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 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), page_address(page) +
 112                                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                                 CDEBUG(D_INODE, "Bad entry: at %p len %d\n",
 129                                        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 }
 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 }
 189
 190
 191 /*
 192  *      ext2_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 static struct page *obdfs_add_entry (struct inode * dir,
 204                                      const char * name, int namelen,
 205                                      struct ext2_dir_entry_2 ** res_dir,
 206                                      int *err)
 207 {
 208         unsigned long offset;
 209         unsigned short rec_len;
 210         struct page *page;
 211         struct ext2_dir_entry_2 * de, * de1;
 212         struct super_block * sb;
 213
 214         ENTRY;
 215         *err = -EINVAL;
 216         *res_dir = NULL;
 217         if (!dir || !dir->i_nlink) {
 218                 EXIT;
 219                 return NULL;
 220         }
 221         sb = dir->i_sb;
 222
 223         if (!namelen) {
 224                 EXIT;
 225                 return NULL;
 226         }
 227         /*
 228          * Is this a busy deleted directory?  Can't create new files if so
 229          */
 230         if (dir->i_size == 0)
 231         {
 232                 EXIT;
 233                 *err = -ENOENT;
 234                 return NULL;
 235         }
 236         page = obdfs_getpage(dir, 0, 0, LOCKED);
 237         if (!page) {
 238                 EXIT;
 239                 return NULL;
 240         }
 241         rec_len = EXT2_DIR_REC_LEN(namelen);
 242         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 243         offset = 0;
 244         de = (struct ext2_dir_entry_2 *) page_address(page);
 245         *err = -ENOSPC;
 246         while (1) {
 247                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 248                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 249                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 250                         UnlockPage(page);
 251                         page_cache_release(page);
 252                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 253                         if (!page) {
 254                                 EXIT;
 255                                 return NULL;
 256                         }
 257                         if (dir->i_size <= offset) {
 258                                 if (dir->i_size == 0) {
 259                                         *err = -ENOENT;
 260                                         EXIT;
 261                                         return NULL;
 262                                 }
 263
 264                                 CDEBUG(D_INODE, "creating next block\n");
 265
 266                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 267                                 de->inode = 0;
 268                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 269                                 dir->i_size = offset + PAGE_SIZE;
 270                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 271                                 mark_inode_dirty(dir);
 272                         } else {
 273
 274                                 ext2_debug ("skipping to next block\n");
 275
 276                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 277                         }
 278                 }
 279                 CDEBUG(D_INODE, "\n");
 280                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 281                                            offset)) {
 282                         *err = -ENOENT;
 283                         UnlockPage(page);
 284                         page_cache_release(page);
 285                         EXIT;
 286                         return NULL;
 287                 }
 288                 CDEBUG(D_INODE, "\n");
 289                 if (ext2_match (namelen, name, de)) {
 290                                 *err = -EEXIST;
 291                                 UnlockPage(page);
 292                                 page_cache_release(page);
 293                                 EXIT;
 294                                 return NULL;
 295                 }
 296                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 297                 if ( (le32_to_cpu(de->inode) == 0 &&
 298                       le16_to_cpu(de->rec_len) >= rec_len) ||
 299                      (le16_to_cpu(de->rec_len) >=
 300                       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  * ext2_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 static void show_dentry(struct list_head * dlist, int subdirs)
 413 {
 414         struct list_head *tmp = dlist;
 415
 416         while ((tmp = tmp->next) != dlist) {
 417                 struct dentry * dentry;
 418                 const char * unhashed = "";
 419
 420                 if ( subdirs )
 421                         dentry  = list_entry(tmp, struct dentry, d_child);
 422                 else
 423                         dentry  = list_entry(tmp, struct dentry, d_alias);
 424
 425                 if (list_empty(&dentry->d_hash))
 426                         unhashed = "(unhashed)";
 427
 428                 if ( dentry->d_inode )
 429                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 430                                dentry->d_parent->d_name.name,
 431                                dentry->d_name.name, dentry->d_count,
 432                                unhashed, dentry->d_inode->i_ino,
 433                                dentry->d_inode->i_dev,
 434                                dentry->d_inode->i_count);
 435                 else
 436                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 437                                dentry->d_parent->d_name.name,
 438                                dentry->d_name.name, dentry->d_count,
 439                                unhashed);
 440         }
 441 }
 442
 443
 444 /*
 445  * By the time this is called, we already have created
 446  * the directory cache entry for the new file, but it
 447  * is so far negative - it has no inode.
