lustre/obdfs/namei.c

   1 /*
   2  *  linux/fs/obdfs/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/ext2/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *
  20  *  Changes for use in OBDFS
  21  *  Copyright (c) 1999, Seagate Technology Inc.
  22  *
  23  */
  24
  25 #include <linux/fs.h>
  26 #include <linux/locks.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/obd_support.h>
  29 #include <linux/obdfs.h>
  30
  31
  32 /*
  33  * define how far ahead to read directories while searching them.
  34  */
  35 #define NAMEI_RA_CHUNKS  2
  36 #define NAMEI_RA_BLOCKS  4
  37 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  38 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  39
  40 /*
  41  * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
  42  *
  43  * `len <= EXT2_NAME_LEN' is guaranteed by caller.
  44  * `de != NULL' is guaranteed by caller.
  45  */
  46 static inline int ext2_match (int len, const char * const name,
  47                        struct ext2_dir_entry_2 * de)
  48 {
  49         if (len != de->name_len)
  50                 return 0;
  51         if (!de->inode)
  52                 return 0;
  53         return !memcmp(name, de->name, len);
  54 }
  55
  56 /*
  57  *      obdfs_find_entry()
  58  *
  59  * finds an entry in the specified directory with the wanted name. It
  60  * returns the cache buffer in which the entry was found, and the entry
  61  * itself (as a parameter - res_dir). It does NOT read the inode of the
  62  * entry - you'll have to do that yourself if you want to.
  63  */
  64 static struct page * obdfs_find_entry (struct inode * dir,
  65                                        const char * const name, int namelen,
  66                                        struct ext2_dir_entry_2 ** res_dir,
  67                                        int lock)
  68 {
  69         struct super_block * sb;
  70         unsigned long offset;
  71         struct page * page;
  72
  73         ENTRY;
  74         CDEBUG(D_INFO, "find entry for %*s\n", namelen, name);
  75
  76         *res_dir = NULL;
  77         sb = dir->i_sb;
  78
  79         if (namelen > EXT2_NAME_LEN)
  80                 return NULL;
  81
  82         CDEBUG(D_INFO, "dirsize is %Ld\n", dir->i_size);
  83
  84         page = 0;
  85         offset = 0;
  86         while ( offset < dir->i_size ) {
  87                 struct ext2_dir_entry_2 * de;
  88                 char * dlimit;
  89
  90                 page = obdfs_getpage(dir, offset, 0, lock);
  91
  92                 if ( !page ) {
  93                         CDEBUG(D_INFO, "No page, offset %lx\n", offset);
  94                         return NULL;
  95                 }
  96
  97                 de = (struct ext2_dir_entry_2 *) page_address(page);
  98                 dlimit = (char *)page_address(page) + PAGE_SIZE;
  99                 while ((char *) de < dlimit) {
 100                         /* this code is executed quadratically often */
 101                         /* do minimal checking `by hand' */
 102                         int de_len;
 103                         /* CDEBUG(D_INFO, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 104                                de, le16_to_cpu(de->rec_len), page_address(page),
 105                                page_address(page) + PAGE_SIZE, offset); */
 106
 107                         if ((char *) de + namelen <= dlimit &&
 108                             ext2_match (namelen, name, de)) {
 109                                 /* found a match -
 110                                    just to be sure, do a full check */
 111                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 112                                                           dir, de, page, offset))
 113                                         goto failure;
 114                                 *res_dir = de;
 115                                 EXIT;
 116                                 return page;
 117                         }
 118                         /* prevent looping on a bad block */
 119                         de_len = le16_to_cpu(de->rec_len);
 120                         if (de_len <= 0) {
 121                                 printk("Bad entry at %p len %d\n", de, de_len);
 122                                 goto failure;
 123                         }
 124                         offset += de_len;
 125                         de = (struct ext2_dir_entry_2 *)
 126                                 ((char *) de + de_len);
 127                         /* CDEBUG(D_INFO, "Next while %lx\n", offset); */
 128                 }
 129                 if ( lock )
 130                         UnlockPage(page);
 131                 page_cache_release(page);
 132                 page = NULL;
 133                 CDEBUG(D_INFO, "Next for %lx\n", offset);
 134         }
 135
 136 failure:
 137         CDEBUG(D_INFO, "Negative case, page %p, offset %ld\n", page, offset);
 138         if (page) {
 139                 if (lock)
 140                         UnlockPage(page);
 141                 page_cache_release(page);
 142         }
 143         EXIT;
 144         return NULL;
 145 } /* obdfs_find_entry */
 146
 147 struct dentry *obdfs_lookup(struct inode *dir, struct dentry *dentry)
 148 {
 149         struct inode * inode;
 150         struct ext2_dir_entry_2 * de;
 151         struct page *page;
 152         ENTRY;
 153
 154         if (dentry->d_name.len > EXT2_NAME_LEN)
 155                 return ERR_PTR(-ENAMETOOLONG);
 156
 157         page = obdfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
 158                                 &de, LOCKED);
 159         inode = NULL;
 160         if ( !page )
 161                 CDEBUG(D_INFO, "No page - negative entry.\n");
 162         if ( page && !de ) {
 163                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 164                 return ERR_PTR(-ENOENT);
 165         }
 166         if (page) {
 167                 unsigned long ino = le32_to_cpu(de->inode);
 168                 UnlockPage(page);
 169                 page_cache_release(page);
 170                 inode = iget(dir->i_sb, ino);
 171
 172                 if (!inode) {
 173                         CDEBUG(D_INODE, "No inode.\n");
 174                         EXIT;
 175                         return ERR_PTR(-EACCES);
 176                 }
 177         }
 178         d_add(dentry, inode);
 179         EXIT;
 180         return NULL;
 181 } /* obdfs_lookup */
 182
 183 /*
 184  *      obdfs_add_entry()
 185  *
 186  * adds a file entry to the specified directory, using the same
 187  * semantics as ext2_find_entry(). It returns NULL if it failed.
