lustre/obdfs/namei.c

   1 /*
   2  *  linux/fs/obdfs/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/ext2/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *
  20  *  Changes for use in OBDFS
  21  *  Copyright (c) 1999, Seagate Technology Inc.
  22  *
  23  */
  24
  25 #include <asm/uaccess.h>
  26
  27 #include <linux/errno.h>
  28 #include <linux/fs.h>
  29 #include <linux/ext2_fs.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/sched.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/locks.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/iobuf.h>
  37
  38 #include <linux/obd_support.h>
  39 #include <linux/obdfs.h>
  40
  41 /*
  42  * define how far ahead to read directories while searching them.
  43  */
  44 #define NAMEI_RA_CHUNKS  2
  45 #define NAMEI_RA_BLOCKS  4
  46 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  47 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  48
  49 /*
  50  * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
  51  *
  52  * `len <= EXT2_NAME_LEN' is guaranteed by caller.
  53  * `de != NULL' is guaranteed by caller.
  54  */
  55 static inline int ext2_match (int len, const char * const name,
  56                        struct ext2_dir_entry_2 * de)
  57 {
  58         if (len != de->name_len)
  59                 return 0;
  60         if (!de->inode)
  61                 return 0;
  62         return !memcmp(name, de->name, len);
  63 }
  64
  65 /*
  66  *      ext2_find_entry()
  67  *
  68  * finds an entry in the specified directory with the wanted name. It
  69  * returns the cache buffer in which the entry was found, and the entry
  70  * itself (as a parameter - res_dir). It does NOT read the inode of the
  71  * entry - you'll have to do that yourself if you want to.
  72  */
  73 static struct page * obdfs_find_entry (struct inode * dir,
  74                                              const char * const name, int namelen,
  75                                              struct ext2_dir_entry_2 ** res_dir, int lock)
  76 {
  77         struct super_block * sb;
  78         unsigned long offset;
  79         struct page * page;
  80         ENTRY;
  81         CDEBUG(D_INODE, "find entry for %*s\n", namelen, name);
  82
  83         *res_dir = NULL;
  84         sb = dir->i_sb;
  85
  86         if (namelen > EXT2_NAME_LEN)
  87                 return NULL;
  88
  89         CDEBUG(D_INODE, "dirsize is %Ld\n", dir->i_size);
  90
  91         page = 0;
  92         offset = 0;
  93         while ( offset < dir->i_size ) {
  94                 struct ext2_dir_entry_2 * de;
  95                 char * dlimit;
  96
  97                 page = obdfs_getpage(dir, offset, 0, lock);
  98
  99                 if ( !page ) {
 100                         CDEBUG(D_INODE, "No page, offset %lx\n", offset);
 101                         return NULL;
 102                 }
 103
 104                 de = (struct ext2_dir_entry_2 *) page_address(page);
 105                 dlimit = (char *)page_address(page) + PAGE_SIZE;
 106                 while ((char *) de < dlimit) {
 107                         /* this code is executed quadratically often */
 108                         /* do minimal checking `by hand' */
 109                         int de_len;
 110                         /* CDEBUG(D_INODE, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 111                                de, le16_to_cpu(de->rec_len), page_address(page),
 112                                page_address(page) + PAGE_SIZE, offset); */
 113
 114                         if ((char *) de + namelen <= dlimit &&
 115                             ext2_match (namelen, name, de)) {
 116                                 /* found a match -
 117                                    just to be sure, do a full check */
 118                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 119                                                           dir, de, page, offset))
 120                                         goto failure;
 121                                 *res_dir = de;
 122                                 EXIT;
 123                                 return page;
 124                         }
 125                         /* prevent looping on a bad block */
 126                         de_len = le16_to_cpu(de->rec_len);
 127                         if (de_len <= 0) {
 128                                 printk("Bad entry at %p len %d\n", de, de_len);
 129                                 goto failure;
 130                         }
 131                         offset += de_len;
 132                         de = (struct ext2_dir_entry_2 *)
 133                                 ((char *) de + de_len);
 134                         /* CDEBUG(D_INODE, "Next while %lx\n", offset); */
 135                 }
 136                 if ( lock )
