 333         dentry=dget((*mnt)->mnt_mountpoint);
 334         spin_unlock(&dcache_lock);
 335         dput(*base);
 336         *base = dentry;
 337         mntput(*mnt);
 338         *mnt = parent;
 339         return 1;
 340 }
 341
 342 int follow_up(struct vfsmount **mnt, struct dentry **dentry)
 343 {
 344         return __follow_up(mnt, dentry);
 345 }
 346
 347 static inline int __follow_down(struct vfsmount **mnt, struct dentry **dentry)
 348 {
 349         struct list_head *p;
 350         spin_lock(&dcache_lock);
 351         p = (*dentry)->d_vfsmnt.next;
 352         while (p != &(*dentry)->d_vfsmnt) {
 353                 struct vfsmount *tmp;
 354                 tmp = list_entry(p, struct vfsmount, mnt_clash);
 355                 if (tmp->mnt_parent == *mnt) {
 356                         *mnt = mntget(tmp);
 357                         spin_unlock(&dcache_lock);
 358                         mntput(tmp->mnt_parent);
 359                         /* tmp holds the mountpoint, so... */
 360                         dput(*dentry);
 361                         *dentry = dget(tmp->mnt_root);
 362                         return 1;
 363                 }
 364                 p = p->next;
 365         }
 366         spin_unlock(&dcache_lock);
 367         return 0;
 368 }
 369
 370 int follow_down(struct vfsmount **mnt, struct dentry **dentry)
 371 {
 372         return __follow_down(mnt,dentry);
 373 }
 374
 375 static inline void follow_dotdot(struct nameidata *nd)
 376 {
 377         while(1) {
 378                 struct vfsmount *parent;
 379                 struct dentry *dentry;
 380                 read_lock(&current->fs->lock);
 381                 if (nd->dentry == current->fs->root &&
 382                     nd->mnt == current->fs->rootmnt)  {
 383                         read_unlock(&current->fs->lock);
 384                         break;
 385                 }
 386                 read_unlock(&current->fs->lock);
 387                 spin_lock(&dcache_lock);
 388                 if (nd->dentry != nd->mnt->mnt_root) {
 389                         dentry = dget(nd->dentry->d_parent);
 390                         spin_unlock(&dcache_lock);
 391                         dput(nd->dentry);
 392                         nd->dentry = dentry;
 393                         break;
 394                 }
 395                 parent=nd->mnt->mnt_parent;
 396                 if (parent == nd->mnt) {
 397                         spin_unlock(&dcache_lock);
 398                         break;
 399                 }
 400                 mntget(parent);
 401                 dentry=dget(nd->mnt->mnt_mountpoint);
 402                 spin_unlock(&dcache_lock);
 403                 dput(nd->dentry);
 404                 nd->dentry = dentry;
 405                 mntput(nd->mnt);
 406                 nd->mnt = parent;
 407         }
 408 }
 409 /*
 410  * Name resolution.
 411  *
 412  * This is the basic name resolution function, turning a pathname
 413  * into the final dentry.
 414  *
 415  * We expect 'base' to be positive and a directory.
 416  */
 417 int path_walk(const char * name, struct nameidata *nd)
 418 {
 419         struct dentry *dentry;
 420         struct inode *inode;
 421         int err;
 422         unsigned int lookup_flags = nd->flags;
 423
 424         while (*name=='/')
 425                 name++;
 426         if (!*name)
 427                 goto return_base;
 428
 429         inode = nd->dentry->d_inode;
 430         if (current->link_count)
 431                 lookup_flags = LOOKUP_FOLLOW;
 432
 433         /* At this point we know we have a real path component. */
 434         for(;;) {
 435                 unsigned long hash;
 436                 struct qstr this;
 437                 unsigned int c;
 438
 439                 err = permission(inode, MAY_EXEC);
 440                 dentry = ERR_PTR(err);
 441                 if (err)
 442                         break;
 443
 444                 this.name = name;
 445                 c = *(const unsigned char *)name;
 446
 447                 hash = init_name_hash();
 448                 do {
 449                         name++;
 450                         hash = partial_name_hash(c, hash);
 451                         c = *(const unsigned char *)name;
 452                 } while (c && (c != '/'));
 453                 this.len = name - (const char *) this.name;
 454                 this.hash = end_name_hash(hash);
 455
 456                 /* remove trailing slashes? */
 457                 if (!c)
 458                         goto last_component;
 459                 while (*++name == '/');
 460                 if (!*name)
 461                         goto last_with_slashes;
 462
 463                 /*
 464                  * "." and ".." are special - ".." especially so because it has
 465                  * to be able to know about the current root directory and
 466                  * parent relationships.
 467                  */
 468                 if (this.name[0] == '.') switch (this.len) {
 469                         default:
 470                                 break;
 471                         case 2:
 472                                 if (this.name[1] != '.')
 473                                         break;
 474                                 follow_dotdot(nd);
 475                                 inode = nd->dentry->d_inode;
 476                                 /* fallthrough */
 477                         case 1:
 478                                 continue;
 479                 }
 480                 /*
 481                  * See if the low-level filesystem might want
 482                  * to use its own hash..
 483                  */
 484                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 485                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 486                         if (err < 0)
 487                                 break;
 488                 }
 489                 /* This does the actual lookups.. */
 490                 dentry = cached_lookup(nd->dentry, &this, LOOKUP_CONTINUE);
 491                 if (!dentry) {
 492                         dentry = real_lookup(nd->dentry, &this, LOOKUP_CONTINUE);
 493                         err = PTR_ERR(dentry);
 494                         if (IS_ERR(dentry))
 495                                 break;
 496                 }
 497                 /* Check mountpoints.. */
 498                 while (d_mountpoint(dentry) && __follow_down(&nd->mnt, &dentry))
 499                         ;
 500
 501                 err = -ENOENT;
 502                 inode = dentry->d_inode;
 503                 if (!inode)
 504                         goto out_dput;
 505                 err = -ENOTDIR;
 506                 if (!inode->i_op)
 507                         goto out_dput;
 508
 509                 if (inode->i_op->follow_link) {
 510                         err = do_follow_link(dentry, nd);
 511                         dput(dentry);
 512                         if (err)
 513                                 goto return_err;
 514                         err = -ENOENT;
 515                         inode = nd->dentry->d_inode;
 516                         if (!inode)
 517                                 break;
 518                         err = -ENOTDIR;
 519                         if (!inode->i_op)
 520                                 break;
 521                 } else {
 522                         dput(nd->dentry);
 523                         nd->dentry = dentry;
 524                 }
 525                 err = -ENOTDIR;
 526                 if (!inode->i_op->lookup)
 527                         break;
 528                 continue;
 529                 /* here ends the main loop */
 530
 531 last_with_slashes:
 532                 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
 533 last_component:
 534                 if (lookup_flags & LOOKUP_PARENT)
 535                         goto lookup_parent;
 536                 if (this.name[0] == '.') switch (this.len) {
 537                         default:
 538                                 break;
 539                         case 2:
 540                                 if (this.name[1] != '.')
