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 <asm/uaccess.h>
  26
  27 #include <linux/errno.h>
  28 #include <linux/fs.h>
  29 #include <linux/ext2_fs.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/sched.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/locks.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/iobuf.h>
  37
  38 #include <linux/obd_support.h>
  39 #include <linux/obdfs.h>
  40
  41 /*
  42  * define how far ahead to read directories while searching them.
  43  */
  44 #define NAMEI_RA_CHUNKS  2
  45 #define NAMEI_RA_BLOCKS  4
  46 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  47 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  48
  49 /*
  50  * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
  51  *
  52  * `len <= EXT2_NAME_LEN' is guaranteed by caller.
  53  * `de != NULL' is guaranteed by caller.
  54  */
  55 static inline int ext2_match (int len, const char * const name,
  56                        struct ext2_dir_entry_2 * de)
  57 {
  58         if (len != de->name_len)
  59                 return 0;
  60         if (!de->inode)
  61                 return 0;
  62         return !memcmp(name, de->name, len);
  63 }
  64
  65 /*
  66  *      ext2_find_entry()
  67  *
  68  * finds an entry in the specified directory with the wanted name. It
  69  * returns the cache buffer in which the entry was found, and the entry
  70  * itself (as a parameter - res_dir). It does NOT read the inode of the
  71  * entry - you'll have to do that yourself if you want to.
  72  */
  73 static struct page * obdfs_find_entry (struct inode * dir,
  74                                              const char * const name, int namelen,
  75                                              struct ext2_dir_entry_2 ** res_dir, int lock)
  76 {
  77         struct super_block * sb;
  78         unsigned long offset;
  79         struct page * page;
  80         ENTRY;
  81         CDEBUG(D_INODE, "find entry for %*s\n", namelen, name);
  82
  83         *res_dir = NULL;
  84         sb = dir->i_sb;
  85
  86         if (namelen > EXT2_NAME_LEN)
  87                 return NULL;
  88
  89         CDEBUG(D_INODE, "dirsize is %Ld\n", dir->i_size);
  90
  91         page = 0;
  92         offset = 0;
  93         while ( offset < dir->i_size ) {
  94                 struct ext2_dir_entry_2 * de;
  95                 char * dlimit;
  96
  97                 page = obdfs_getpage(dir, offset, 0, lock);
  98
  99                 if ( !page ) {
 100                         CDEBUG(D_INODE, "No page, offset %lx\n", offset);
 101                         return NULL;
 102                 }
 103
 104                 de = (struct ext2_dir_entry_2 *) page_address(page);
 105                 dlimit = (char *)page_address(page) + PAGE_SIZE;
 106                 while ((char *) de < dlimit) {
 107                         /* this code is executed quadratically often */
 108                         /* do minimal checking `by hand' */
 109                         int de_len;
 110                         /* CDEBUG(D_INODE, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 111                                de, le16_to_cpu(de->rec_len), page_address(page),
 112                                page_address(page) + PAGE_SIZE, offset); */
 113
 114                         if ((char *) de + namelen <= dlimit &&
 115                             ext2_match (namelen, name, de)) {
 116                                 /* found a match -
 117                                    just to be sure, do a full check */
 118                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 119                                                           dir, de, page, offset))
 120                                         goto failure;
 121                                 *res_dir = de;
 122                                 EXIT;
 123                                 return page;
 124                         }
 125                         /* prevent looping on a bad block */
 126                         de_len = le16_to_cpu(de->rec_len);
 127                         if (de_len <= 0) {
 128                                 printk("Bad entry at %p len %d\n", de, de_len);
 129                                 goto failure;
 130                         }
 131                         offset += de_len;
 132                         de = (struct ext2_dir_entry_2 *)
 133                                 ((char *) de + de_len);
 134                         /* CDEBUG(D_INODE, "Next while %lx\n", offset); */
 135                 }
 136                 if ( lock )
 137                         UnlockPage(page);
 138                 page_cache_release(page);
 139                 page = NULL;
 140                 CDEBUG(D_INODE, "Next for %lx\n", offset);
 141         }
 142
 143 failure:
 144         CDEBUG(D_INODE, "Negative case, page %p, offset %ld\n", page, offset);
 145         if (page) {
 146                 if (lock)
 147                         UnlockPage(page);
 148                 page_cache_release(page);
 149         }
 150         EXIT;
 151         return NULL;
 152 }
 153
 154 struct dentry *obdfs_lookup(struct inode * dir, struct dentry *dentry)
 155 {
 156         struct inode * inode;
 157         struct ext2_dir_entry_2 * de;
 158         struct page * page;
 159         ENTRY;
 160
 161         if (dentry->d_name.len > EXT2_NAME_LEN)
