lustre/obdfs/namei.c

   1 /*
   2  *  linux/fs/obdfs/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/ext2/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *
  20  *  Changes for use in OBDFS
  21  *  Copyright (c) 1999, Seagate Technology Inc.
  22  *
  23  */
  24
  25 #include <asm/uaccess.h>
  26
  27 #include <linux/errno.h>
  28 #include <linux/fs.h>
  29 #include <linux/ext2_fs.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/sched.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/locks.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/iobuf.h>
  37
  38 #include <linux/obd_support.h>
  39 #include <linux/obdfs.h>
  40
  41 /*
  42  * define how far ahead to read directories while searching them.
  43  */
  44 #define NAMEI_RA_CHUNKS  2
  45 #define NAMEI_RA_BLOCKS  4
  46 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  47 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  48
  49 /*
  50  * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
  51  *
  52  * `len <= EXT2_NAME_LEN' is guaranteed by caller.
  53  * `de != NULL' is guaranteed by caller.
  54  */
  55 static inline int ext2_match (int len, const char * const name,
  56                        struct ext2_dir_entry_2 * de)
  57 {
  58         if (len != de->name_len)
  59                 return 0;
  60         if (!de->inode)
  61                 return 0;
  62         return !memcmp(name, de->name, len);
  63 }
  64
  65 /*
  66  *      obdfs_find_entry()
  67  *
  68  * finds an entry in the specified directory with the wanted name. It
  69  * returns the cache buffer in which the entry was found, and the entry
  70  * itself (as a parameter - res_dir).  It does NOT read the inode of the
  71  * entry - you'll have to do that yourself if you want to.
  72  */
  73 static struct page * obdfs_find_entry (struct inode * dir,
  74                                        const char * const name, int namelen,
  75                                        struct ext2_dir_entry_2 ** res_dir,
  76                                        int lock)
  77 {
  78         struct super_block * sb;
  79         unsigned long offset;
  80         struct page * page;
  81         ENTRY;
  82         CDEBUG(D_INODE, "find entry for %*s\n", namelen, name);
  83
  84         *res_dir = NULL;
  85         sb = dir->i_sb;
  86
  87         if (namelen > EXT2_NAME_LEN)
  88                 return NULL;
  89
  90         CDEBUG(D_INODE, "dirsize is %Ld\n", dir->i_size);
  91
  92         page = 0;
  93         offset = 0;
  94         while ( offset < dir->i_size ) {
  95                 struct ext2_dir_entry_2 * de;
  96                 char * dlimit;
  97
  98                 page = obdfs_getpage(dir, offset, 0, lock);
  99
 100                 if ( !page ) {
 101                         CDEBUG(D_INODE, "No page, offset %lx\n", offset);
 102                         return NULL;
 103                 }
 104
 105                 de = (struct ext2_dir_entry_2 *) page_address(page);
 106                 dlimit = (char *)page_address(page) + PAGE_SIZE;
 107                 while ((char *) de < dlimit) {
 108                         /* this code is executed quadratically often */
 109                         /* do minimal checking `by hand' */
 110                         int de_len;
 111                         /* CDEBUG(D_INODE, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 112                                de, le16_to_cpu(de->rec_len), page_address(page),
 113                                page_address(page) + PAGE_SIZE, offset); */
 114
 115                         if ((char *) de + namelen <= dlimit &&
 116                             ext2_match (namelen, name, de)) {
 117                                 /* found a match -
 118                                    just to be sure, do a full check */
 119                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 120                                                           dir, de, page, offset))
 121                                         goto failure;
 122                                 *res_dir = de;
 123                                 EXIT;
 124                                 return page;
 125                         }
 126                         /* prevent looping on a bad block */
 127                         de_len = le16_to_cpu(de->rec_len);
 128                         if (de_len <= 0) {
 129                                 printk("Bad entry at %p len %d\n", de, de_len);
 130                                 goto failure;
 131                         }
 132                         offset += de_len;
 133                         de = (struct ext2_dir_entry_2 *)
 134                                 ((char *) de + de_len);
 135                         /* CDEBUG(D_INODE, "Next while %lx\n", offset); */
 136                 }
 137                 if ( lock )
 138                         UnlockPage(page);
 139                 page_cache_release(page);
 140                 page = NULL;
 141                 CDEBUG(D_INODE, "Next for %lx\n", offset);
 142         }
 143
 144 failure:
 145         CDEBUG(D_INODE, "Negative case, page %p, offset %ld\n", page, offset);
 146         if (page) {
 147                 if (lock)
 148                         UnlockPage(page);
 149                 page_cache_release(page);
 150         }
 151         EXIT;
 152         return NULL;
 153 } /* obdfs_find_entry */
 154
 155 struct dentry *obdfs_lookup(struct inode *dir, struct dentry *dentry)
 156 {