 188  *
 189  * NOTE!! The inode part of 'de' is left at 0 - which means you
 190  * may not sleep between calling this and putting something into
 191  * the entry, as someone else might have used it while you slept.
 192  *
 193  * returns a locked and held page upon success
 194  */
 195
 196
 197 /* We do this with a locked page: that's not necessary, since the semaphore
 198  * on the inode protects this page as well.
 199  */
 200 static struct page *obdfs_add_entry (struct inode * dir,
 201                                      const char * name, int namelen,
 202                                      struct ext2_dir_entry_2 ** res_dir,
 203                                      int *err)
 204 {
 205         unsigned long offset;
 206         unsigned short rec_len;
 207         struct page *page;
 208         struct ext2_dir_entry_2 * de, * de1;
 209         struct super_block * sb;
 210
 211         ENTRY;
 212         *err = -EINVAL;
 213         *res_dir = NULL;
 214         if (!dir || !dir->i_nlink) {
 215                 CDEBUG(D_INODE, "bad directory\n");
 216                 EXIT;
 217                 return NULL;
 218         }
 219         sb = dir->i_sb;
 220
 221         if (!namelen) {
 222                 CDEBUG(D_INODE, "bad directory\n");
 223                 EXIT;
 224                 return NULL;
 225         }
 226         /*
 227          * Is this a busy deleted directory?  Can't create new files if so
 228          */
 229         if (dir->i_size == 0)
 230         {
 231                 OIDEBUG(dir);
 232                 *err = -ENOENT;
 233                 EXIT;
 234                 return NULL;
 235         }
 236         page = obdfs_getpage(dir, 0, 0, LOCKED);
 237         if (!page) {
 238                 EXIT;
 239                 return NULL;
 240         }
 241         rec_len = EXT2_DIR_REC_LEN(namelen);
 242         /* CDEBUG(D_INFO, "reclen: %d\n", rec_len); */
 243         /* PDEBUG(page, "starting search"); */
 244         offset = 0;
 245         de = (struct ext2_dir_entry_2 *) page_address(page);
 246         *err = -ENOSPC;
 247         while (1) {
 248                 /* CDEBUG(D_INFO,
 249                        "Entry at %p, (page at %#lx - %#lx), offset %ld\n",
 250                        de, page_address(page), page_address(page) + PAGE_SIZE,
 251                        offset); */
 252                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 253                         UnlockPage(page);
 254                         page_cache_release(page);
 255                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 256                         if (!page) {
 257                                 EXIT;
 258                                 return NULL;
 259                         }
 260                         /* PDEBUG(page, "new directory page"); */
 261                         if (dir->i_size <= offset) {
 262                                 if (dir->i_size == 0) {
 263                                         *err = -ENOENT;
 264                                         EXIT;
 265                                         return NULL;
 266                                 }
 267
 268                                 CDEBUG(D_INFO, "creating next block\n");
 269
 270                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 271                                 de->inode = 0;
 272                                 de->rec_len = cpu_to_le16(PAGE_SIZE);
 273                                 dir->i_size = offset + PAGE_SIZE;
 274                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 275                                 mark_inode_dirty(dir);
 276                         } else {
 277
 278                                 CDEBUG(D_INFO, "skipping to next block\n");
 279
 280                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 281                         }
 282                 }
 283                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 284                                            offset)) {
 285                         *err = -ENOENT;
 286                         UnlockPage(page);
 287                         page_cache_release(page);
 288                         EXIT;
 289                         return NULL;
 290                 }
 291                 CDEBUG(D_INFO, "\n");
 292                 if (ext2_match (namelen, name, de)) {
 293                                 *err = -EEXIST;
 294                                 UnlockPage(page);
 295                                 page_cache_release(page);
 296                                 EXIT;
 297                                 return NULL;
 298                 }
 299                 /* CDEBUG(D_INFO, "Testing for enough space at de %p\n", de);*/
 300                 if ((le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 301                     (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 302                         offset += le16_to_cpu(de->rec_len);
 303                         /* CDEBUG(D_INFO,
 304                                "Found enough space de %p, offset %#lx\n",
 305                                de, offset); */
