lustre/obdfs/namei.c

   1 /*
   2  *  linux/fs/obdfs/namei.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/ext2/namei.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  *  Directory entry file type support and forward compatibility hooks
  18  *      for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
  19  *
  20  *  Changes for use in OBDFS
  21  *  Copyright (c) 1999, Seagate Technology Inc.
  22  *
  23  */
  24
  25 #include <asm/uaccess.h>
  26
  27 #include <linux/errno.h>
  28 #include <linux/fs.h>
  29 #include <linux/ext2_fs.h>
  30 #include <linux/fcntl.h>
  31 #include <linux/sched.h>
  32 #include <linux/stat.h>
  33 #include <linux/string.h>
  34 #include <linux/locks.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/iobuf.h>
  37
  38 #include <linux/obd_support.h>
  39 #include <linux/obdfs.h>
  40
  41 /*
  42  * define how far ahead to read directories while searching them.
  43  */
  44 #define NAMEI_RA_CHUNKS  2
  45 #define NAMEI_RA_BLOCKS  4
  46 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  47 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  48
  49 /*
  50  * NOTE! unlike strncmp, ext2_match returns 1 for success, 0 for failure.
  51  *
  52  * `len <= EXT2_NAME_LEN' is guaranteed by caller.
  53  * `de != NULL' is guaranteed by caller.
  54  */
  55 static inline int ext2_match (int len, const char * const name,
  56                        struct ext2_dir_entry_2 * de)
  57 {
  58         if (len != de->name_len)
  59                 return 0;
  60         if (!de->inode)
  61                 return 0;
  62         return !memcmp(name, de->name, len);
  63 }
  64
  65 /*
  66  *      obdfs_find_entry()
  67  *
  68  * finds an entry in the specified directory with the wanted name. It
  69  * returns the cache buffer in which the entry was found, and the entry
  70  * itself (as a parameter - res_dir).  It does NOT read the inode of the
  71  * entry - you'll have to do that yourself if you want to.
  72  */
  73 static struct page * obdfs_find_entry (struct inode * dir,
  74                                        const char * const name, int namelen,
  75                                        struct ext2_dir_entry_2 ** res_dir,
  76                                        int lock)
  77 {
  78         struct super_block * sb;
  79         unsigned long offset;
  80         struct page * page;
  81         ENTRY;
  82         CDEBUG(D_INODE, "find entry for %*s\n", namelen, name);
  83
  84         *res_dir = NULL;
  85         sb = dir->i_sb;
  86
  87         if (namelen > EXT2_NAME_LEN)
  88                 return NULL;
  89
  90         CDEBUG(D_INODE, "dirsize is %Ld\n", dir->i_size);
  91
  92         page = 0;
  93         offset = 0;
  94         while ( offset < dir->i_size ) {
  95                 struct ext2_dir_entry_2 * de;
  96                 char * dlimit;
  97
  98                 page = obdfs_getpage(dir, offset, 0, lock);
  99
 100                 if ( !page ) {
 101                         CDEBUG(D_INODE, "No page, offset %lx\n", offset);
 102                         return NULL;
 103                 }
 104
 105                 de = (struct ext2_dir_entry_2 *) page_address(page);
 106                 dlimit = (char *)page_address(page) + PAGE_SIZE;
 107                 while ((char *) de < dlimit) {
 108                         /* this code is executed quadratically often */
 109                         /* do minimal checking `by hand' */
 110                         int de_len;
 111                         /* CDEBUG(D_INODE, "Entry %p len %d, page at %#lx - %#lx , offset %lx\n",
 112                                de, le16_to_cpu(de->rec_len), page_address(page),
 113                                page_address(page) + PAGE_SIZE, offset); */
 114
 115                         if ((char *) de + namelen <= dlimit &&
 116                             ext2_match (namelen, name, de)) {
 117                                 /* found a match -
 118                                    just to be sure, do a full check */
 119                                 if (!obdfs_check_dir_entry("ext2_find_entry",
 120                                                           dir, de, page, offset))
 121                                         goto failure;
 122                                 *res_dir = de;
 123                                 EXIT;
 124                                 return page;
 125                         }
 126                         /* prevent looping on a bad block */
 127                         de_len = le16_to_cpu(de->rec_len);
 128                         if (de_len <= 0) {
 129                                 printk("Bad entry at %p len %d\n", de, de_len);
 130                                 goto failure;
 131                         }
 132                         offset += de_len;
 133                         de = (struct ext2_dir_entry_2 *)
 134                                 ((char *) de + de_len);
 135                         /* CDEBUG(D_INODE, "Next while %lx\n", offset); */
 136                 }
 137                 if ( lock )
 138                         UnlockPage(page);
 139                 page_cache_release(page);
 140                 page = NULL;
 141                 CDEBUG(D_INODE, "Next for %lx\n", offset);
 142         }
 143
 144 failure:
 145         CDEBUG(D_INODE, "Negative case, page %p, offset %ld\n", page, offset);
 146         if (page) {
 147                 if (lock)
 148                         UnlockPage(page);
 149                 page_cache_release(page);
 150         }
 151         EXIT;
 152         return NULL;
 153 } /* obdfs_find_entry */