 157         struct inode *inode;
 158         struct ext2_dir_entry_2 *de;
 159         struct page *page;
 160         ENTRY;
 161
 162         if (dentry->d_name.len > EXT2_NAME_LEN)
 163                 return ERR_PTR(-ENAMETOOLONG);
 164
 165         page = obdfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
 166                                 &de, LOCKED);
 167         inode = NULL;
 168         if ( !page )
 169                 CDEBUG(D_INODE, "No page - negative entry.\n");
 170         if ( page && !de ) {
 171                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 172                 return ERR_PTR(-ENOENT);
 173         }
 174         if (page) {
 175                 unsigned long ino = le32_to_cpu(de->inode);
 176                 UnlockPage(page);
 177                 page_cache_release(page);
 178                 inode = iget(dir->i_sb, ino);
 179
 180                 if (!inode) {
 181                         CDEBUG(D_INODE, "No inode.\n");
 182                         EXIT;
 183                         return ERR_PTR(-EACCES);
 184                 }
 185         }
 186         d_add(dentry, inode);
 187         EXIT;
 188         return NULL;
 189 } /* obdfs_lookup */
 190
 191
 192 /*
 193  *      obdfs_add_entry()
 194  *
 195  * adds a file entry to the specified directory, using the same
 196  * semantics as ext2_find_entry(). It returns NULL if it failed.
 197  *
 198  * NOTE!! The inode part of 'de' is left at 0 - which means you
 199  * may not sleep between calling this and putting something into
 200  * the entry, as someone else might have used it while you slept.
 201
 202  * returns a locked and held page upon success
 203  */
 204
 205 /* XXX I believe these pages should in fact NOT be locked */
 206
 207 static struct page *obdfs_add_entry (struct inode * dir,
 208                                      const char * name, int namelen,
 209                                      struct ext2_dir_entry_2 ** res_dir,
 210                                      int *err)
 211 {
 212         unsigned long offset;
 213         unsigned short rec_len;
 214         struct page *page;
 215         struct ext2_dir_entry_2 * de, * de1;
 216         struct super_block * sb;
 217
 218         ENTRY;
 219         *err = -EINVAL;
 220         *res_dir = NULL;
 221         if (!dir || !dir->i_nlink) {
 222                 EXIT;
 223                 return NULL;
 224         }
 225         sb = dir->i_sb;
 226
 227         if (!namelen) {
 228                 EXIT;
 229                 return NULL;
 230         }
 231         /*
 232          * Is this a busy deleted directory?  Can't create new files if so
 233          */
 234         if (dir->i_size == 0)
 235         {
 236                 EXIT;
 237                 *err = -ENOENT;
 238                 return NULL;
 239         }
 240         page = obdfs_getpage(dir, 0, 0, LOCKED);
 241         if (!page) {
 242                 EXIT;
 243                 return NULL;
 244         }
 245         rec_len = EXT2_DIR_REC_LEN(namelen);
 246         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 247         PDEBUG(page, "starting search");
 248         offset = 0;
 249         de = (struct ext2_dir_entry_2 *) page_address(page);
 250         *err = -ENOSPC;
 251         while (1) {
 252                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 253                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 254                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 255                         UnlockPage(page);
 256                         page_cache_release(page);
 257                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 258                         if (!page) {
 259                                 EXIT;
 260                                 return NULL;
 261                         }
 262                         PDEBUG(page, "new directory page");
 263                         if (dir->i_size <= offset) {
 264                                 if (dir->i_size == 0) {
 265                                         *err = -ENOENT;
 266                                         EXIT;
 267                                         return NULL;
 268                                 }
 269
 270                                 CDEBUG(D_INODE, "creating next block\n");
 271
 272                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 273                                 de->inode = 0;
 274                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 275                                 dir->i_size = offset + PAGE_SIZE;
 276                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 277                                 mark_inode_dirty(dir);
 278                         } else {
 279
 280                                 ext2_debug ("skipping to next block\n");
 281
 282                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 283                         }
 284                 }
 285                 CDEBUG(D_INODE, "\n");
 286                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 287                                            offset)) {
 288                         *err = -ENOENT;
 289                         UnlockPage(page);
 290                         page_cache_release(page);
 291                         EXIT;
 292                         return NULL;
 293                 }
 294                 CDEBUG(D_INODE, "\n");
 295                 if (ext2_match (namelen, name, de)) {