 137                         UnlockPage(page);
 138                 page_cache_release(page);
 139                 page = NULL;
 140                 CDEBUG(D_INODE, "Next for %lx\n", offset);
 141         }
 142
 143 failure:
 144         CDEBUG(D_INODE, "Negative case, page %p, offset %ld\n", page, offset);
 145         if (page) {
 146                 if (lock)
 147                         UnlockPage(page);
 148                 page_cache_release(page);
 149         }
 150         EXIT;
 151         return NULL;
 152 }
 153
 154 struct dentry *obdfs_lookup(struct inode * dir, struct dentry *dentry)
 155 {
 156         struct inode * inode;
 157         struct ext2_dir_entry_2 * de;
 158         struct page * page;
 159         ENTRY;
 160
 161         if (dentry->d_name.len > EXT2_NAME_LEN)
 162                 return ERR_PTR(-ENAMETOOLONG);
 163
 164         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 165         inode = NULL;
 166         if ( !page )
 167                 CDEBUG(D_INODE, "No page - negative entry.\n");
 168         if ( page && !de ) {
 169                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 170                 return ERR_PTR(-ENOENT);
 171         }
 172         if (page) {
 173                 unsigned long ino = le32_to_cpu(de->inode);
 174                 UnlockPage(page);
 175                 page_cache_release(page);
 176                 inode = iget(dir->i_sb, ino);
 177
 178                 if (!inode) {
 179                         CDEBUG(D_INODE, "No inode.\n");
 180                         EXIT;
 181                         return ERR_PTR(-EACCES);
 182                 }
 183         }
 184         d_add(dentry, inode);
 185         EXIT;
 186         return NULL;
 187 }
 188
 189
 190 /*
 191  *      ext2_add_entry()
 192  *
 193  * adds a file entry to the specified directory, using the same
 194  * semantics as ext2_find_entry(). It returns NULL if it failed.
 195  *
 196  * NOTE!! The inode part of 'de' is left at 0 - which means you
 197  * may not sleep between calling this and putting something into
 198  * the entry, as someone else might have used it while you slept.
 199
 200  * returns a locked and held page upon success
 201  */
 202
 203 /* XXX I believe these pages should in fact NOT be locked */
 204
 205 static struct page *obdfs_add_entry (struct inode * dir,
 206                                      const char * name, int namelen,
 207                                      struct ext2_dir_entry_2 ** res_dir,
 208                                      int *err)
 209 {
 210         unsigned long offset;
 211         unsigned short rec_len;
 212         struct page *page;
 213         struct ext2_dir_entry_2 * de, * de1;
 214         struct super_block * sb;
 215
 216         ENTRY;
 217         *err = -EINVAL;
 218         *res_dir = NULL;
 219         if (!dir || !dir->i_nlink) {
 220                 EXIT;
 221                 return NULL;
 222         }
 223         sb = dir->i_sb;
 224
 225         if (!namelen) {
 226                 EXIT;
 227                 return NULL;
 228         }
 229         /*
 230          * Is this a busy deleted directory?  Can't create new files if so
 231          */
 232         if (dir->i_size == 0)
 233         {
 234                 EXIT;
 235                 *err = -ENOENT;
 236                 return NULL;
 237         }
 238         page = obdfs_getpage(dir, 0, 0, LOCKED);
 239         if (!page) {
 240                 EXIT;
 241                 return NULL;
 242         }
 243         rec_len = EXT2_DIR_REC_LEN(namelen);
 244         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 245         offset = 0;
 246         de = (struct ext2_dir_entry_2 *) page_address(page);
 247         *err = -ENOSPC;
 248         while (1) {
 249                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 250                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 251                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 252                         UnlockPage(page);
 253                         page_cache_release(page);
 254                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 255                         if (!page) {
 256                                 EXIT;
 257                                 return NULL;
 258                         }
 259                         if (dir->i_size <= offset) {
 260                                 if (dir->i_size == 0) {
 261                                         *err = -ENOENT;
 262                                         EXIT;
 263                                         return NULL;
 264                                 }
 265
 266                                 CDEBUG(D_INODE, "creating next block\n");
 267
 268                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 269                                 de->inode = 0;