 154
 155 struct dentry *obdfs_lookup(struct inode * dir, struct dentry *dentry)
 156 {
 157         struct inode * inode;
 158         struct ext2_dir_entry_2 * de;
 159         struct page * page;
 160         ENTRY;
 161
 162         if (dentry->d_name.len > EXT2_NAME_LEN)
 163                 return ERR_PTR(-ENAMETOOLONG);
 164
 165         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 166         inode = NULL;
 167         if ( !page )
 168                 CDEBUG(D_INODE, "No page - negative entry.\n");
 169         if ( page && !de ) {
 170                 CDEBUG(D_INODE, "Danger: PAGE but de.\n");
 171                 return ERR_PTR(-ENOENT);
 172         }
 173         if (page) {
 174                 unsigned long ino = le32_to_cpu(de->inode);
 175                 UnlockPage(page);
 176                 page_cache_release(page);
 177                 inode = iget(dir->i_sb, ino);
 178
 179                 if (!inode) {
 180                         CDEBUG(D_INODE, "No inode.\n");
 181                         EXIT;
 182                         return ERR_PTR(-EACCES);
 183                 }
 184         }
 185         d_add(dentry, inode);
 186         EXIT;
 187         return NULL;
 188 } /* obdfs_lookup */
 189
 190
 191 /*
 192  *      obdfs_add_entry()
 193  *
 194  * adds a file entry to the specified directory, using the same
 195  * semantics as ext2_find_entry(). It returns NULL if it failed.
 196  *
 197  * NOTE!! The inode part of 'de' is left at 0 - which means you
 198  * may not sleep between calling this and putting something into
 199  * the entry, as someone else might have used it while you slept.
 200
 201  * returns a locked and held page upon success
 202  */
 203
 204 /* XXX I believe these pages should in fact NOT be locked */
 205
 206 static struct page *obdfs_add_entry (struct inode * dir,
 207                                      const char * name, int namelen,
 208                                      struct ext2_dir_entry_2 ** res_dir,
 209                                      int *err)
 210 {
 211         unsigned long offset;
 212         unsigned short rec_len;
 213         struct page *page;
 214         struct ext2_dir_entry_2 * de, * de1;
 215         struct super_block * sb;
 216
 217         ENTRY;
 218         *err = -EINVAL;
 219         *res_dir = NULL;
 220         if (!dir || !dir->i_nlink) {
 221                 EXIT;
 222                 return NULL;
 223         }
 224         sb = dir->i_sb;
 225
 226         if (!namelen) {
 227                 EXIT;
 228                 return NULL;
 229         }
 230         /*
 231          * Is this a busy deleted directory?  Can't create new files if so
 232          */
 233         if (dir->i_size == 0)
 234         {
 235                 EXIT;
 236                 *err = -ENOENT;
 237                 return NULL;
 238         }
 239         page = obdfs_getpage(dir, 0, 0, LOCKED);
 240         if (!page) {
 241                 EXIT;
 242                 return NULL;
 243         }
 244         rec_len = EXT2_DIR_REC_LEN(namelen);
 245         CDEBUG(D_INODE, "reclen: %d\n", rec_len);
 246         PDEBUG(page, "starting search");
 247         offset = 0;
 248         de = (struct ext2_dir_entry_2 *) page_address(page);
 249         *err = -ENOSPC;
 250         while (1) {
 251                 CDEBUG(D_INODE, "Considering entry at %p, (page at %#lx - %#lx), offset %ld\n",
 252                        de, page_address(page), page_address(page) + PAGE_SIZE, offset);
 253                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 254                         UnlockPage(page);
 255                         page_cache_release(page);
 256                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 257                         if (!page) {
 258                                 EXIT;
 259                                 return NULL;
 260                         }
 261                         PDEBUG(page, "new directory page");
 262                         if (dir->i_size <= offset) {
 263                                 if (dir->i_size == 0) {
 264                                         *err = -ENOENT;
 265                                         EXIT;
 266                                         return NULL;
 267                                 }
 268
 269                                 CDEBUG(D_INODE, "creating next block\n");
 270
 271                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 272                                 de->inode = 0;
 273                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 274                                 dir->i_size = offset + PAGE_SIZE;
 275                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 276                                 mark_inode_dirty(dir);
 277                         } else {
 278
 279                                 ext2_debug ("skipping to next block\n");
 280
 281                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 282                         }
 283                 }
 284                 CDEBUG(D_INODE, "\n");
 285                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 286                                            offset)) {
 287                         *err = -ENOENT;
 288                         UnlockPage(page);
 289                         page_cache_release(page);
 290                         EXIT;
 291                         return NULL;
 292                 }
 293                 CDEBUG(D_INODE, "\n");
 294                 if (ext2_match (namelen, name, de)) {
 295                                 *err = -EEXIST;
