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
 206 /* We do this with a locked page: that's not necessary, since the semaphore
 207  * on the inode protects this page as well.
 208  */
 209 static struct page *obdfs_add_entry (struct inode * dir,
 210                                      const char * name, int namelen,
 211                                      struct ext2_dir_entry_2 ** res_dir,
 212                                      int *err)
 213 {
 214         unsigned long offset;
 215         unsigned short rec_len;
 216         struct page *page;
 217         struct ext2_dir_entry_2 * de, * de1;
 218         struct super_block * sb;
 219
 220         ENTRY;
 221         *err = -EINVAL;
 222         *res_dir = NULL;
 223         if (!dir || !dir->i_nlink) {
 224                 EXIT;
 225                 return NULL;
 226         }
 227         sb = dir->i_sb;
 228
 229         if (!namelen) {
 230                 EXIT;
 231                 return NULL;
 232         }
 233         /*
 234          * Is this a busy deleted directory?  Can't create new files if so
 235          */
 236         if (dir->i_size == 0)
 237         {
 238                 OIDEBUG(dir);
 239                 *err = -ENOENT;
 240                 EXIT;
 241                 return NULL;
 242         }
 243         page = obdfs_getpage(dir, 0, 0, LOCKED);
 244         if (!page) {
 245                 EXIT;
 246                 return NULL;
 247         }
 248         rec_len = EXT2_DIR_REC_LEN(namelen);
 249         /* CDEBUG(D_INODE, "reclen: %d\n", rec_len); */
 250         PDEBUG(page, "starting search");
 251         offset = 0;
 252         de = (struct ext2_dir_entry_2 *) page_address(page);
 253         *err = -ENOSPC;
 254         while (1) {
 255                 /* CDEBUG(D_INODE,
 256                        "Entry at %p, (page at %#lx - %#lx), offset %ld\n",
 257                        de, page_address(page), page_address(page) + PAGE_SIZE,
 258                        offset); */
 259                 if ((char *)de >= PAGE_SIZE + (char *)page_address(page)) {
 260                         UnlockPage(page);
 261                         page_cache_release(page);
 262                         page = obdfs_getpage(dir, offset, 1, LOCKED);
 263                         if (!page) {
 264                                 EXIT;
 265                                 return NULL;
 266                         }
 267                         PDEBUG(page, "new directory page");
 268                         if (dir->i_size <= offset) {
 269                                 if (dir->i_size == 0) {
 270                                         *err = -ENOENT;
 271                                         EXIT;
 272                                         return NULL;
 273                                 }
 274
 275                                 CDEBUG(D_INODE, "creating next block\n");
 276
 277                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 278                                 de->inode = 0;
 279                                 de->rec_len = le16_to_cpu(PAGE_SIZE);
 280                                 dir->i_size = offset + PAGE_SIZE;
 281                                 dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 282                                 mark_inode_dirty(dir);
 283                         } else {
 284
 285                                 ext2_debug ("skipping to next block\n");
 286
 287                                 de = (struct ext2_dir_entry_2 *) page_address(page);
 288                         }
 289                 }
 290                 if (!obdfs_check_dir_entry ("ext2_add_entry", dir, de, page,
 291                                            offset)) {
 292                         *err = -ENOENT;
 293                         UnlockPage(page);
 294                         page_cache_release(page);
 295                         EXIT;
 296                         return NULL;
 297                 }
 298                 CDEBUG(D_INODE, "\n");
 299                 if (ext2_match (namelen, name, de)) {
 300                                 *err = -EEXIST;
 301                                 UnlockPage(page);
 302                                 page_cache_release(page);
 303                                 EXIT;
 304                                 return NULL;
 305                 }
 306                 /* CDEBUG(D_INODE, "Testing for enough space at de %p\n", de);*/
 307                 if ( (le32_to_cpu(de->inode) == 0 && le16_to_cpu(de->rec_len) >= rec_len) ||
 308                      (le16_to_cpu(de->rec_len) >= EXT2_DIR_REC_LEN(de->name_len) + rec_len)) {
 309                         offset += le16_to_cpu(de->rec_len);
 310                         /* CDEBUG(D_INODE,
 311                                "Found enough space de %p, offset %#lx\n",
 312                                de, offset); */
 313                         if (le32_to_cpu(de->inode)) {
 314                                 /*CDEBUG(D_INODE, "Insert new in %p\n", de);*/
 315                                 de1 = (struct ext2_dir_entry_2 *) ((char *) de +
 316                                         EXT2_DIR_REC_LEN(de->name_len));
 317                                 /*CDEBUG(D_INODE, "-- de1 at %p\n", de1);*/
 318                                 de1->rec_len = cpu_to_le16(le16_to_cpu(de->rec_len) -
 319                                         EXT2_DIR_REC_LEN(de->name_len));
 320                                 de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 321                                 de = de1;
 322                         }
 323                         /* CDEBUG(D_INODE,
 324                                "Reclen adjusted; copy %d bytes to %p, "
 325                                "page at %#lx EOP at %#lx\n",
 326                                namelen, de->name, page_address(page),
 327                                page_address(page) + PAGE_SIZE); */
 328                         de->inode = 0;
 329                         de->name_len = namelen;
 330                         de->file_type = 0;
 331                         memcpy (de->name, name, namelen);
 332                         /*
 333                          * XXX shouldn't update any times until successful
 334                          * completion of syscall, but too many callers depend
 335                          * on this.
