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  *      obdfs_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,
  76                                        int lock)
  77 {
  78         struct super_block * sb;
  79         unsigned long offset;
  80         struct page * page;
  81         ENTRY;
  82         CDEBUG(D_INODE, "find entry for %*s\n", namelen, name);
  83
  84         *res_dir = NULL;
  85         sb = dir->i_sb;
  86
  87         if (namelen > EXT2_NAME_LEN)
  88                 return NULL;
  89
  90         CDEBUG(D_INODE, "dirsize is %Ld\n", dir->i_size);
  91
  92         page = 0;
  93         offset = 0;
  94         while ( offset < dir->i_size ) {
  95                 struct ext2_dir_entry_2 * de;
  96                 char * dlimit;
  97
  98                 page = obdfs_getpage(dir, offset, 0, lock);
  99
 100                 if ( !page ) {
 101                         CDEBUG(D_INODE, "No page, offset %lx\n", offset);
 102                         return NULL;
 103                 }
 104
 105                 de = (struct ext2_dir_entry_2 *) page_address(page);
 106                 dlimit = (char *)page_address(page) + PAGE_SIZE;
 107                 while ((char *) de < dlimit) {
 108                         /* this code is executed quadratically often */
 109                         /* do minimal checking `by hand' */
 110                         int de_len;
 111                         /* CDEBUG(D_INODE, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 112                                de, le16_to_cpu(de->rec_len), page_address(page),
 113                                page_address(page) + PAGE_SIZE, offset); */
 114
 115                         if ((char *) de + namelen <= dlimit &&
 116                             ext2_match (namelen, name, de)) {
 117                                 /* found a match -
 118                                    just to be sure, do a full check */
 119                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 120                                                           dir, de, page, offset))
 121                                         goto failure;
 122                                 *res_dir = de;
 123                                 EXIT;
 124                                 return page;
 125                         }
 126                         /* prevent looping on a bad block */
 127                         de_len = le16_to_cpu(de->rec_len);
 128                         if (de_len <= 0) {
 129                                 printk("Bad entry at %p len %d\n", de, de_len);
 130                                 goto failure;
 131                         }
 132                         offset += de_len;
 133                         de = (struct ext2_dir_entry_2 *)
 134                                 ((char *) de + de_len);
 135                         /* CDEBUG(D_INODE, "Next while %lx\n", offset); */
 136                 }
 137                 if ( lock )
 138                         UnlockPage(page);
 139                 page_cache_release(page);
 140                 page = NULL;
 141                 CDEBUG(D_INODE, "Next for %lx\n", offset);
 142         }
 143
 144 failure:
 145         CDEBUG(D_INODE, "Negative case, page %p, offset %ld\n", page, offset);
 146         if (page) {
 147                 if (lock)
 148                         UnlockPage(page);
 149                 page_cache_release(page);
 150         }
 151         EXIT;
 152         return NULL;
 153 } /* obdfs_find_entry */
 154
 155 struct dentry *obdfs_lookup(struct inode * dir, struct dentry *dentry)
 156 {
 157         struct inode * inode;
 158         struct ext2_dir_entry_2 * de;
 159         struct page * page;
 160         ENTRY;
 161
 162         if (dentry->d_name.len > EXT2_NAME_LEN)
 163                 return ERR_PTR(-ENAMETOOLONG);
 164
 165         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 166         inode = NULL;
 167         if ( !page )
 168                 CDEBUG(D_INODE, "No page - negative entry.\n");
 169         if ( page && !de ) {
 170                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 171                 return ERR_PTR(-ENOENT);
 172         }
 173         if (page) {
 174                 unsigned long ino = le32_to_cpu(de->inode);
 175                 UnlockPage(page);
 176                 page_cache_release(page);
 177                 inode = iget(dir->i_sb, ino);
 178
 179                 if (!inode) {
 180                         CDEBUG(D_INODE, "No inode.\n");
 181                         EXIT;
 182                         return ERR_PTR(-EACCES);
 183                 }
 184         }
 185         d_add(dentry, inode);
 186         EXIT;
 187         return NULL;
 188 } /* obdfs_lookup */
 189
 190
 191 /*
 192  *      obdfs_add_entry()
 193  *
 194  * adds a file entry to the specified directory, using the same
 195  * semantics as ext2_find_entry(). It returns NULL if it failed.
