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                 OIDEBUG(dir);
 237                 *err = -ENOENT;
 238                 EXIT;
 239                 return NULL;
 240         }
 241         page = obdfs_getpage(dir, 0, 0, LOCKED);
 242         if (!page) {
 243                 EXIT;
 244                 return NULL;
 245         }
 246         rec_len = EXT2_DIR_REC_LEN(namelen);
 247         /* CDEBUG(D_INODE, "reclen: %d\n", rec_len); */
 248         PDEBUG(page, "starting search");
 249         offset = 0;
 250         de = (struct ext2_dir_entry_2 *) page_address(page);
 251         *err = -ENOSPC;
 252         while (1) {
 253                 /* CDEBUG(D_INODE,
 254                        "Entry at %p, (page at %#lx - %#lx), offset %ld\n",
 255                        de, page_address(page), page_address(page) + PAGE_SIZE,
 256                        offset); */
 257                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 258                         UnlockPage(page);
 259                         page_cache_release(page);
 260                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 261                         if (!page) {
 262                                 EXIT;
 263                                 return NULL;
 264                         }
 265                         PDEBUG(page, "new directory page");
 266                         if (dir->i_size <= offset) {
 267                                 if (dir->i_size == 0) {
 268                                         *err = -ENOENT;
 269                                         EXIT;
 270                                         return NULL;
 271                                 }
 272
 273                                 CDEBUG(D_INODE, "creating next block\n");
 274
 275                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 276                                 de->inode = 0;
 277                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 278                                 dir->i_size = offset + PAGE_SIZE;
 279                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 280                                 mark_inode_dirty(dir);
 281                         } else {
 282
 283                                 ext2_debug ("skipping to next block\n");
 284
 285                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 286                         }
 287                 }
 288                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 289                                            offset)) {
 290                         *err = -ENOENT;
 291                         UnlockPage(page);
 292                         page_cache_release(page);
 293                         EXIT;
 294                         return NULL;
 295                 }
 296                 CDEBUG(D_INODE, "\n");
 297                 if (ext2_match (namelen, name, de)) {
 298                                 *err = -EEXIST;
 299                                 UnlockPage(page);
 300                                 page_cache_release(page);
 301                                 EXIT;
 302                                 return NULL;
 303                 }
 304                 /* CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);*/
 305                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 306                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 307                         offset += le16_to_cpu(de->rec_len);
 308                         /* CDEBUG(D_INODE,
 309                                "Found enough space de %p, offset %#lx\n",
 310                                de, offset); */
 311                         if (le32_to_cpu(de->inode)) {
 312                                 /*CDEBUG(D_INODE, "Insert new in %p\n", de);*/
 313                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 314                                         EXT2_DIR_REC_LEN(de->name_len));
 315                                 /*CDEBUG(D_INODE, "-- de1 at %p\n", de1);*/
 316                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 317                                         EXT2_DIR_REC_LEN(de->name_len));
 318                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 319                                 de = de1;
 320                         }
 321                         /* CDEBUG(D_INODE,
 322                                "Reclen adjusted; copy %d bytes to %p, "
 323                                "page at %#lx EOP at %#lx\n",
 324                                namelen, de->name, page_address(page),
 325                                page_address(page) + PAGE_SIZE); */
 326                         de->inode = 0;
 327                         de->name_len = namelen;
 328                         de->file_type = 0;
 329                         memcpy (de->name, name, namelen);
 330                         /*
 331                          * XXX shouldn't update any times until successful
 332                          * completion of syscall, but too many callers depend
 333                          * on this.
 334                          *
 335                          * XXX similarly, too many callers depend on
 336                          * ext2_new_inode() setting the times, but error
 337                          * recovery deletes the inode, so the worst that can
 338                          * happen is that the times are slightly out of date
 339                          * and/or different from the directory change time.
