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 static struct page *obdfs_add_entry (struct inode * dir,
 203                                      const char * name, int namelen,
 204                                      struct ext2_dir_entry_2 ** res_dir,
 205                                      int *err)
 206 {
 207         unsigned long offset;
 208         unsigned short rec_len;
 209         struct page *page;
 210         struct ext2_dir_entry_2 * de, * de1;
 211         struct super_block * sb;
 212
 213         ENTRY;
 214         *err = -EINVAL;
 215         *res_dir = NULL;
 216         if (!dir || !dir->i_nlink) {
 217                 EXIT;
 218                 return NULL;
 219         }
 220         sb = dir->i_sb;
 221
 222         if (!namelen) {
 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                 EXIT;
 232                 *err = -ENOENT;
 233                 return NULL;
 234         }
 235         page = obdfs_getpage(dir, 0, 0, LOCKED);
 236         if (!page) {
 237                 EXIT;
 238                 return NULL;
 239         }
 240         rec_len = EXT2_DIR_REC_LEN(namelen);
 241         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 242         offset = 0;
 243         de = (struct ext2_dir_entry_2 *) page_address(page);
 244         *err = -ENOSPC;
 245         while (1) {
 246                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 247                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 248                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 249                         UnlockPage(page);
 250                         page_cache_release(page);
 251                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 252                         if (!page) {
 253                                 EXIT;
 254                                 return NULL;
 255                         }
 256                         if (dir->i_size <= offset) {
 257                                 if (dir->i_size == 0) {
 258                                         *err = -ENOENT;
 259                                         EXIT;
 260                                         return NULL;
 261                                 }
 262
 263                                 CDEBUG(D_INODE, "creating next block\n");
 264
 265                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 266                                 de->inode = 0;
 267                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 268                                 dir->i_size = offset + PAGE_SIZE;
 269                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 270                                 mark_inode_dirty(dir);
 271                         } else {
 272
 273                                 ext2_debug ("skipping to next block\n");
 274
 275                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 276                         }
 277                 }
 278                 CDEBUG(D_INODE, "\n");
 279                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 280                                            offset)) {
 281                         *err = -ENOENT;
 282                         UnlockPage(page);
 283                         page_cache_release(page);
 284                         EXIT;
 285                         return NULL;
 286                 }
 287                 CDEBUG(D_INODE, "\n");
 288                 if (ext2_match (namelen, name, de)) {
 289                                 *err = -EEXIST;
 290                                 UnlockPage(page);
 291                                 page_cache_release(page);
 292                                 EXIT;
 293                                 return NULL;
 294                 }
 295                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 296                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 297                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 298                         offset += le16_to_cpu(de->rec_len);
 299                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 300                         if (le32_to_cpu(de->inode)) {
 301                                 CDEBUG(D_INODE, "Inserting new in %p\n", de);
 302                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 303                                         EXT2_DIR_REC_LEN(de->name_len));
 304                                 CDEBUG(D_INODE, "-- de1 at %p\n", de1);
 305                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 306                                         EXT2_DIR_REC_LEN(de->name_len));
 307                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 308                                 de = de1;
 309                         }
 310                         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);
 311                         de->inode = 0;
 312                         de->name_len = namelen;
 313                         de->file_type = 0;
 314                         memcpy (de->name, name, namelen);
 315                         CDEBUG(D_INODE, "Copy done\n");
 316                         /*
 317                          * XXX shouldn't update any times until successful
 318                          * completion of syscall, but too many callers depend
 319                          * on this.
 320                          *
 321                          * XXX similarly, too many callers depend on
 322                          * ext2_new_inode() setting the times, but error
 323                          * recovery deletes the inode, so the worst that can
 324                          * happen is that the times are slightly out of date
 325                          * and/or different from the directory change time.