 196  *
 197  * NOTE!! The inode part of 'de' is left at 0 - which means you
 198  * may not sleep between calling this and putting something into
 199  * the entry, as someone else might have used it while you slept.
 200
 201  * returns a locked and held page upon success
 202  */
 203
 204 /* XXX I believe these pages should in fact NOT be locked */
 205
 206 static struct page *obdfs_add_entry (struct inode * dir,
 207                                      const char * name, int namelen,
 208                                      struct ext2_dir_entry_2 ** res_dir,
 209                                      int *err)
 210 {
 211         unsigned long offset;
 212         unsigned short rec_len;
 213         struct page *page;
 214         struct ext2_dir_entry_2 * de, * de1;
 215         struct super_block * sb;
 216
 217         ENTRY;
 218         *err = -EINVAL;
 219         *res_dir = NULL;
 220         if (!dir || !dir->i_nlink) {
 221                 EXIT;
 222                 return NULL;
 223         }
 224         sb = dir->i_sb;
 225
 226         if (!namelen) {
 227                 EXIT;
 228                 return NULL;
 229         }
 230         /*
 231          * Is this a busy deleted directory?  Can't create new files if so
 232          */
 233         if (dir->i_size == 0)
 234         {
 235                 EXIT;
 236                 *err = -ENOENT;
 237                 return NULL;
 238         }
 239         page = obdfs_getpage(dir, 0, 0, LOCKED);
 240         if (!page) {
 241                 EXIT;
 242                 return NULL;
 243         }
 244         rec_len = EXT2_DIR_REC_LEN(namelen);
 245         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 246         PDEBUG(page, "starting search");
 247         offset = 0;
 248         de = (struct ext2_dir_entry_2 *) page_address(page);
 249         *err = -ENOSPC;
 250         while (1) {
 251                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 252                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 253                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 254                         UnlockPage(page);
 255                         page_cache_release(page);
 256                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 257                         if (!page) {
 258                                 EXIT;
 259                                 return NULL;
 260                         }
 261                         PDEBUG(page, "new directory page");
 262                         if (dir->i_size <= offset) {
 263                                 if (dir->i_size == 0) {
 264                                         *err = -ENOENT;
 265                                         EXIT;
 266                                         return NULL;
 267                                 }
 268
 269                                 CDEBUG(D_INODE, "creating next block\n");
 270
 271                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 272                                 de->inode = 0;
 273                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 274                                 dir->i_size = offset + PAGE_SIZE;
 275                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 276                                 mark_inode_dirty(dir);
 277                         } else {
 278
 279                                 ext2_debug ("skipping to next block\n");
 280
 281                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 282                         }
 283                 }
 284                 CDEBUG(D_INODE, "\n");
 285                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 286                                            offset)) {
 287                         *err = -ENOENT;
 288                         UnlockPage(page);
 289                         page_cache_release(page);
 290                         EXIT;
 291                         return NULL;
 292                 }
 293                 CDEBUG(D_INODE, "\n");
 294                 if (ext2_match (namelen, name, de)) {
 295                                 *err = -EEXIST;
 296                                 UnlockPage(page);
 297                                 page_cache_release(page);
 298                                 EXIT;
 299                                 return NULL;
 300                 }
 301                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 302                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 303                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 304                         offset += le16_to_cpu(de->rec_len);
 305                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 306                         if (le32_to_cpu(de->inode)) {
 307                                 CDEBUG(D_INODE, "Inserting 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_INODE, "-- 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_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);
 317                         de->inode = 0;
 318                         de->name_len = namelen;
 319                         de->file_type = 0;
 320                         memcpy (de->name, name, namelen);
 321                         CDEBUG(D_INODE, "Copy done\n");
 322                         /*
 323                          * XXX shouldn't update any times until successful
 324                          * completion of syscall, but too many callers depend
 325                          * on this.
 326                          *
 327                          * XXX similarly, too many callers depend on
 328                          * ext2_new_inode() setting the times, but error
 329                          * recovery deletes the inode, so the worst that can
 330                          * happen is that the times are slightly out of date
 331                          * and/or different from the directory change time.