 306                         if (le32_to_cpu(de->inode)) {
 307                                 /*CDEBUG(D_INFO, "Insert new in %p\n", de);*/
 308                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 309                                         EXT2_DIR_REC_LEN(de->name_len));
 310                                 /*CDEBUG(D_INFO, "-- de1 at %p\n", de1);*/
 311                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 312                                         EXT2_DIR_REC_LEN(de->name_len));
 313                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 314                                 de = de1;
 315                         }
 316                         /* CDEBUG(D_INFO,
 317                                "Reclen adjusted; copy %d bytes to %p, "
 318                                "page at %#lx EOP at %#lx\n",
 319                                namelen, de->name, page_address(page),
 320                                page_address(page) + PAGE_SIZE); */
 321                         de->inode = 0;
 322                         de->name_len = namelen;
 323                         de->file_type = 0;
 324                         memcpy (de->name, name, namelen);
 325                         /*
 326                          * XXX shouldn't update any times until successful
 327                          * completion of syscall, but too many callers depend
 328                          * on this.
 329                          *
 330                          * XXX similarly, too many callers depend on
 331                          * ext2_new_inode() setting the times, but error
 332                          * recovery deletes the inode, so the worst that can
 333                          * happen is that the times are slightly out of date
 334                          * and/or different from the directory change time.
 335                          */
 336                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 337                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 338                         mark_inode_dirty(dir);
 339                         dir->i_version = ++event;
 340                         *res_dir = de;
 341                         *err = 0;
 342                         /* PDEBUG(page, "add_entry"); */
 343                         /* XXX unlock page here */
 344                         EXIT;
 345                         return page;
 346                 }
 347                 offset += le16_to_cpu(de->rec_len);
 348                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 349         }
 350
 351         UnlockPage(page);
 352         page_cache_release(page);
 353         /* PDEBUG(page, "add_entry"); */
 354         EXIT;
 355         return NULL;
 356 } /* obdfs_add_entry */
 357
 358 /*
 359  * obdfs_delete_entry deletes a directory entry by merging it with the
 360  * previous entry
 361  */
 362 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 363                               struct page * page)
 364 {
 365         struct ext2_dir_entry_2 * de, * pde;
 366         int i;
 367
 368         i = 0;
 369         pde = NULL;
 370         de = (struct ext2_dir_entry_2 *) page_address(page);
 371         while (i < PAGE_SIZE) {
 372                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 373                                            de, page, i))
 374                         return -EIO;
 375                 if (de == dir)  {
 376                         if (pde)
 377                                 pde->rec_len =
 378                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 379                                                     le16_to_cpu(dir->rec_len));
 380                         else
 381                                 dir->inode = 0;
 382                         return 0;
 383                 }
 384                 i += le16_to_cpu(de->rec_len);
 385                 pde = de;
 386                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 387         }
 388         return -ENOENT;
 389 } /* obdfs_delete_entry */
 390
 391
 392 static inline void ext2_set_de_type(struct super_block *sb,
 393                                 struct ext2_dir_entry_2 *de,
 394                                 umode_t mode) {
 395         /* XXX fix this to check for obdfs feature, not ext2 feature */
 396         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 397                 return;
 398         if (S_ISREG(mode))
 399                 de->file_type = EXT2_FT_REG_FILE;
 400         else if (S_ISDIR(mode))
 401                 de->file_type = EXT2_FT_DIR;
 402         else if (S_ISLNK(mode))
 403                 de->file_type = EXT2_FT_SYMLINK;
 404         else if (S_ISSOCK(mode))
 405                 de->file_type = EXT2_FT_SOCK;
 406         else if (S_ISFIFO(mode))
 407                 de->file_type = EXT2_FT_FIFO;
 408         else if (S_ISCHR(mode))
 409                 de->file_type = EXT2_FT_CHRDEV;
 410         else if (S_ISBLK(mode))
 411                 de->file_type = EXT2_FT_BLKDEV;
 412 }
 413
 414
 415 /*
 416  * Display all dentries holding the specified inode.