 270                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 271                                 dir->i_size = offset + PAGE_SIZE;
 272                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 273                                 mark_inode_dirty(dir);
 274                         } else {
 275
 276                                 ext2_debug ("skipping to next block\n");
 277
 278                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 279                         }
 280                 }
 281                 CDEBUG(D_INODE, "\n");
 282                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 283                                            offset)) {
 284                         *err = -ENOENT;
 285                         UnlockPage(page);
 286                         page_cache_release(page);
 287                         EXIT;
 288                         return NULL;
 289                 }
 290                 CDEBUG(D_INODE, "\n");
 291                 if (ext2_match (namelen, name, de)) {
 292                                 *err = -EEXIST;
 293                                 UnlockPage(page);
 294                                 page_cache_release(page);
 295                                 EXIT;
 296                                 return NULL;
 297                 }
 298                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 299                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 300                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 301                         offset += le16_to_cpu(de->rec_len);
 302                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 303                         if (le32_to_cpu(de->inode)) {
 304                                 CDEBUG(D_INODE, "Inserting new in %p\n", de);
 305                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 306                                         EXT2_DIR_REC_LEN(de->name_len));
 307                                 CDEBUG(D_INODE, "-- de1 at %p\n", de1);
 308                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 309                                         EXT2_DIR_REC_LEN(de->name_len));
 310                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 311                                 de = de1;
 312                         }
 313                         CDEBUG(D_INODE, "Reclen adjusted; copy %d bytes to %p, page at %#lx EOP at %#lx\n", namelen, de->name, page_address(page), page_address(page) + PAGE_SIZE);
 314                         de->inode = 0;
 315                         de->name_len = namelen;
 316                         de->file_type = 0;
 317                         memcpy (de->name, name, namelen);
 318                         CDEBUG(D_INODE, "Copy done\n");
 319                         /*
 320                          * XXX shouldn't update any times until successful
 321                          * completion of syscall, but too many callers depend
 322                          * on this.
 323                          *
 324                          * XXX similarly, too many callers depend on
 325                          * ext2_new_inode() setting the times, but error
 326                          * recovery deletes the inode, so the worst that can
 327                          * happen is that the times are slightly out of date
 328                          * and/or different from the directory change time.
 329                          */
 330                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 331                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 332                         mark_inode_dirty(dir);
 333                         dir->i_version = ++event;
 334                         *res_dir = de;
 335                         *err = 0;
 336                         PDEBUG(page, "addentry");
 337                         CDEBUG(D_INODE, "Regular exit from add_entry");
 338                         EXIT;
 339                         return page;
 340                 }
 341                 CDEBUG(D_INODE, "\n");
 342                 offset += le16_to_cpu(de->rec_len);
 343                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 344
 345         }
 346         CDEBUG(D_INODE, "\n");
 347
 348         UnlockPage(page);
 349         page_cache_release(page);
 350         PDEBUG(page, "addentry");
 351         EXIT;
 352         return NULL;
 353 }
 354
 355 /*
 356  * 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 #if 0
 413 static void show_dentry(struct list_head * dlist, int subdirs)
 414 {
 415         struct list_head *tmp = dlist;
 416
 417         while ((tmp = tmp->next) != dlist) {
 418                 struct dentry * dentry;
 419                 const char * unhashed = "";