 296                                 UnlockPage(page);
 297                                 page_cache_release(page);
 298                                 EXIT;
 299                                 return NULL;
 300                 }
 301                 CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);
 302                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 303                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 304                         offset += le16_to_cpu(de->rec_len);
 305                         CDEBUG(D_INODE, "Found enough space de %p, offset %#lx\n", de, offset);
 306                         if (le32_to_cpu(de->inode)) {
 307                                 CDEBUG(D_INODE, "Inserting new in %p\n", de);
 308                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 309                                         EXT2_DIR_REC_LEN(de->name_len));
 310                                 CDEBUG(D_INODE, "-- de1 at %p\n", de1);
 311                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 312                                         EXT2_DIR_REC_LEN(de->name_len));
 313                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 314                                 de = de1;
 315                         }
 316                         CDEBUG(D_INODE, "Reclen adjusted; copy %d bytes to %p, page at %#lx EOP at %#lx\n", namelen, de->name, page_address(page), page_address(page) + PAGE_SIZE);
 317                         de->inode = 0;
 318                         de->name_len = namelen;
 319                         de->file_type = 0;
 320                         memcpy (de->name, name, namelen);
 321                         CDEBUG(D_INODE, "Copy done\n");
 322                         /*
 323                          * XXX shouldn't update any times until successful
 324                          * completion of syscall, but too many callers depend
 325                          * on this.
 326                          *
 327                          * XXX similarly, too many callers depend on
 328                          * ext2_new_inode() setting the times, but error
 329                          * recovery deletes the inode, so the worst that can
 330                          * happen is that the times are slightly out of date
 331                          * and/or different from the directory change time.
 332                          */
 333                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 334                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 335                         mark_inode_dirty(dir);
 336                         dir->i_version = ++event;
 337                         *res_dir = de;
 338                         *err = 0;
 339                         PDEBUG(page, "addentry");
 340                         CDEBUG(D_INODE, "Regular exit from add_entry");
 341                         EXIT;
 342                         return page;
 343                 }
 344                 CDEBUG(D_INODE, "\n");
 345                 offset += le16_to_cpu(de->rec_len);
 346                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 347
 348         }
 349         CDEBUG(D_INODE, "\n");
 350
 351         UnlockPage(page);
 352         page_cache_release(page);
 353         PDEBUG(page, "addentry");
 354         EXIT;
 355         return NULL;
 356 } /* obdfs_add_entry */
 357
 358 /*
 359  * obdfs_delete_entry deletes a directory entry by merging it with the
 360  * previous entry
 361  */
 362 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 363                               struct page * page)
 364 {
 365         struct ext2_dir_entry_2 * de, * pde;
 366         int i;
 367
 368         i = 0;
 369         pde = NULL;
 370         de = (struct ext2_dir_entry_2 *) page_address(page);
 371         while (i < PAGE_SIZE) {
 372                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 373                                            de, page, i))
 374                         return -EIO;
 375                 if (de == dir)  {
 376                         if (pde)
 377                                 pde->rec_len =
 378                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 379                                                     le16_to_cpu(dir->rec_len));
 380                         else
 381                                 dir->inode = 0;
 382                         return 0;
 383                 }
 384                 i += le16_to_cpu(de->rec_len);
 385                 pde = de;
 386                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 387         }
 388         return -ENOENT;
 389 } /* obdfs_delete_entry */
 390
 391
 392 static inline void ext2_set_de_type(struct super_block *sb,
 393                                 struct ext2_dir_entry_2 *de,
 394                                 umode_t mode) {
 395         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 396                 return;
 397         if (S_ISCHR(mode))
 398                 de->file_type = EXT2_FT_CHRDEV;
 399         else if (S_ISBLK(mode))
 400                 de->file_type = EXT2_FT_BLKDEV;
 401         else if (S_ISFIFO(mode))
 402                 de->file_type = EXT2_FT_FIFO;
 403         else if (S_ISLNK(mode))
 404                 de->file_type = EXT2_FT_SYMLINK;
 405         else if (S_ISREG(mode))
 406                 de->file_type = EXT2_FT_REG_FILE;
 407         else if (S_ISDIR(mode))
 408                 de->file_type = EXT2_FT_DIR;
 409 }
 410
 411
 412 /*
 413  * Display all dentries holding the specified inode.