 340                          */
 341                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 342                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 343                         mark_inode_dirty(dir);
 344                         dir->i_version = ++event;
 345                         *res_dir = de;
 346                         *err = 0;
 347                         PDEBUG(page, "add_entry");
 348                         EXIT;
 349                         return page;
 350                 }
 351                 offset += le16_to_cpu(de->rec_len);
 352                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 353
 354         }
 355
 356         UnlockPage(page);
 357         page_cache_release(page);
 358         PDEBUG(page, "add_entry");
 359         EXIT;
 360         return NULL;
 361 } /* obdfs_add_entry */
 362
 363 /*
 364  * obdfs_delete_entry deletes a directory entry by merging it with the
 365  * previous entry
 366  */
 367 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 368                               struct page * page)
 369 {
 370         struct ext2_dir_entry_2 * de, * pde;
 371         int i;
 372
 373         i = 0;
 374         pde = NULL;
 375         de = (struct ext2_dir_entry_2 *) page_address(page);
 376         while (i < PAGE_SIZE) {
 377                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 378                                            de, page, i))
 379                         return -EIO;
 380                 if (de == dir)  {
 381                         if (pde)
 382                                 pde->rec_len =
 383                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 384                                                     le16_to_cpu(dir->rec_len));
 385                         else
 386                                 dir->inode = 0;
 387                         return 0;
 388                 }
 389                 i += le16_to_cpu(de->rec_len);
 390                 pde = de;
 391                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 392         }
 393         return -ENOENT;
 394 } /* obdfs_delete_entry */
 395
 396
 397 static inline void ext2_set_de_type(struct super_block *sb,
 398                                 struct ext2_dir_entry_2 *de,
 399                                 umode_t mode) {
 400         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 401                 return;
 402         if (S_ISCHR(mode))
 403                 de->file_type = EXT2_FT_CHRDEV;
 404         else if (S_ISBLK(mode))
 405                 de->file_type = EXT2_FT_BLKDEV;
 406         else if (S_ISFIFO(mode))
 407                 de->file_type = EXT2_FT_FIFO;
 408         else if (S_ISLNK(mode))
 409                 de->file_type = EXT2_FT_SYMLINK;
 410         else if (S_ISREG(mode))
 411                 de->file_type = EXT2_FT_REG_FILE;
 412         else if (S_ISDIR(mode))
 413                 de->file_type = EXT2_FT_DIR;
 414 }
 415
 416
 417 /*
 418  * Display all dentries holding the specified inode.
 419  */
 420 #if 0
 421 static void show_dentry(struct list_head * dlist, int subdirs)
 422 {
 423         struct list_head *tmp = dlist;
 424
 425         while ((tmp = tmp->next) != dlist) {
 426                 struct dentry * dentry;
 427                 const char * unhashed = "";
 428
 429                 if ( subdirs )
 430                         dentry  = list_entry(tmp, struct dentry, d_child);
 431                 else
 432                         dentry  = list_entry(tmp, struct dentry, d_alias);
 433
 434                 if (list_empty(&dentry->d_hash))
 435                         unhashed = "(unhashed)";
 436
 437                 if ( dentry->d_inode )
 438                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 439                                dentry->d_parent->d_name.name,
 440                                dentry->d_name.name, dentry->d_count,
 441                                unhashed, dentry->d_inode->i_ino,
 442                                dentry->d_inode->i_dev,
 443                                dentry->d_inode->i_count);
 444                 else
 445                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 446                                dentry->d_parent->d_name.name,
 447                                dentry->d_name.name, dentry->d_count,
 448                                unhashed);
 449         }
 450 } /* show_dentry */
 451 #endif
 452
 453
 454 static struct inode *obdfs_new_inode(struct inode *dir, int mode)
 455 {
 456         struct obdo *oa;
 457         struct inode *inode;
 458         int err;
 459
 460         ENTRY;
 461         oa = obdo_alloc();
 462         if (!oa) {
 463                 EXIT;
 464                 return ERR_PTR(-ENOMEM);
 465         }
 466
 467         /* Send a hint to the create method on the type of file to create */
 468         oa->o_mode = mode;
 469         oa->o_valid |= OBD_MD_FLMODE;
 470
 471         err = IOPS(dir, create)(IID(dir), oa);
 472
 473         if ( err ) {
 474                 obdo_free(oa);
 475                 EXIT;
 476                 return ERR_PTR(err);
 477         }
 478
 479         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 480
 481         if (!inode) {
 482                 IOPS(dir, destroy)(IID(dir), oa);
 483                 obdo_free(oa);
 484                 EXIT;
 485                 return ERR_PTR(-EIO);
 486         }
 487
 488         if (!list_empty(&inode->i_dentry)) {
 489                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 490                 IOPS(dir, destroy)(IID(dir), oa);
 491                 obdo_free(oa);
 492                 iput(inode);
 493                 EXIT;
 494                 return ERR_PTR(-EIO);
 495         }
 496         obdo_free(oa);
 497         INIT_LIST_HEAD(&OBDFS_LIST(inode));
 498
 499         EXIT;
 500         return inode;
 501 } /* obdfs_new_inode */
 502
 503
 504 /*
 505  * By the time this is called, we already have created
 506  * the directory cache entry for the new file, but it
 507  * is so far negative - it has no inode.