 420
 421                 if ( subdirs )
 422                         dentry  = list_entry(tmp, struct dentry, d_child);
 423                 else
 424                         dentry  = list_entry(tmp, struct dentry, d_alias);
 425
 426                 if (list_empty(&dentry->d_hash))
 427                         unhashed = "(unhashed)";
 428
 429                 if ( dentry->d_inode )
 430                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 431                                dentry->d_parent->d_name.name,
 432                                dentry->d_name.name, dentry->d_count,
 433                                unhashed, dentry->d_inode->i_ino,
 434                                dentry->d_inode->i_dev,
 435                                dentry->d_inode->i_count);
 436                 else
 437                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 438                                dentry->d_parent->d_name.name,
 439                                dentry->d_name.name, dentry->d_count,
 440                                unhashed);
 441         }
 442 }
 443 #endif
 444
 445
 446 struct inode *obdfs_new_inode(struct inode *dir)
 447 {
 448         struct obdo *obdo;
 449         struct inode *inode;
 450         struct obdfs_inode_info *oinfo;
 451         int err;
 452
 453         obdo = obdo_alloc();
 454         if (!obdo) {
 455                 EXIT;
 456                 return ERR_PTR(-ENOMEM);
 457         }
 458
 459         err = IOPS(dir, create)(IID(dir), obdo);
 460         if ( err )
 461                 return ERR_PTR(err);
 462
 463         inode = iget(dir->i_sb, (unsigned long)obdo->o_id);
 464         if (!inode) {
 465                 obdo_free(obdo);
 466                 EXIT;
 467                 return ERR_PTR(-EIO);
 468         }
 469
 470         if (!list_empty(&inode->i_dentry)) {
 471                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n",
 472                        inode->i_ino);
 473                 iput(inode);
 474                 EXIT;
 475                 return ERR_PTR(-EIO);
 476         }
 477
 478         obdo_free(obdo);
 479
 480         oinfo = inode->u.generic_ip;
 481         INIT_LIST_HEAD(&oinfo->oi_list);
 482         EXIT;
 483         return inode;
 484 }
 485
 486
 487 /*
 488  * By the time this is called, we already have created
 489  * the directory cache entry for the new file, but it
 490  * is so far negative - it has no inode.
 491  *
 492  * If the create succeeds, we fill in the inode information
 493  * with d_instantiate().
 494  */
 495 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 496 {
 497         struct inode * inode;
 498         struct page *page;
 499         struct ext2_dir_entry_2 * de;
 500         int err = -EIO;
 501
 502         ENTRY;
 503
 504         inode = obdfs_new_inode(dir);
 505         if ( IS_ERR(inode) ) {
 506                 EXIT;
 507                 return PTR_ERR(inode);
 508         }
 509
 510         inode->i_op = &obdfs_file_inode_operations;
 511         inode->i_mode = mode;
 512         mark_inode_dirty(inode);
 513         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 514         if (!page) {
 515                 inode->i_nlink--;
 516                 mark_inode_dirty(inode);
 517                 iput (inode);
 518                 EXIT;
 519                 return err;
 520         }
 521         de->inode = cpu_to_le32(inode->i_ino);
 522         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 523         dir->i_version = ++event;
 524
 525         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 526         UnlockPage(page);
 527
 528         page_cache_release(page);
 529         d_instantiate(dentry, inode);
 530         EXIT;
 531         return err;
 532 }
 533
 534 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 535 {
 536         struct inode * inode;
 537         struct page *page;
 538         struct ext2_dir_entry_2 * de;
 539         int err;
 540
 541         ENTRY;
 542
 543         inode = obdfs_new_inode(dir);
 544         if ( IS_ERR(inode) ) {
 545                 EXIT;
 546                 return PTR_ERR(inode);
 547         }
 548
 549         inode->i_uid = current->fsuid;
 550         init_special_inode(inode, mode, rdev);
 551         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 552         if (!page)
 553                 goto out_no_entry;
 554         de->inode = cpu_to_le32(inode->i_ino);
 555         dir->i_version = ++event;
 556         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 557         mark_inode_dirty(inode);
 558
 559         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 560         UnlockPage(page);
 561
 562         d_instantiate(dentry, inode);
 563         page_cache_release(page);
 564         err = 0;
 565 out:
 566         return err;
 567
 568 out_no_entry:
 569         inode->i_nlink--;
 570         mark_inode_dirty(inode);
 571         iput(inode);
 572         goto out;
 573 }
 574
 575 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 576 {
 577         struct inode * inode;
 578         struct page *page, *inode_page;
 579         struct ext2_dir_entry_2 * de;
 580         int err;
 581
 582         ENTRY;
 583
 584         err = -EMLINK;
 585         if (dir->i_nlink >= EXT2_LINK_MAX)
 586                 goto out;
 587
 588         inode = obdfs_new_inode(dir);
 589         if ( IS_ERR(inode) ) {
 590                 EXIT;
 591                 return PTR_ERR(inode);
 592         }
 593
 594         inode->i_op = &obdfs_dir_inode_operations;
 595         inode->i_blocks = 0;
 596         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 597         if (!inode_page) {
 598                 inode->i_nlink--; /* is this nlink == 0? */
 599                 mark_inode_dirty(inode);
 600                 iput (inode);
 601                 return err;
 602         }
 603
 604         /* create . and .. */
 605         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 606         de->inode = cpu_to_le32(inode->i_ino);
 607         de->name_len = 1;
 608         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 609         strcpy (de->name, ".");
 610         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 611
 612         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 613         de->inode = cpu_to_le32(dir->i_ino);
 614         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 615         de->name_len = 2;
 616         strcpy (de->name, "..");
 617         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 618
 619         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 620         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 621         inode->i_size = PAGE_SIZE;
 622         UnlockPage(inode_page);
 623         page_cache_release(inode_page);
 624         /* XXX handle err */
 625
 626         inode->i_nlink = 2;
 627         inode->i_mode = S_IFDIR | mode;
 628         if (dir->i_mode & S_ISGID)
 629                 inode->i_mode |= S_ISGID;
 630         mark_inode_dirty(inode);
 631
 632         /* now deal with the parent */
 633         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 634         if (!page) {
 635                 goto out_no_entry;
 636         }
 637
 638         de->inode = cpu_to_le32(inode->i_ino);
 639         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 640         dir->i_version = ++event;
 641
 642         dir->i_nlink++;
 643         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 644         mark_inode_dirty(dir);
 645         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 646
 647         UnlockPage(page);
 648
 649         page_cache_release(page);
 650         d_instantiate(dentry, inode);
 651         err = 0;
 652 out:
 653         EXIT;
 654         return err;
 655
 656 out_no_entry:
 657         inode->i_nlink = 0;
 658         mark_inode_dirty(inode);
 659         iput (inode);
 660         EXIT;
 661         goto out;
 662 }
 663
 664
 665 /*
 666  * routine to check that the specified directory is empty (for rmdir)
 667  */
 668 static int empty_dir (struct inode * inode)
 669 {
 670         unsigned long offset;
 671         struct page *page;
 672         struct ext2_dir_entry_2 * de, * de1;
 673         struct super_block * sb;
 674
 675         sb = inode->i_sb;
 676         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 677             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 678                 ext2_warning (inode->i_sb, "empty_dir",
 679                               "bad directory (dir #%lu) - no data block",
 680                               inode->i_ino);
 681                 return 1;
 682         }
 683         de = (struct ext2_dir_entry_2 *) page_address(page);
 684         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 685         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 686             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 687                 ext2_warning (inode->i_sb, "empty_dir",
 688                               "bad directory (dir #%lu) - no `.' or `..'",
 689                               inode->i_ino);
 690                 page_cache_release(page);
 691                 return 1;
 692         }
 693         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 694         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 695         while (offset < inode->i_size ) {
 696                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 697                         if (page) {
 698                                 UnlockPage(page);
 699                                 page_cache_release(page);