 448  *
 449  * If the create succeeds, we fill in the inode information
 450  * with d_instantiate().
 451  */
 452 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 453 {
 454         struct inode * inode;
 455         struct page *page;
 456         struct ext2_dir_entry_2 * de;
 457         int err = -EIO;
 458         ino_t ino;
 459
 460         ENTRY;
 461
 462         /*
 463          * N.B. Several error exits in ext2_new_inode don't set err.
 464          */
 465         ino = iops(dir)->o_create(iid(dir), 0, &err);
 466         if ( err )
 467                 return err;
 468         err = -EIO;
 469         inode =  iget(dir->i_sb, ino);
 470         if (!inode || !list_empty(&inode->i_dentry)) {
 471                 CDEBUG(D_INODE, "No inode, ino %ld\n", ino);
 472                 EXIT;
 473                 return err;
 474         }
 475
 476         inode->i_op = &obdfs_inode_ops;
 477         inode->i_mode = mode;
 478         mark_inode_dirty(inode);
 479         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 480         if (!page) {
 481                 inode->i_nlink--;
 482                 mark_inode_dirty(inode);
 483                 iput (inode);
 484                 EXIT;
 485                 return err;
 486         }
 487         de->inode = cpu_to_le32(inode->i_ino);
 488         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 489         dir->i_version = ++event;
 490         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 491         UnlockPage(page);
 492 #if 0
 493         if (IS_SYNC(dir)) {
 494                 ll_rw_block (WRITE, 1, &bh);
 495                 wait_on_buffer (bh);
 496         }
 497 #endif
 498         page_cache_release(page);
 499         d_instantiate(dentry, inode);
 500         EXIT;
 501         return 0;
 502 }
 503
 504 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 505 {
 506         struct inode * inode;
 507         struct page *page;
 508         struct ext2_dir_entry_2 * de;
 509         int err = -EIO;
 510
 511         ino_t ino;
 512
 513         ENTRY;
 514
 515         /*
 516          * N.B. Several error exits in ext2_new_inode don't set err.
 517          */
 518         ino = iops(dir)->o_create(iid(dir), 0, &err);
 519         if ( ino == -1 )
 520                 return -1;
 521         inode =  iget(dir->i_sb, ino);
 522         if (!inode)
 523                 return err;
 524
 525         inode->i_uid = current->fsuid;
 526         init_special_inode(inode, mode, rdev);
 527         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 528         if (!page)
 529                 goto out_no_entry;
 530         de->inode = cpu_to_le32(inode->i_ino);
 531         dir->i_version = ++event;
 532         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 533         mark_inode_dirty(inode);
 534         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 535         UnlockPage(page);
 536
 537 #if 0
 538         if (IS_SYNC(dir)) {
 539                 ll_rw_block (WRITE, 1, &bh);
 540                 wait_on_buffer (bh);
 541         }
 542 #endif
 543         d_instantiate(dentry, inode);
 544         page_cache_release(page);
 545         err = 0;
 546 out:
 547         return err;
 548
 549 out_no_entry:
 550         inode->i_nlink--;
 551         mark_inode_dirty(inode);
 552         iput(inode);
 553         goto out;
 554 }
 555
 556 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 557 {
 558         struct inode * child;
 559         struct page *page, *child_page;
 560         struct ext2_dir_entry_2 * de;
 561         int err;
 562         ino_t ino;
 563
 564         ENTRY;
 565
 566         err = -EMLINK;
 567         if (dir->i_nlink >= EXT2_LINK_MAX)
 568                 goto out;
 569
 570         err = -EIO;
 571         ino = iops(dir)->o_create(iid(dir), 0, &err);
 572         if ( ino == -1 )
 573                 return -1;
 574         child =  iget(dir->i_sb, ino);
 575         if (!child)
 576                 return err;
 577
 578
 579         child->i_op = &obdfs_inode_ops;
 580         child->i_blocks = 0;
 581         child_page = obdfs_getpage(child, 0, 1, LOCKED);
 582         if (!child_page) {
 583                 child->i_nlink--; /* is this nlink == 0? */
 584                 mark_inode_dirty(child);
 585                 iput (child);
 586                 return err;
 587         }
 588
 589         /* create . and .. */
 590         de = (struct ext2_dir_entry_2 *) page_address(child_page);
 591         de->inode = cpu_to_le32(child->i_ino);
 592         de->name_len = 1;