 162                 return ERR_PTR(-ENAMETOOLONG);
 163
 164         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 165         inode = NULL;
 166         if ( !page )
 167                 CDEBUG(D_INODE, "No page - negative entry.\n");
 168         if ( page && !de ) {
 169                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 170                 return ERR_PTR(-ENOENT);
 171         }
 172         if (page) {
 173                 unsigned long ino = le32_to_cpu(de->inode);
 174                 UnlockPage(page);
 175                 page_cache_release(page);
 176                 inode = iget(dir->i_sb, ino);
 177
 178                 if (!inode) {
 179                         CDEBUG(D_INODE, "No inode.\n");
 180                         EXIT;
 181                         return ERR_PTR(-EACCES);
 182                 }
 183         }
 184         d_add(dentry, inode);
 185         EXIT;
 186         return NULL;
 187 }
 188
 189
 190 /*
 191  *      ext2_add_entry()
 192  *
 193  * adds a file entry to the specified directory, using the same
 194  * semantics as ext2_find_entry(). It returns NULL if it failed.
 195  *
 196  * NOTE!! The inode part of 'de' is left at 0 - which means you
 197  * may not sleep between calling this and putting something into
 198  * the entry, as someone else might have used it while you slept.
 199
 200  * returns a locked and held page upon success
 201  */
 202
 203 /* XXX I believe these pages should in fact NOT be locked */
 204
 205 static struct page *obdfs_add_entry (struct inode * dir,
 206                                      const char * name, int namelen,
 207                                      struct ext2_dir_entry_2 ** res_dir,
 208                                      int *err)
 209 {
 210         unsigned long offset;
 211         unsigned short rec_len;
 212         struct page *page;
 213         struct ext2_dir_entry_2 * de, * de1;
 214         struct super_block * sb;
 215
 216         ENTRY;
 217         *err = -EINVAL;
 218         *res_dir = NULL;
 219         if (!dir || !dir->i_nlink) {
 220                 EXIT;
 221                 return NULL;
 222         }
 223         sb = dir->i_sb;
 224
 225         if (!namelen) {
 226                 EXIT;
 227                 return NULL;
 228         }
 229         /*
 230          * Is this a busy deleted directory?  Can't create new files if so
 231          */
 232         if (dir->i_size == 0)
 233         {
 234                 EXIT;
 235                 *err = -ENOENT;
 236                 return NULL;
 237         }
 238         page = obdfs_getpage(dir, 0, 0, LOCKED);
 239         if (!page) {
 240                 EXIT;
 241                 return NULL;
 242         }
 243         rec_len = EXT2_DIR_REC_LEN(namelen);
 244         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 245         offset = 0;
 246         de = (struct ext2_dir_entry_2 *) page_address(page);
 247         *err = -ENOSPC;
 248         while (1) {
 249                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 250                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 251                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 252                         UnlockPage(page);
 253                         page_cache_release(page);
 254                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 255                         if (!page) {
 256                                 EXIT;
 257                                 return NULL;
 258                         }
 259                         if (dir->i_size <= offset) {
 260                                 if (dir->i_size == 0) {
 261                                         *err = -ENOENT;
 262                                         EXIT;
 263                                         return NULL;
 264                                 }
 265
 266                                 CDEBUG(D_INODE, "creating next block\n");
 267
 268                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 269                                 de->inode = 0;
 270                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 271                                 dir->i_size = offset + PAGE_SIZE;
 272                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 273                                 mark_inode_dirty(dir);
 274                         } else {
 275
 276                                 ext2_debug ("skipping to next block\n");
 277
 278                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 279                         }
 280                 }
 281                 CDEBUG(D_INODE, "\n");
 282                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 283                                            offset)) {
 284                         *err = -ENOENT;
 285                         UnlockPage(page);
 286                         page_cache_release(page);
 287                         EXIT;
 288                         return NULL;
 289                 }
 290                 CDEBUG(D_INODE, "\n");
 291                 if (ext2_match (namelen, name, de)) {
 292                                 *err = -EEXIST;
 293                                 UnlockPage(page);
 294                                 page_cache_release(page);
 295                                 EXIT;