 508  *
 509  * If the create succeeds, we fill in the inode information
 510  * with d_instantiate().
 511  */
 512 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 513 {
 514         struct inode * inode;
 515         struct page *page;
 516         struct ext2_dir_entry_2 * de;
 517         int err = -EIO;
 518
 519         ENTRY;
 520         inode = obdfs_new_inode(dir, mode);
 521         if ( IS_ERR(inode) ) {
 522                 EXIT;
 523                 return PTR_ERR(inode);
 524         }
 525
 526         inode->i_op = &obdfs_file_inode_operations;
 527         mark_inode_dirty(inode);
 528         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 529         if (!page) {
 530                 inode->i_nlink--;
 531                 mark_inode_dirty(inode);
 532                 iput (inode);
 533                 EXIT;
 534                 return err;
 535         }
 536         de->inode = cpu_to_le32(inode->i_ino);
 537         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 538         dir->i_version = ++event;
 539
 540         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 541         UnlockPage(page);
 542
 543         page_cache_release(page);
 544         d_instantiate(dentry, inode);
 545         EXIT;
 546         return err;
 547 } /* obdfs_create */
 548
 549 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 550 {
 551         struct inode * inode;
 552         struct page *page;
 553         struct ext2_dir_entry_2 * de;
 554         int err;
 555
 556         ENTRY;
 557         inode = obdfs_new_inode(dir, mode);
 558         if ( IS_ERR(inode) ) {
 559                 EXIT;
 560                 return PTR_ERR(inode);
 561         }
 562
 563         inode->i_uid = current->fsuid;
 564         init_special_inode(inode, mode, rdev);
 565         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 566         if (!page)
 567                 goto out_no_entry;
 568         de->inode = cpu_to_le32(inode->i_ino);
 569         dir->i_version = ++event;
 570         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 571         mark_inode_dirty(inode);
 572
 573         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 574         UnlockPage(page);
 575
 576         d_instantiate(dentry, inode);
 577         page_cache_release(page);
 578         err = 0;
 579 out:
 580         return err;
 581
 582 out_no_entry:
 583         inode->i_nlink--;
 584         mark_inode_dirty(inode);
 585         iput(inode);
 586         goto out;
 587 } /* obdfs_mknod */
 588
 589 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 590 {
 591         struct inode * inode;
 592         struct page *page, *inode_page;
 593         struct ext2_dir_entry_2 * de;
 594         int err;
 595
 596         ENTRY;
 597
 598         err = -EMLINK;
 599         if (dir->i_nlink >= EXT2_LINK_MAX)
 600                 goto out;
 601
 602         mode |= S_IFDIR;
 603         if (dir->i_mode & S_ISGID)
 604                 mode |= S_ISGID;
 605
 606         inode = obdfs_new_inode(dir, mode);
 607         if ( IS_ERR(inode) ) {
 608                 EXIT;
 609                 return PTR_ERR(inode);
 610         }
 611
 612         inode->i_op = &obdfs_dir_inode_operations;
 613         inode->i_blocks = 0;
 614         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 615         if (!inode_page) {
 616                 inode->i_nlink--; /* is this nlink == 0? */
 617                 mark_inode_dirty(inode);
 618                 iput (inode);
 619                 return err;
 620         }
 621
 622         /* create . and .. */
 623         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 624         de->inode = cpu_to_le32(inode->i_ino);
 625         de->name_len = 1;
 626         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 627         strcpy (de->name, ".");
 628         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 629
 630         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 631         de->inode = cpu_to_le32(dir->i_ino);
 632         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 633         de->name_len = 2;
 634         strcpy (de->name, "..");
 635         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 636
 637         /* XXX handle err */
 638         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 639         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 640         inode->i_size = PAGE_SIZE;
 641         UnlockPage(inode_page);
 642         page_cache_release(inode_page);
 643
 644         inode->i_nlink = 2;
 645         mark_inode_dirty(inode);
 646
 647         /* now deal with the parent */
 648         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 649         if (!page) {
 650                 goto out_no_entry;
 651         }
 652
 653         de->inode = cpu_to_le32(inode->i_ino);
 654         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 655         dir->i_version = ++event;
 656
 657         dir->i_nlink++;
 658         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 659         mark_inode_dirty(dir);
 660         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 661         /* XXX handle err? */
 662
 663         UnlockPage(page);
 664
 665         page_cache_release(page);
 666         d_instantiate(dentry, inode);
 667 out:
 668         EXIT;
 669         return err;
 670
 671 out_no_entry:
 672         inode->i_nlink = 0;
 673         mark_inode_dirty(inode);