 296                                 *err = -EEXIST;
 297                                 UnlockPage(page);
 298                                 page_cache_release(page);
 299                                 EXIT;
 300                                 return NULL;
 301                 }
 302                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 303                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 304                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 305                         offset += le16_to_cpu(de->rec_len);
 306                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 307                         if (le32_to_cpu(de->inode)) {
 308                                 CDEBUG(D_INODE, "Inserting new in %p\n", de);
 309                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 310                                         EXT2_DIR_REC_LEN(de->name_len));
 311                                 CDEBUG(D_INODE, "-- de1 at %p\n", de1);
 312                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 313                                         EXT2_DIR_REC_LEN(de->name_len));
 314                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 315                                 de = de1;
 316                         }
 317                         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);
 318                         de->inode = 0;
 319                         de->name_len = namelen;
 320                         de->file_type = 0;
 321                         memcpy (de->name, name, namelen);
 322                         CDEBUG(D_INODE, "Copy done\n");
 323                         /*
 324                          * XXX shouldn't update any times until successful
 325                          * completion of syscall, but too many callers depend
 326                          * on this.
 327                          *
 328                          * XXX similarly, too many callers depend on
 329                          * ext2_new_inode() setting the times, but error
 330                          * recovery deletes the inode, so the worst that can
 331                          * happen is that the times are slightly out of date
 332                          * and/or different from the directory change time.
 333                          */
 334                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 335                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 336                         mark_inode_dirty(dir);
 337                         dir->i_version = ++event;
 338                         *res_dir = de;
 339                         *err = 0;
 340                         PDEBUG(page, "add_entry");
 341                         CDEBUG(D_INODE, "Regular exit from add_entry");
 342                         EXIT;
 343                         return page;
 344                 }
 345                 CDEBUG(D_INODE, "\n");
 346                 offset += le16_to_cpu(de->rec_len);
 347                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 348
 349         }
 350         CDEBUG(D_INODE, "\n");
 351
 352         UnlockPage(page);
 353         page_cache_release(page);
 354         PDEBUG(page, "add_entry");
 355         EXIT;
 356         return NULL;
 357 } /* obdfs_add_entry */
 358
 359 /*
 360  * obdfs_delete_entry deletes a directory entry by merging it with the
 361  * previous entry
 362  */
 363 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 364                               struct page * page)
 365 {
 366         struct ext2_dir_entry_2 * de, * pde;
 367         int i;
 368
 369         i = 0;
 370         pde = NULL;
 371         de = (struct ext2_dir_entry_2 *) page_address(page);
 372         while (i < PAGE_SIZE) {
 373                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 374                                            de, page, i))
 375                         return -EIO;
 376                 if (de == dir)  {
 377                         if (pde)
 378                                 pde->rec_len =
 379                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 380                                                     le16_to_cpu(dir->rec_len));
 381                         else
 382                                 dir->inode = 0;
 383                         return 0;
 384                 }
 385                 i += le16_to_cpu(de->rec_len);
 386                 pde = de;
 387                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 388         }
 389         return -ENOENT;
 390 } /* obdfs_delete_entry */
 391
 392
 393 static inline void ext2_set_de_type(struct super_block *sb,
 394                                 struct ext2_dir_entry_2 *de,
 395                                 umode_t mode) {
 396         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 397                 return;
 398         if (S_ISCHR(mode))
 399                 de->file_type = EXT2_FT_CHRDEV;
 400         else if (S_ISBLK(mode))
 401                 de->file_type = EXT2_FT_BLKDEV;
 402         else if (S_ISFIFO(mode))
 403                 de->file_type = EXT2_FT_FIFO;
 404         else if (S_ISLNK(mode))
 405                 de->file_type = EXT2_FT_SYMLINK;
 406         else if (S_ISREG(mode))
 407                 de->file_type = EXT2_FT_REG_FILE;
 408         else if (S_ISDIR(mode))
 409                 de->file_type = EXT2_FT_DIR;
 410 }
 411
 412
 413 /*
 414  * Display all dentries holding the specified inode.