 417  */
 418 #if 0
 419 static void show_dentry(struct list_head * dlist, int subdirs)
 420 {
 421         struct list_head *tmp = dlist;
 422
 423         while ((tmp = tmp->next) != dlist) {
 424                 struct dentry * dentry;
 425                 const char * unhashed = "";
 426
 427                 if ( subdirs )
 428                         dentry  = list_entry(tmp, struct dentry, d_child);
 429                 else
 430                         dentry  = list_entry(tmp, struct dentry, d_alias);
 431
 432                 if (list_empty(&dentry->d_hash))
 433                         unhashed = "(unhashed)";
 434
 435                 if ( dentry->d_inode )
 436                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 437                                dentry->d_parent->d_name.name,
 438                                dentry->d_name.name, dentry->d_count,
 439                                unhashed, dentry->d_inode->i_ino,
 440                                dentry->d_inode->i_dev,
 441                                dentry->d_inode->i_count);
 442                 else
 443                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 444                                dentry->d_parent->d_name.name,
 445                                dentry->d_name.name, dentry->d_count,
 446                                unhashed);
 447         }
 448 } /* show_dentry */
 449 #endif
 450
 451
 452 static struct inode *obdfs_new_inode(struct inode *dir, int mode)
 453 {
 454         struct obdo *oa;
 455         struct inode *inode;
 456         int err;
 457
 458         ENTRY;
 459         if (IOPS(dir, create) == NULL) {
 460                 printk(KERN_ERR __FUNCTION__ ": no create method!\n");
 461                 EXIT;
 462                 return ERR_PTR(-EIO);
 463         }
 464         oa = obdo_alloc();
 465         if (!oa) {
 466                 EXIT;
 467                 return ERR_PTR(-ENOMEM);
 468         }
 469
 470         /* Send a hint to the create method on the type of file to create */
 471         oa->o_mode = mode;
 472         oa->o_valid |= OBD_MD_FLMODE;
 473
 474         err = IOPS(dir, create)(IID(dir), oa);
 475
 476         if ( err ) {
 477                 CDEBUG(D_INODE, "fatal: creating new inode (err %d)\n", err);
 478                 obdo_free(oa);
 479                 EXIT;
 480                 return ERR_PTR(err);
 481         }
 482
 483         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 484
 485         if (!inode) {
 486                 CDEBUG(D_INODE, "fatal: get new inode %ld\n", (long)oa->o_id);
 487                 IOPS(dir, destroy)(IID(dir), oa);
 488                 obdo_free(oa);
 489                 EXIT;
 490                 return ERR_PTR(-EIO);
 491         }
 492
 493         if (!list_empty(&inode->i_dentry)) {
 494                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 495                 IOPS(dir, destroy)(IID(dir), oa);
 496                 obdo_free(oa);
 497                 iput(inode);
 498                 EXIT;
 499                 return ERR_PTR(-EIO);
 500         }
 501         obdo_free(oa);
 502
 503         EXIT;
 504         return inode;
 505 } /* obdfs_new_inode */
 506
 507
 508 /*
 509  * By the time this is called, we already have created
 510  * the directory cache entry for the new file, but it
 511  * is so far negative - it has no inode.
 512  *
 513  * If the create succeeds, we fill in the inode information
 514  * with d_instantiate().