 332                          */
 333                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 334                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 335                         mark_inode_dirty(dir);
 336                         dir->i_version = ++event;
 337                         *res_dir = de;
 338                         *err = 0;
 339                         PDEBUG(page, "addentry");
 340                         CDEBUG(D_INODE, "Regular exit from add_entry");
 341                         EXIT;
 342                         return page;
 343                 }
 344                 CDEBUG(D_INODE, "\n");
 345                 offset += le16_to_cpu(de->rec_len);
 346                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 347
 348         }
 349         CDEBUG(D_INODE, "\n");
 350
 351         UnlockPage(page);
 352         page_cache_release(page);
 353         PDEBUG(page, "addentry");
 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         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 396                 return;
 397         if (S_ISCHR(mode))
 398                 de->file_type = EXT2_FT_CHRDEV;
 399         else if (S_ISBLK(mode))
 400                 de->file_type = EXT2_FT_BLKDEV;
 401         else if (S_ISFIFO(mode))
 402                 de->file_type = EXT2_FT_FIFO;
 403         else if (S_ISLNK(mode))
 404                 de->file_type = EXT2_FT_SYMLINK;
 405         else if (S_ISREG(mode))
 406                 de->file_type = EXT2_FT_REG_FILE;
 407         else if (S_ISDIR(mode))
 408                 de->file_type = EXT2_FT_DIR;
 409 }
 410
 411
 412 /*
 413  * Display all dentries holding the specified inode.
 414  */
 415 #if 0
 416 static void show_dentry(struct list_head * dlist, int subdirs)
 417 {
 418         struct list_head *tmp = dlist;
 419
 420         while ((tmp = tmp->next) != dlist) {
 421                 struct dentry * dentry;
 422                 const char * unhashed = "";
 423
 424                 if ( subdirs )
 425                         dentry  = list_entry(tmp, struct dentry, d_child);
 426                 else
 427                         dentry  = list_entry(tmp, struct dentry, d_alias);
 428
 429                 if (list_empty(&dentry->d_hash))
 430                         unhashed = "(unhashed)";
 431
 432                 if ( dentry->d_inode )
 433                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 434                                dentry->d_parent->d_name.name,
 435                                dentry->d_name.name, dentry->d_count,
 436                                unhashed, dentry->d_inode->i_ino,
 437                                dentry->d_inode->i_dev,
 438                                dentry->d_inode->i_count);
 439                 else
 440                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 441                                dentry->d_parent->d_name.name,
 442                                dentry->d_name.name, dentry->d_count,
 443                                unhashed);
 444         }
 445 } /* show_dentry */
 446 #endif
 447
 448
 449 static struct inode *obdfs_new_inode(struct inode *dir, int mode)
 450 {
 451         struct obdo *oa;
 452         struct inode *inode;
 453         int err;
 454
 455         ENTRY;
 456         oa = obdo_alloc();
 457         if (!oa) {
 458                 EXIT;
 459                 return ERR_PTR(-ENOMEM);
 460         }
 461
 462         /* Send a hint to the create method on the type of file to create */
 463         oa->o_mode = mode;
 464         oa->o_valid |= OBD_MD_FLMODE;
 465
 466         err = IOPS(dir, create)(IID(dir), oa);
 467
 468         if ( err ) {
 469                 obdo_free(oa);
 470                 EXIT;
 471                 return ERR_PTR(err);
 472         }
 473
 474         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 475
 476         if (!inode) {
 477                 IOPS(dir, destroy)(IID(dir), oa);
 478                 obdo_free(oa);
 479                 EXIT;
 480                 return ERR_PTR(-EIO);
 481         }
 482
 483         if (!list_empty(&inode->i_dentry)) {
 484                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 485                 IOPS(dir, destroy)(IID(dir), oa);
 486                 obdo_free(oa);
 487                 iput(inode);
 488                 EXIT;
 489                 return ERR_PTR(-EIO);
 490         }
 491         obdo_free(oa);
 492         INIT_LIST_HEAD(&OBDFS_LIST(inode));
 493
 494         EXIT;
 495         return inode;
 496 } /* obdfs_new_inode */
 497
 498
 499 /*
 500  * By the time this is called, we already have created
 501  * the directory cache entry for the new file, but it
 502  * is so far negative - it has no inode.
 503  *
 504  * If the create succeeds, we fill in the inode information
 505  * with d_instantiate().