 296                                 return NULL;
 297                 }
 298                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 299                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 300                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 301                         offset += le16_to_cpu(de->rec_len);
 302                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 303                         if (le32_to_cpu(de->inode)) {
 304                                 CDEBUG(D_INODE, "Inserting new in %p\n", de);
 305                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 306                                         EXT2_DIR_REC_LEN(de->name_len));
 307                                 CDEBUG(D_INODE, "-- de1 at %p\n", de1);
 308                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 309                                         EXT2_DIR_REC_LEN(de->name_len));
 310                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 311                                 de = de1;
 312                         }
 313                         CDEBUG(D_INODE, "Reclen adjusted; copy %d bytes to %p, page at %#lx EOP at %#lx\n", namelen, de->name, page_address(page), page_address(page) + PAGE_SIZE);
 314                         de->inode = 0;
 315                         de->name_len = namelen;
 316                         de->file_type = 0;
 317                         memcpy (de->name, name, namelen);
 318                         CDEBUG(D_INODE, "Copy done\n");
 319                         /*
 320                          * XXX shouldn't update any times until successful
 321                          * completion of syscall, but too many callers depend
 322                          * on this.
 323                          *
 324                          * XXX similarly, too many callers depend on
 325                          * ext2_new_inode() setting the times, but error
 326                          * recovery deletes the inode, so the worst that can
 327                          * happen is that the times are slightly out of date
 328                          * and/or different from the directory change time.
 329                          */
 330                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 331                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 332                         mark_inode_dirty(dir);
 333                         dir->i_version = ++event;
 334                         *res_dir = de;
 335                         *err = 0;
 336                         PDEBUG(page, "addentry");
 337                         CDEBUG(D_INODE, "Regular exit from add_entry");
 338                         EXIT;
 339                         return page;
 340                 }
 341                 CDEBUG(D_INODE, "\n");
 342                 offset += le16_to_cpu(de->rec_len);
 343                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 344
 345         }
 346         CDEBUG(D_INODE, "\n");
 347
 348         UnlockPage(page);
 349         page_cache_release(page);
 350         PDEBUG(page, "addentry");
 351         EXIT;
 352         return NULL;
 353 }
 354
 355 /*
 356  * obdfs_delete_entry deletes a directory entry by merging it with the
 357  * previous entry
 358  */
 359 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 360                               struct page * page)
 361 {
 362         struct ext2_dir_entry_2 * de, * pde;
 363         int i;
 364
 365         i = 0;
 366         pde = NULL;
 367         de = (struct ext2_dir_entry_2 *) page_address(page);
 368         while (i < PAGE_SIZE) {
 369                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 370                                            de, page, i))
 371                         return -EIO;
 372                 if (de == dir)  {
 373                         if (pde)
 374                                 pde->rec_len =
 375                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 376                                                     le16_to_cpu(dir->rec_len));
 377                         else
 378                                 dir->inode = 0;
 379                         return 0;
 380                 }
 381                 i += le16_to_cpu(de->rec_len);
 382                 pde = de;
 383                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 384         }
 385         return -ENOENT;
 386 }
 387
 388
 389 static inline void ext2_set_de_type(struct super_block *sb,
 390                                 struct ext2_dir_entry_2 *de,
 391                                 umode_t mode) {
 392         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 393                 return;
 394         if (S_ISCHR(mode))
 395                 de->file_type = EXT2_FT_CHRDEV;
 396         else if (S_ISBLK(mode))
 397                 de->file_type = EXT2_FT_BLKDEV;
 398         else if (S_ISFIFO(mode))
 399                 de->file_type = EXT2_FT_FIFO;
 400         else if (S_ISLNK(mode))
 401                 de->file_type = EXT2_FT_SYMLINK;
 402         else if (S_ISREG(mode))
 403                 de->file_type = EXT2_FT_REG_FILE;
 404         else if (S_ISDIR(mode))
 405                 de->file_type = EXT2_FT_DIR;
 406 }
 407
 408
 409 /*
 410  * Display all dentries holding the specified inode.