 415  */
 416 #if 0
 417 static void show_dentry(struct list_head * dlist, int subdirs)
 418 {
 419         struct list_head *tmp = dlist;
 420
 421         while ((tmp = tmp->next) != dlist) {
 422                 struct dentry * dentry;
 423                 const char * unhashed = "";
 424
 425                 if ( subdirs )
 426                         dentry  = list_entry(tmp, struct dentry, d_child);
 427                 else
 428                         dentry  = list_entry(tmp, struct dentry, d_alias);
 429
 430                 if (list_empty(&dentry->d_hash))
 431                         unhashed = "(unhashed)";
 432
 433                 if ( dentry->d_inode )
 434                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 435                                dentry->d_parent->d_name.name,
 436                                dentry->d_name.name, dentry->d_count,
 437                                unhashed, dentry->d_inode->i_ino,
 438                                dentry->d_inode->i_dev,
 439                                dentry->d_inode->i_count);
 440                 else
 441                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 442                                dentry->d_parent->d_name.name,
 443                                dentry->d_name.name, dentry->d_count,
 444                                unhashed);
 445         }
 446 } /* show_dentry */
 447 #endif
 448
 449
 450 static struct inode *obdfs_new_inode(struct inode *dir, int mode)
 451 {
 452         struct obdo *oa;
 453         struct inode *inode;
 454         int err;
 455
 456         ENTRY;
 457         oa = obdo_alloc();
 458         if (!oa) {
 459                 EXIT;
 460                 return ERR_PTR(-ENOMEM);
 461         }
 462
 463         /* Send a hint to the create method on the type of file to create */
 464         oa->o_mode = mode;
 465         oa->o_valid |= OBD_MD_FLMODE;
 466
 467         err = IOPS(dir, create)(IID(dir), oa);
 468
 469         if ( err ) {
 470                 obdo_free(oa);
 471                 EXIT;
 472                 return ERR_PTR(err);
 473         }
 474
 475         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 476
 477         if (!inode) {
 478                 IOPS(dir, destroy)(IID(dir), oa);
 479                 obdo_free(oa);
 480                 EXIT;
 481                 return ERR_PTR(-EIO);
 482         }
 483
 484         if (!list_empty(&inode->i_dentry)) {
 485                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 486                 IOPS(dir, destroy)(IID(dir), oa);
 487                 obdo_free(oa);
 488                 iput(inode);
 489                 EXIT;
 490                 return ERR_PTR(-EIO);
 491         }
 492         obdo_free(oa);
 493         INIT_LIST_HEAD(&OBDFS_LIST(inode));
 494
 495         EXIT;
 496         return inode;
 497 } /* obdfs_new_inode */
 498
 499
 500 /*
 501  * By the time this is called, we already have created
 502  * the directory cache entry for the new file, but it
 503  * is so far negative - it has no inode.
 504  *
 505  * If the create succeeds, we fill in the inode information
 506  * with d_instantiate().
 507  */
 508 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 509 {
 510         struct inode * inode;
 511         struct page *page;
 512         struct ext2_dir_entry_2 * de;
 513         int err = -EIO;
 514
 515         ENTRY;
 516         inode = obdfs_new_inode(dir, mode);
 517         if ( IS_ERR(inode) ) {
 518                 EXIT;
 519                 return PTR_ERR(inode);
 520         }
 521
 522         inode->i_op = &obdfs_file_inode_operations;
 523         mark_inode_dirty(inode);
 524         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 525         if (!page) {
 526                 inode->i_nlink--;
 527                 mark_inode_dirty(inode);
 528                 iput (inode);
 529                 EXIT;
 530                 return err;
 531         }
 532         de->inode = cpu_to_le32(inode->i_ino);
 533         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 534         dir->i_version = ++event;
 535
 536         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 537         UnlockPage(page);
 538
 539         page_cache_release(page);
 540         d_instantiate(dentry, inode);
 541         EXIT;
 542         return err;
 543 } /* obdfs_create */
 544
 545 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 546 {
 547         struct inode * inode;
 548         struct page *page;
 549         struct ext2_dir_entry_2 * de;
 550         int err;
 551
 552         ENTRY;
 553         inode = obdfs_new_inode(dir, mode);
 554         if ( IS_ERR(inode) ) {
 555                 EXIT;
 556                 return PTR_ERR(inode);
 557         }
 558
 559         inode->i_uid = current->fsuid;
 560         init_special_inode(inode, mode, rdev);
 561         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 562         if (!page)
 563                 goto out_no_entry;
 564         de->inode = cpu_to_le32(inode->i_ino);
 565         dir->i_version = ++event;
 566         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 567         mark_inode_dirty(inode);
 568
 569         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 570         UnlockPage(page);
 571
 572         d_instantiate(dentry, inode);