 593         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 594         strcpy (de->name, ".");
 595         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 596
 597         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 598         de->inode = cpu_to_le32(dir->i_ino);
 599         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 600         de->name_len = 2;
 601         strcpy (de->name, "..");
 602         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 603
 604         iops(dir)->o_brw(WRITE, iid(dir), child, child_page, 1);
 605         child->i_blocks = PAGE_SIZE/child->i_sb->s_blocksize;
 606         child->i_size = PAGE_SIZE;
 607         UnlockPage(child_page);
 608         page_cache_release(child_page);
 609
 610         child->i_nlink = 2;
 611         child->i_mode = S_IFDIR | mode;
 612         if (dir->i_mode & S_ISGID)
 613                 child->i_mode |= S_ISGID;
 614         mark_inode_dirty(child);
 615
 616         /* now deal with the parent */
 617         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 618         if (!page) {
 619                 goto out_no_entry;
 620         }
 621
 622         de->inode = cpu_to_le32(child->i_ino);
 623         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 624         dir->i_version = ++event;
 625
 626 #if 0
 627         if (IS_SYNC(dir)) {
 628                 ll_rw_block (WRITE, 1, &bh);
 629                 wait_on_buffer (bh);
 630         }
 631 #endif
 632         dir->i_nlink++;
 633         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 634         mark_inode_dirty(dir);
 635         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 1);
 636         UnlockPage(page);
 637         page_cache_release(page);
 638         d_instantiate(dentry, child);
 639         err = 0;
 640 out:
 641         EXIT;
 642         return err;
 643
 644 out_no_entry:
 645         child->i_nlink = 0;
 646         mark_inode_dirty(child);
 647         iput (child);
 648         EXIT;
 649         goto out;
 650 }
 651
 652
 653 /*
 654  * routine to check that the specified directory is empty (for rmdir)
 655  */
 656 static int empty_dir (struct inode * inode)
 657 {
 658         unsigned long offset;
 659         struct page *page;
 660         struct ext2_dir_entry_2 * de, * de1;
 661         struct super_block * sb;
 662
 663         sb = inode->i_sb;
 664         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 665             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 666                 ext2_warning (inode->i_sb, "empty_dir",
 667                               "bad directory (dir #%lu) - no data block",
 668                               inode->i_ino);
 669                 return 1;
 670         }
 671         de = (struct ext2_dir_entry_2 *) page_address(page);
 672         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 673         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 674             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 675                 ext2_warning (inode->i_sb, "empty_dir",
 676                               "bad directory (dir #%lu) - no `.' or `..'",
 677                               inode->i_ino);
 678                 page_cache_release(page);
 679                 return 1;
 680         }
 681         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 682         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 683         while (offset < inode->i_size ) {
 684                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 685                         if (page) {
 686                                 UnlockPage(page);
 687                                 page_cache_release(page);
 688                         }
 689                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 690                         if (!page) {
 691 #if 0
 692                                 ext2_error (sb, "empty_dir",
 693                                             "directory #%lu contains a hole at offset %lu",
 694                                             inode->i_ino, offset);
 695 #endif
 696                                 offset += sb->s_blocksize;
 697                                 continue;
 698                         }
 699                         de = (struct ext2_dir_entry_2 *) page_address(page);
 700                 }
 701                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 702                                            offset)) {
 703                         UnlockPage(page);
 704                         page_cache_release(page);
 705                         return 1;