 414  */
 415 #if 0
 416 static void show_dentry(struct list_head * dlist, int subdirs)
 417 {
 418         struct list_head *tmp = dlist;
 419
 420         while ((tmp = tmp->next) != dlist) {
 421                 struct dentry * dentry;
 422                 const char * unhashed = "";
 423
 424                 if ( subdirs )
 425                         dentry  = list_entry(tmp, struct dentry, d_child);
 426                 else
 427                         dentry  = list_entry(tmp, struct dentry, d_alias);
 428
 429                 if (list_empty(&dentry->d_hash))
 430                         unhashed = "(unhashed)";
 431
 432                 if ( dentry->d_inode )
 433                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 434                                dentry->d_parent->d_name.name,
 435                                dentry->d_name.name, dentry->d_count,
 436                                unhashed, dentry->d_inode->i_ino,
 437                                dentry->d_inode->i_dev,
 438                                dentry->d_inode->i_count);
 439                 else
 440                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 441                                dentry->d_parent->d_name.name,
 442                                dentry->d_name.name, dentry->d_count,
 443                                unhashed);
 444         }
 445 } /* show_dentry */
 446 #endif
 447
 448
 449 struct inode *obdfs_new_inode(struct inode *dir)
 450 {
 451         struct obdo *oa;
 452         struct inode *inode;
 453         int err;
 454
 455         ENTRY;
 456         oa = obdo_alloc();
 457         if (!oa) {
 458                 EXIT;
 459                 return ERR_PTR(-ENOMEM);
 460         }
 461
 462         err = IOPS(dir, create)(IID(dir), oa);
 463
 464         if ( err ) {
 465                 obdo_free(oa);
 466                 EXIT;
 467                 return ERR_PTR(err);
 468         }
 469
 470         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 471
 472         if (!inode) {
 473                 IOPS(dir, destroy)(IID(dir), oa);
 474                 obdo_free(oa);
 475                 EXIT;
 476                 return ERR_PTR(-EIO);
 477         }
 478
 479         if (!list_empty(&inode->i_dentry)) {
 480                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 481                 IOPS(dir, destroy)(IID(dir), oa);
 482                 obdo_free(oa);
 483                 iput(inode);
 484                 EXIT;
 485                 return ERR_PTR(-EIO);
 486         }
 487
 488         INIT_LIST_HEAD(&OBD_LIST(inode));
 489
 490         EXIT;
 491         return inode;
 492 } /* obdfs_new_inode */
 493
 494
 495 /*
 496  * By the time this is called, we already have created
 497  * the directory cache entry for the new file, but it
 498  * is so far negative - it has no inode.
 499  *
 500  * If the create succeeds, we fill in the inode information
 501  * with d_instantiate().