 515  */
 516 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 517 {
 518         struct inode * inode;
 519         struct page *page;
 520         struct ext2_dir_entry_2 * de;
 521         int err = -EIO;
 522
 523         ENTRY;
 524         inode = obdfs_new_inode(dir, mode);
 525         if ( IS_ERR(inode) ) {
 526                 EXIT;
 527                 return PTR_ERR(inode);
 528         }
 529
 530         inode->i_op = &obdfs_file_inode_operations;
 531         mark_inode_dirty(inode);
 532         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 533         if (!page) {
 534                 inode->i_nlink--;
 535                 mark_inode_dirty(inode);
 536                 iput (inode);
 537                 EXIT;
 538                 return err;
 539         }
 540         de->inode = cpu_to_le32(inode->i_ino);
 541         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 542         dir->i_version = ++event;
 543
 544         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 545         UnlockPage(page);
 546
 547         page_cache_release(page);
 548         d_instantiate(dentry, inode);
 549         EXIT;
 550         return err;
 551 } /* obdfs_create */
 552
 553 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 554 {
 555         struct inode * inode;
 556         struct page *page;
 557         struct ext2_dir_entry_2 * de;
 558         struct obdfs_inode_info *oinfo;
 559         int err;
 560
 561         ENTRY;
 562         inode = obdfs_new_inode(dir, mode);
 563         if ( IS_ERR(inode) ) {
 564                 EXIT;
 565                 return PTR_ERR(inode);
 566         }
 567
 568         inode->i_uid = current->fsuid;
 569         init_special_inode(inode, mode, rdev);
 570         oinfo = obdfs_i2info(inode);
 571         ((obd_count *)oinfo->oi_inline)[0] = rdev;
 572         oinfo->oi_flags |= OBD_FL_INLINEDATA;
 573
 574         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 575         if (!page)
 576                 goto out_no_entry;
 577         de->inode = cpu_to_le32(inode->i_ino);
 578         dir->i_version = ++event;
 579         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 580         mark_inode_dirty(inode);
 581
 582         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 583         UnlockPage(page);
 584
 585         d_instantiate(dentry, inode);
 586         page_cache_release(page);
 587         err = 0;
 588 out:
 589         return err;
 590
 591 out_no_entry:
 592         inode->i_nlink--;
 593         mark_inode_dirty(inode);
 594         iput(inode);
 595         goto out;
 596 } /* obdfs_mknod */
 597
 598 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 599 {
 600         struct inode * inode;
 601         struct page *page, *inode_page;
 602         struct ext2_dir_entry_2 * de;
 603         int err;
 604
 605         ENTRY;
 606
 607         err = -EMLINK;
 608         if (dir->i_nlink >= EXT2_LINK_MAX)
 609                 goto out;
 610
 611         mode |= S_IFDIR;
 612         if (dir->i_mode & S_ISGID)
 613                 mode |= S_ISGID;
 614
 615         inode = obdfs_new_inode(dir, mode);
 616         if ( IS_ERR(inode) ) {
 617                 EXIT;
 618                 return PTR_ERR(inode);
 619         }
 620
 621         inode->i_op = &obdfs_dir_inode_operations;
 622         inode->i_blocks = 0;
 623         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 624         if (!inode_page) {
 625                 inode->i_nlink--; /* is this nlink == 0? */
 626                 mark_inode_dirty(inode);
 627                 iput (inode);
 628                 return -EIO;
 629         }
 630         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 631
 632         /* create . and .. */
 633         de->inode = cpu_to_le32(inode->i_ino);
 634         de->name_len = 1;
 635         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 636         strcpy (de->name, ".");
 637         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 638         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 639         de->inode = cpu_to_le32(dir->i_ino);
 640         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 641         de->name_len = 2;
 642         strcpy (de->name, "..");
 643         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 644         inode->i_nlink = 2;
 645
 646         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 647         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 648         inode->i_size = PAGE_SIZE;
 649         UnlockPage(inode_page);
 650         page_cache_release(inode_page);
 651         mark_inode_dirty(inode);
 652         if (err) {
 653                 EXIT;
 654                 goto out_no_entry;
 655         }
 656
 657         /* now deal with the parent */
 658         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 659         if (!page) {
 660                 EXIT;
 661                 goto out_no_entry;
 662         }
 663
 664         de->inode = cpu_to_le32(inode->i_ino);
 665         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 666         dir->i_version = ++event;
 667
 668         dir->i_nlink++;
 669         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 670         mark_inode_dirty(dir);
 671         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 672
 673         UnlockPage(page);
 674
 675         page_cache_release(page);
 676         d_instantiate(dentry, inode);
 677         EXIT;
 678 out:
 679         return err;
 680
 681 out_no_entry:
 682         inode->i_nlink = 0;
 683         mark_inode_dirty(inode);
 684         iput (inode);
 685         goto out;
 686 } /* obdfs_mkdir */
 687
 688
 689 /*
 690  * routine to check that the specified directory is empty (for rmdir)