 573         page_cache_release(page);
 574         err = 0;
 575 out:
 576         return err;
 577
 578 out_no_entry:
 579         inode->i_nlink--;
 580         mark_inode_dirty(inode);
 581         iput(inode);
 582         goto out;
 583 } /* obdfs_mknod */
 584
 585 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 586 {
 587         struct inode * inode;
 588         struct page *page, *inode_page;
 589         struct ext2_dir_entry_2 * de;
 590         int err;
 591
 592         ENTRY;
 593
 594         err = -EMLINK;
 595         if (dir->i_nlink >= EXT2_LINK_MAX)
 596                 goto out;
 597
 598         mode |= S_IFDIR;
 599         if (dir->i_mode & S_ISGID)
 600                 mode |= S_ISGID;
 601
 602         inode = obdfs_new_inode(dir, mode);
 603         if ( IS_ERR(inode) ) {
 604                 EXIT;
 605                 return PTR_ERR(inode);
 606         }
 607
 608         inode->i_op = &obdfs_dir_inode_operations;
 609         inode->i_blocks = 0;
 610         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 611         if (!inode_page) {
 612                 inode->i_nlink--; /* is this nlink == 0? */
 613                 mark_inode_dirty(inode);
 614                 iput (inode);
 615                 return err;
 616         }
 617
 618         /* create . and .. */
 619         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 620         de->inode = cpu_to_le32(inode->i_ino);
 621         de->name_len = 1;
 622         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 623         strcpy (de->name, ".");
 624         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 625
 626         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 627         de->inode = cpu_to_le32(dir->i_ino);
 628         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 629         de->name_len = 2;
 630         strcpy (de->name, "..");
 631         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 632
 633         /* XXX handle err */
 634         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 635         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 636         inode->i_size = PAGE_SIZE;
 637         UnlockPage(inode_page);
 638         page_cache_release(inode_page);
 639
 640         inode->i_nlink = 2;
 641         mark_inode_dirty(inode);
 642
 643         /* now deal with the parent */
 644         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 645         if (!page) {
 646                 goto out_no_entry;
 647         }
 648
 649         de->inode = cpu_to_le32(inode->i_ino);
 650         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 651         dir->i_version = ++event;
 652
 653         dir->i_nlink++;
 654         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 655         mark_inode_dirty(dir);
 656         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 657         /* XXX handle err? */
 658
 659         UnlockPage(page);
 660
 661         page_cache_release(page);
 662         d_instantiate(dentry, inode);
 663 out:
 664         EXIT;
 665         return err;
 666
 667 out_no_entry:
 668         inode->i_nlink = 0;
 669         mark_inode_dirty(inode);
 670         iput (inode);
 671         EXIT;
 672         goto out;
 673 } /* obdfs_mkdir */
 674
 675
 676 /*
 677  * routine to check that the specified directory is empty (for rmdir)
 678  */
 679 static int empty_dir (struct inode * inode)
 680 {
 681         unsigned long offset;
 682         struct page *page;
 683         struct ext2_dir_entry_2 * de, * de1;
 684         struct super_block * sb;
 685
 686         sb = inode->i_sb;
 687         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 688             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 689                 ext2_warning (inode->i_sb, "empty_dir",
 690                               "bad directory (dir #%lu) - no data block",
 691                               inode->i_ino);
 692                 return 1;
 693         }
 694         de = (struct ext2_dir_entry_2 *) page_address(page);
 695         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 696         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 697             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 698                 ext2_warning (inode->i_sb, "empty_dir",
 699                               "bad directory (dir #%lu) - no `.' or `..'",
 700                               inode->i_ino);
 701                 page_cache_release(page);
 702                 return 1;
 703         }
 704         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 705         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 706         while (offset < inode->i_size ) {
 707                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 708                         if (page) {
 709                                 UnlockPage(page);
 710                                 page_cache_release(page);
 711                         }
 712                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 713                         if (!page) {
 714 #if 0
 715                                 ext2_error (sb, "empty_dir",
 716                                             "directory #%lu contains a hole at offset %lu",
 717                                             inode->i_ino, offset);
 718 #endif