 326                          */
 327                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 328                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 329                         mark_inode_dirty(dir);
 330                         dir->i_version = ++event;
 331                         *res_dir = de;
 332                         *err = 0;
 333                         PDEBUG(page, "addentry");
 334                         CDEBUG(D_INODE, "Regular exit from add_entry");
 335                         EXIT;
 336                         return page;
 337                 }
 338                 CDEBUG(D_INODE, "\n");
 339                 offset += le16_to_cpu(de->rec_len);
 340                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 341
 342         }
 343         CDEBUG(D_INODE, "\n");
 344
 345         UnlockPage(page);
 346         page_cache_release(page);
 347         PDEBUG(page, "addentry");
 348         EXIT;
 349         return NULL;
 350 }
 351
 352 /*
 353  * ext2_delete_entry deletes a directory entry by merging it with the
 354  * previous entry
 355  */
 356 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 357                               struct page * page)
 358 {
 359         struct ext2_dir_entry_2 * de, * pde;
 360         int i;
 361
 362         i = 0;
 363         pde = NULL;
 364         de = (struct ext2_dir_entry_2 *) page_address(page);
 365         while (i < PAGE_SIZE) {
 366                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 367                                            de, page, i))
 368                         return -EIO;
 369                 if (de == dir)  {
 370                         if (pde)
 371                                 pde->rec_len =
 372                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 373                                                     le16_to_cpu(dir->rec_len));
 374                         else
 375                                 dir->inode = 0;
 376                         return 0;
 377                 }
 378                 i += le16_to_cpu(de->rec_len);
 379                 pde = de;
 380                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 381         }
 382         return -ENOENT;
 383 }
 384
 385
 386 static inline void ext2_set_de_type(struct super_block *sb,
 387                                 struct ext2_dir_entry_2 *de,
 388                                 umode_t mode) {
 389         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 390                 return;
 391         if (S_ISCHR(mode))
 392                 de->file_type = EXT2_FT_CHRDEV;
 393         else if (S_ISBLK(mode))
 394                 de->file_type = EXT2_FT_BLKDEV;
 395         else if (S_ISFIFO(mode))
 396                 de->file_type = EXT2_FT_FIFO;
 397         else if (S_ISLNK(mode))
 398                 de->file_type = EXT2_FT_SYMLINK;
 399         else if (S_ISREG(mode))
 400                 de->file_type = EXT2_FT_REG_FILE;
 401         else if (S_ISDIR(mode))
 402                 de->file_type = EXT2_FT_DIR;
 403 }
 404
 405
 406 /*
 407  * Display all dentries holding the specified inode.
 408  */
 409 #if 0
 410 static void show_dentry(struct list_head * dlist, int subdirs)
 411 {
 412         struct list_head *tmp = dlist;
 413
 414         while ((tmp = tmp->next) != dlist) {
 415                 struct dentry * dentry;
 416                 const char * unhashed = "";
 417
 418                 if ( subdirs )
 419                         dentry  = list_entry(tmp, struct dentry, d_child);
 420                 else
 421                         dentry  = list_entry(tmp, struct dentry, d_alias);
 422
 423                 if (list_empty(&dentry->d_hash))
 424                         unhashed = "(unhashed)";
 425
 426                 if ( dentry->d_inode )
 427                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 428                                dentry->d_parent->d_name.name,
 429                                dentry->d_name.name, dentry->d_count,
 430                                unhashed, dentry->d_inode->i_ino,
 431                                dentry->d_inode->i_dev,
 432                                dentry->d_inode->i_count);
 433                 else
 434                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 435                                dentry->d_parent->d_name.name,
 436                                dentry->d_name.name, dentry->d_count,
 437                                unhashed);
 438         }
 439 }
 440 #endif
 441
 442
 443 /*
 444  * By the time this is called, we already have created
 445  * the directory cache entry for the new file, but it
 446  * is so far negative - it has no inode.
 447  *
 448  * If the create succeeds, we fill in the inode information
 449  * with d_instantiate().