 674         iput (inode);
 675         EXIT;
 676         goto out;
 677 } /* obdfs_mkdir */
 678
 679
 680 /*
 681  * routine to check that the specified directory is empty (for rmdir)
 682  */
 683 static int empty_dir (struct inode * inode)
 684 {
 685         unsigned long offset;
 686         struct page *page;
 687         struct ext2_dir_entry_2 * de, * de1;
 688         struct super_block * sb;
 689
 690         sb = inode->i_sb;
 691         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 692             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 693                 ext2_warning (inode->i_sb, "empty_dir",
 694                               "bad directory (dir #%lu) - no data block",
 695                               inode->i_ino);
 696                 return 1;
 697         }
 698         de = (struct ext2_dir_entry_2 *) page_address(page);
 699         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 700         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 701             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 702                 ext2_warning (inode->i_sb, "empty_dir",
 703                               "bad directory (dir #%lu) - no `.' or `..'",
 704                               inode->i_ino);
 705                 page_cache_release(page);
 706                 return 1;
 707         }
 708         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 709         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 710         while (offset < inode->i_size ) {
 711                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 712                         if (page) {
 713                                 UnlockPage(page);
 714                                 page_cache_release(page);
 715                         }
 716                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 717                         if (!page) {
 718 #if 0
 719                                 ext2_error (sb, "empty_dir",
 720                                             "directory #%lu contains a hole at offset %lu",
 721                                             inode->i_ino, offset);
 722 #endif
 723                                 offset += sb->s_blocksize;
 724                                 continue;
 725                         }
 726                         de = (struct ext2_dir_entry_2 *) page_address(page);
 727                 }
 728                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 729                                            offset)) {
 730                         UnlockPage(page);
 731                         page_cache_release(page);
 732                         return 1;
 733                 }
 734                 if (le32_to_cpu(de->inode)) {
 735                         UnlockPage(page);
 736                         page_cache_release(page);
 737                         return 0;
 738                 }
 739                 offset += le16_to_cpu(de->rec_len);
 740                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 741         }
 742         UnlockPage(page);
 743         page_cache_release(page);
 744         return 1;
 745 } /* empty_dir */
 746
 747 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 748 {
 749         int retval;
 750         struct inode * inode;
 751         struct page *page;
 752         struct ext2_dir_entry_2 * de;
 753         int err;
 754
 755         ENTRY;
 756
 757         retval = -ENOENT;
 758         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 759         if (!page)
 760                 goto end_rmdir;
 761
 762         inode = dentry->d_inode;
 763         DQUOT_INIT(inode);
 764
 765         retval = -EIO;
 766         if (le32_to_cpu(de->inode) != inode->i_ino)
 767                 goto end_rmdir;
 768
 769         retval = -ENOTEMPTY;
 770         if (!empty_dir (inode))
 771                 goto end_rmdir;
 772
 773         retval = obdfs_delete_entry (de, page);
 774         dir->i_version = ++event;
 775         if (retval)
 776                 goto end_rmdir;
 777         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 778         /* XXX handle err? */
 779         UnlockPage(page);
 780
 781         if (inode->i_nlink != 2)
 782                 ext2_warning (inode->i_sb, "ext2_rmdir",
 783                               "empty directory has nlink!=2 (%d)",
 784                               inode->i_nlink);
 785         inode->i_version = ++event;
 786         inode->i_nlink = 0;
 787         inode->i_size = 0;
 788         mark_inode_dirty(inode);
 789         dir->i_nlink--;
 790         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 791         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 792         mark_inode_dirty(dir);
 793         d_delete(dentry);
 794
 795 end_rmdir:
 796         if ( page )
 797                 page_cache_release(page);
 798         EXIT;
 799         return retval;
 800 } /* obdfs_rmdir */
 801
 802 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 803 {
 804         int retval;
 805         struct inode * inode;
 806         struct page *page;
 807         struct ext2_dir_entry_2 * de;
 808         int err;
 809
 810         ENTRY;
 811
 812         retval = -ENOENT;
 813         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 814         if (!page)
 815                 goto end_unlink;
 816
 817         inode = dentry->d_inode;
 818         DQUOT_INIT(inode);
 819
 820         retval = -EIO;