 719                                 offset += sb->s_blocksize;
 720                                 continue;
 721                         }
 722                         de = (struct ext2_dir_entry_2 *) page_address(page);
 723                 }
 724                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 725                                            offset)) {
 726                         UnlockPage(page);
 727                         page_cache_release(page);
 728                         return 1;
 729                 }
 730                 if (le32_to_cpu(de->inode)) {
 731                         UnlockPage(page);
 732                         page_cache_release(page);
 733                         return 0;
 734                 }
 735                 offset += le16_to_cpu(de->rec_len);
 736                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 737         }
 738         UnlockPage(page);
 739         page_cache_release(page);
 740         return 1;
 741 } /* empty_dir */
 742
 743 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 744 {
 745         int retval;
 746         struct inode * inode;
 747         struct page *page;
 748         struct ext2_dir_entry_2 * de;
 749         int err;
 750
 751         ENTRY;
 752
 753         retval = -ENOENT;
 754         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 755         if (!page)
 756                 goto end_rmdir;
 757
 758         inode = dentry->d_inode;
 759         DQUOT_INIT(inode);
 760
 761         retval = -EIO;
 762         if (le32_to_cpu(de->inode) != inode->i_ino)
 763                 goto end_rmdir;
 764
 765         retval = -ENOTEMPTY;
 766         if (!empty_dir (inode))
 767                 goto end_rmdir;
 768
 769         retval = obdfs_delete_entry (de, page);
 770         dir->i_version = ++event;
 771         if (retval)
 772                 goto end_rmdir;
 773         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 774         /* XXX handle err? */
 775         UnlockPage(page);
 776
 777         if (inode->i_nlink != 2)
 778                 ext2_warning (inode->i_sb, "ext2_rmdir",
 779                               "empty directory has nlink!=2 (%d)",
 780                               inode->i_nlink);
 781         inode->i_version = ++event;
 782         inode->i_nlink = 0;
 783         inode->i_size = 0;
 784         mark_inode_dirty(inode);
 785         dir->i_nlink--;
 786         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 787         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 788         mark_inode_dirty(dir);
 789         d_delete(dentry);
 790
 791 end_rmdir:
 792         if ( page )
 793                 page_cache_release(page);
 794         EXIT;
 795         return retval;
 796 } /* obdfs_rmdir */
 797
 798 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 799 {
 800         int retval;
 801         struct inode * inode;
 802         struct page *page;
 803         struct ext2_dir_entry_2 * de;
 804         int err;
 805
 806         ENTRY;
 807
 808         retval = -ENOENT;
 809         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 810         if (!page)
 811                 goto end_unlink;
 812
 813         inode = dentry->d_inode;
 814         DQUOT_INIT(inode);
 815
 816         retval = -EIO;
 817         if (le32_to_cpu(de->inode) != inode->i_ino)
 818                 goto end_unlink;
 819
 820         if (!inode->i_nlink) {
 821                 ext2_warning (inode->i_sb, "ext2_unlink",
 822                               "Deleting nonexistent file (%lu), %d",
 823                               inode->i_ino, inode->i_nlink);
 824                 inode->i_nlink = 1;
 825         }
 826         retval = obdfs_delete_entry (de, page);
 827         if (retval)
 828                 goto end_unlink;
 829         dir->i_version = ++event;
 830         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 831         /* XXX handle err? */
 832         UnlockPage(page);
 833
 834         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 835         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 836         mark_inode_dirty(dir);
 837         inode->i_nlink--;
 838         mark_inode_dirty(inode);
 839         inode->i_ctime = dir->i_ctime;
 840         retval = 0;
 841         d_delete(dentry);       /* This also frees the inode */
 842
 843 end_unlink:
 844         if (page)
 845                 page_cache_release(page);
 846         EXIT;
 847         return retval;
 848 } /* obdfs_unlink */
 849
 850 int obdfs_symlink (struct inode * dir, struct dentry *dentry,
 851                    const char * symname)
 852 {
 853         struct ext2_dir_entry_2 * de;
 854         struct inode * inode;
 855         struct obdfs_inode_info *oinfo;
 856         struct page* page = NULL, * name_page = NULL;
 857         char * link;
 858         int i, l, err = -EIO;
 859         char c;
 860
 861         ENTRY;
 862         inode = obdfs_new_inode(dir, S_IFLNK | S_IRWXUGO);
 863         if ( IS_ERR(inode) ) {
 864                 EXIT;
 865                 return PTR_ERR(inode);
 866         }
 867
 868         inode->i_op = &obdfs_symlink_inode_operations;
 869         for (l = 0; l < inode->i_sb->s_blocksize - 1 && symname [l]; l++)
 870                 ;
 871
 872         oinfo = OBDFS_INFO(inode);
 873
 874         if (l >= sizeof(oinfo->oi_inline)) {