 541                                         break;
 542                                 follow_dotdot(nd);
 543                                 inode = nd->dentry->d_inode;
 544                                 /* fallthrough */
 545                         case 1:
 546                                 goto return_base;
 547                 }
 548                 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
 549                         err = nd->dentry->d_op->d_hash(nd->dentry, &this);
 550                         if (err < 0)
 551                                 break;
 552                 }
 553                 dentry = cached_lookup(nd->dentry, &this, 0);
 554                 if (!dentry) {
 555                         dentry = real_lookup(nd->dentry, &this, 0);
 556                         err = PTR_ERR(dentry);
 557                         if (IS_ERR(dentry))
 558                                 break;
 559                 }
 560                 while (d_mountpoint(dentry) && __follow_down(&nd->mnt, &dentry))
 561                         ;
 562                 inode = dentry->d_inode;
 563                 if ((lookup_flags & LOOKUP_FOLLOW)
 564                     && inode && inode->i_op && inode->i_op->follow_link) {
 565                         err = do_follow_link(dentry, nd);
 566                         dput(dentry);
 567                         if (err)
 568                                 goto return_err;
 569                         inode = nd->dentry->d_inode;
 570                 } else {
 571                         dput(nd->dentry);
 572                         nd->dentry = dentry;
 573                 }
 574                 err = -ENOENT;
 575                 if (!inode)
 576                         goto no_inode;
 577                 if (lookup_flags & LOOKUP_DIRECTORY) {
 578                         err = -ENOTDIR;
 579                         if (!inode->i_op || !inode->i_op->lookup)
 580                                 break;
 581                 }
 582                 goto return_base;
 583 no_inode:
 584                 err = -ENOENT;
 585                 if (lookup_flags & (LOOKUP_POSITIVE|LOOKUP_DIRECTORY))
 586                         break;
 587                 goto return_base;
 588 lookup_parent:
 589                 nd->last = this;
 590                 nd->last_type = LAST_NORM;
 591                 if (this.name[0] != '.')
 592                         goto return_base;
 593                 if (this.len == 1)
 594                         nd->last_type = LAST_DOT;
 595                 else if (this.len == 2 && this.name[1] == '.')
 596                         nd->last_type = LAST_DOTDOT;
 597 return_base:
 598                 return 0;
 599 out_dput:
 600                 dput(dentry);
 601                 break;
 602         }
 603         path_release(nd);
 604 return_err:
 605         return err;
 606 }
 607
 608 /* SMP-safe */
 609 /* returns 1 if everything is done */
 610 static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
 611 {
 612         if (path_walk(name, nd))
 613                 return 0;
 614
 615         if (!nd->dentry->d_inode) {
 616                 struct nameidata nd_root;
 617                 nd_root.last_type = LAST_ROOT;
 618                 nd_root.flags = nd->flags;
 619                 read_lock(&current->fs->lock);
 620                 nd_root.mnt = mntget(current->fs->rootmnt);
 621                 nd_root.dentry = dget(current->fs->root);
 622                 read_unlock(&current->fs->lock);
 623                 if (path_walk(name, &nd_root))
 624                         return 1;
 625                 if (nd_root.dentry->d_inode) {
 626                         path_release(nd);
 627                         nd->dentry = nd_root.dentry;
 628                         nd->mnt = nd_root.mnt;
 629                         nd->last = nd_root.last;
 630                         return 1;
 631                 }
 632                 path_release(&nd_root);
 633         }
 634         return 1;
 635 }
 636
 637 void set_fs_altroot(void)
 638 {
 639         char *emul = __emul_prefix();
 640         struct nameidata nd;
 641         struct vfsmount *mnt = NULL, *oldmnt;
 642         struct dentry *dentry = NULL, *olddentry;
 643         if (emul) {
 644                 read_lock(&current->fs->lock);
 645                 nd.mnt = mntget(current->fs->rootmnt);
 646                 nd.dentry = dget(current->fs->root);
 647                 read_unlock(&current->fs->lock);
 648                 nd.flags = LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_POSITIVE;
 649                 if (path_walk(emul,&nd) == 0) {
 650                         mnt = nd.mnt;
 651                         dentry = nd.dentry;
 652                 }
 653         }
 654         write_lock(&current->fs->lock);
 655         oldmnt = current->fs->altrootmnt;
 656         olddentry = current->fs->altroot;
 657         current->fs->altrootmnt = mnt;
 658         current->fs->altroot = dentry;
 659         write_unlock(&current->fs->lock);
 660         if (olddentry) {
 661                 dput(olddentry);
 662                 mntput(oldmnt);
 663         }
 664 }
 665
 666 /* SMP-safe */
 667 static inline int
 668 walk_init_root(const char *name, struct nameidata *nd)
 669 {
 670         read_lock(&current->fs->lock);
 671         if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
 672                 nd->mnt = mntget(current->fs->altrootmnt);
 673                 nd->dentry = dget(current->fs->altroot);
 674                 read_unlock(&current->fs->lock);
 675                 if (__emul_lookup_dentry(name,nd))
 676                         return 0;
 677                 read_lock(&current->fs->lock);
 678         }
 679         nd->mnt = mntget(current->fs->rootmnt);
 680         nd->dentry = dget(current->fs->root);
 681         read_unlock(&current->fs->lock);
 682         return 1;
 683 }
 684
 685 /* SMP-safe */
 686 int path_init(const char *name,unsigned int flags,struct nameidata *nd)
 687 {
 688         nd->last_type = LAST_ROOT; /* if there are only slashes... */
 689         nd->flags = flags;
 690         if (*name=='/')
 691                 return walk_init_root(name,nd);
 692         read_lock(&current->fs->lock);
 693         nd->mnt = mntget(current->fs->pwdmnt);
 694         nd->dentry = dget(current->fs->pwd);
 695         read_unlock(&current->fs->lock);
 696         return 1;
 697 }
 698
 699 /*
 700  * Restricted form of lookup. Doesn't follow links, single-component only,
 701  * needs parent already locked. Doesn't follow mounts.
 702  * SMP-safe.
 703  */
 704 struct dentry * lookup_hash(struct qstr *name, struct dentry * base)
 705 {
 706         struct dentry * dentry;
 707         struct inode *inode;
 708         int err;
 709
 710         inode = base->d_inode;
 711         err = permission(inode, MAY_EXEC);
 712         dentry = ERR_PTR(err);
 713         if (err)
 714                 goto out;
 715
 716         /*
 717          * See if the low-level filesystem might want
 718          * to use its own hash..