 821         if (le32_to_cpu(de->inode) != inode->i_ino)
 822                 goto end_unlink;
 823
 824         if (!inode->i_nlink) {
 825                 ext2_warning (inode->i_sb, "ext2_unlink",
 826                               "Deleting nonexistent file (%lu), %d",
 827                               inode->i_ino, inode->i_nlink);
 828                 inode->i_nlink = 1;
 829         }
 830         retval = obdfs_delete_entry (de, page);
 831         if (retval)
 832                 goto end_unlink;
 833         dir->i_version = ++event;
 834         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 835         /* XXX handle err? */
 836         UnlockPage(page);
 837
 838         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 839         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 840         mark_inode_dirty(dir);
 841         inode->i_nlink--;
 842         mark_inode_dirty(inode);
 843         inode->i_ctime = dir->i_ctime;
 844         retval = 0;
 845         d_delete(dentry);       /* This also frees the inode */
 846
 847 end_unlink:
 848         if (page)
 849                 page_cache_release(page);
 850         EXIT;
 851         return retval;
 852 } /* obdfs_unlink */
 853
 854 int obdfs_symlink (struct inode * dir, struct dentry *dentry,
 855                    const char * symname)
 856 {
 857         struct ext2_dir_entry_2 * de;
 858         struct inode * inode;
 859         struct obdfs_inode_info *oinfo;
 860         struct page* page = NULL, * name_page = NULL;
 861         char * link;
 862         int i, l, err = -EIO;
 863         char c;
 864
 865         ENTRY;
 866         inode = obdfs_new_inode(dir, S_IFLNK | S_IRWXUGO);
 867         if ( IS_ERR(inode) ) {
 868                 EXIT;
 869                 return PTR_ERR(inode);
 870         }
 871
 872         inode->i_op = &obdfs_symlink_inode_operations;
 873         for (l = 0; l < inode->i_sb->s_blocksize - 1 && symname [l]; l++)
 874                 ;
 875
 876         oinfo = OBDFS_INFO(inode);
 877
 878         if (l >= sizeof(oinfo->oi_inline)) {
 879                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 880
 881                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 882                 if (!name_page) {
 883                         inode->i_nlink--;
 884                         mark_inode_dirty(inode);
 885                         iput (inode);
 886                         EXIT;
 887                         return err;
 888                 }
 889                 link = (char *)page_address(name_page);
 890         } else {
 891                 link = oinfo->oi_inline;
 892                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 893
 894                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 895         }
 896         i = 0;
 897         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 898                 link[i++] = c;
 899         link[i] = 0;
 900         if (name_page) {
 901                 err = obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 902                 /* XXX handle err */
 903                 PDEBUG(name_page, "symlink");
 904                 UnlockPage(name_page);
 905                 page_cache_release(name_page);
 906         }
 907         inode->i_size = i;
 908         mark_inode_dirty(inode);
 909
 910         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 911                                 &de, &err);
 912         if (!page)
 913                 goto out_no_entry;
 914         de->inode = cpu_to_le32(inode->i_ino);
 915         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 916         dir->i_version = ++event;
 917         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 918         UnlockPage(page);
 919
 920         d_instantiate(dentry, inode);
 921 out:
 922         EXIT;
 923         return err;
 924
 925 out_no_entry:
 926         inode->i_nlink--;
 927         mark_inode_dirty(inode);
 928         iput (inode);
 929         goto out;
 930 } /* obdfs_symlink */
 931
 932 int obdfs_link (struct dentry * old_dentry,
 933                 struct inode * dir, struct dentry *dentry)
 934 {
 935         struct inode *inode = old_dentry->d_inode;
 936         struct ext2_dir_entry_2 * de;
 937         struct page *page;
 938         int err;
 939
 940         ENTRY;
 941
 942         if (S_ISDIR(inode->i_mode))
 943                 return -EPERM;
 944
 945         if (inode->i_nlink >= EXT2_LINK_MAX)
 946                 return -EMLINK;
 947
 948         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 949         if (!page)
 950                 return err;
 951
 952         de->inode = cpu_to_le32(inode->i_ino);
 953         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 954         dir->i_version = ++event;
 955
 956         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 957         UnlockPage(page);
 958
 959         page_cache_release(page);
 960         inode->i_nlink++;
 961         inode->i_ctime = CURRENT_TIME;
 962         mark_inode_dirty(inode);
 963         inode->i_count++;
 964         d_instantiate(dentry, inode);
 965         return err;
 966 } /* obdfs_link */
 967
 968 #define PARENT_INO(buffer) \
 969         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 970         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 971
 972 /*
 973  * Anybody can rename anything with this: the permission checks are left to the
 974  * higher-level routines.