 450  */
 451 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 452 {
 453         struct inode * inode;
 454         struct page *page;
 455         struct ext2_dir_entry_2 * de;
 456         int err = -EIO;
 457         struct obdo *oa;
 458
 459         ENTRY;
 460
 461         oa = obdo_alloc(oa);
 462         if (!oa) {
 463                 EXIT;
 464                 return -ENOMEM;
 465         }
 466
 467         err = IOPS(dir, create)(IID(dir), 0, oa);
 468
 469         if ( err )
 470                 return err;
 471         obdo_to_inode(inode, oa);
 472         inode = iget(dir->i_sb, inode->i_ino);
 473         if (!inode || !list_empty(&inode->i_dentry)) {
 474                 CDEBUG(D_INODE, "No inode, ino %ld\n", inode->i_ino);
 475                 obdo_free(oa);
 476                 EXIT;
 477                 return -EIO;
 478         }
 479
 480         inode->i_op = &obdfs_file_inode_operations;
 481         inode->i_mode = mode;
 482         mark_inode_dirty(inode);
 483         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 484                                 &de, &err);
 485         if (!page) {
 486                 inode->i_nlink--;
 487                 mark_inode_dirty(inode);
 488                 iput (inode);
 489                 obdo_free(oa);
 490                 EXIT;
 491                 return err;
 492         }
 493         de->inode = cpu_to_le32(inode->i_ino);
 494         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 495         dir->i_version = ++event;
 496         oa->o_id = dir->i_ino; /* Note: this is a different OA than above */
 497         err = IOPS(dir, brw)(WRITE, IID(dir), oa, (char *)page_address(page),
 498                              PAGE_SIZE, (page->index) << PAGE_SHIFT, 0);
 499         obdo_free(oa);
 500         UnlockPage(page);
 501 #if 0
 502         if (IS_SYNC(dir)) {
 503                 ll_rw_block (WRITE, 1, &bh);
 504                 wait_on_buffer (bh);
 505         }
 506 #endif
 507         page_cache_release(page);
 508         d_instantiate(dentry, inode);
 509         EXIT;
 510         return 0;
 511 }
 512
 513 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 514 {
 515         struct inode * inode;
 516         struct page *page;
 517         struct ext2_dir_entry_2 * de;
 518         struct obdo *oa;
 519         int err;
 520
 521         ENTRY;
 522
 523         oa = obdo_alloc();
 524         if (!oa) {
 525                 EXIT;
 526                 return -ENOMEM;
 527         }
 528         err = IOPS(dir, create)(IID(dir), 0, oa);
 529         if ( err )
 530                 return err;
 531         obdo_to_inode(inode, oa);
 532         inode = iget(dir->i_sb, inode->i_ino);
 533         if (!inode) {
 534                 obdo_free(oa);
 535                 EXIT;
 536                 return -EIO;
 537         }
 538
 539         inode->i_uid = current->fsuid;
 540         init_special_inode(inode, mode, rdev);
 541         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 542                                 &de, &err);
 543         if (!page)
 544                 goto out_no_entry;
 545         de->inode = cpu_to_le32(inode->i_ino);
 546         dir->i_version = ++event;
 547         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 548         mark_inode_dirty(inode);
 549         oa->o_id = dir->i_ino; /* Note: this is a different OA than above */
 550         err = IOPS(dir, brw)(WRITE, IID(dir), oa, (char *)page_address(page),
 551                                PAGE_SIZE, (page->index) << PAGE_SHIFT, 0);
 552         UnlockPage(page);
 553         obdo_free(oa);
 554
 555 #if 0
 556         if (IS_SYNC(dir)) {
 557                 ll_rw_block (WRITE, 1, &bh);
 558                 wait_on_buffer (bh);
 559         }
 560 #endif
 561         d_instantiate(dentry, inode);
 562         page_cache_release(page);
 563         err = 0;
 564 out:
 565         return err;
 566
 567 out_no_entry:
 568         inode->i_nlink--;
 569         mark_inode_dirty(inode);
 570         iput(inode);
 571         goto out;
 572 }
 573
 574 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 575 {
 576         struct inode * inode;
 577         struct page *page, *inode_page;
 578         struct ext2_dir_entry_2 * de;
 579         struct obdo *ioa, *doa;
 580         int err;
 581
 582         ENTRY;
 583
 584         err = -EMLINK;
 585         if (dir->i_nlink >= EXT2_LINK_MAX)
 586                 goto out;
 587
 588         ioa = obdo_alloc();
 589         if (!ioa) {
 590                 EXIT;
 591                 return -ENOMEM;
 592         }
 593         doa = obdo_alloc();
 594         if (!ioa) {
 595                 EXIT;
 596                 return -ENOMEM;
 597         }
 598
 599         err = IOPS(dir, create)(IID(dir), 0, ioa);
 600         if ( err ) {
 601                 obdo_free(oa);
 602                 EXIT;
 603                 return err;
 604         }
 605         obdo_to_inode(inode, ioa);
 606         inode = iget(dir->i_sb, inode->i_ino);
 607         if (!inode) {