 719          */
 720         if (base->d_op && base->d_op->d_hash) {
 721                 err = base->d_op->d_hash(base, name);
 722                 dentry = ERR_PTR(err);
 723                 if (err < 0)
 724                         goto out;
 725         }
 726
 727         dentry = cached_lookup(base, name, 0);
 728         if (!dentry) {
 729                 struct dentry *new = d_alloc(base, name);
 730                 dentry = ERR_PTR(-ENOMEM);
 731                 if (!new)
 732                         goto out;
 733                 lock_kernel();
 734                 dentry = inode->i_op->lookup(inode, new);
 735                 unlock_kernel();
 736                 if (!dentry)
 737                         dentry = new;
 738                 else
 739                         dput(new);
 740         }
 741 out:
 742         return dentry;
 743 }
 744
 745 /* SMP-safe */
 746 struct dentry * lookup_one(const char * name, struct dentry * base)
 747 {
 748         unsigned long hash;
 749         struct qstr this;
 750         unsigned int c;
 751
 752         this.name = name;
 753         c = *(const unsigned char *)name;
 754         if (!c)
 755                 goto access;
 756
 757         hash = init_name_hash();
 758         do {
 759                 name++;
 760                 if (c == '/')
 761                         goto access;
 762                 hash = partial_name_hash(c, hash);
 763                 c = *(const unsigned char *)name;
 764         } while (c);
 765         this.len = name - (const char *) this.name;
 766         this.hash = end_name_hash(hash);
 767
 768         return lookup_hash(&this, base);
 769 access:
 770         return ERR_PTR(-EACCES);
 771 }
 772
 773 /*
 774  *      namei()
 775  *
 776  * is used by most simple commands to get the inode of a specified name.
 777  * Open, link etc use their own routines, but this is enough for things
 778  * like 'chmod' etc.
 779  *
 780  * namei exists in two versions: namei/lnamei. The only difference is
 781  * that namei follows links, while lnamei does not.
 782  * SMP-safe
 783  */
 784 int __user_walk(const char *name, unsigned flags, struct nameidata *nd)
 785 {
 786         char *tmp;
 787         int err;
 788
 789         tmp = getname(name);
 790         err = PTR_ERR(tmp);
 791         if (!IS_ERR(tmp)) {
 792                 err = 0;
 793                 if (path_init(tmp, flags, nd))
 794                         err = path_walk(tmp, nd);
 795                 putname(tmp);
 796         }
 797         return err;
 798 }
 799
 800 /*
 801  * It's inline, so penalty for filesystems that don't use sticky bit is
 802  * minimal.
 803  */
 804 static inline int check_sticky(struct inode *dir, struct inode *inode)
 805 {
 806         if (!(dir->i_mode & S_ISVTX))
 807                 return 0;
 808         if (inode->i_uid == current->fsuid)
 809                 return 0;
 810         if (dir->i_uid == current->fsuid)
 811                 return 0;
 812         return !capable(CAP_FOWNER);
 813 }
 814
 815 /*
 816  *      Check whether we can remove a link victim from directory dir, check
 817  *  whether the type of victim is right.
 818  *  1. We can't do it if dir is read-only (done in permission())
 819  *  2. We should have write and exec permissions on dir
 820  *  3. We can't remove anything from append-only dir
 821  *  4. We can't do anything with immutable dir (done in permission())
 822  *  5. If the sticky bit on dir is set we should either
 823  *      a. be owner of dir, or
 824  *      b. be owner of victim, or
 825  *      c. have CAP_FOWNER capability
 826  *  6. If the victim is append-only or immutable we can't do antyhing with
 827  *     links pointing to it.
 828  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
 829  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
 830  *  9. We can't remove a root or mountpoint.
 831  */
 832 static inline int may_delete(struct inode *dir,struct dentry *victim, int isdir)
 833 {
 834         int error;
 835         if (!victim->d_inode || victim->d_parent->d_inode != dir)
 836                 return -ENOENT;
 837         error = permission(dir,MAY_WRITE | MAY_EXEC);
 838         if (error)
 839                 return error;
 840         if (IS_APPEND(dir))
 841                 return -EPERM;
 842         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
 843             IS_IMMUTABLE(victim->d_inode))
 844                 return -EPERM;
 845         if (isdir) {
 846                 if (!S_ISDIR(victim->d_inode->i_mode))
 847                         return -ENOTDIR;
 848                 if (IS_ROOT(victim))
 849                         return -EBUSY;
 850         } else if (S_ISDIR(victim->d_inode->i_mode))
 851                 return -EISDIR;
 852         return 0;
 853 }
 854
 855 /*      Check whether we can create an object with dentry child in directory
 856  *  dir.
 857  *  1. We can't do it if child already exists (open has special treatment for
 858  *     this case, but since we are inlined it's OK)
 859  *  2. We can't do it if dir is read-only (done in permission())
 860  *  3. We should have write and exec permissions on dir
 861  *  4. We can't do it if dir is immutable (done in permission())
 862  */
 863 static inline int may_create(struct inode *dir, struct dentry *child) {
 864         if (child->d_inode)
 865                 return -EEXIST;
 866         if (IS_DEADDIR(dir))
 867                 return -ENOENT;
 868         return permission(dir,MAY_WRITE | MAY_EXEC);
 869 }
 870
 871 /*
 872  * Special case: O_CREAT|O_EXCL implies O_NOFOLLOW for security
 873  * reasons.
 874  *
 875  * O_DIRECTORY translates into forcing a directory lookup.
 876  */
 877 static inline int lookup_flags(unsigned int f)
 878 {
 879         unsigned long retval = LOOKUP_FOLLOW;
 880
 881         if (f & O_NOFOLLOW)
 882                 retval &= ~LOOKUP_FOLLOW;
 883
 884         if ((f & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 885                 retval &= ~LOOKUP_FOLLOW;
 886
 887         if (f & O_DIRECTORY)
 888                 retval |= LOOKUP_DIRECTORY;
 889
 890         return retval;
 891 }
 892
 893 int vfs_create(struct inode *dir, struct dentry *dentry, int mode)
 894 {
 895         int error;
 896
 897         mode &= S_IALLUGO & ~current->fs->umask;
 898         mode |= S_IFREG;
 899
 900         down(&dir->i_zombie);
 901         error = may_create(dir, dentry);
 902         if (error)
 903                 goto exit_lock;
 904
 905         error = -EACCES;        /* shouldn't it be ENOSYS? */
 906         if (!dir->i_op || !dir->i_op->create)
 907                 goto exit_lock;
 908
 909         DQUOT_INIT(dir);
 910         lock_kernel();
 911         error = dir->i_op->create(dir, dentry, mode);
 912         unlock_kernel();
 913 exit_lock:
 914         up(&dir->i_zombie);
 915         return error;
 916 }
 917
 918 /*
 919  *      open_namei()
 920  *
 921  * namei for open - this is in fact almost the whole open-routine.