 506  */
 507 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 508 {
 509         struct inode * inode;
 510         struct page *page;
 511         struct ext2_dir_entry_2 * de;
 512         int err = -EIO;
 513
 514         ENTRY;
 515         inode = obdfs_new_inode(dir, mode);
 516         if ( IS_ERR(inode) ) {
 517                 EXIT;
 518                 return PTR_ERR(inode);
 519         }
 520
 521         inode->i_op = &obdfs_file_inode_operations;
 522         mark_inode_dirty(inode);
 523         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 524         if (!page) {
 525                 inode->i_nlink--;
 526                 mark_inode_dirty(inode);
 527                 iput (inode);
 528                 EXIT;
 529                 return err;
 530         }
 531         de->inode = cpu_to_le32(inode->i_ino);
 532         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 533         dir->i_version = ++event;
 534
 535         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 536         UnlockPage(page);
 537
 538         page_cache_release(page);
 539         d_instantiate(dentry, inode);
 540         EXIT;
 541         return err;
 542 } /* obdfs_create */
 543
 544 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 545 {
 546         struct inode * inode;
 547         struct page *page;
 548         struct ext2_dir_entry_2 * de;
 549         int err;
 550
 551         ENTRY;
 552
 553         inode = obdfs_new_inode(dir, mode);
 554         if ( IS_ERR(inode) ) {
 555                 EXIT;
 556                 return PTR_ERR(inode);
 557         }
 558
 559         inode->i_uid = current->fsuid;
 560         init_special_inode(inode, mode, rdev);
 561         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 562         if (!page)
 563                 goto out_no_entry;
 564         de->inode = cpu_to_le32(inode->i_ino);
 565         dir->i_version = ++event;
 566         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 567         mark_inode_dirty(inode);
 568
 569         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 570         UnlockPage(page);
 571
 572         d_instantiate(dentry, inode);
 573         page_cache_release(page);
 574         err = 0;
 575 out:
 576         return err;
 577
 578 out_no_entry:
 579         inode->i_nlink--;
 580         mark_inode_dirty(inode);
 581         iput(inode);
 582         goto out;
 583 } /* obdfs_mknod */
 584
 585 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 586 {
 587         struct inode * inode;
 588         struct page *page, *inode_page;
 589         struct ext2_dir_entry_2 * de;
 590         int err;
 591
 592         ENTRY;
 593
 594         err = -EMLINK;
 595         if (dir->i_nlink >= EXT2_LINK_MAX)
 596                 goto out;
 597
 598         mode |= S_IFDIR;
 599         if (dir->i_mode & S_ISGID)
 600                 mode |= S_ISGID;
 601
 602         inode = obdfs_new_inode(dir, mode);
 603         if ( IS_ERR(inode) ) {
 604                 EXIT;
 605                 return PTR_ERR(inode);
 606         }
 607
 608         inode->i_op = &obdfs_dir_inode_operations;
 609         inode->i_blocks = 0;
 610         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 611         if (!inode_page) {
 612                 inode->i_nlink--; /* is this nlink == 0? */
 613                 mark_inode_dirty(inode);
 614                 iput (inode);
 615                 return err;
 616         }
 617
 618         /* create . and .. */
 619         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 620         de->inode = cpu_to_le32(inode->i_ino);
 621         de->name_len = 1;
 622         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 623         strcpy (de->name, ".");
 624         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 625
 626         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 627         de->inode = cpu_to_le32(dir->i_ino);
 628         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 629         de->name_len = 2;
 630         strcpy (de->name, "..");
 631         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 632
 633         /* XXX handle err */
 634         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 635         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 636         inode->i_size = PAGE_SIZE;
 637         UnlockPage(inode_page);
 638         page_cache_release(inode_page);
 639
 640         inode->i_nlink = 2;
 641         mark_inode_dirty(inode);
 642
 643         /* now deal with the parent */
 644         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 645         if (!page) {
 646                 goto out_no_entry;
 647         }
 648
 649         de->inode = cpu_to_le32(inode->i_ino);
 650         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 651         dir->i_version = ++event;
 652
 653         dir->i_nlink++;
 654         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 655         mark_inode_dirty(dir);
 656         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 657         /* XXX handle err? */
 658
 659         UnlockPage(page);
 660
 661         page_cache_release(page);
 662         d_instantiate(dentry, inode);
 663 out:
 664         EXIT;
 665         return err;
 666
 667 out_no_entry:
 668         inode->i_nlink = 0;
 669         mark_inode_dirty(inode);
 670         iput (inode);
 671         EXIT;
 672         goto out;