 975  */
 976 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 977                            struct inode * new_dir, struct dentry *new_dentry)
 978 {
 979         struct inode * old_inode, * new_inode;
 980         struct page * old_page, * new_page, * dir_page;
 981         struct ext2_dir_entry_2 * old_de, * new_de;
 982         int err;
 983
 984         ENTRY;
 985
 986         new_page = dir_page = NULL;
 987
 988         /* does the old entry exist? - if not get out */
 989         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 990         PDEBUG(old_page, "rename - old page");
 991         /*
 992          *  Check for inode number is _not_ due to possible IO errors.
 993          *  We might rmdir the source, keep it as pwd of some process
 994          *  and merrily kill the link to whatever was created under the
 995          *  same name. Goodbye sticky bit ;-<
 996          */
 997         old_inode = old_dentry->d_inode;
 998         err = -ENOENT;
 999         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
1000                 goto end_rename;
1001
1002         /* find new inode */
1003         new_inode = new_dentry->d_inode;
1004         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1005                                 new_dentry->d_name.len, &new_de, NOLOCK);
1006         PDEBUG(new_page, "rename - new page ");
1007         if (new_page) {
1008                 if (!new_inode) {
1009                         page_cache_release(new_page);
1010                         new_page = NULL;
1011                 } else {
1012                         DQUOT_INIT(new_inode);
1013                 }
1014         }
1015         /* in this case we to check more ... */
1016         if (S_ISDIR(old_inode->i_mode)) {
1017                 /* can only rename into empty new directory */
1018                 if (new_inode) {
1019                         err = -ENOTEMPTY;
1020                         if (!empty_dir (new_inode))
1021                                 goto end_rename;
1022                 }
1023                 err = -EIO;
1024                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1025                 PDEBUG(dir_page, "rename dir page");
1026
1027                 if (!dir_page)
1028                         goto end_rename;
1029                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1030                         goto end_rename;
1031                 err = -EMLINK;
1032                 if (!new_inode && new_dir!=old_dir &&
1033                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1034                         goto end_rename;
1035         }
1036         /* create the target dir entry */
1037         if (!new_page) {
1038                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1039                                         new_dentry->d_name.len, &new_de,
1040                                         &err);
1041                 PDEBUG(new_page, "rename new page");
1042                 if (!new_page)
1043                         goto end_rename;
1044         }
1045         new_dir->i_version = ++event;
1046
1047         /*
1048          * remove the old entry
1049          */
1050         new_de->inode = le32_to_cpu(old_inode->i_ino);
1051         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1052                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1053                 new_de->file_type = old_de->file_type;
1054
1055         obdfs_delete_entry (old_de, old_page);
1056
1057         old_dir->i_version = ++event;
1058         if (new_inode) {
1059                 new_inode->i_nlink--;
1060                 new_inode->i_ctime = CURRENT_TIME;
1061                 mark_inode_dirty(new_inode);
1062         }
1063         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1064         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1065         mark_inode_dirty(old_dir);
1066         if (dir_page) {
1067                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1068                 /* XXX handle err */
1069                 err = obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1070                 old_dir->i_nlink--;
1071                 mark_inode_dirty(old_dir);
1072                 if (new_inode) {
1073                         new_inode->i_nlink--;
1074                         mark_inode_dirty(new_inode);
1075                 } else {
1076                         new_dir->i_nlink++;
1077                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1078                         mark_inode_dirty(new_dir);
1079                 }
1080         }
1081         if ( old_page != new_page ) {
1082                 unsigned long index = old_page->index;
1083                 /* lock the old_page and release unlocked copy */
1084                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1085                 page_cache_release(old_page);
1086                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1087                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1088                 /* XXX handle err */
1089                 err = obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1090         }
1091
1092         err = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1093
1094 end_rename:
1095         if (old_page && PageLocked(old_page) )
1096                 UnlockPage(old_page);
1097         if (old_page)
1098                 page_cache_release(old_page);
1099         if (new_page && PageLocked(new_page) )
1100                 UnlockPage(new_page);
1101         if (new_page)
1102                 page_cache_release(new_page);
1103         if (dir_page && PageLocked(dir_page) )
1104                 UnlockPage(dir_page);
1105         if (dir_page)
1106                 page_cache_release(dir_page);
1107
1108
1109         return err;
1110 } /* obdfs_rename */