 502  */
 503 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 504 {
 505         struct inode * inode;
 506         struct page *page;
 507         struct ext2_dir_entry_2 * de;
 508         int err = -EIO;
 509
 510         ENTRY;
 511         inode = obdfs_new_inode(dir);
 512         if ( IS_ERR(inode) ) {
 513                 EXIT;
 514                 return PTR_ERR(inode);
 515         }
 516
 517         inode->i_op = &obdfs_file_inode_operations;
 518         inode->i_mode = mode;
 519         mark_inode_dirty(inode);
 520         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 521         if (!page) {
 522                 inode->i_nlink--;
 523                 mark_inode_dirty(inode);
 524                 iput (inode);
 525                 EXIT;
 526                 return err;
 527         }
 528         de->inode = cpu_to_le32(inode->i_ino);
 529         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 530         dir->i_version = ++event;
 531
 532         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 533         UnlockPage(page);
 534
 535         page_cache_release(page);
 536         d_instantiate(dentry, inode);
 537         EXIT;
 538         return err;
 539 } /* obdfs_create */
 540
 541 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 542 {
 543         struct inode * inode;
 544         struct page *page;
 545         struct ext2_dir_entry_2 * de;
 546         int err;
 547
 548         ENTRY;
 549
 550         inode = obdfs_new_inode(dir);
 551         if ( IS_ERR(inode) ) {
 552                 EXIT;
 553                 return PTR_ERR(inode);
 554         }
 555
 556         inode->i_uid = current->fsuid;
 557         init_special_inode(inode, mode, rdev);
 558         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 559         if (!page)
 560                 goto out_no_entry;
 561         de->inode = cpu_to_le32(inode->i_ino);
 562         dir->i_version = ++event;
 563         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 564         mark_inode_dirty(inode);
 565
 566         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 567         UnlockPage(page);
 568
 569         d_instantiate(dentry, inode);
 570         page_cache_release(page);
 571         err = 0;
 572 out:
 573         return err;
 574
 575 out_no_entry:
 576         inode->i_nlink--;
 577         mark_inode_dirty(inode);
 578         iput(inode);
 579         goto out;
 580 } /* obdfs_mknod */
 581
 582 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 583 {
 584         struct inode * inode;
 585         struct page *page, *inode_page;
 586         struct ext2_dir_entry_2 * de;
 587         int err;
 588
 589         ENTRY;
 590
 591         err = -EMLINK;
 592         if (dir->i_nlink >= EXT2_LINK_MAX)
 593                 goto out;
 594
 595         inode = obdfs_new_inode(dir);
 596         if ( IS_ERR(inode) ) {
 597                 EXIT;
 598                 return PTR_ERR(inode);
 599         }
 600
 601         inode->i_op = &obdfs_dir_inode_operations;
 602         inode->i_blocks = 0;
 603         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 604         if (!inode_page) {
 605                 inode->i_nlink--; /* is this nlink == 0? */
 606                 mark_inode_dirty(inode);
 607                 iput (inode);
 608                 return err;
 609         }
 610
 611         /* create . and .. */
 612         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 613         de->inode = cpu_to_le32(inode->i_ino);
 614         de->name_len = 1;
 615         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 616         strcpy (de->name, ".");
 617         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 618
 619         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 620         de->inode = cpu_to_le32(dir->i_ino);
 621         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 622         de->name_len = 2;
 623         strcpy (de->name, "..");
 624         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 625
 626         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 627         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 628         inode->i_size = PAGE_SIZE;
 629         UnlockPage(inode_page);
 630         page_cache_release(inode_page);
 631         /* XXX handle err */
 632
 633         inode->i_nlink = 2;
 634         inode->i_mode = S_IFDIR | mode;
 635         if (dir->i_mode & S_ISGID)
 636                 inode->i_mode |= S_ISGID;
 637         mark_inode_dirty(inode);
 638
 639         /* now deal with the parent */
 640         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 641         if (!page) {
 642                 goto out_no_entry;
 643         }
 644
 645         de->inode = cpu_to_le32(inode->i_ino);
 646         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 647         dir->i_version = ++event;
 648
 649         dir->i_nlink++;
 650         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 651         mark_inode_dirty(dir);
 652         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 653
 654         UnlockPage(page);
 655
 656         page_cache_release(page);
 657         d_instantiate(dentry, inode);
 658         err = 0;
 659 out:
 660         EXIT;
 661         return err;
 662
 663 out_no_entry:
 664         inode->i_nlink = 0;
 665         mark_inode_dirty(inode);