 922  *
 923  * Note that the low bits of "flag" aren't the same as in the open
 924  * system call - they are 00 - no permissions needed
 925  *                        01 - read permission needed
 926  *                        10 - write permission needed
 927  *                        11 - read/write permissions needed
 928  * which is a lot more logical, and also allows the "no perm" needed
 929  * for symlinks (where the permissions are checked later).
 930  * SMP-safe
 931  */
 932 int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
 933 {
 934         int acc_mode, error = 0;
 935         struct inode *inode;
 936         struct dentry *dentry;
 937         struct dentry *dir;
 938         int count = 0;
 939
 940         acc_mode = ACC_MODE(flag);
 941
 942         /*
 943          * The simplest case - just a plain lookup.
 944          */
 945         if (!(flag & O_CREAT)) {
 946                 if (path_init(pathname, lookup_flags(flag), nd))
 947                         error = path_walk(pathname, nd);
 948                 if (error)
 949                         return error;
 950                 dentry = nd->dentry;
 951                 goto ok;
 952         }
 953
 954         /*
 955          * Create - we need to know the parent.
 956          */
 957         if (path_init(pathname, LOOKUP_PARENT, nd))
 958                 error = path_walk(pathname, nd);
 959         if (error)
 960                 return error;
 961
 962         /*
 963          * We have the parent and last component. First of all, check
 964          * that we are not asked to creat(2) an obvious directory - that
 965          * will not do.
 966          */
 967         error = -EISDIR;
 968         if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
 969                 goto exit;
 970
 971         dir = nd->dentry;
 972         down(&dir->d_inode->i_sem);
 973         dentry = lookup_hash(&nd->last, nd->dentry);
 974
 975 do_last:
 976         error = PTR_ERR(dentry);
 977         if (IS_ERR(dentry)) {
 978                 up(&dir->d_inode->i_sem);
 979                 goto exit;
 980         }
 981
 982         /* Negative dentry, just create the file */
 983         if (!dentry->d_inode) {
 984                 error = vfs_create(dir->d_inode, dentry, mode);
 985                 up(&dir->d_inode->i_sem);
 986                 dput(nd->dentry);
 987                 nd->dentry = dentry;
 988                 if (error)
 989                         goto exit;
 990                 /* Don't check for write permission, don't truncate */
 991                 acc_mode = 0;
 992                 flag &= ~O_TRUNC;
 993                 goto ok;
 994         }
 995
 996         /*
 997          * It already exists.
 998          */
 999         up(&dir->d_inode->i_sem);
1000
1001         error = -EEXIST;
1002         if (flag & O_EXCL)
1003                 goto exit_dput;
1004
1005         if (d_mountpoint(dentry)) {
1006                 error = -ELOOP;
1007                 if (flag & O_NOFOLLOW)
1008                         goto exit_dput;
1009                 do __follow_down(&nd->mnt,&dentry); while(d_mountpoint(dentry));
1010         }
1011         error = -ENOENT;
1012         if (!dentry->d_inode)
1013                 goto exit_dput;
1014         if (dentry->d_inode->i_op && dentry->d_inode->i_op->follow_link)
1015                 goto do_link;
1016
1017         dput(nd->dentry);
1018         nd->dentry = dentry;
1019         error = -EISDIR;
1020         if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode))
1021                 goto exit;
1022 ok:
1023         error = -ENOENT;
1024         inode = dentry->d_inode;
1025         if (!inode)
1026                 goto exit;
1027
1028         error = -ELOOP;
1029         if (S_ISLNK(inode->i_mode))
1030                 goto exit;
1031
1032         error = -EISDIR;
1033         if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1034                 goto exit;
1035
1036         error = permission(inode,acc_mode);
1037         if (error)
1038                 goto exit;
1039
1040         /*
1041          * FIFO's, sockets and device files are special: they don't
1042          * actually live on the filesystem itself, and as such you
1043          * can write to them even if the filesystem is read-only.
1044          */
1045         if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1046                 flag &= ~O_TRUNC;
1047         } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1048                 error = -EACCES;
1049                 if (IS_NODEV(inode))
1050                         goto exit;
1051
1052                 flag &= ~O_TRUNC;
1053         } else {
1054                 error = -EROFS;
1055                 if (IS_RDONLY(inode) && (flag & 2))
1056                         goto exit;
1057         }
1058         /*
1059          * An append-only file must be opened in append mode for writing.
1060          */
1061         error = -EPERM;
1062         if (IS_APPEND(inode)) {
1063                 if  ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1064                         goto exit;
1065                 if (flag & O_TRUNC)
1066                         goto exit;
1067         }
1068
1069         if (flag & O_TRUNC) {
1070                 error = get_write_access(inode);
1071                 if (error)
1072                         goto exit;
1073
1074                 /*
1075                  * Refuse to truncate files with mandatory locks held on them.
1076                  */
1077                 error = locks_verify_locked(inode);
1078                 if (!error) {
1079                         DQUOT_INIT(inode);
1080
1081                         error = do_truncate(dentry, 0);
1082                 }
1083                 put_write_access(inode);
1084                 if (error)
1085                         goto exit;
1086         } else
1087                 if (flag & FMODE_WRITE)
1088                         DQUOT_INIT(inode);
1089
1090         return 0;
1091
1092 exit_dput:
1093         dput(dentry);
1094 exit:
1095         path_release(nd);
1096         return error;
1097
1098 do_link:
1099         error = -ELOOP;
1100         if (flag & O_NOFOLLOW)
1101                 goto exit_dput;
1102         /*
1103          * This is subtle. Instead of calling do_follow_link() we do the
1104          * thing by hands. The reason is that this way we have zero link_count
1105          * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1106          * After that we have the parent and last component, i.e.
1107          * we are in the same situation as after the first path_walk().
1108          * Well, almost - if the last component is normal we get its copy
1109          * stored in nd->last.name and we will have to putname() it when we
1110          * are done. Procfs-like symlinks just set LAST_BIND.