 673 } /* obdfs_mkdir */
 674
 675
 676 /*
 677  * routine to check that the specified directory is empty (for rmdir)
 678  */
 679 static int empty_dir (struct inode * inode)
 680 {
 681         unsigned long offset;
 682         struct page *page;
 683         struct ext2_dir_entry_2 * de, * de1;
 684         struct super_block * sb;
 685
 686         sb = inode->i_sb;
 687         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 688             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 689                 ext2_warning (inode->i_sb, "empty_dir",
 690                               "bad directory (dir #%lu) - no data block",
 691                               inode->i_ino);
 692                 return 1;
 693         }
 694         de = (struct ext2_dir_entry_2 *) page_address(page);
 695         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 696         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 697             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 698                 ext2_warning (inode->i_sb, "empty_dir",
 699                               "bad directory (dir #%lu) - no `.' or `..'",
 700                               inode->i_ino);
 701                 page_cache_release(page);
 702                 return 1;
 703         }
 704         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 705         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 706         while (offset < inode->i_size ) {
 707                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 708                         if (page) {
 709                                 UnlockPage(page);
 710                                 page_cache_release(page);
 711                         }
 712                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 713                         if (!page) {
 714 #if 0
 715                                 ext2_error (sb, "empty_dir",
 716                                             "directory #%lu contains a hole at offset %lu",
 717                                             inode->i_ino, offset);
 718 #endif
 719                                 offset += sb->s_blocksize;
 720                                 continue;
 721                         }
 722                         de = (struct ext2_dir_entry_2 *) page_address(page);
 723                 }
 724                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 725                                            offset)) {
 726                         UnlockPage(page);
 727                         page_cache_release(page);
 728                         return 1;
 729                 }
 730                 if (le32_to_cpu(de->inode)) {
 731                         UnlockPage(page);
 732                         page_cache_release(page);
 733                         return 0;
 734                 }
 735                 offset += le16_to_cpu(de->rec_len);
 736                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 737         }
 738         UnlockPage(page);
 739         page_cache_release(page);
 740         return 1;
 741 } /* empty_dir */
 742
 743 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 744 {
 745         int retval;
 746         struct inode * inode;
 747         struct page *page;
 748         struct ext2_dir_entry_2 * de;
 749         int err;
 750
 751         ENTRY;
 752
 753         retval = -ENOENT;
 754         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 755         if (!page)
 756                 goto end_rmdir;
 757
 758         inode = dentry->d_inode;
 759         DQUOT_INIT(inode);
 760
 761         retval = -EIO;
 762         if (le32_to_cpu(de->inode) != inode->i_ino)
 763                 goto end_rmdir;
 764
 765         retval = -ENOTEMPTY;
 766         if (!empty_dir (inode))
 767                 goto end_rmdir;
 768
 769         retval = obdfs_delete_entry (de, page);
 770         dir->i_version = ++event;
 771         if (retval)
 772                 goto end_rmdir;
 773         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 774         /* XXX handle err? */
 775         UnlockPage(page);
 776
 777         if (inode->i_nlink != 2)
 778                 ext2_warning (inode->i_sb, "ext2_rmdir",
 779                               "empty directory has nlink!=2 (%d)",
 780                               inode->i_nlink);
 781         inode->i_version = ++event;
 782         inode->i_nlink = 0;
 783         inode->i_size = 0;
 784         mark_inode_dirty(inode);
 785         dir->i_nlink--;
 786         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 787         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 788         mark_inode_dirty(dir);
 789         d_delete(dentry);
 790
 791 end_rmdir:
 792         if ( page )
 793                 page_cache_release(page);
 794         EXIT;
 795         return retval;
 796 } /* obdfs_rmdir */
 797
 798 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 799 {
 800         int retval;
 801         struct inode * inode;
 802         struct page *page;
 803         struct ext2_dir_entry_2 * de;
 804         int err;
 805
 806         ENTRY;
 807
 808         retval = -ENOENT;
 809         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 810         if (!page)
 811                 goto end_unlink;
 812
 813         inode = dentry->d_inode;
 814         DQUOT_INIT(inode);
 815
 816         retval = -EIO;
 817         if (le32_to_cpu(de->inode) != inode->i_ino)