 411  */
 412 #if 0
 413 static void show_dentry(struct list_head * dlist, int subdirs)
 414 {
 415         struct list_head *tmp = dlist;
 416
 417         while ((tmp = tmp->next) != dlist) {
 418                 struct dentry * dentry;
 419                 const char * unhashed = "";
 420
 421                 if ( subdirs )
 422                         dentry  = list_entry(tmp, struct dentry, d_child);
 423                 else
 424                         dentry  = list_entry(tmp, struct dentry, d_alias);
 425
 426                 if (list_empty(&dentry->d_hash))
 427                         unhashed = "(unhashed)";
 428
 429                 if ( dentry->d_inode )
 430                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 431                                dentry->d_parent->d_name.name,
 432                                dentry->d_name.name, dentry->d_count,
 433                                unhashed, dentry->d_inode->i_ino,
 434                                dentry->d_inode->i_dev,
 435                                dentry->d_inode->i_count);
 436                 else
 437                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 438                                dentry->d_parent->d_name.name,
 439                                dentry->d_name.name, dentry->d_count,
 440                                unhashed);
 441         }
 442 }
 443 #endif
 444
 445
 446 struct inode *obdfs_new_inode(struct inode *dir)
 447 {
 448         struct obdo *obdo;
 449         struct inode *inode;
 450         struct obdfs_inode_info *oinfo;
 451         ino_t ino;
 452         int err;
 453
 454         ENTRY;
 455         obdo = obdo_alloc();
 456         if (!obdo) {
 457                 EXIT;
 458                 return ERR_PTR(-ENOMEM);
 459         }
 460
 461         err = IOPS(dir, create)(IID(dir), obdo);
 462         ino = (ino_t)obdo->o_id;
 463         obdo_free(obdo);
 464
 465         if ( err ) {
 466                 EXIT;
 467                 return ERR_PTR(err);
 468         }
 469
 470         inode = iget(dir->i_sb, ino);
 471
 472         if (!inode) {
 473                 EXIT;
 474                 return ERR_PTR(-EIO);
 475         }
 476
 477         if (!list_empty(&inode->i_dentry)) {
 478                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 479                 iput(inode);
 480                 EXIT;
 481                 return ERR_PTR(-EIO);
 482         }
 483
 484         oinfo = OBD_INFO(inode);
 485         INIT_LIST_HEAD(&oinfo->oi_list);
 486         EXIT;
 487         return inode;
 488 }
 489
 490
 491 /*
 492  * By the time this is called, we already have created
 493  * the directory cache entry for the new file, but it
 494  * is so far negative - it has no inode.
 495  *
 496  * If the create succeeds, we fill in the inode information
 497  * with d_instantiate().