 706                 }
 707                 if (le32_to_cpu(de->inode)) {
 708                         UnlockPage(page);
 709                         page_cache_release(page);
 710                         return 0;
 711                 }
 712                 offset += le16_to_cpu(de->rec_len);
 713                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 714         }
 715         UnlockPage(page);
 716         page_cache_release(page);
 717         return 1;
 718 }
 719
 720 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 721 {
 722         int retval;
 723         struct inode * inode;
 724         struct page *page;
 725         struct ext2_dir_entry_2 * de;
 726
 727         ENTRY;
 728
 729         retval = -ENOENT;
 730         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 731         if (!page)
 732                 goto end_rmdir;
 733
 734         inode = dentry->d_inode;
 735         DQUOT_INIT(inode);
 736
 737         retval = -EIO;
 738         if (le32_to_cpu(de->inode) != inode->i_ino)
 739                 goto end_rmdir;
 740
 741         retval = -ENOTEMPTY;
 742         if (!empty_dir (inode))
 743                 goto end_rmdir;
 744
 745         retval = obdfs_delete_entry (de, page);
 746         dir->i_version = ++event;
 747         if (retval)
 748                 goto end_rmdir;
 749         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 750         UnlockPage(page);
 751 #if 0
 752         if (IS_SYNC(dir)) {
 753                 ll_rw_block (WRITE, 1, &bh);
 754                 wait_on_buffer (bh);
 755         }
 756 #endif
 757         if (inode->i_nlink != 2)
 758                 ext2_warning (inode->i_sb, "ext2_rmdir",
 759                               "empty directory has nlink!=2 (%d)",
 760                               inode->i_nlink);
 761         inode->i_version = ++event;
 762         inode->i_nlink = 0;
 763         inode->i_size = 0;
 764         mark_inode_dirty(inode);
 765         dir->i_nlink--;
 766         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 767         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 768         mark_inode_dirty(dir);
 769         d_delete(dentry);
 770
 771 end_rmdir:
 772         if ( page )
 773                 page_cache_release(page);
 774         EXIT;
 775         return retval;
 776 }
 777
 778 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 779 {
 780         int retval;
 781         struct inode * inode;
 782         struct page *page;
 783         struct ext2_dir_entry_2 * de;
 784
 785         ENTRY;
 786
 787         retval = -ENOENT;
 788         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 789         if (!page)
 790                 goto end_unlink;
 791
 792         inode = dentry->d_inode;
 793         DQUOT_INIT(inode);
 794
 795         retval = -EIO;
 796         if (le32_to_cpu(de->inode) != inode->i_ino)
 797                 goto end_unlink;
 798
 799         if (!inode->i_nlink) {
 800                 ext2_warning (inode->i_sb, "ext2_unlink",
 801                               "Deleting nonexistent file (%lu), %d",
 802                               inode->i_ino, inode->i_nlink);
 803                 inode->i_nlink = 1;
 804         }
 805         retval = obdfs_delete_entry (de, page);
 806         if (retval)
 807                 goto end_unlink;
 808         dir->i_version = ++event;
 809         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 810         UnlockPage(page);
 811 #if 0
 812         if (IS_SYNC(dir)) {
 813                 ll_rw_block (WRITE, 1, &bh);
 814                 wait_on_buffer (bh);
 815         }
 816 #endif
 817         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 818         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 819         mark_inode_dirty(dir);
 820         inode->i_nlink--;
 821         mark_inode_dirty(inode);
 822         inode->i_ctime = dir->i_ctime;
 823         retval = 0;
 824         d_delete(dentry);       /* This also frees the inode */
 825
 826 end_unlink:
 827         if (page)
 828                 page_cache_release(page);
 829         EXIT;
 830         return retval;
 831 }
 832
 833 int obdfs_symlink (struct inode * dir, struct dentry *dentry, const char * symname)
 834 {
 835         struct ext2_dir_entry_2 * de;
 836         struct inode * inode;
 837         struct page* page = NULL, * name_page = NULL;
 838         char * link;
 839         int i, l, err = -EIO;
 840         char c;
 841         ino_t ino;
 842
 843         ENTRY;
 844         /*
 845          * N.B. Several error exits in ext2_new_inode don't set err.