 818                 goto end_unlink;
 819
 820         if (!inode->i_nlink) {
 821                 ext2_warning (inode->i_sb, "ext2_unlink",
 822                               "Deleting nonexistent file (%lu), %d",
 823                               inode->i_ino, inode->i_nlink);
 824                 inode->i_nlink = 1;
 825         }
 826         retval = obdfs_delete_entry (de, page);
 827         if (retval)
 828                 goto end_unlink;
 829         dir->i_version = ++event;
 830         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 831         /* XXX handle err? */
 832         UnlockPage(page);
 833
 834         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 835         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 836         mark_inode_dirty(dir);
 837         inode->i_nlink--;
 838         mark_inode_dirty(inode);
 839         inode->i_ctime = dir->i_ctime;
 840         retval = 0;
 841         d_delete(dentry);       /* This also frees the inode */
 842
 843 end_unlink:
 844         if (page)
 845                 page_cache_release(page);
 846         EXIT;
 847         return retval;
 848 } /* obdfs_unlink */
 849
 850 int obdfs_symlink (struct inode * dir, struct dentry *dentry,
 851                    const char * symname)
 852 {
 853         struct ext2_dir_entry_2 * de;
 854         struct inode * inode;
 855         struct obdfs_inode_info *oinfo;
 856         struct page* page = NULL, * name_page = NULL;
 857         char * link;
 858         int i, l, err = -EIO;
 859         char c;
 860
 861         ENTRY;
 862         inode = obdfs_new_inode(dir, S_IFLNK | S_IRWXUGO);
 863         if ( IS_ERR(inode) ) {
 864                 EXIT;
 865                 return PTR_ERR(inode);
 866         }
 867
 868         inode->i_op = &obdfs_symlink_inode_operations;
 869         for (l = 0; l < inode->i_sb->s_blocksize - 1 && symname [l]; l++)
 870                 ;
 871
 872         oinfo = OBDFS_INFO(inode);
 873
 874         if (l >= sizeof(oinfo->oi_inline)) {
 875                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 876
 877                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 878                 if (!name_page) {
 879                         inode->i_nlink--;
 880                         mark_inode_dirty(inode);
 881                         iput (inode);
 882                         EXIT;
 883                         return err;
 884                 }
 885                 link = (char *)page_address(name_page);
 886         } else {
 887                 link = oinfo->oi_inline;
 888                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 889
 890                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 891
 892         }
 893         i = 0;
 894         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 895                 link[i++] = c;
 896         link[i] = 0;
 897         if (name_page) {
 898                 err = obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 899                 /* XXX handle err */
 900                 PDEBUG(name_page, "symlink");
 901                 UnlockPage(name_page);
 902                 page_cache_release(name_page);
 903         }
 904         inode->i_size = i;
 905         mark_inode_dirty(inode);
 906
 907         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 908                                 &de, &err);
 909         if (!page)
 910                 goto out_no_entry;
 911         de->inode = cpu_to_le32(inode->i_ino);
 912         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 913         dir->i_version = ++event;
 914         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 915         UnlockPage(page);
 916
 917         d_instantiate(dentry, inode);
 918 out:
 919         EXIT;
 920         return err;
 921
 922 out_no_entry:
 923         inode->i_nlink--;
 924         mark_inode_dirty(inode);
 925         iput (inode);
 926         goto out;
 927 } /* obdfs_symlink */
 928
 929 int obdfs_link (struct dentry * old_dentry,
 930                 struct inode * dir, struct dentry *dentry)
 931 {
 932         struct inode *inode = old_dentry->d_inode;
 933         struct ext2_dir_entry_2 * de;
 934         struct page *page;
 935         int err;
 936
 937         ENTRY;
 938
 939         if (S_ISDIR(inode->i_mode))
 940                 return -EPERM;
 941
 942         if (inode->i_nlink >= EXT2_LINK_MAX)
 943                 return -EMLINK;
 944
 945         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 946         if (!page)
 947                 return err;
 948
 949         de->inode = cpu_to_le32(inode->i_ino);
 950         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 951         dir->i_version = ++event;
 952
 953         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 954         UnlockPage(page);
 955
 956         page_cache_release(page);
 957         inode->i_nlink++;
 958         inode->i_ctime = CURRENT_TIME;
 959         mark_inode_dirty(inode);
 960         inode->i_count++;
 961         d_instantiate(dentry, inode);
 962         return err;
 963 } /* obdfs_link */
 964
 965 #define PARENT_INO(buffer) \
 966         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 967         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 968
 969 /*
 970  * Anybody can rename anything with this: the permission checks are left to the
 971  * higher-level routines.