 608                 obdo_free(oa);
 609                 EXIT;
 610                 return -EIO;
 611         }
 612
 613         inode->i_op = &obdfs_dir_inode_operations;
 614         inode->i_blocks = 0;
 615         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 616         if (!inode_page) {
 617                 inode->i_nlink--; /* is this nlink == 0? */
 618                 mark_inode_dirty(inode);
 619                 iput (inode);
 620                 obdo_free(oa);
 621                 return err;
 622         }
 623
 624         /* create . and .. */
 625         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 626         de->inode = cpu_to_le32(inode->i_ino);
 627         de->name_len = 1;
 628         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 629         strcpy (de->name, ".");
 630         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 631
 632         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 633         de->inode = cpu_to_le32(dir->i_ino);
 634         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 635         de->name_len = 2;
 636         strcpy (de->name, "..");
 637         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 638
 639         oa->o_id = dir->i_ino;
 640         err = IOPS(dir, brw)(WRITE, IID(dir), oa, (char *)page_address(inode_page),
 641                              PAGE_SIZE, (inode_page->index) << PAGE_SHIFT, 1);
 642
 643         obdo_to_inode(dir, doa); /* copy o_blocks to i_blocks */
 644         obdo_free(doa);
 645
 646         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 647         inode->i_size = PAGE_SIZE;
 648         UnlockPage(inode_page);
 649         page_cache_release(inode_page);
 650
 651         inode->i_nlink = 2;
 652         inode->i_mode = S_IFDIR | mode;
 653         if (dir->i_mode & S_ISGID)
 654                 inode->i_mode |= S_ISGID;
 655         mark_inode_dirty(inode);
 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                 goto out_no_entry;
 661         }
 662
 663         de->inode = cpu_to_le32(inode->i_ino);
 664         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 665         dir->i_version = ++event;
 666
 667 #if 0
 668         if (IS_SYNC(dir)) {
 669                 ll_rw_block (WRITE, 1, &bh);
 670                 wait_on_buffer (bh);
 671         }
 672 #endif
 673         dir->i_nlink++;
 674         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 675         mark_inode_dirty(dir);
 676         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 1);
 677         UnlockPage(page);
 678         page_cache_release(page);
 679         d_instantiate(dentry, inode);
 680         err = 0;
 681 out:
 682         EXIT;
 683         return err;
 684
 685 out_no_entry:
 686         inode->i_nlink = 0;
 687         mark_inode_dirty(inode);
 688         iput (inode);
 689         EXIT;
 690         goto out;
 691 }
 692
 693
 694 /*
 695  * routine to check that the specified directory is empty (for rmdir)
 696  */
 697 static int empty_dir (struct inode * inode)
 698 {
 699         unsigned long offset;
 700         struct page *page;
 701         struct ext2_dir_entry_2 * de, * de1;
 702         struct super_block * sb;
 703
 704         sb = inode->i_sb;
 705         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 706             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 707                 ext2_warning (inode->i_sb, "empty_dir",
 708                               "bad directory (dir #%lu) - no data block",
 709                               inode->i_ino);
 710                 return 1;
 711         }
 712         de = (struct ext2_dir_entry_2 *) page_address(page);
 713         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 714         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 715             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 716                 ext2_warning (inode->i_sb, "empty_dir",
 717                               "bad directory (dir #%lu) - no `.' or `..'",
 718                               inode->i_ino);
 719                 page_cache_release(page);
 720                 return 1;
 721         }
 722         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 723         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 724         while (offset < inode->i_size ) {
 725                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 726                         if (page) {
 727                                 UnlockPage(page);
 728                                 page_cache_release(page);
 729                         }
 730                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 731                         if (!page) {
 732 #if 0
 733                                 ext2_error (sb, "empty_dir",
 734                                             "directory #%lu contains a hole at offset %lu",
 735                                             inode->i_ino, offset);
 736 #endif
 737                                 offset += sb->s_blocksize;