 666         iput (inode);
 667         EXIT;
 668         goto out;
 669 } /* obdfs_mkdir */
 670
 671
 672 /*
 673  * routine to check that the specified directory is empty (for rmdir)
 674  */
 675 static int empty_dir (struct inode * inode)
 676 {
 677         unsigned long offset;
 678         struct page *page;
 679         struct ext2_dir_entry_2 * de, * de1;
 680         struct super_block * sb;
 681
 682         sb = inode->i_sb;
 683         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 684             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 685                 ext2_warning (inode->i_sb, "empty_dir",
 686                               "bad directory (dir #%lu) - no data block",
 687                               inode->i_ino);
 688                 return 1;
 689         }
 690         de = (struct ext2_dir_entry_2 *) page_address(page);
 691         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 692         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 693             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 694                 ext2_warning (inode->i_sb, "empty_dir",
 695                               "bad directory (dir #%lu) - no `.' or `..'",
 696                               inode->i_ino);
 697                 page_cache_release(page);
 698                 return 1;
 699         }
 700         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 701         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 702         while (offset < inode->i_size ) {
 703                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 704                         if (page) {
 705                                 UnlockPage(page);
 706                                 page_cache_release(page);
 707                         }
 708                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 709                         if (!page) {
 710 #if 0
 711                                 ext2_error (sb, "empty_dir",
 712                                             "directory #%lu contains a hole at offset %lu",
 713                                             inode->i_ino, offset);
 714 #endif
 715                                 offset += sb->s_blocksize;
 716                                 continue;
 717                         }
 718                         de = (struct ext2_dir_entry_2 *) page_address(page);
 719                 }
 720                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 721                                            offset)) {
 722                         UnlockPage(page);
 723                         page_cache_release(page);
 724                         return 1;
 725                 }
 726                 if (le32_to_cpu(de->inode)) {
 727                         UnlockPage(page);
 728                         page_cache_release(page);
 729                         return 0;
 730                 }
 731                 offset += le16_to_cpu(de->rec_len);
 732                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 733         }
 734         UnlockPage(page);
 735         page_cache_release(page);
 736         return 1;
 737 } /* empty_dir */
 738
 739 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 740 {
 741         int retval;
 742         struct inode * inode;
 743         struct page *page;
 744         struct ext2_dir_entry_2 * de;
 745         int err;
 746
 747         ENTRY;
 748
 749         retval = -ENOENT;
 750         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 751         if (!page)
 752                 goto end_rmdir;
 753
 754         inode = dentry->d_inode;
 755         DQUOT_INIT(inode);
 756
 757         retval = -EIO;
 758         if (le32_to_cpu(de->inode) != inode->i_ino)
 759                 goto end_rmdir;
 760
 761         retval = -ENOTEMPTY;
 762         if (!empty_dir (inode))
 763                 goto end_rmdir;
 764
 765         retval = obdfs_delete_entry (de, page);
 766         dir->i_version = ++event;
 767         if (retval)
 768                 goto end_rmdir;
 769         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 770         UnlockPage(page);
 771
 772         if (inode->i_nlink != 2)
 773                 ext2_warning (inode->i_sb, "ext2_rmdir",
 774                               "empty directory has nlink!=2 (%d)",
 775                               inode->i_nlink);
 776         inode->i_version = ++event;
 777         inode->i_nlink = 0;
 778         inode->i_size = 0;
 779         mark_inode_dirty(inode);
 780         dir->i_nlink--;
 781         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 782         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 783         mark_inode_dirty(dir);
 784         d_delete(dentry);
 785
 786 end_rmdir:
 787         if ( page )
 788                 page_cache_release(page);
 789         EXIT;
 790         return retval;
 791 } /* obdfs_rmdir */
 792
 793 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 794 {
 795         int retval;
 796         struct inode * inode;
 797         struct page *page;
 798         struct ext2_dir_entry_2 * de;
 799         int err;
 800
 801         ENTRY;
 802
 803         retval = -ENOENT;
 804         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 805         if (!page)
 806                 goto end_unlink;
 807
 808         inode = dentry->d_inode;
 809         DQUOT_INIT(inode);
 810