 498  */
 499 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 500 {
 501         struct inode * inode;
 502         struct page *page;
 503         struct ext2_dir_entry_2 * de;
 504         int err = -EIO;
 505
 506         ENTRY;
 507         inode = obdfs_new_inode(dir);
 508         if ( IS_ERR(inode) ) {
 509                 EXIT;
 510                 return PTR_ERR(inode);
 511         }
 512
 513         inode->i_op = &obdfs_file_inode_operations;
 514         inode->i_mode = mode;
 515         mark_inode_dirty(inode);
 516         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 517         if (!page) {
 518                 inode->i_nlink--;
 519                 mark_inode_dirty(inode);
 520                 iput (inode);
 521                 EXIT;
 522                 return err;
 523         }
 524         de->inode = cpu_to_le32(inode->i_ino);
 525         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 526         dir->i_version = ++event;
 527
 528         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 529         UnlockPage(page);
 530
 531         page_cache_release(page);
 532         d_instantiate(dentry, inode);
 533         EXIT;
 534         return err;
 535 }
 536
 537 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 538 {
 539         struct inode * inode;
 540         struct page *page;
 541         struct ext2_dir_entry_2 * de;
 542         int err;
 543
 544         ENTRY;
 545
 546         inode = obdfs_new_inode(dir);
 547         if ( IS_ERR(inode) ) {
 548                 EXIT;
 549                 return PTR_ERR(inode);
 550         }
 551
 552         inode->i_uid = current->fsuid;
 553         init_special_inode(inode, mode, rdev);
 554         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 555         if (!page)
 556                 goto out_no_entry;
 557         de->inode = cpu_to_le32(inode->i_ino);
 558         dir->i_version = ++event;
 559         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 560         mark_inode_dirty(inode);
 561
 562         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 563         UnlockPage(page);
 564
 565         d_instantiate(dentry, inode);
 566         page_cache_release(page);
 567         err = 0;
 568 out:
 569         return err;
 570
 571 out_no_entry:
 572         inode->i_nlink--;
 573         mark_inode_dirty(inode);
 574         iput(inode);
 575         goto out;
 576 }
 577
 578 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 579 {
 580         struct inode * inode;
 581         struct page *page, *inode_page;
 582         struct ext2_dir_entry_2 * de;
 583         int err;
 584
 585         ENTRY;
 586
 587         err = -EMLINK;
 588         if (dir->i_nlink >= EXT2_LINK_MAX)
 589                 goto out;
 590
 591         inode = obdfs_new_inode(dir);
 592         if ( IS_ERR(inode) ) {
 593                 EXIT;
 594                 return PTR_ERR(inode);
 595         }
 596
 597         inode->i_op = &obdfs_dir_inode_operations;
 598         inode->i_blocks = 0;
 599         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 600         if (!inode_page) {
 601                 inode->i_nlink--; /* is this nlink == 0? */
 602                 mark_inode_dirty(inode);
 603                 iput (inode);
 604                 return err;
 605         }
 606
 607         /* create . and .. */
 608         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 609         de->inode = cpu_to_le32(inode->i_ino);
 610         de->name_len = 1;
 611         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 612         strcpy (de->name, ".");
 613         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 614
 615         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 616         de->inode = cpu_to_le32(dir->i_ino);
 617         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 618         de->name_len = 2;
 619         strcpy (de->name, "..");
 620         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 621
 622         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 623         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 624         inode->i_size = PAGE_SIZE;
 625         UnlockPage(inode_page);
 626         page_cache_release(inode_page);
 627         /* XXX handle err */
 628
 629         inode->i_nlink = 2;
 630         inode->i_mode = S_IFDIR | mode;
 631         if (dir->i_mode & S_ISGID)
 632                 inode->i_mode |= S_ISGID;
 633         mark_inode_dirty(inode);
 634
 635         /* now deal with the parent */
 636         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 637         if (!page) {
 638                 goto out_no_entry;
 639         }
 640
 641         de->inode = cpu_to_le32(inode->i_ino);
 642         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 643         dir->i_version = ++event;
 644
 645         dir->i_nlink++;
 646         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 647         mark_inode_dirty(dir);
 648         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 649
 650         UnlockPage(page);
 651
 652         page_cache_release(page);
 653         d_instantiate(dentry, inode);
 654         err = 0;
 655 out:
 656         EXIT;
 657         return err;
 658
 659 out_no_entry:
 660         inode->i_nlink = 0;
 661         mark_inode_dirty(inode);