 336                          *
 337                          * XXX similarly, too many callers depend on
 338                          * ext2_new_inode() setting the times, but error
 339                          * recovery deletes the inode, so the worst that can
 340                          * happen is that the times are slightly out of date
 341                          * and/or different from the directory change time.
 342                          */
 343                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 344                         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 345                         mark_inode_dirty(dir);
 346                         dir->i_version = ++event;
 347                         *res_dir = de;
 348                         *err = 0;
 349                         PDEBUG(page, "add_entry");
 350                         /* XXX unlock page here */
 351                         EXIT;
 352                         return page;
 353                 }
 354                 offset += le16_to_cpu(de->rec_len);
 355                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 356
 357         }
 358
 359         UnlockPage(page);
 360         page_cache_release(page);
 361         PDEBUG(page, "add_entry");
 362         EXIT;
 363         return NULL;
 364 } /* obdfs_add_entry */
 365
 366 /*
 367  * obdfs_delete_entry deletes a directory entry by merging it with the
 368  * previous entry
 369  */
 370 static int obdfs_delete_entry (struct ext2_dir_entry_2 * dir,
 371                               struct page * page)
 372 {
 373         struct ext2_dir_entry_2 * de, * pde;
 374         int i;
 375
 376         i = 0;
 377         pde = NULL;
 378         de = (struct ext2_dir_entry_2 *) page_address(page);
 379         while (i < PAGE_SIZE) {
 380                 if (!obdfs_check_dir_entry ("ext2_delete_entry", NULL,
 381                                            de, page, i))
 382                         return -EIO;
 383                 if (de == dir)  {
 384                         if (pde)
 385                                 pde->rec_len =
 386                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 387                                                     le16_to_cpu(dir->rec_len));
 388                         else
 389                                 dir->inode = 0;
 390                         return 0;
 391                 }
 392                 i += le16_to_cpu(de->rec_len);
 393                 pde = de;
 394                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 395         }
 396         return -ENOENT;
 397 } /* obdfs_delete_entry */
 398
 399
 400 static inline void ext2_set_de_type(struct super_block *sb,
 401                                 struct ext2_dir_entry_2 *de,
 402                                 umode_t mode) {
 403         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_FILETYPE))
 404                 return;
 405         if (S_ISCHR(mode))
 406                 de->file_type = EXT2_FT_CHRDEV;
 407         else if (S_ISBLK(mode))
 408                 de->file_type = EXT2_FT_BLKDEV;
 409         else if (S_ISFIFO(mode))
 410                 de->file_type = EXT2_FT_FIFO;
 411         else if (S_ISLNK(mode))
 412                 de->file_type = EXT2_FT_SYMLINK;
 413         else if (S_ISREG(mode))
 414                 de->file_type = EXT2_FT_REG_FILE;
 415         else if (S_ISDIR(mode))
 416                 de->file_type = EXT2_FT_DIR;
 417 }
 418
 419
 420 /*
 421  * Display all dentries holding the specified inode.