 738                                 continue;
 739                         }
 740                         de = (struct ext2_dir_entry_2 *) page_address(page);
 741                 }
 742                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 743                                            offset)) {
 744                         UnlockPage(page);
 745                         page_cache_release(page);
 746                         return 1;
 747                 }
 748                 if (le32_to_cpu(de->inode)) {
 749                         UnlockPage(page);
 750                         page_cache_release(page);
 751                         return 0;
 752                 }
 753                 offset += le16_to_cpu(de->rec_len);
 754                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 755         }
 756         UnlockPage(page);
 757         page_cache_release(page);
 758         return 1;
 759 }
 760
 761 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 762 {
 763         int retval;
 764         struct inode * inode;
 765         struct page *page;
 766         struct ext2_dir_entry_2 * de;
 767
 768         ENTRY;
 769
 770         retval = -ENOENT;
 771         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 772         if (!page)
 773                 goto end_rmdir;
 774
 775         inode = dentry->d_inode;
 776         DQUOT_INIT(inode);
 777
 778         retval = -EIO;
 779         if (le32_to_cpu(de->inode) != inode->i_ino)
 780                 goto end_rmdir;
 781
 782         retval = -ENOTEMPTY;
 783         if (!empty_dir (inode))
 784                 goto end_rmdir;
 785
 786         retval = obdfs_delete_entry (de, page);
 787         dir->i_version = ++event;
 788         if (retval)
 789                 goto end_rmdir;
 790         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 791         UnlockPage(page);
 792 #if 0
 793         if (IS_SYNC(dir)) {
 794                 ll_rw_block (WRITE, 1, &bh);
 795                 wait_on_buffer (bh);
 796         }
 797 #endif
 798         if (inode->i_nlink != 2)
 799                 ext2_warning (inode->i_sb, "ext2_rmdir",
 800                               "empty directory has nlink!=2 (%d)",
 801                               inode->i_nlink);
 802         inode->i_version = ++event;
 803         inode->i_nlink = 0;
 804         inode->i_size = 0;
 805         mark_inode_dirty(inode);
 806         dir->i_nlink--;
 807         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 808         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 809         mark_inode_dirty(dir);
 810         d_delete(dentry);
 811
 812 end_rmdir:
 813         if ( page )
 814                 page_cache_release(page);
 815         EXIT;
 816         return retval;
 817 }
 818
 819 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 820 {
 821         int retval;
 822         struct inode * inode;
 823         struct page *page;
 824         struct ext2_dir_entry_2 * de;
 825
 826         ENTRY;
 827
 828         retval = -ENOENT;
 829         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 830         if (!page)
 831                 goto end_unlink;
 832
 833         inode = dentry->d_inode;
 834         DQUOT_INIT(inode);
 835
 836         retval = -EIO;
 837         if (le32_to_cpu(de->inode) != inode->i_ino)
 838                 goto end_unlink;
 839
 840         if (!inode->i_nlink) {
 841                 ext2_warning (inode->i_sb, "ext2_unlink",
 842                               "Deleting nonexistent file (%lu), %d",
 843                               inode->i_ino, inode->i_nlink);
 844                 inode->i_nlink = 1;
 845         }
 846         retval = obdfs_delete_entry (de, page);
 847         if (retval)
 848                 goto end_unlink;
 849         dir->i_version = ++event;
 850         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 851         UnlockPage(page);
 852 #if 0
 853         if (IS_SYNC(dir)) {
 854                 ll_rw_block (WRITE, 1, &bh);
 855                 wait_on_buffer (bh);
 856         }
 857 #endif
 858         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 859         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 860         mark_inode_dirty(dir);
 861         inode->i_nlink--;
 862         mark_inode_dirty(inode);
 863         inode->i_ctime = dir->i_ctime;
 864         retval = 0;
 865         d_delete(dentry);       /* This also frees the inode */
 866
 867 end_unlink:
 868         if (page)
 869                 page_cache_release(page);
 870         EXIT;
 871         return retval;
 872 }
 873
 874 int obdfs_symlink (struct inode * dir, struct dentry *dentry, const char * symname)
 875 {
 876         struct ext2_dir_entry_2 * de;
 877         struct inode * inode;
 878         struct page* page = NULL, * name_page = NULL;
 879         char * link;
 880         int i, l, err = -EIO;
 881         char c;
 882         obd_id id;
 883
 884         ENTRY;
 885         /*
 886          * N.B. Several error exits in ext2_new_inode don't set err.