 846          */
 847         ino = iops(dir)->o_create(iid(dir), 0, &err);
 848         if ( ino == -1 )  {
 849                 EXIT;
 850                 return -1;
 851         }
 852         inode =  iget(dir->i_sb, ino);
 853         if (!inode) {
 854                 EXIT;
 855                 return err;
 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         /* For obdfs we always use normal (not fast) symlinks
 865         if (l >= sizeof (inode->u.ext2_i.i_data)) { */
 866                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 867
 868                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 869                 if (!name_page) {
 870                         inode->i_nlink--;
 871                         mark_inode_dirty(inode);
 872                         iput (inode);
 873                         EXIT;
 874                         return err;
 875                 }
 876                 link = (char *)page_address(name_page);
 877         /* } else {
 878                 link = (char *) inode->u.ext2_i.i_data;
 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                 iops(inode)->o_brw(WRITE, iid(inode), inode, name_page, 1);
 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         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 1);
 903         UnlockPage(page);
 904 #if 0
 905         if (IS_SYNC(dir)) {
 906                 ll_rw_block (WRITE, 1, &bh);
 907                 wait_on_buffer (bh);
 908         }
 909 #endif
 910         d_instantiate(dentry, inode);
 911         err = 0;
 912 out:
 913         EXIT;
 914         return err;
 915
 916 out_no_entry:
 917         inode->i_nlink--;
 918         mark_inode_dirty(inode);
 919         iput (inode);
 920         goto out;
 921 }
 922
 923 int obdfs_link (struct dentry * old_dentry,
 924                 struct inode * dir, struct dentry *dentry)
 925 {
 926         struct inode *inode = old_dentry->d_inode;
 927         struct ext2_dir_entry_2 * de;
 928         struct page *page;
 929         int err;
 930
 931         ENTRY;
 932
 933         if (S_ISDIR(inode->i_mode))
 934                 return -EPERM;
 935
 936         if (inode->i_nlink >= EXT2_LINK_MAX)
 937                 return -EMLINK;
 938
 939         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 940         if (!page)
 941                 return err;
 942
 943         de->inode = cpu_to_le32(inode->i_ino);
 944         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 945         dir->i_version = ++event;
 946         iops(dir)->o_brw(WRITE, iid(dir), dir, page, 0);
 947         UnlockPage(page);
 948
 949 #if 0
 950         if (IS_SYNC(dir)) {
 951                 ll_rw_block (WRITE, 1, &bh);
 952                 wait_on_buffer (bh);
 953         }
 954 #endif
 955         page_cache_release(page);
 956         inode->i_nlink++;
 957         inode->i_ctime = CURRENT_TIME;
 958         mark_inode_dirty(inode);
 959         inode->i_count++;
 960         d_instantiate(dentry, inode);
 961         return 0;
 962 }
 963
 964 #define PARENT_INO(buffer) \
 965         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 966         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 967
 968 /*
 969  * Anybody can rename anything with this: the permission checks are left to the
 970  * higher-level routines.
 971  */
 972 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 973                            struct inode * new_dir, struct dentry *new_dentry)
 974 {
 975         struct inode * old_inode, * new_inode;
 976         struct page * old_page, * new_page, * dir_page;
 977         struct ext2_dir_entry_2 * old_de, * new_de;
 978         int retval;
 979
 980         ENTRY;
 981
 982         new_page = dir_page = NULL;
 983
 984         /* does the old entry exist? - if not get out */
 985         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 986         PDEBUG(old_page, "rename - old page");
 987         /*
 988          *  Check for inode number is _not_ due to possible IO errors.