1111          */
1112         UPDATE_ATIME(dentry->d_inode);
1113         error = dentry->d_inode->i_op->follow_link(dentry, nd);
1114         dput(dentry);
1115         if (error)
1116                 return error;
1117         if (nd->last_type == LAST_BIND) {
1118                 dentry = nd->dentry;
1119                 goto ok;
1120         }
1121         error = -EISDIR;
1122         if (nd->last_type != LAST_NORM)
1123                 goto exit;
1124         if (nd->last.name[nd->last.len]) {
1125                 putname(nd->last.name);
1126                 goto exit;
1127         }
1128         if (count++==32) {
1129                 dentry = nd->dentry;
1130                 putname(nd->last.name);
1131                 goto ok;
1132         }
1133         dir = nd->dentry;
1134         down(&dir->d_inode->i_sem);
1135         dentry = lookup_hash(&nd->last, nd->dentry);
1136         putname(nd->last.name);
1137         goto do_last;
1138 }
1139
1140 /* SMP-safe */
1141 static struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1142 {
1143         struct dentry *dentry;
1144
1145         down(&nd->dentry->d_inode->i_sem);
1146         dentry = ERR_PTR(-EEXIST);
1147         if (nd->last_type != LAST_NORM)
1148                 goto fail;
1149         dentry = lookup_hash(&nd->last, nd->dentry);
1150         if (IS_ERR(dentry))
1151                 goto fail;
1152         if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1153                 goto enoent;
1154         return dentry;
1155 enoent:
1156         dput(dentry);
1157         dentry = ERR_PTR(-ENOENT);
1158 fail:
1159         return dentry;
1160 }
1161
1162 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1163 {
1164         int error = -EPERM;
1165
1166         mode &= ~current->fs->umask;
1167
1168         down(&dir->i_zombie);
1169         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1170                 goto exit_lock;
1171
1172         error = may_create(dir, dentry);
1173         if (error)
1174                 goto exit_lock;
1175
1176         error = -EPERM;
1177         if (!dir->i_op || !dir->i_op->mknod)
1178                 goto exit_lock;
1179
1180         DQUOT_INIT(dir);
1181         lock_kernel();
1182         error = dir->i_op->mknod(dir, dentry, mode, dev);
1183         unlock_kernel();
1184 exit_lock:
1185         up(&dir->i_zombie);
1186         return error;
1187 }
1188
1189 asmlinkage long sys_mknod(const char * filename, int mode, dev_t dev)
1190 {
1191         int error = 0;
1192         char * tmp;
1193         struct dentry * dentry;
1194         struct nameidata nd;
1195
1196         if (S_ISDIR(mode))
1197                 return -EPERM;
1198         tmp = getname(filename);
1199         if (IS_ERR(tmp))
1200                 return PTR_ERR(tmp);
1201
1202         if (path_init(tmp, LOOKUP_PARENT, &nd))
1203                 error = path_walk(tmp, &nd);
1204         if (error)
1205                 goto out;
1206         dentry = lookup_create(&nd, 0);
1207         error = PTR_ERR(dentry);
1208         if (!IS_ERR(dentry)) {
1209                 switch (mode & S_IFMT) {
1210                 case 0: case S_IFREG:
1211                         error = vfs_create(nd.dentry->d_inode,dentry,mode);
1212                         break;
1213                 case S_IFCHR: case S_IFBLK: case S_IFIFO: case S_IFSOCK:
1214                         error = vfs_mknod(nd.dentry->d_inode,dentry,mode,dev);
1215                         break;
1216                 case S_IFDIR:
1217                         error = -EPERM;
1218                         break;
1219                 default:
1220                         error = -EINVAL;
1221                 }
1222                 dput(dentry);
1223         }
1224         up(&nd.dentry->d_inode->i_sem);
1225         path_release(&nd);
1226 out:
1227         putname(tmp);
1228
1229         return error;
1230 }
1231
1232 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1233 {
1234         int error;
1235
1236         down(&dir->i_zombie);
1237         error = may_create(dir, dentry);
1238         if (error)
1239                 goto exit_lock;
1240
1241         error = -EPERM;
1242         if (!dir->i_op || !dir->i_op->mkdir)
1243                 goto exit_lock;
1244
1245         DQUOT_INIT(dir);
1246         mode &= (S_IRWXUGO|S_ISVTX) & ~current->fs->umask;
1247         lock_kernel();
1248         error = dir->i_op->mkdir(dir, dentry, mode);
1249         unlock_kernel();
1250
1251 exit_lock:
1252         up(&dir->i_zombie);
1253         return error;
1254 }
1255
1256 asmlinkage long sys_mkdir(const char * pathname, int mode)
1257 {
1258         int error = 0;
1259         char * tmp;
1260
1261         tmp = getname(pathname);
1262         error = PTR_ERR(tmp);
1263         if (!IS_ERR(tmp)) {
1264                 struct dentry *dentry;
1265                 struct nameidata nd;
1266
1267                 if (path_init(tmp, LOOKUP_PARENT, &nd))
1268                         error = path_walk(tmp, &nd);
1269                 if (error)
1270                         goto out;
1271                 dentry = lookup_create(&nd, 1);
1272                 error = PTR_ERR(dentry);
1273                 if (!IS_ERR(dentry)) {
1274                         error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1275                         dput(dentry);
1276                 }
1277                 up(&nd.dentry->d_inode->i_sem);
1278                 path_release(&nd);
1279 out:
1280                 putname(tmp);
1281         }
1282
1283         return error;
1284 }
1285
1286 /*
1287  * We try to drop the dentry early: we should have
1288  * a usage count of 2 if we're the only user of this
1289  * dentry, and if that is true (possibly after pruning
1290  * the dcache), then we drop the dentry now.
1291  *
1292  * A low-level filesystem can, if it choses, legally
1293  * do a
1294  *
1295  *      if (!d_unhashed(dentry))
1296  *              return -EBUSY;
1297  *
1298  * if it cannot handle the case of removing a directory
1299  * that is still in use by something else..