 811         retval = -EIO;
 812         if (le32_to_cpu(de->inode) != inode->i_ino)
 813                 goto end_unlink;
 814
 815         if (!inode->i_nlink) {
 816                 ext2_warning (inode->i_sb, "ext2_unlink",
 817                               "Deleting nonexistent file (%lu), %d",
 818                               inode->i_ino, inode->i_nlink);
 819                 inode->i_nlink = 1;
 820         }
 821         retval = obdfs_delete_entry (de, page);
 822         if (retval)
 823                 goto end_unlink;
 824         dir->i_version = ++event;
 825         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 826         UnlockPage(page);
 827
 828         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 829         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 830         mark_inode_dirty(dir);
 831         inode->i_nlink--;
 832         mark_inode_dirty(inode);
 833         inode->i_ctime = dir->i_ctime;
 834         retval = 0;
 835         d_delete(dentry);       /* This also frees the inode */
 836
 837 end_unlink:
 838         if (page)
 839                 page_cache_release(page);
 840         EXIT;
 841         return retval;
 842 } /* obdfs_unlink */
 843
 844 int obdfs_symlink (struct inode * dir, struct dentry *dentry, const char * symname)
 845 {
 846         struct ext2_dir_entry_2 * de;
 847         struct inode * inode;
 848         struct obdfs_inode_info *oinfo;
 849         struct page* page = NULL, * name_page = NULL;
 850         char * link;
 851         int i, l, err = -EIO;
 852         char c;
 853
 854         ENTRY;
 855         inode = obdfs_new_inode(dir);
 856         oinfo = OBD_INFO(inode);
 857         if ( IS_ERR(inode) ) {
 858                 EXIT;
 859                 return PTR_ERR(inode);
 860         }
 861
 862         inode->i_mode = S_IFLNK | S_IRWXUGO;
 863         inode->i_op = &obdfs_symlink_inode_operations;
 864         for (l = 0; l < inode->i_sb->s_blocksize - 1 &&
 865              symname [l]; l++)
 866                 ;
 867
 868         if (l >= sizeof (oinfo->oi_inline)) {
 869                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 870
 871                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 872                 if (!name_page) {
 873                         inode->i_nlink--;
 874                         mark_inode_dirty(inode);
 875                         iput (inode);
 876                         EXIT;
 877                         return err;
 878                 }
 879                 link = (char *)page_address(name_page);
 880         } else {
 881                 link = oinfo->oi_inline;
 882                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 883
 884                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 885
 886         }
 887         i = 0;
 888         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 889                 link[i++] = c;
 890         link[i] = 0;
 891         if (name_page) {
 892                 obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 893                 PDEBUG(name_page, "symlink");
 894                 UnlockPage(name_page);
 895                 page_cache_release(name_page);
 896         }
 897         inode->i_size = i;
 898         mark_inode_dirty(inode);
 899
 900         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 901         if (!page)
 902                 goto out_no_entry;
 903         de->inode = cpu_to_le32(inode->i_ino);
 904         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 905         dir->i_version = ++event;
 906         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 907         UnlockPage(page);
 908
 909         d_instantiate(dentry, inode);
 910 out:
 911         EXIT;
 912         return err;
 913
 914 out_no_entry:
 915         inode->i_nlink--;
 916         mark_inode_dirty(inode);
 917         iput (inode);
 918         goto out;
 919 } /* obdfs_symlink */
 920
 921 int obdfs_link (struct dentry * old_dentry,
 922                 struct inode * dir, struct dentry *dentry)
 923 {
 924         struct inode *inode = old_dentry->d_inode;
 925         struct ext2_dir_entry_2 * de;
 926         struct page *page;
 927         int err;
 928
 929         ENTRY;
 930
 931         if (S_ISDIR(inode->i_mode))
 932                 return -EPERM;
 933
 934         if (inode->i_nlink >= EXT2_LINK_MAX)
 935                 return -EMLINK;
 936
 937         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 938         if (!page)
 939                 return err;
 940
 941         de->inode = cpu_to_le32(inode->i_ino);
 942         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 943         dir->i_version = ++event;
 944
 945         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 946         UnlockPage(page);
 947
 948         page_cache_release(page);
 949         inode->i_nlink++;
 950         inode->i_ctime = CURRENT_TIME;
 951         mark_inode_dirty(inode);
 952         inode->i_count++;
 953         d_instantiate(dentry, inode);
 954         return err;
 955 } /* obdfs_link */
 956
 957 #define PARENT_INO(buffer) \
 958         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 959         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 960
 961 /*
 962  * Anybody can rename anything with this: the permission checks are left to the
 963  * higher-level routines.