 691  */
 692 static int empty_dir (struct inode * inode)
 693 {
 694         unsigned long offset;
 695         struct page *page;
 696         struct ext2_dir_entry_2 * de, * de1;
 697         struct super_block * sb;
 698
 699         sb = inode->i_sb;
 700         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 701             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 702                 ext2_warning (inode->i_sb, "empty_dir",
 703                               "bad directory (dir #%lu) - no data block",
 704                               inode->i_ino);
 705                 return 1;
 706         }
 707         de = (struct ext2_dir_entry_2 *) page_address(page);
 708         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 709         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 710             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 711                 ext2_warning (inode->i_sb, "empty_dir",
 712                               "bad directory (dir #%lu) - no `.' or `..'",
 713                               inode->i_ino);
 714                 page_cache_release(page);
 715                 return 1;
 716         }
 717         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 718         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 719         while (offset < inode->i_size ) {
 720                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 721                         if (page) {
 722                                 UnlockPage(page);
 723                                 page_cache_release(page);
 724                         }
 725                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 726                         if (!page) {
 727 #if 0
 728                                 ext2_error (sb, "empty_dir",
 729                                             "directory #%lu contains a hole at offset %lu",
 730                                             inode->i_ino, offset);
 731 #endif
 732                                 offset += sb->s_blocksize;
 733                                 continue;
 734                         }
 735                         de = (struct ext2_dir_entry_2 *) page_address(page);
 736                 }
 737                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 738                                            offset)) {
 739                         UnlockPage(page);
 740                         page_cache_release(page);
 741                         return 1;
 742                 }
 743                 if (le32_to_cpu(de->inode)) {
 744                         UnlockPage(page);
 745                         page_cache_release(page);
 746                         return 0;
 747                 }
 748                 offset += le16_to_cpu(de->rec_len);
 749                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 750         }
 751         UnlockPage(page);
 752         page_cache_release(page);
 753         return 1;
 754 } /* empty_dir */
 755
 756 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 757 {
 758         int retval;
 759         struct inode * inode;
 760         struct page *page;
 761         struct ext2_dir_entry_2 * de;
 762
 763         ENTRY;
 764
 765         retval = -ENOENT;
 766         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 767         if (!page)
 768                 goto end_rmdir;
 769
 770         inode = dentry->d_inode;
 771         DQUOT_INIT(inode);
 772
 773         retval = -EIO;
 774         if (le32_to_cpu(de->inode) != inode->i_ino)
 775                 goto end_rmdir;
 776
 777         retval = -ENOTEMPTY;
 778         if (!empty_dir (inode))
 779                 goto end_rmdir;
 780
 781         retval = obdfs_delete_entry (de, page);
 782         dir->i_version = ++event;
 783         if (retval)
 784                 goto end_rmdir;
 785         retval = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 786         /* XXX handle err? */
 787         UnlockPage(page);
 788
 789         if (inode->i_nlink != 2)
 790                 ext2_warning (inode->i_sb, "ext2_rmdir",
 791                               "empty directory has nlink!=2 (%d)",
 792                               inode->i_nlink);
 793         inode->i_version = ++event;
 794         inode->i_nlink = 0;
 795         inode->i_size = 0;
 796         mark_inode_dirty(inode);
 797         dir->i_nlink--;
 798         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 799         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 800         mark_inode_dirty(dir);
 801         d_delete(dentry);
 802
 803 end_rmdir:
 804         if ( page )
 805                 page_cache_release(page);
 806         EXIT;
 807         return retval;
 808 } /* obdfs_rmdir */
 809
 810 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 811 {
 812         int retval;
 813         struct inode * inode;
 814         struct page *page;
 815         struct ext2_dir_entry_2 * de;
 816
 817         ENTRY;
 818
 819         retval = -ENOENT;
 820         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 821         if (!page)
 822                 goto end_unlink;
 823
 824         inode = dentry->d_inode;
 825         DQUOT_INIT(inode);
 826
 827         retval = -EIO;
 828         if (le32_to_cpu(de->inode) != inode->i_ino)
 829                 goto end_unlink;
 830
 831         if (!inode->i_nlink) {
 832                 ext2_warning (inode->i_sb, "ext2_unlink",
 833                               "Deleting nonexistent file (%lu), %d",
 834                               inode->i_ino, inode->i_nlink);
 835                 inode->i_nlink = 1;
 836         }
 837         retval = obdfs_delete_entry (de, page);