1300  */
1301 static void d_unhash(struct dentry *dentry)
1302 {
1303         dget(dentry);
1304         switch (atomic_read(&dentry->d_count)) {
1305         default:
1306                 shrink_dcache_parent(dentry);
1307                 if (atomic_read(&dentry->d_count) != 2)
1308                         break;
1309         case 2:
1310                 d_drop(dentry);
1311         }
1312 }
1313
1314 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1315 {
1316         int error;
1317
1318         error = may_delete(dir, dentry, 1);
1319         if (error)
1320                 return error;
1321
1322         if (!dir->i_op || !dir->i_op->rmdir)
1323                 return -EPERM;
1324
1325         DQUOT_INIT(dir);
1326
1327         double_down(&dir->i_zombie, &dentry->d_inode->i_zombie);
1328         d_unhash(dentry);
1329         if (IS_DEADDIR(dir))
1330                 error = -ENOENT;
1331         else if (d_mountpoint(dentry))
1332                 error = -EBUSY;
1333         else {
1334                 lock_kernel();
1335                 error = dir->i_op->rmdir(dir, dentry);
1336                 unlock_kernel();
1337                 if (!error)
1338                         dentry->d_inode->i_flags |= S_DEAD;
1339         }
1340         double_up(&dir->i_zombie, &dentry->d_inode->i_zombie);
1341         if (!error)
1342                 d_delete(dentry);
1343         dput(dentry);
1344
1345         return error;
1346 }
1347
1348 asmlinkage long sys_rmdir(const char * pathname)
1349 {
1350         int error = 0;
1351         char * name;
1352         struct dentry *dentry;
1353         struct nameidata nd;
1354
1355         name = getname(pathname);
1356         if(IS_ERR(name))
1357                 return PTR_ERR(name);
1358
1359         if (path_init(name, LOOKUP_PARENT, &nd))
1360                 error = path_walk(name, &nd);
1361         if (error)
1362                 goto exit;
1363
1364         switch(nd.last_type) {
1365                 case LAST_DOTDOT:
1366                         error = -ENOTEMPTY;
1367                         goto exit1;
1368                 case LAST_ROOT: case LAST_DOT:
1369                         error = -EBUSY;
1370                         goto exit1;
1371         }
1372         down(&nd.dentry->d_inode->i_sem);
1373         dentry = lookup_hash(&nd.last, nd.dentry);
1374         error = PTR_ERR(dentry);
1375         if (!IS_ERR(dentry)) {
1376                 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1377                 dput(dentry);
1378         }
1379         up(&nd.dentry->d_inode->i_sem);
1380 exit1:
1381         path_release(&nd);
1382 exit:
1383         putname(name);
1384         return error;
1385 }
1386
1387 int vfs_unlink(struct inode *dir, struct dentry *dentry)
1388 {
1389         int error;
1390
1391         down(&dir->i_zombie);
1392         error = may_delete(dir, dentry, 0);
1393         if (!error) {
1394                 error = -EPERM;
1395                 if (dir->i_op && dir->i_op->unlink) {
1396                         DQUOT_INIT(dir);
1397                         if (d_mountpoint(dentry))
1398                                 error = -EBUSY;
1399                         else {
1400                                 lock_kernel();
1401                                 error = dir->i_op->unlink(dir, dentry);
1402                                 unlock_kernel();
1403                                 if (!error)
1404                                         d_delete(dentry);
1405                         }
1406                 }
1407         }
1408         up(&dir->i_zombie);
1409         return error;
1410 }
1411
1412 asmlinkage long sys_unlink(const char * pathname)
1413 {
1414         int error = 0;
1415         char * name;
1416         struct dentry *dentry;
1417         struct nameidata nd;
1418
1419         name = getname(pathname);
1420         if(IS_ERR(name))
1421                 return PTR_ERR(name);
1422
1423         if (path_init(name, LOOKUP_PARENT, &nd))
1424                 error = path_walk(name, &nd);
1425         if (error)
1426                 goto exit;
1427         error = -EISDIR;
1428         if (nd.last_type != LAST_NORM)
1429                 goto exit1;
1430         down(&nd.dentry->d_inode->i_sem);
1431         dentry = lookup_hash(&nd.last, nd.dentry);
1432         error = PTR_ERR(dentry);
1433         if (!IS_ERR(dentry)) {
1434                 /* Why not before? Because we want correct error value */
1435                 if (nd.last.name[nd.last.len])
1436                         goto slashes;
1437                 error = vfs_unlink(nd.dentry->d_inode, dentry);
1438         exit2:
1439                 dput(dentry);
1440         }
1441         up(&nd.dentry->d_inode->i_sem);
1442 exit1:
1443         path_release(&nd);
1444 exit:
1445         putname(name);
1446
1447         return error;
1448
1449 slashes:
1450         error = !dentry->d_inode ? -ENOENT :
1451                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
1452         goto exit2;
1453 }
1454
1455 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
1456 {
1457         int error;
1458
1459         down(&dir->i_zombie);
1460         error = may_create(dir, dentry);
1461         if (error)
1462                 goto exit_lock;
1463
1464         error = -EPERM;
1465         if (!dir->i_op || !dir->i_op->symlink)
1466                 goto exit_lock;
1467
1468         DQUOT_INIT(dir);
1469         lock_kernel();
1470         error = dir->i_op->symlink(dir, dentry, oldname);
1471         unlock_kernel();
1472
1473 exit_lock:
1474         up(&dir->i_zombie);
1475         return error;
1476 }
1477
1478 asmlinkage long sys_symlink(const char * oldname, const char * newname)
1479 {
1480         int error = 0;
1481         char * from;
1482         char * to;
1483
1484         from = getname(oldname);
1485         if(IS_ERR(from))
1486                 return PTR_ERR(from);
1487         to = getname(newname);
1488         error = PTR_ERR(to);
1489         if (!IS_ERR(to)) {
1490                 struct dentry *dentry;
1491                 struct nameidata nd;
1492
1493                 if (path_init(to, LOOKUP_PARENT, &nd))
1494                         error = path_walk(to, &nd);
1495                 if (error)
1496                         goto out;
1497                 dentry = lookup_create(&nd, 0);
1498                 error = PTR_ERR(dentry);
1499                 if (!IS_ERR(dentry)) {
1500                         error = vfs_symlink(nd.dentry->d_inode, dentry, from);
1501                         dput(dentry);
1502                 }
1503                 up(&nd.dentry->d_inode->i_sem);
1504                 path_release(&nd);
1505 out:
1506                 putname(to);
1507         }
1508         putname(from);
1509         return error;
1510 }
1511
1512 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
1513 {
1514         struct inode *inode;
1515         int error;
1516
1517         down(&dir->i_zombie);
1518         error = -ENOENT;
1519         inode = old_dentry->d_inode;
1520         if (!inode)
1521                 goto exit_lock;
1522
1523         error = may_create(dir, new_dentry);
1524         if (error)
1525                 goto exit_lock;
1526
1527         error = -EXDEV;
1528         if (dir->i_dev != inode->i_dev)
1529                 goto exit_lock;
1530
1531         /*
1532          * A link to an append-only or immutable file cannot be created.