 887          */
 888         err = iops(dir)->o_create(iid(dir), 0, &id);
 889         if ( err )  {
 890                 EXIT;
 891                 return err;
 892         }
 893         inode =  iget(dir->i_sb, (ino_t)id);
 894         if (!inode) {
 895                 EXIT;
 896                 return err;
 897         }
 898         inode->i_mode = S_IFLNK | S_IRWXUGO;
 899         inode->i_op = &obdfs_symlink_inode_operations;
 900         for (l = 0; l < inode->i_sb->s_blocksize - 1 &&
 901              symname [l]; l++)
 902                 ;
 903
 904         if (l >= sizeof (inode->u.ext2_i.i_data)) {
 905                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 906
 907                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 908                 if (!name_page) {
 909                         inode->i_nlink--;
 910                         mark_inode_dirty(inode);
 911                         iput (inode);
 912                         EXIT;
 913                         return err;
 914                 }
 915                 link = (char *)page_address(name_page);
 916         } else {
 917                 link = (char *) inode->u.ext2_i.i_data;
 918
 919                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 920
 921         }
 922         i = 0;
 923         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 924                 link[i++] = c;
 925         link[i] = 0;
 926         if (name_page) {
 927                 iops(inode)->o_brw(WRITE, iid(inode), inode, name_page, 1);
 928                 PDEBUG(name_page, "symlink");
 929                 UnlockPage(name_page);
 930                 page_cache_release(name_page);
 931         }
 932         inode->i_size = i;
 933         mark_inode_dirty(inode);
 934
 935         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 936         if (!page)
 937                 goto out_no_entry;
 938         de->inode = cpu_to_le32(inode->i_ino);
 939         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 940         dir->i_version = ++event;
 941         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 1);
 942         UnlockPage(page);
 943 #if 0
 944         if (IS_SYNC(dir)) {
 945                 ll_rw_block (WRITE, 1, &bh);
 946                 wait_on_buffer (bh);
 947         }
 948 #endif
 949         d_instantiate(dentry, inode);
 950         err = 0;
 951 out:
 952         EXIT;
 953         return err;
 954
 955 out_no_entry:
 956         inode->i_nlink--;
 957         mark_inode_dirty(inode);
 958         iput (inode);
 959         goto out;
 960 }
 961
 962 int obdfs_link (struct dentry * old_dentry,
 963                 struct inode * dir, struct dentry *dentry)
 964 {
 965         struct inode *inode = old_dentry->d_inode;
 966         struct ext2_dir_entry_2 * de;
 967         struct page *page;
 968         int err;
 969
 970         ENTRY;
 971
 972         if (S_ISDIR(inode->i_mode))
 973                 return -EPERM;
 974
 975         if (inode->i_nlink >= EXT2_LINK_MAX)
 976                 return -EMLINK;
 977
 978         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 979         if (!page)
 980                 return err;
 981
 982         de->inode = cpu_to_le32(inode->i_ino);
 983         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 984         dir->i_version = ++event;
 985         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 986         UnlockPage(page);
 987
 988 #if 0
 989         if (IS_SYNC(dir)) {
 990                 ll_rw_block (WRITE, 1, &bh);
 991                 wait_on_buffer (bh);
 992         }
 993 #endif
 994         page_cache_release(page);
 995         inode->i_nlink++;
 996         inode->i_ctime = CURRENT_TIME;
 997         mark_inode_dirty(inode);
 998         inode->i_count++;
 999         d_instantiate(dentry, inode);
1000         return 0;
1001 }
1002
1003 #define PARENT_INO(buffer) \
1004         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
1005         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
1006
1007 /*
1008  * Anybody can rename anything with this: the permission checks are left to the
1009  * higher-level routines.
1010  */
1011 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
1012                            struct inode * new_dir, struct dentry *new_dentry)
1013 {
1014         struct inode * old_inode, * new_inode;
1015         struct page * old_page, * new_page, * dir_page;
1016         struct ext2_dir_entry_2 * old_de, * new_de;
1017         int retval;
1018
1019         ENTRY;
1020
1021         new_page = dir_page = NULL;
1022
1023         /* does the old entry exist? - if not get out */
1024         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
1025         PDEBUG(old_page, "rename - old page");
1026         /*
1027          *  Check for inode number is _not_ due to possible IO errors.