 972  */
 973 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 974                            struct inode * new_dir, struct dentry *new_dentry)
 975 {
 976         struct inode * old_inode, * new_inode;
 977         struct page * old_page, * new_page, * dir_page;
 978         struct ext2_dir_entry_2 * old_de, * new_de;
 979         int err;
 980
 981         ENTRY;
 982
 983         new_page = dir_page = NULL;
 984
 985         /* does the old entry exist? - if not get out */
 986         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 987         PDEBUG(old_page, "rename - old page");
 988         /*
 989          *  Check for inode number is _not_ due to possible IO errors.
 990          *  We might rmdir the source, keep it as pwd of some process
 991          *  and merrily kill the link to whatever was created under the
 992          *  same name. Goodbye sticky bit ;-<
 993          */
 994         old_inode = old_dentry->d_inode;
 995         err = -ENOENT;
 996         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
 997                 goto end_rename;
 998
 999         /* find new inode */
1000         new_inode = new_dentry->d_inode;
1001         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1002                                 new_dentry->d_name.len, &new_de, NOLOCK);
1003         PDEBUG(new_page, "rename - new page ");
1004         if (new_page) {
1005                 if (!new_inode) {
1006                         page_cache_release(new_page);
1007                         new_page = NULL;
1008                 } else {
1009                         DQUOT_INIT(new_inode);
1010                 }
1011         }
1012         /* in this case we to check more ... */
1013         if (S_ISDIR(old_inode->i_mode)) {
1014                 /* can only rename into empty new directory */
1015                 if (new_inode) {
1016                         err = -ENOTEMPTY;
1017                         if (!empty_dir (new_inode))
1018                                 goto end_rename;
1019                 }
1020                 err = -EIO;
1021                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1022                 PDEBUG(dir_page, "rename dir page");
1023
1024                 if (!dir_page)
1025                         goto end_rename;
1026                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1027                         goto end_rename;
1028                 err = -EMLINK;
1029                 if (!new_inode && new_dir!=old_dir &&
1030                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1031                         goto end_rename;
1032         }
1033         /* create the target dir entry */
1034         if (!new_page) {
1035                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1036                                         new_dentry->d_name.len, &new_de,
1037                                         &err);
1038                 PDEBUG(new_page, "rename new page");
1039                 if (!new_page)
1040                         goto end_rename;
1041         }
1042         new_dir->i_version = ++event;
1043
1044         /*
1045          * remove the old entry
1046          */
1047         new_de->inode = le32_to_cpu(old_inode->i_ino);
1048         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1049                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1050                 new_de->file_type = old_de->file_type;
1051
1052         obdfs_delete_entry (old_de, old_page);
1053
1054         old_dir->i_version = ++event;
1055         if (new_inode) {
1056                 new_inode->i_nlink--;
1057                 new_inode->i_ctime = CURRENT_TIME;
1058                 mark_inode_dirty(new_inode);
1059         }
1060         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1061         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1062         mark_inode_dirty(old_dir);
1063         if (dir_page) {
1064                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1065                 /* XXX handle err */
1066                 err = obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1067                 old_dir->i_nlink--;
1068                 mark_inode_dirty(old_dir);
1069                 if (new_inode) {
1070                         new_inode->i_nlink--;
1071                         mark_inode_dirty(new_inode);
1072                 } else {
1073                         new_dir->i_nlink++;
1074                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1075                         mark_inode_dirty(new_dir);
1076                 }
1077         }
1078         if ( old_page != new_page ) {
1079                 unsigned long index = old_page->index;
1080                 /* lock the old_page and release unlocked copy */
1081                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1082                 page_cache_release(old_page);
1083                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1084                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1085                 /* XXX handle err */
1086                 err = obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1087         }
1088
1089         err = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1090
1091 end_rename:
1092         if (old_page && PageLocked(old_page) )
1093                 UnlockPage(old_page);
1094         if (old_page)
1095                 page_cache_release(old_page);
1096         if (new_page && PageLocked(new_page) )
1097                 UnlockPage(new_page);
1098         if (new_page)
1099                 page_cache_release(new_page);
1100         if (dir_page && PageLocked(dir_page) )
1101                 UnlockPage(dir_page);
1102         if (dir_page)
1103                 page_cache_release(dir_page);
1104
1105
1106         return err;
1107 } /* obdfs_rename */