1533          */
1534         error = -EPERM;
1535         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1536                 goto exit_lock;
1537         if (!dir->i_op || !dir->i_op->link)
1538                 goto exit_lock;
1539
1540         DQUOT_INIT(dir);
1541         lock_kernel();
1542         error = dir->i_op->link(old_dentry, dir, new_dentry);
1543         unlock_kernel();
1544
1545 exit_lock:
1546         up(&dir->i_zombie);
1547         return error;
1548 }
1549
1550 /*
1551  * Hardlinks are often used in delicate situations.  We avoid
1552  * security-related surprises by not following symlinks on the
1553  * newname.  --KAB
1554  *
1555  * We don't follow them on the oldname either to be compatible
1556  * with linux 2.0, and to avoid hard-linking to directories
1557  * and other special files.  --ADM
1558  */
1559 asmlinkage long sys_link(const char * oldname, const char * newname)
1560 {
1561         int error;
1562         char * from;
1563         char * to;
1564
1565         from = getname(oldname);
1566         if(IS_ERR(from))
1567                 return PTR_ERR(from);
1568         to = getname(newname);
1569         error = PTR_ERR(to);
1570         if (!IS_ERR(to)) {
1571                 struct dentry *new_dentry;
1572                 struct nameidata nd, old_nd;
1573
1574                 error = 0;
1575                 if (path_init(from, LOOKUP_POSITIVE, &old_nd))
1576                         error = path_walk(from, &old_nd);
1577                 if (error)
1578                         goto exit;
1579                 if (path_init(to, LOOKUP_PARENT, &nd))
1580                         error = path_walk(to, &nd);
1581                 if (error)
1582                         goto out;
1583                 error = -EXDEV;
1584                 if (old_nd.mnt != nd.mnt)
1585                         goto out;
1586                 new_dentry = lookup_create(&nd, 0);
1587                 error = PTR_ERR(new_dentry);
1588                 if (!IS_ERR(new_dentry)) {
1589                         error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
1590                         dput(new_dentry);
1591                 }
1592                 up(&nd.dentry->d_inode->i_sem);
1593                 path_release(&nd);
1594 out:
1595                 path_release(&old_nd);
1596 exit:
1597                 putname(to);
1598         }
1599         putname(from);
1600
1601         return error;
1602 }
1603
1604 /*
1605  * The worst of all namespace operations - renaming directory. "Perverted"
1606  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
1607  * Problems:
1608  *      a) we can get into loop creation. Check is done in is_subdir().
1609  *      b) race potential - two innocent renames can create a loop together.
1610  *         That's where 4.4 screws up. Current fix: serialization on
1611  *         sb->s_vfs_rename_sem. We might be more accurate, but that's another
1612  *         story.
1613  *      c) we have to lock _three_ objects - parents and victim (if it exists).
1614  *         And that - after we got ->i_sem on parents (until then we don't know
1615  *         whether the target exists at all, let alone whether it is a directory
1616  *         or not). Solution: ->i_zombie. Taken only after ->i_sem. Always taken
1617  *         on link creation/removal of any kind. And taken (without ->i_sem) on
1618  *         directory that will be removed (both in rmdir() and here).
1619  *      d) some filesystems don't support opened-but-unlinked directories,
1620  *         either because of layout or because they are not ready to deal with
1621  *         all cases correctly. The latter will be fixed (taking this sort of
1622  *         stuff into VFS), but the former is not going away. Solution: the same
1623  *         trick as in rmdir().
1624  *      e) conversion from fhandle to dentry may come in the wrong moment - when
1625  *         we are removing the target. Solution: we will have to grab ->i_zombie
1626  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
1627  *         ->i_sem on parents, which works but leads to some truely excessive
1628  *         locking].
1629  */
1630 int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
1631                struct inode *new_dir, struct dentry *new_dentry)
1632 {
1633         int error;
1634         struct inode *target;
1635
1636         if (old_dentry->d_inode == new_dentry->d_inode)
1637                 return 0;
1638
1639         error = may_delete(old_dir, old_dentry, 1);
1640         if (error)
1641                 return error;
1642
1643         if (new_dir->i_dev != old_dir->i_dev)
1644                 return -EXDEV;
1645
1646         if (!new_dentry->d_inode)
1647                 error = may_create(new_dir, new_dentry);
1648         else
1649                 error = may_delete(new_dir, new_dentry, 1);
1650         if (error)
1651                 return error;
1652
1653         if (!old_dir->i_op || !old_dir->i_op->rename)
1654                 return -EPERM;
1655
1656         /*
1657          * If we are going to change the parent - check write permissions,
1658          * we'll need to flip '..'.
1659          */
1660         if (new_dir != old_dir) {
1661                 error = permission(old_dentry->d_inode, MAY_WRITE);
1662         }
1663         if (error)
1664                 return error;
1665
1666         DQUOT_INIT(old_dir);
1667         DQUOT_INIT(new_dir);
1668         down(&old_dir->i_sb->s_vfs_rename_sem);
1669         error = -EINVAL;
1670         if (is_subdir(new_dentry, old_dentry))
1671                 goto out_unlock;
1672         target = new_dentry->d_inode;
1673         if (target) { /* Hastur! Hastur! Hastur! */
1674                 triple_down(&old_dir->i_zombie,
1675                             &new_dir->i_zombie,
1676                             &target->i_zombie);
1677                 d_unhash(new_dentry);
1678         } else
1679                 double_down(&old_dir->i_zombie,
1680                             &new_dir->i_zombie);
1681         if (IS_DEADDIR(old_dir)||IS_DEADDIR(new_dir))
1682                 error = -ENOENT;
1683         else if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1684                 error = -EBUSY;
1685         else
1686                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1687         if (target) {
1688                 if (!error)
1689                         target->i_flags |= S_DEAD;
1690                 triple_up(&old_dir->i_zombie,
1691                           &new_dir->i_zombie,
1692                           &target->i_zombie);
1693                 d_rehash(new_dentry);
1694                 dput(new_dentry);
1695         } else
1696                 double_up(&old_dir->i_zombie,
1697                           &new_dir->i_zombie);
1698
1699         if (!error)
1700                 d_move(old_dentry,new_dentry);
1701 out_unlock:
1702         up(&old_dir->i_sb->s_vfs_rename_sem);
1703         return error;
1704 }
1705
1706 int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
1707                struct inode *new_dir, struct dentry *new_dentry)
1708 {
1709         int error;
1710
1711         if (old_dentry->d_inode == new_dentry->d_inode)
1712                 return 0;
1713
1714         error = may_delete(old_dir, old_dentry, 0);
1715         if (error)
1716                 return error;
1717
1718         if (new_dir->i_dev != old_dir->i_dev)
1719                 return -EXDEV;
1720
1721         if (!new_dentry->d_inode)
1722                 error = may_create(new_dir, new_dentry);
1723         else
1724                 error = may_delete(new_dir, new_dentry, 0);
1725         if (error)