 662         iput (inode);
 663         EXIT;
 664         goto out;
 665 }
 666
 667
 668 /*
 669  * routine to check that the specified directory is empty (for rmdir)
 670  */
 671 static int empty_dir (struct inode * inode)
 672 {
 673         unsigned long offset;
 674         struct page *page;
 675         struct ext2_dir_entry_2 * de, * de1;
 676         struct super_block * sb;
 677
 678         sb = inode->i_sb;
 679         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 680             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 681                 ext2_warning (inode->i_sb, "empty_dir",
 682                               "bad directory (dir #%lu) - no data block",
 683                               inode->i_ino);
 684                 return 1;
 685         }
 686         de = (struct ext2_dir_entry_2 *) page_address(page);
 687         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 688         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 689             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 690                 ext2_warning (inode->i_sb, "empty_dir",
 691                               "bad directory (dir #%lu) - no `.' or `..'",
 692                               inode->i_ino);
 693                 page_cache_release(page);
 694                 return 1;
 695         }
 696         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 697         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 698         while (offset < inode->i_size ) {
 699                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 700                         if (page) {
 701                                 UnlockPage(page);
 702                                 page_cache_release(page);
 703                         }
 704                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 705                         if (!page) {
 706 #if 0
 707                                 ext2_error (sb, "empty_dir",
 708                                             "directory #%lu contains a hole at offset %lu",
 709                                             inode->i_ino, offset);
 710 #endif
 711                                 offset += sb->s_blocksize;
 712                                 continue;
 713                         }
 714                         de = (struct ext2_dir_entry_2 *) page_address(page);
 715                 }
 716                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 717                                            offset)) {
 718                         UnlockPage(page);
 719                         page_cache_release(page);
 720                         return 1;
 721                 }
 722                 if (le32_to_cpu(de->inode)) {
 723                         UnlockPage(page);
 724                         page_cache_release(page);
 725                         return 0;
 726                 }
 727                 offset += le16_to_cpu(de->rec_len);
 728                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 729         }
 730         UnlockPage(page);
 731         page_cache_release(page);
 732         return 1;
 733 }
 734
 735 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 736 {
 737         int retval;
 738         struct inode * inode;
 739         struct page *page;
 740         struct ext2_dir_entry_2 * de;
 741         int err;
 742
 743         ENTRY;
 744
 745         retval = -ENOENT;
 746         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 747         if (!page)
 748                 goto end_rmdir;
 749
 750         inode = dentry->d_inode;
 751         DQUOT_INIT(inode);
 752
 753         retval = -EIO;
 754         if (le32_to_cpu(de->inode) != inode->i_ino)
 755                 goto end_rmdir;
 756
 757         retval = -ENOTEMPTY;
 758         if (!empty_dir (inode))
 759                 goto end_rmdir;
 760
 761         retval = obdfs_delete_entry (de, page);
 762         dir->i_version = ++event;
 763         if (retval)
 764                 goto end_rmdir;
 765         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 766         UnlockPage(page);
 767
 768         if (inode->i_nlink != 2)
 769                 ext2_warning (inode->i_sb, "ext2_rmdir",
 770                               "empty directory has nlink!=2 (%d)",
 771                               inode->i_nlink);
 772         inode->i_version = ++event;
 773         inode->i_nlink = 0;
 774         inode->i_size = 0;
 775         mark_inode_dirty(inode);
 776         dir->i_nlink--;
 777         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 778         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 779         mark_inode_dirty(dir);
 780         d_delete(dentry);
 781
 782 end_rmdir:
 783         if ( page )
 784                 page_cache_release(page);
 785         EXIT;
 786         return retval;
 787 }
 788
 789 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 790 {
 791         int retval;
 792         struct inode * inode;
 793         struct page *page;
 794         struct ext2_dir_entry_2 * de;
 795         int err;
 796
 797         ENTRY;
 798
 799         retval = -ENOENT;
 800         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 801         if (!page)
 802                 goto end_unlink;
 803
 804         inode = dentry->d_inode;
 805         DQUOT_INIT(inode);