 964  */
 965 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 966                            struct inode * new_dir, struct dentry *new_dentry)
 967 {
 968         struct inode * old_inode, * new_inode;
 969         struct page * old_page, * new_page, * dir_page;
 970         struct ext2_dir_entry_2 * old_de, * new_de;
 971         int err;
 972
 973         ENTRY;
 974
 975         new_page = dir_page = NULL;
 976
 977         /* does the old entry exist? - if not get out */
 978         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 979         PDEBUG(old_page, "rename - old page");
 980         /*
 981          *  Check for inode number is _not_ due to possible IO errors.
 982          *  We might rmdir the source, keep it as pwd of some process
 983          *  and merrily kill the link to whatever was created under the
 984          *  same name. Goodbye sticky bit ;-<
 985          */
 986         old_inode = old_dentry->d_inode;
 987         err = -ENOENT;
 988         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
 989                 goto end_rename;
 990
 991         /* find new inode */
 992         new_inode = new_dentry->d_inode;
 993         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
 994                                 new_dentry->d_name.len, &new_de, NOLOCK);
 995         PDEBUG(new_page, "rename - new page ");
 996         if (new_page) {
 997                 if (!new_inode) {
 998                         page_cache_release(new_page);
 999                         new_page = NULL;
1000                 } else {
1001                         DQUOT_INIT(new_inode);
1002                 }
1003         }
1004         /* in this case we to check more ... */
1005         if (S_ISDIR(old_inode->i_mode)) {
1006                 /* can only rename into empty new directory */
1007                 if (new_inode) {
1008                         err = -ENOTEMPTY;
1009                         if (!empty_dir (new_inode))
1010                                 goto end_rename;
1011                 }
1012                 err = -EIO;
1013                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1014                 PDEBUG(dir_page, "rename dir page");
1015
1016                 if (!dir_page)
1017                         goto end_rename;
1018                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1019                         goto end_rename;
1020                 err = -EMLINK;
1021                 if (!new_inode && new_dir!=old_dir &&
1022                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1023                         goto end_rename;
1024         }
1025         /* create the target dir entry */
1026         if (!new_page) {
1027                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1028                                         new_dentry->d_name.len, &new_de,
1029                                         &err);
1030                 PDEBUG(new_page, "rename new page");
1031                 if (!new_page)
1032                         goto end_rename;
1033         }
1034         new_dir->i_version = ++event;
1035
1036         /*
1037          * remove the old entry
1038          */
1039         new_de->inode = le32_to_cpu(old_inode->i_ino);
1040         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1041                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1042                 new_de->file_type = old_de->file_type;
1043
1044         obdfs_delete_entry (old_de, old_page);
1045
1046         old_dir->i_version = ++event;
1047         if (new_inode) {
1048                 new_inode->i_nlink--;
1049                 new_inode->i_ctime = CURRENT_TIME;
1050                 mark_inode_dirty(new_inode);
1051         }
1052         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1053         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1054         mark_inode_dirty(old_dir);
1055         if (dir_page) {
1056                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1057                 /* XXX handle err */
1058                 err = obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1059                 old_dir->i_nlink--;
1060                 mark_inode_dirty(old_dir);
1061                 if (new_inode) {
1062                         new_inode->i_nlink--;
1063                         mark_inode_dirty(new_inode);
1064                 } else {
1065                         new_dir->i_nlink++;
1066                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1067                         mark_inode_dirty(new_dir);
1068                 }
1069         }
1070         if ( old_page != new_page ) {
1071                 unsigned long index = old_page->index;
1072                 /* lock the old_page and release unlocked copy */
1073                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1074                 page_cache_release(old_page);
1075                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1076                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1077                 /* XXX handle err */
1078                 err = obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1079         }
1080
1081         err = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1082
1083 end_rename:
1084         if (old_page && PageLocked(old_page) )
1085                 UnlockPage(old_page);
1086         if (old_page)
1087                 page_cache_release(old_page);
1088         if (new_page && PageLocked(new_page) )
1089                 UnlockPage(new_page);
1090         if (new_page)
1091                 page_cache_release(new_page);
1092         if (dir_page && PageLocked(dir_page) )
1093                 UnlockPage(dir_page);
1094         if (dir_page)
1095                 page_cache_release(dir_page);
1096
1097
1098         return err;
1099 } /* obdfs_rename */