1726                 return error;
1727
1728         if (!old_dir->i_op || !old_dir->i_op->rename)
1729                 return -EPERM;
1730
1731         DQUOT_INIT(old_dir);
1732         DQUOT_INIT(new_dir);
1733         double_down(&old_dir->i_zombie, &new_dir->i_zombie);
1734         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
1735                 error = -EBUSY;
1736         else
1737                 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
1738         double_up(&old_dir->i_zombie, &new_dir->i_zombie);
1739         if (error)
1740                 return error;
1741         /* The following d_move() should become unconditional */
1742         if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME)) {
1743                 d_move(old_dentry, new_dentry);
1744         }
1745         return 0;
1746 }
1747
1748 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1749                struct inode *new_dir, struct dentry *new_dentry)
1750 {
1751         if (S_ISDIR(old_dentry->d_inode->i_mode))
1752                 return vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
1753         else
1754                 return vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
1755 }
1756
1757 static inline int do_rename(const char * oldname, const char * newname)
1758 {
1759         int error = 0;
1760         struct dentry * old_dir, * new_dir;
1761         struct dentry * old_dentry, *new_dentry;
1762         struct nameidata oldnd, newnd;
1763
1764         if (path_init(oldname, LOOKUP_PARENT, &oldnd))
1765                 error = path_walk(oldname, &oldnd);
1766
1767         if (error)
1768                 goto exit;
1769
1770         if (path_init(newname, LOOKUP_PARENT, &newnd))
1771                 error = path_walk(newname, &newnd);
1772         if (error)
1773                 goto exit1;
1774
1775         error = -EXDEV;
1776         if (oldnd.mnt != newnd.mnt)
1777                 goto exit2;
1778
1779         old_dir = oldnd.dentry;
1780         error = -EBUSY;
1781         if (oldnd.last_type != LAST_NORM)
1782                 goto exit2;
1783
1784         new_dir = newnd.dentry;
1785         if (newnd.last_type != LAST_NORM)
1786                 goto exit2;
1787
1788         double_lock(new_dir, old_dir);
1789
1790         old_dentry = lookup_hash(&oldnd.last, old_dir);
1791         error = PTR_ERR(old_dentry);
1792         if (IS_ERR(old_dentry))
1793                 goto exit3;
1794         /* source must exist */
1795         error = -ENOENT;
1796         if (!old_dentry->d_inode)
1797                 goto exit4;
1798         /* unless the source is a directory trailing slashes give -ENOTDIR */
1799         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
1800                 error = -ENOTDIR;
1801                 if (oldnd.last.name[oldnd.last.len])
1802                         goto exit4;
1803                 if (newnd.last.name[newnd.last.len])
1804                         goto exit4;
1805         }
1806         new_dentry = lookup_hash(&newnd.last, new_dir);
1807         error = PTR_ERR(new_dentry);
1808         if (IS_ERR(new_dentry))
1809                 goto exit4;
1810
1811         lock_kernel();
1812         error = vfs_rename(old_dir->d_inode, old_dentry,
1813                                    new_dir->d_inode, new_dentry);
1814         unlock_kernel();
1815
1816         dput(new_dentry);
1817 exit4:
1818         dput(old_dentry);
1819 exit3:
1820         double_up(&new_dir->d_inode->i_sem, &old_dir->d_inode->i_sem);
1821 exit2:
1822         path_release(&newnd);
1823 exit1:
1824         path_release(&oldnd);
1825 exit:
1826         return error;
1827 }
1828
1829 asmlinkage long sys_rename(const char * oldname, const char * newname)
1830 {
1831         int error;
1832         char * from;
1833         char * to;
1834
1835         from = getname(oldname);
1836         if(IS_ERR(from))
1837                 return PTR_ERR(from);
1838         to = getname(newname);
1839         error = PTR_ERR(to);
1840         if (!IS_ERR(to)) {
1841                 error = do_rename(from,to);
1842                 putname(to);
1843         }
1844         putname(from);
1845         return error;
1846 }
1847
1848 int vfs_readlink(struct dentry *dentry, char *buffer, int buflen, const char *link)
1849 {
1850         int len;
1851
1852         len = PTR_ERR(link);
1853         if (IS_ERR(link))
1854                 goto out;
1855
1856         len = strlen(link);
1857         if (len > (unsigned) buflen)
1858                 len = buflen;
1859         if (copy_to_user(buffer, link, len))
1860                 len = -EFAULT;
1861 out:
1862         return len;
1863 }
1864
1865 static inline int
1866 __vfs_follow_link(struct nameidata *nd, const char *link)
1867 {
1868         int res = 0;
1869         char *name;
1870         if (IS_ERR(link))
1871                 goto fail;
1872
1873         if (*link == '/') {
1874                 path_release(nd);
1875                 if (!walk_init_root(link, nd))
1876                         /* weird __emul_prefix() stuff did it */
1877                         goto out;
1878         }
1879         res = path_walk(link, nd);
1880 out:
1881         if (current->link_count || res || nd->last_type!=LAST_NORM)
1882                 return res;
1883         /*
1884          * If it is an iterative symlinks resolution in open_namei() we
1885          * have to copy the last component. And all that crap because of
1886          * bloody create() on broken symlinks. Furrfu...
1887          */
1888         name = __getname();
1889         if (IS_ERR(name))
1890                 goto fail_name;
1891         strcpy(name, nd->last.name);
1892         nd->last.name = name;
1893         return 0;
1894 fail_name:
1895         link = name;
1896 fail:
1897         path_release(nd);
1898         return PTR_ERR(link);
1899 }
1900
1901 int vfs_follow_link(struct nameidata *nd, const char *link)
1902 {
1903         return __vfs_follow_link(nd, link);
1904 }
1905
1906 /* get the link contents into pagecache */
1907 static char *page_getlink(struct dentry * dentry, struct page **ppage)
1908 {
1909         struct page * page;
1910         struct address_space *mapping = dentry->d_inode->i_mapping;
1911         page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
1912                                 NULL);
1913         if (IS_ERR(page))
1914                 goto sync_fail;
1915         wait_on_page(page);
1916         if (!Page_Uptodate(page))
1917                 goto async_fail;
1918         *ppage = page;
1919         return (char*) kmap(page);
1920
1921 async_fail:
1922         page_cache_release(page);
1923         return ERR_PTR(-EIO);
1924
1925 sync_fail:
1926         return (char*)page;
1927 }
1928
1929 int page_readlink(struct dentry *dentry, char *buffer, int buflen)
1930 {
1931         struct page *page = NULL;
1932         char *s = page_getlink(dentry, &page);
1933         int res = vfs_readlink(dentry,buffer,buflen,s);
1934         if (page) {
1935                 kunmap(page);
1936                 page_cache_release(page);
1937         }
1938         return res;
1939 }
1940
1941 int page_follow_link(struct dentry *dentry, struct nameidata *nd)
1942 {
1943         struct page *page = NULL;
1944         char *s = page_getlink(dentry, &page);
1945         int res = __vfs_follow_link(nd, s);
1946         if (page) {
1947                 kunmap(page);
1948                 page_cache_release(page);
1949         }
1950         return res;
1951 }
1952
1953 struct inode_operations page_symlink_inode_operations = {
1954         readlink:       page_readlink,
1955         follow_link:    page_follow_link,
1956 };