 806
 807         retval = -EIO;
 808         if (le32_to_cpu(de->inode) != inode->i_ino)
 809                 goto end_unlink;
 810
 811         if (!inode->i_nlink) {
 812                 ext2_warning (inode->i_sb, "ext2_unlink",
 813                               "Deleting nonexistent file (%lu), %d",
 814                               inode->i_ino, inode->i_nlink);
 815                 inode->i_nlink = 1;
 816         }
 817         retval = obdfs_delete_entry (de, page);
 818         if (retval)
 819                 goto end_unlink;
 820         dir->i_version = ++event;
 821         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 822         UnlockPage(page);
 823
 824         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 825         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 826         mark_inode_dirty(dir);
 827         inode->i_nlink--;
 828         mark_inode_dirty(inode);
 829         inode->i_ctime = dir->i_ctime;
 830         retval = 0;
 831         d_delete(dentry);       /* This also frees the inode */
 832
 833 end_unlink:
 834         if (page)
 835                 page_cache_release(page);
 836         EXIT;
 837         return retval;
 838 }
 839
 840 int obdfs_symlink (struct inode * dir, struct dentry *dentry, const char * symname)
 841 {
 842         struct ext2_dir_entry_2 * de;
 843         struct inode * inode;
 844         struct obdfs_inode_info *oinfo;
 845         struct page* page = NULL, * name_page = NULL;
 846         char * link;
 847         int i, l, err = -EIO;
 848         char c;
 849
 850         ENTRY;
 851         inode = obdfs_new_inode(dir);
 852         oinfo = OBD_INFO(inode);
 853         if ( IS_ERR(inode) ) {
 854                 EXIT;
 855                 return PTR_ERR(inode);
 856         }
 857
 858         inode->i_mode = S_IFLNK | S_IRWXUGO;
 859         inode->i_op = &obdfs_symlink_inode_operations;
 860         for (l = 0; l < inode->i_sb->s_blocksize - 1 &&
 861              symname [l]; l++)
 862                 ;
 863
 864         if (l >= sizeof (oinfo->oi_inline)) {
 865                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 866
 867                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 868                 if (!name_page) {
 869                         inode->i_nlink--;
 870                         mark_inode_dirty(inode);
 871                         iput (inode);
 872                         EXIT;
 873                         return err;
 874                 }
 875                 link = (char *)page_address(name_page);
 876         } else {
 877                 link = oinfo->oi_inline;
 878                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 879
 880                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 881
 882         }
 883         i = 0;
 884         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 885                 link[i++] = c;
 886         link[i] = 0;
 887         if (name_page) {
 888                 obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 889                 PDEBUG(name_page, "symlink");
 890                 UnlockPage(name_page);
 891                 page_cache_release(name_page);
 892         }
 893         inode->i_size = i;
 894         mark_inode_dirty(inode);
 895
 896         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 897         if (!page)
 898                 goto out_no_entry;
 899         de->inode = cpu_to_le32(inode->i_ino);
 900         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 901         dir->i_version = ++event;
 902         obdfs_do_writepage(dir, page, IS_SYNC(dir));
 903         UnlockPage(page);
 904
 905         d_instantiate(dentry, inode);
 906         err = 0;
 907 out:
 908         EXIT;
 909         return err;
 910
 911 out_no_entry:
 912         inode->i_nlink--;
 913         mark_inode_dirty(inode);
 914         iput (inode);
 915         goto out;
 916 }
 917
 918 int obdfs_link (struct dentry * old_dentry,
 919                 struct inode * dir, struct dentry *dentry)
 920 {
 921         struct inode *inode = old_dentry->d_inode;
 922         struct ext2_dir_entry_2 * de;
 923         struct page *page;
 924         int err;
 925
 926         ENTRY;
 927
 928         if (S_ISDIR(inode->i_mode))
 929                 return -EPERM;
 930
 931         if (inode->i_nlink >= EXT2_LINK_MAX)
 932                 return -EMLINK;
 933
 934         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 935         if (!page)
 936                 return err;
 937
 938         de->inode = cpu_to_le32(inode->i_ino);
 939         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 940         dir->i_version = ++event;
 941
 942         obdfs_do_writepage(dir, page, IS_SYNC(dir));
 943         UnlockPage(page);
 944
 945         page_cache_release(page);
 946         inode->i_nlink++;
 947         inode->i_ctime = CURRENT_TIME;
 948         mark_inode_dirty(inode);
 949         inode->i_count++;
 950         d_instantiate(dentry, inode);
 951         return 0;
 952 }
 953
 954 #define PARENT_INO(buffer) \
 955         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 956         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 957
 958 /*
 959  * Anybody can rename anything with this: the permission checks are left to the
 960  * higher-level routines.