1028          *  We might rmdir the source, keep it as pwd of some process
1029          *  and merrily kill the link to whatever was created under the
1030          *  same name. Goodbye sticky bit ;-<
1031          */
1032         old_inode = old_dentry->d_inode;
1033         retval = -ENOENT;
1034         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
1035                 goto end_rename;
1036
1037         /* find new inode */
1038         new_inode = new_dentry->d_inode;
1039         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1040                                 new_dentry->d_name.len, &new_de, NOLOCK);
1041         PDEBUG(new_page, "rename - new page ");
1042         if (new_page) {
1043                 if (!new_inode) {
1044                         page_cache_release(new_page);
1045                         new_page = NULL;
1046                 } else {
1047                         DQUOT_INIT(new_inode);
1048                 }
1049         }
1050         /* in this case we to check more ... */
1051         if (S_ISDIR(old_inode->i_mode)) {
1052                 /* can only rename into empty new directory */
1053                 if (new_inode) {
1054                         retval = -ENOTEMPTY;
1055                         if (!empty_dir (new_inode))
1056                                 goto end_rename;
1057                 }
1058                 retval = -EIO;
1059                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1060                 PDEBUG(dir_page, "rename dir page");
1061
1062                 if (!dir_page)
1063                         goto end_rename;
1064                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1065                         goto end_rename;
1066                 retval = -EMLINK;
1067                 if (!new_inode && new_dir!=old_dir &&
1068                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1069                         goto end_rename;
1070         }
1071         /* create the target dir entry */
1072         if (!new_page) {
1073                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1074                                         new_dentry->d_name.len, &new_de,
1075                                         &retval);
1076                 PDEBUG(new_page, "rename new page");
1077                 if (!new_page)
1078                         goto end_rename;
1079         }
1080         new_dir->i_version = ++event;
1081
1082         /*
1083          * remove the old entry
1084          */
1085         new_de->inode = le32_to_cpu(old_inode->i_ino);
1086         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1087                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1088                 new_de->file_type = old_de->file_type;
1089
1090         obdfs_delete_entry (old_de, old_page);
1091
1092         old_dir->i_version = ++event;
1093         if (new_inode) {
1094                 new_inode->i_nlink--;
1095                 new_inode->i_ctime = CURRENT_TIME;
1096                 mark_inode_dirty(new_inode);
1097         }
1098         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1099         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1100         mark_inode_dirty(old_dir);
1101         if (dir_page) {
1102                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1103                 iops(old_inode)->o_brw(WRITE, iid(old_inode), old_inode, dir_page, 0);
1104                 old_dir->i_nlink--;
1105                 mark_inode_dirty(old_dir);
1106                 if (new_inode) {
1107                         new_inode->i_nlink--;
1108                         mark_inode_dirty(new_inode);
1109                 } else {
1110                         new_dir->i_nlink++;
1111                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1112                         mark_inode_dirty(new_dir);
1113                 }
1114         }
1115         if ( old_page != new_page ) {
1116                 unsigned long index = old_page->index;
1117                 /* lock the old_page and release unlocked copy */
1118                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1119                 page_cache_release(old_page);
1120                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1121                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1122                 iops(old_dir)->o_brw(WRITE, iid(old_dir), old_dir, old_page,0);
1123         }
1124 #if 0
1125         if (IS_SYNC(old_dir)) {
1126                 ll_rw_block (WRITE, 1, &old_bh);
1127                 wait_on_buffer (old_bh);
1128         }
1129 #endif
1130         iops(new_dir)->o_brw(WRITE, iid(new_dir), new_dir, new_page, 0);
1131 #if 0
1132         if (IS_SYNC(new_dir)) {
1133                 ll_rw_block (WRITE, 1, &new_bh);
1134                 wait_on_buffer (new_bh);
1135         }
1136 #endif
1137
1138         retval = 0;
1139
1140 end_rename:
1141         if (old_page && PageLocked(old_page) )
1142                 UnlockPage(old_page);
1143         if (old_page)
1144                 page_cache_release(old_page);
1145         if (new_page && PageLocked(new_page) )
1146                 UnlockPage(new_page);
1147         if (new_page)
1148                 page_cache_release(new_page);
1149         if (dir_page && PageLocked(dir_page) )
1150                 UnlockPage(dir_page);
1151         if (dir_page)
1152                 page_cache_release(dir_page);
1153
1154
1155         return retval;
1156 }