 989          *  We might rmdir the source, keep it as pwd of some process
 990          *  and merrily kill the link to whatever was created under the
 991          *  same name. Goodbye sticky bit ;-<
 992          */
 993         old_inode = old_dentry->d_inode;
 994         retval = -ENOENT;
 995         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
 996                 goto end_rename;
 997
 998         /* find new inode */
 999         new_inode = new_dentry->d_inode;
1000         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1001                                 new_dentry->d_name.len, &new_de, NOLOCK);
1002         PDEBUG(new_page, "rename - new page ");
1003         if (new_page) {
1004                 if (!new_inode) {
1005                         page_cache_release(new_page);
1006                         new_page = NULL;
1007                 } else {
1008                         DQUOT_INIT(new_inode);
1009                 }
1010         }
1011         /* in this case we to check more ... */
1012         if (S_ISDIR(old_inode->i_mode)) {
1013                 /* can only rename into empty new directory */
1014                 if (new_inode) {
1015                         retval = -ENOTEMPTY;
1016                         if (!empty_dir (new_inode))
1017                                 goto end_rename;
1018                 }
1019                 retval = -EIO;
1020                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1021                 PDEBUG(dir_page, "rename dir page");
1022
1023                 if (!dir_page)
1024                         goto end_rename;
1025                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1026                         goto end_rename;
1027                 retval = -EMLINK;
1028                 if (!new_inode && new_dir!=old_dir &&
1029                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1030                         goto end_rename;
1031         }
1032         /* create the target dir entry */
1033         if (!new_page) {
1034                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1035                                         new_dentry->d_name.len, &new_de,
1036                                         &retval);
1037                 PDEBUG(new_page, "rename new page");
1038                 if (!new_page)
1039                         goto end_rename;
1040         }
1041         new_dir->i_version = ++event;
1042
1043         /*
1044          * remove the old entry
1045          */
1046         new_de->inode = le32_to_cpu(old_inode->i_ino);
1047         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1048                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1049                 new_de->file_type = old_de->file_type;
1050
1051         obdfs_delete_entry (old_de, old_page);
1052
1053         old_dir->i_version = ++event;
1054         if (new_inode) {
1055                 new_inode->i_nlink--;
1056                 new_inode->i_ctime = CURRENT_TIME;
1057                 mark_inode_dirty(new_inode);
1058         }
1059         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1060         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1061         mark_inode_dirty(old_dir);
1062         if (dir_page) {
1063                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1064                 iops(old_inode)->o_brw(WRITE, iid(old_inode), old_inode, dir_page, 0);
1065                 old_dir->i_nlink--;
1066                 mark_inode_dirty(old_dir);
1067                 if (new_inode) {
1068                         new_inode->i_nlink--;
1069                         mark_inode_dirty(new_inode);
1070                 } else {
1071                         new_dir->i_nlink++;
1072                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1073                         mark_inode_dirty(new_dir);
1074                 }
1075         }
1076         if ( old_page != new_page ) {
1077                 unsigned long index = old_page->index;
1078                 /* lock the old_page and release unlocked copy */
1079                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1080                 page_cache_release(old_page);
1081                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1082                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1083                 iops(old_dir)->o_brw(WRITE, iid(old_dir), old_dir, old_page,0);
1084         }
1085 #if 0
1086         if (IS_SYNC(old_dir)) {
1087                 ll_rw_block (WRITE, 1, &old_bh);
1088                 wait_on_buffer (old_bh);
1089         }
1090 #endif
1091         iops(new_dir)->o_brw(WRITE, iid(new_dir), new_dir, new_page, 0);
1092 #if 0
1093         if (IS_SYNC(new_dir)) {
1094                 ll_rw_block (WRITE, 1, &new_bh);
1095                 wait_on_buffer (new_bh);
1096         }
1097 #endif
1098
1099         retval = 0;
1100
1101 end_rename:
1102         if (old_page && PageLocked(old_page) )
1103                 UnlockPage(old_page);
1104         if (old_page)
1105                 page_cache_release(old_page);
1106         if (new_page && PageLocked(new_page) )
1107                 UnlockPage(new_page);
1108         if (new_page)
1109                 page_cache_release(new_page);
1110         if (dir_page && PageLocked(dir_page) )
1111                 UnlockPage(dir_page);
1112         if (dir_page)
1113                 page_cache_release(dir_page);
1114
1115
1116         return retval;
1117 }