 838         if (retval)
 839                 goto end_unlink;
 840         dir->i_version = ++event;
 841         retval = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 842         /* XXX handle err? */
 843         UnlockPage(page);
 844
 845         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 846         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 847         mark_inode_dirty(dir);
 848         inode->i_nlink--;
 849         mark_inode_dirty(inode);
 850         inode->i_ctime = dir->i_ctime;
 851         d_delete(dentry);       /* This also frees the inode */
 852
 853 end_unlink:
 854         if (page)
 855                 page_cache_release(page);
 856         EXIT;
 857         return retval;
 858 } /* obdfs_unlink */
 859
 860 int obdfs_symlink (struct inode * dir, struct dentry *dentry,
 861                    const char * symname)
 862 {
 863         struct inode * inode;
 864         struct ext2_dir_entry_2 * de;
 865         struct obdfs_inode_info *oinfo;
 866         struct page* page = NULL, * name_page = NULL;
 867         char * link;
 868         int l, err;
 869
 870         ENTRY;
 871         err = -ENAMETOOLONG;
 872         l = strlen(symname)+1;
 873         if (l > PAGE_SIZE) {
 874                 EXIT;
 875                 goto out;
 876         }
 877
 878         inode = obdfs_new_inode(dir, S_IFLNK);
 879         if ( IS_ERR(inode) ) {
 880                 EXIT;
 881                 goto out;
 882         }
 883
 884         inode->i_mode = S_IFLNK | S_IRWXUGO;
 885         oinfo = obdfs_i2info(inode);
 886
 887         if (l >= sizeof(oinfo->oi_inline)) {
 888                 CDEBUG(D_INFO, "l=%d, normal symlink\n", l);
 889                 inode->i_op = &obdfs_symlink_inode_operations;
 890
 891                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 892                 if (!name_page) {
 893                         EXIT;
 894                         err = -ENOMEM;
 895                         goto out_no_entry;
 896                 }
 897                 link = (char *)page_address(name_page);
 898         } else {
 899                 CDEBUG(D_INFO, "l=%d, fast symlink\n", l);
 900                 inode->i_op = &obdfs_fast_symlink_inode_operations;
 901                 link = oinfo->oi_inline;
 902                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 903         }
 904         memcpy(link, symname, l);
 905         if (name_page) {
 906                 err = obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 907                 /* PDEBUG(name_page, "symlink"); */
 908                 UnlockPage(name_page);
 909                 page_cache_release(name_page);
 910                 if (err) {
 911                         EXIT;
 912                         goto out_no_entry;
 913                 }
 914         }
 915         inode->i_size = l-1;
 916         mark_inode_dirty(inode);
 917
 918         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 919                                 &de, &err);
 920         if (!page)
 921                 goto out_no_entry;
 922         de->inode = cpu_to_le32(inode->i_ino);
 923         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 924         dir->i_version = ++event;
 925         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 926         UnlockPage(page);
 927
 928         d_instantiate(dentry, inode);
 929 out:
 930         EXIT;
 931         return err;
 932
 933 out_no_entry:
 934         inode->i_nlink--;
 935         mark_inode_dirty(inode);
 936         iput (inode);
 937         goto out;
 938 } /* obdfs_symlink */
 939
 940 int obdfs_link (struct dentry * old_dentry,
 941                 struct inode * dir, struct dentry *dentry)
 942 {
 943         struct inode *inode = old_dentry->d_inode;
 944         struct ext2_dir_entry_2 * de;
 945         struct page *page;
 946         int err;
 947
 948         ENTRY;
 949
 950         if (S_ISDIR(inode->i_mode))
 951                 return -EPERM;
 952
 953         if (inode->i_nlink >= EXT2_LINK_MAX)
 954                 return -EMLINK;
 955
 956         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 957         if (!page)
 958                 return err;
 959
 960         de->inode = cpu_to_le32(inode->i_ino);
 961         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 962         dir->i_version = ++event;
 963
 964         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 965         UnlockPage(page);
 966
 967         page_cache_release(page);
 968         inode->i_nlink++;
 969         inode->i_ctime = CURRENT_TIME;
 970         mark_inode_dirty(inode);
 971         inode->i_count++;
 972         d_instantiate(dentry, inode);
 973         return err;
 974 } /* obdfs_link */
 975
 976 #define PARENT_INO(buffer) \
 977         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 978         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 979
 980 /*
 981  * Anybody can rename anything with this: the permission checks are left to the
 982  * higher-level routines.
 983  */
 984 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 985                            struct inode * new_dir, struct dentry *new_dentry)
 986 {
 987         struct inode * old_inode, * new_inode;
 988         struct page * old_page, * new_page, * dir_page;
 989         struct ext2_dir_entry_2 * old_de, * new_de;
 990         int retval;
 991
 992         ENTRY;
 993
 994         new_page = dir_page = NULL;
 995
 996         /* does the old entry exist? - if not get out */
 997         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 998         /* PDEBUG(old_page, "rename - old page"); */
 999         /*
1000          *  Check for inode number is _not_ due to possible IO errors.