 700                         }
 701                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 702                         if (!page) {
 703 #if 0
 704                                 ext2_error (sb, "empty_dir",
 705                                             "directory #%lu contains a hole at offset %lu",
 706                                             inode->i_ino, offset);
 707 #endif
 708                                 offset += sb->s_blocksize;
 709                                 continue;
 710                         }
 711                         de = (struct ext2_dir_entry_2 *) page_address(page);
 712                 }
 713                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 714                                            offset)) {
 715                         UnlockPage(page);
 716                         page_cache_release(page);
 717                         return 1;
 718                 }
 719                 if (le32_to_cpu(de->inode)) {
 720                         UnlockPage(page);
 721                         page_cache_release(page);
 722                         return 0;
 723                 }
 724                 offset += le16_to_cpu(de->rec_len);
 725                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 726         }
 727         UnlockPage(page);
 728         page_cache_release(page);
 729         return 1;
 730 }
 731
 732 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 733 {
 734         int retval;
 735         struct inode * inode;
 736         struct page *page;
 737         struct ext2_dir_entry_2 * de;
 738         int err;
 739
 740         ENTRY;
 741
 742         retval = -ENOENT;
 743         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 744         if (!page)
 745                 goto end_rmdir;
 746
 747         inode = dentry->d_inode;
 748         DQUOT_INIT(inode);
 749
 750         retval = -EIO;
 751         if (le32_to_cpu(de->inode) != inode->i_ino)
 752                 goto end_rmdir;
 753
 754         retval = -ENOTEMPTY;
 755         if (!empty_dir (inode))
 756                 goto end_rmdir;
 757
 758         retval = obdfs_delete_entry (de, page);
 759         dir->i_version = ++event;
 760         if (retval)
 761                 goto end_rmdir;
 762         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 763         UnlockPage(page);
 764
 765         if (inode->i_nlink != 2)
 766                 ext2_warning (inode->i_sb, "ext2_rmdir",
 767                               "empty directory has nlink!=2 (%d)",
 768                               inode->i_nlink);
 769         inode->i_version = ++event;
 770         inode->i_nlink = 0;
 771         inode->i_size = 0;
 772         mark_inode_dirty(inode);
 773         dir->i_nlink--;
 774         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 775         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 776         mark_inode_dirty(dir);
 777         d_delete(dentry);
 778
 779 end_rmdir:
 780         if ( page )
 781                 page_cache_release(page);
 782         EXIT;
 783         return retval;
 784 }
 785
 786 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 787 {
 788         int retval;
 789         struct inode * inode;
 790         struct page *page;
 791         struct ext2_dir_entry_2 * de;
 792         int err;
 793
 794         ENTRY;
 795
 796         retval = -ENOENT;
 797         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 798         if (!page)
 799                 goto end_unlink;
 800
 801         inode = dentry->d_inode;
 802         DQUOT_INIT(inode);
 803
 804         retval = -EIO;
 805         if (le32_to_cpu(de->inode) != inode->i_ino)
 806                 goto end_unlink;
 807
 808         if (!inode->i_nlink) {
 809                 ext2_warning (inode->i_sb, "ext2_unlink",
 810                               "Deleting nonexistent file (%lu), %d",
 811                               inode->i_ino, inode->i_nlink);
 812                 inode->i_nlink = 1;
 813         }
 814         retval = obdfs_delete_entry (de, page);
 815         if (retval)
 816                 goto end_unlink;
 817         dir->i_version = ++event;
 818         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 819         UnlockPage(page);
 820
 821         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 822         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 823         mark_inode_dirty(dir);
 824         inode->i_nlink--;
 825         mark_inode_dirty(inode);
 826         inode->i_ctime = dir->i_ctime;
 827         retval = 0;
 828         d_delete(dentry);       /* This also frees the inode */
 829
 830 end_unlink:
 831         if (page)
 832                 page_cache_release(page);
 833         EXIT;
 834         return retval;
 835 }
 836