 961  */
 962 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 963                            struct inode * new_dir, struct dentry *new_dentry)
 964 {
 965         struct inode * old_inode, * new_inode;
 966         struct page * old_page, * new_page, * dir_page;
 967         struct ext2_dir_entry_2 * old_de, * new_de;
 968         int retval;
 969
 970         ENTRY;
 971
 972         new_page = dir_page = NULL;
 973
 974         /* does the old entry exist? - if not get out */
 975         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 976         PDEBUG(old_page, "rename - old page");
 977         /*
 978          *  Check for inode number is _not_ due to possible IO errors.
 979          *  We might rmdir the source, keep it as pwd of some process
 980          *  and merrily kill the link to whatever was created under the
 981          *  same name. Goodbye sticky bit ;-<
 982          */
 983         old_inode = old_dentry->d_inode;
 984         retval = -ENOENT;
 985         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
 986                 goto end_rename;
 987
 988         /* find new inode */
 989         new_inode = new_dentry->d_inode;
 990         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
 991                                 new_dentry->d_name.len, &new_de, NOLOCK);
 992         PDEBUG(new_page, "rename - new page ");
 993         if (new_page) {
 994                 if (!new_inode) {
 995                         page_cache_release(new_page);
 996                         new_page = NULL;
 997                 } else {
 998                         DQUOT_INIT(new_inode);
 999                 }
1000         }
1001         /* in this case we to check more ... */
1002         if (S_ISDIR(old_inode->i_mode)) {
1003                 /* can only rename into empty new directory */
1004                 if (new_inode) {
1005                         retval = -ENOTEMPTY;
1006                         if (!empty_dir (new_inode))
1007                                 goto end_rename;
1008                 }
1009                 retval = -EIO;
1010                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1011                 PDEBUG(dir_page, "rename dir page");
1012
1013                 if (!dir_page)
1014                         goto end_rename;
1015                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1016                         goto end_rename;
1017                 retval = -EMLINK;
1018                 if (!new_inode && new_dir!=old_dir &&
1019                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1020                         goto end_rename;
1021         }
1022         /* create the target dir entry */
1023         if (!new_page) {
1024                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1025                                         new_dentry->d_name.len, &new_de,
1026                                         &retval);
1027                 PDEBUG(new_page, "rename new page");
1028                 if (!new_page)
1029                         goto end_rename;
1030         }
1031         new_dir->i_version = ++event;
1032
1033         /*
1034          * remove the old entry
1035          */
1036         new_de->inode = le32_to_cpu(old_inode->i_ino);
1037         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1038                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1039                 new_de->file_type = old_de->file_type;
1040
1041         obdfs_delete_entry (old_de, old_page);
1042
1043         old_dir->i_version = ++event;
1044         if (new_inode) {
1045                 new_inode->i_nlink--;
1046                 new_inode->i_ctime = CURRENT_TIME;
1047                 mark_inode_dirty(new_inode);
1048         }
1049         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1050         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1051         mark_inode_dirty(old_dir);
1052         if (dir_page) {
1053                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1054                 obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1055                 old_dir->i_nlink--;
1056                 mark_inode_dirty(old_dir);
1057                 if (new_inode) {
1058                         new_inode->i_nlink--;
1059                         mark_inode_dirty(new_inode);
1060                 } else {
1061                         new_dir->i_nlink++;
1062                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1063                         mark_inode_dirty(new_dir);
1064                 }
1065         }
1066         if ( old_page != new_page ) {
1067                 unsigned long index = old_page->index;
1068                 /* lock the old_page and release unlocked copy */
1069                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1070                 page_cache_release(old_page);
1071                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1072                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1073                 obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1074         }
1075
1076         obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1077
1078         retval = 0;
1079
1080 end_rename:
1081         if (old_page && PageLocked(old_page) )
1082                 UnlockPage(old_page);
1083         if (old_page)
1084                 page_cache_release(old_page);
1085         if (new_page && PageLocked(new_page) )
1086                 UnlockPage(new_page);
1087         if (new_page)
1088                 page_cache_release(new_page);
1089         if (dir_page && PageLocked(dir_page) )
1090                 UnlockPage(dir_page);
1091         if (dir_page)
1092                 page_cache_release(dir_page);
1093
1094
1095         return retval;
1096 }