1001          *  We might rmdir the source, keep it as pwd of some process
1002          *  and merrily kill the link to whatever was created under the
1003          *  same name. Goodbye sticky bit ;-<
1004          */
1005         old_inode = old_dentry->d_inode;
1006         retval = -ENOENT;
1007         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino) {
1008                 EXIT;
1009                 goto end_rename;
1010         }
1011
1012         /* find new inode */
1013         new_inode = new_dentry->d_inode;
1014         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1015                                 new_dentry->d_name.len, &new_de, NOLOCK);
1016         /* PDEBUG(new_page, "rename - new page "); */
1017         if (new_page) {
1018                 if (!new_inode) {
1019                         page_cache_release(new_page);
1020                         new_page = NULL;
1021                 } else {
1022                         DQUOT_INIT(new_inode);
1023                 }
1024         }
1025         /* in this case we to check more ... */
1026         if (S_ISDIR(old_inode->i_mode)) {
1027                 /* can only rename into empty new directory */
1028                 if (new_inode) {
1029                         retval = -ENOTEMPTY;
1030                         if (!empty_dir (new_inode)) {
1031                                 EXIT;
1032                                 goto end_rename;
1033                         }
1034                 }
1035                 retval = -EIO;
1036                 dir_page = obdfs_getpage (old_inode, 0, 0, LOCKED);
1037                 /* PDEBUG(dir_page, "rename dir page"); */
1038
1039                 if (!dir_page) {
1040                         EXIT;
1041                         goto end_rename;
1042                 }
1043                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) !=
1044                     old_dir->i_ino) {
1045                         EXIT;
1046                         goto end_rename;
1047                 }
1048                 retval = -EMLINK;
1049                 if (!new_inode && new_dir!=old_dir &&
1050                                 new_dir->i_nlink >= EXT2_LINK_MAX) {
1051                         EXIT;
1052                         goto end_rename;
1053                 }
1054         }
1055         /* create the target dir entry */
1056         if (!new_page) {
1057                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1058                                             new_dentry->d_name.len, &new_de,
1059                                             &retval);
1060                 /* PDEBUG(new_page, "rename new page"); */
1061                 if (!new_page) {
1062                         EXIT;
1063                         goto end_rename;
1064                 }
1065         }
1066         new_dir->i_version = ++event;
1067
1068         /*
1069          * remove the old entry
1070          */
1071         new_de->inode = le32_to_cpu(old_inode->i_ino);
1072         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1073                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1074                 new_de->file_type = old_de->file_type;
1075
1076         obdfs_delete_entry (old_de, old_page);
1077
1078         old_dir->i_version = ++event;
1079         if (new_inode) {
1080                 new_inode->i_nlink--;
1081                 new_inode->i_ctime = CURRENT_TIME;
1082                 mark_inode_dirty(new_inode);
1083         }
1084         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1085         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1086         mark_inode_dirty(old_dir);
1087         if (dir_page) {
1088                 PARENT_INO(page_address(dir_page)) =le32_to_cpu(new_dir->i_ino);
1089                 retval = obdfs_do_writepage(old_inode, dir_page,
1090                                             IS_SYNC(old_inode));
1091                 /* XXX handle err - not sure if this is correct */
1092                 if (retval) {
1093                         EXIT;
1094                         goto end_rename;
1095                 }
1096                 old_dir->i_nlink--;
1097                 mark_inode_dirty(old_dir);
1098                 if (new_inode) {
1099                         new_inode->i_nlink--;
1100                         mark_inode_dirty(new_inode);
1101                 } else {
1102                         new_dir->i_nlink++;
1103                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1104                         mark_inode_dirty(new_dir);
1105                 }
1106         }
1107         if ( old_page != new_page ) {
1108                 unsigned long index = old_page->index;
1109                 /* lock the old_page and release unlocked copy */
1110                 CDEBUG(D_INFO, "old_page at %p\n", old_page);
1111                 page_cache_release(old_page);
1112                 old_page = obdfs_getpage(old_dir, index << PAGE_SHIFT, 0,
1113                                          LOCKED);
1114                 CDEBUG(D_INFO, "old_page at %p\n", old_page);
1115                 retval = obdfs_do_writepage(old_dir, old_page,IS_SYNC(old_dir));
1116                 /* XXX handle err - not sure if this is correct */
1117                 if (retval) {
1118                         EXIT;
1119                         goto end_rename;
1120                 }
1121         }
1122
1123         retval = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1124
1125 end_rename:
1126         if (old_page && PageLocked(old_page) )
1127                 UnlockPage(old_page);
1128         if (old_page)
1129                 page_cache_release(old_page);
1130         if (new_page && PageLocked(new_page) )
1131                 UnlockPage(new_page);
1132         if (new_page)
1133                 page_cache_release(new_page);
1134         if (dir_page && PageLocked(dir_page) )
1135                 UnlockPage(dir_page);
1136         if (dir_page)
1137                 page_cache_release(dir_page);
1138
1139         return retval;
1140 } /* obdfs_rename */