 837 int obdfs_symlink (struct inode * dir, struct dentry *dentry, const char * symname)
 838 {
 839         struct ext2_dir_entry_2 * de;
 840         struct inode * inode;
 841         struct obdfs_inode_info *oinfo;
 842         struct page* page = NULL, * name_page = NULL;
 843         char * link;
 844         int i, l, err = -EIO;
 845         char c;
 846
 847         ENTRY;
 848         inode = obdfs_new_inode(dir);
 849         oinfo = inode->u.generic_ip;
 850         if ( IS_ERR(inode) ) {
 851                 EXIT;
 852                 return PTR_ERR(inode);
 853         }
 854
 855         inode->i_mode = S_IFLNK | S_IRWXUGO;
 856         inode->i_op = &obdfs_symlink_inode_operations;
 857         for (l = 0; l < inode->i_sb->s_blocksize - 1 &&
 858              symname [l]; l++)
 859                 ;
 860
 861         if (l >= sizeof (oinfo->oi_inline)) {
 862                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 863
 864                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 865                 if (!name_page) {
 866                         inode->i_nlink--;
 867                         mark_inode_dirty(inode);
 868                         iput (inode);
 869                         EXIT;
 870                         return err;
 871                 }
 872                 link = (char *)page_address(name_page);
 873         } else {
 874                 link = oinfo->oi_inline;
 875                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 876
 877                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 878
 879         }
 880         i = 0;
 881         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 882                 link[i++] = c;
 883         link[i] = 0;
 884         if (name_page) {
 885                 obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 886                 PDEBUG(name_page, "symlink");
 887                 UnlockPage(name_page);
 888                 page_cache_release(name_page);
 889         }
 890         inode->i_size = i;
 891         mark_inode_dirty(inode);
 892
 893         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 894         if (!page)
 895                 goto out_no_entry;
 896         de->inode = cpu_to_le32(inode->i_ino);
 897         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 898         dir->i_version = ++event;
 899         obdfs_do_writepage(dir, page, IS_SYNC(dir));
 900         UnlockPage(page);
 901
 902         d_instantiate(dentry, inode);
 903         err = 0;
 904 out:
 905         EXIT;
 906         return err;
 907
 908 out_no_entry:
 909         inode->i_nlink--;
 910         mark_inode_dirty(inode);
 911         iput (inode);
 912         goto out;
 913 }
 914
 915 int obdfs_link (struct dentry * old_dentry,
 916                 struct inode * dir, struct dentry *dentry)
 917 {
 918         struct inode *inode = old_dentry->d_inode;
 919         struct ext2_dir_entry_2 * de;
 920         struct page *page;
 921         int err;
 922
 923         ENTRY;
 924
 925         if (S_ISDIR(inode->i_mode))
 926                 return -EPERM;
 927
 928         if (inode->i_nlink >= EXT2_LINK_MAX)
 929                 return -EMLINK;
 930
 931         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 932         if (!page)
 933                 return err;
 934
 935         de->inode = cpu_to_le32(inode->i_ino);
 936         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 937         dir->i_version = ++event;
 938
 939         obdfs_do_writepage(dir, page, IS_SYNC(dir));
 940         UnlockPage(page);
 941
 942         page_cache_release(page);
 943         inode->i_nlink++;
 944         inode->i_ctime = CURRENT_TIME;
 945         mark_inode_dirty(inode);
 946         inode->i_count++;
 947         d_instantiate(dentry, inode);
 948         return 0;
 949 }
 950
 951 #define PARENT_INO(buffer) \
 952         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 953         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 954
 955 /*
 956  * Anybody can rename anything with this: the permission checks are left to the
 957  * higher-level routines.
 958  */
 959 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 960                            struct inode * new_dir, struct dentry *new_dentry)
 961 {
 962         struct inode * old_inode, * new_inode;
 963         struct page * old_page, * new_page, * dir_page;
 964         struct ext2_dir_entry_2 * old_de, * new_de;
 965         int retval;
 966
 967         ENTRY;
 968
 969         new_page = dir_page = NULL;
 970
 971         /* does the old entry exist? - if not get out */
 972         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 973         PDEBUG(old_page, "rename - old page");
 974         /*
 975          *  Check for inode number is _not_ due to possible IO errors.