 875                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 876
 877                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 878                 if (!name_page) {
 879                         inode->i_nlink--;
 880                         mark_inode_dirty(inode);
 881                         iput (inode);
 882                         EXIT;
 883                         return err;
 884                 }
 885                 link = (char *)page_address(name_page);
 886         } else {
 887                 link = oinfo->oi_inline;
 888                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 889
 890                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 891         }
 892         i = 0;
 893         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 894                 link[i++] = c;
 895         link[i] = 0;
 896         if (name_page) {
 897                 err = obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 898                 /* XXX handle err */
 899                 PDEBUG(name_page, "symlink");
 900                 UnlockPage(name_page);
 901                 page_cache_release(name_page);
 902         }
 903         inode->i_size = i;
 904         mark_inode_dirty(inode);
 905
 906         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 907                                 &de, &err);
 908         if (!page)
 909                 goto out_no_entry;
 910         de->inode = cpu_to_le32(inode->i_ino);
 911         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 912         dir->i_version = ++event;
 913         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 914         UnlockPage(page);
 915
 916         d_instantiate(dentry, inode);
 917 out:
 918         EXIT;
 919         return err;
 920
 921 out_no_entry:
 922         inode->i_nlink--;
 923         mark_inode_dirty(inode);
 924         iput (inode);
 925         goto out;
 926 } /* obdfs_symlink */
 927
 928 int obdfs_link (struct dentry * old_dentry,
 929                 struct inode * dir, struct dentry *dentry)
 930 {
 931         struct inode *inode = old_dentry->d_inode;
 932         struct ext2_dir_entry_2 * de;
 933         struct page *page;
 934         int err;
 935
 936         ENTRY;
 937
 938         if (S_ISDIR(inode->i_mode))
 939                 return -EPERM;
 940
 941         if (inode->i_nlink >= EXT2_LINK_MAX)
 942                 return -EMLINK;
 943
 944         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 945         if (!page)
 946                 return err;
 947
 948         de->inode = cpu_to_le32(inode->i_ino);
 949         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 950         dir->i_version = ++event;
 951
 952         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 953         UnlockPage(page);
 954
 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 err;
 962 } /* obdfs_link */
 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 err;
 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         err = -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                         err = -ENOTEMPTY;
1016                         if (!empty_dir (new_inode))
1017                                 goto end_rename;
1018                 }
1019                 err = -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                 err = -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                                         &err);
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                 /* XXX handle err */
1065                 err = obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1066                 old_dir->i_nlink--;
1067                 mark_inode_dirty(old_dir);
1068                 if (new_inode) {
1069                         new_inode->i_nlink--;
1070                         mark_inode_dirty(new_inode);
1071                 } else {
1072                         new_dir->i_nlink++;
1073                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1074                         mark_inode_dirty(new_dir);
1075                 }
1076         }
1077         if ( old_page != new_page ) {
1078                 unsigned long index = old_page->index;
1079                 /* lock the old_page and release unlocked copy */
1080                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1081                 page_cache_release(old_page);
1082                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1083                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1084                 /* XXX handle err */
1085                 err = obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1086         }
1087
1088         err = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1089
1090 end_rename:
1091         if (old_page && PageLocked(old_page) )
1092                 UnlockPage(old_page);
1093         if (old_page)
1094                 page_cache_release(old_page);
1095         if (new_page && PageLocked(new_page) )
1096                 UnlockPage(new_page);
1097         if (new_page)
1098                 page_cache_release(new_page);
1099         if (dir_page && PageLocked(dir_page) )
1100                 UnlockPage(dir_page);
1101         if (dir_page)
1102                 page_cache_release(dir_page);
1103
1104
1105         return err;
1106 } /* obdfs_rename */