 976          *  We might rmdir the source, keep it as pwd of some process
 977          *  and merrily kill the link to whatever was created under the
 978          *  same name. Goodbye sticky bit ;-<
 979          */
 980         old_inode = old_dentry->d_inode;
 981         retval = -ENOENT;
 982         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
 983                 goto end_rename;
 984
 985         /* find new inode */
 986         new_inode = new_dentry->d_inode;
 987         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
 988                                 new_dentry->d_name.len, &new_de, NOLOCK);
 989         PDEBUG(new_page, "rename - new page ");
 990         if (new_page) {
 991                 if (!new_inode) {
 992                         page_cache_release(new_page);
 993                         new_page = NULL;
 994                 } else {
 995                         DQUOT_INIT(new_inode);
 996                 }
 997         }
 998         /* in this case we to check more ... */
 999         if (S_ISDIR(old_inode->i_mode)) {
1000                 /* can only rename into empty new directory */
1001                 if (new_inode) {
1002                         retval = -ENOTEMPTY;
1003                         if (!empty_dir (new_inode))
1004                                 goto end_rename;
1005                 }
1006                 retval = -EIO;
1007                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1008                 PDEBUG(dir_page, "rename dir page");
1009
1010                 if (!dir_page)
1011                         goto end_rename;
1012                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1013                         goto end_rename;
1014                 retval = -EMLINK;
1015                 if (!new_inode && new_dir!=old_dir &&
1016                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1017                         goto end_rename;
1018         }
1019         /* create the target dir entry */
1020         if (!new_page) {
1021                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1022                                         new_dentry->d_name.len, &new_de,
1023                                         &retval);
1024                 PDEBUG(new_page, "rename new page");
1025                 if (!new_page)
1026                         goto end_rename;
1027         }
1028         new_dir->i_version = ++event;
1029
1030         /*
1031          * remove the old entry
1032          */
1033         new_de->inode = le32_to_cpu(old_inode->i_ino);
1034         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1035                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1036                 new_de->file_type = old_de->file_type;
1037
1038         obdfs_delete_entry (old_de, old_page);
1039
1040         old_dir->i_version = ++event;
1041         if (new_inode) {
1042                 new_inode->i_nlink--;
1043                 new_inode->i_ctime = CURRENT_TIME;
1044                 mark_inode_dirty(new_inode);
1045         }
1046         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1047         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1048         mark_inode_dirty(old_dir);
1049         if (dir_page) {
1050                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1051                 obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1052                 old_dir->i_nlink--;
1053                 mark_inode_dirty(old_dir);
1054                 if (new_inode) {
1055                         new_inode->i_nlink--;
1056                         mark_inode_dirty(new_inode);
1057                 } else {
1058                         new_dir->i_nlink++;
1059                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1060                         mark_inode_dirty(new_dir);
1061                 }
1062         }
1063         if ( old_page != new_page ) {
1064                 unsigned long index = old_page->index;
1065                 /* lock the old_page and release unlocked copy */
1066                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1067                 page_cache_release(old_page);
1068                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1069                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1070                 obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1071         }
1072
1073         obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1074
1075         retval = 0;
1076
1077 end_rename:
1078         if (old_page && PageLocked(old_page) )
1079                 UnlockPage(old_page);
1080         if (old_page)
1081                 page_cache_release(old_page);
1082         if (new_page && PageLocked(new_page) )
1083                 UnlockPage(new_page);
1084         if (new_page)
1085                 page_cache_release(new_page);
1086         if (dir_page && PageLocked(dir_page) )
1087                 UnlockPage(dir_page);
1088         if (dir_page)
1089                 page_cache_release(dir_page);
1090
1091
1092         return retval;
1093 }