 422  */
 423 #if 0
 424 static void show_dentry(struct list_head * dlist, int subdirs)
 425 {
 426         struct list_head *tmp = dlist;
 427
 428         while ((tmp = tmp->next) != dlist) {
 429                 struct dentry * dentry;
 430                 const char * unhashed = "";
 431
 432                 if ( subdirs )
 433                         dentry  = list_entry(tmp, struct dentry, d_child);
 434                 else
 435                         dentry  = list_entry(tmp, struct dentry, d_alias);
 436
 437                 if (list_empty(&dentry->d_hash))
 438                         unhashed = "(unhashed)";
 439
 440                 if ( dentry->d_inode )
 441                         printk("show_dentry: %s/%s, d_count=%d%s (ino %ld, dev %d, ct %d)\n",
 442                                dentry->d_parent->d_name.name,
 443                                dentry->d_name.name, dentry->d_count,
 444                                unhashed, dentry->d_inode->i_ino,
 445                                dentry->d_inode->i_dev,
 446                                dentry->d_inode->i_count);
 447                 else
 448                         printk("show_dentry: %s/%s, d_count=%d%s \n",
 449                                dentry->d_parent->d_name.name,
 450                                dentry->d_name.name, dentry->d_count,
 451                                unhashed);
 452         }
 453 } /* show_dentry */
 454 #endif
 455
 456
 457 static struct inode *obdfs_new_inode(struct inode *dir, int mode)
 458 {
 459         struct obdo *oa;
 460         struct inode *inode;
 461         int err;
 462
 463         ENTRY;
 464         oa = obdo_alloc();
 465         if (!oa) {
 466                 EXIT;
 467                 return ERR_PTR(-ENOMEM);
 468         }
 469
 470         /* Send a hint to the create method on the type of file to create */
 471         oa->o_mode = mode;
 472         oa->o_valid |= OBD_MD_FLMODE;
 473
 474         err = IOPS(dir, create)(IID(dir), oa);
 475
 476         if ( err ) {
 477                 obdo_free(oa);
 478                 EXIT;
 479                 return ERR_PTR(err);
 480         }
 481
 482         inode = iget(dir->i_sb, (ino_t)oa->o_id);
 483
 484         if (!inode) {
 485                 IOPS(dir, destroy)(IID(dir), oa);
 486                 obdo_free(oa);
 487                 EXIT;
 488                 return ERR_PTR(-EIO);
 489         }
 490
 491         if (!list_empty(&inode->i_dentry)) {
 492                 CDEBUG(D_INODE, "New inode (%ld) has aliases!\n", inode->i_ino);
 493                 IOPS(dir, destroy)(IID(dir), oa);
 494                 obdo_free(oa);
 495                 iput(inode);
 496                 EXIT;
 497                 return ERR_PTR(-EIO);
 498         }
 499         obdo_free(oa);
 500
 501         EXIT;
 502         return inode;
 503 } /* obdfs_new_inode */
 504
 505
 506 /*
 507  * By the time this is called, we already have created
 508  * the directory cache entry for the new file, but it
 509  * is so far negative - it has no inode.
 510  *
 511  * If the create succeeds, we fill in the inode information
 512  * with d_instantiate().
 513  */
 514 int obdfs_create (struct inode * dir, struct dentry * dentry, int mode)
 515 {
 516         struct inode * inode;
 517         struct page *page;
 518         struct ext2_dir_entry_2 * de;
 519         int err = -EIO;
 520
 521         ENTRY;
 522         inode = obdfs_new_inode(dir, mode);
 523         if ( IS_ERR(inode) ) {
 524                 EXIT;
 525                 return PTR_ERR(inode);
 526         }
 527
 528         inode->i_op = &obdfs_file_inode_operations;
 529         mark_inode_dirty(inode);
 530         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 531         if (!page) {
 532                 inode->i_nlink--;
 533                 mark_inode_dirty(inode);
 534                 iput (inode);
 535                 EXIT;
 536                 return err;
 537         }
 538         de->inode = cpu_to_le32(inode->i_ino);
 539         ext2_set_de_type(dir->i_sb, de, S_IFREG);
 540         dir->i_version = ++event;
 541
 542         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 543         UnlockPage(page);
 544
 545         page_cache_release(page);
 546         d_instantiate(dentry, inode);
 547         EXIT;
 548         return err;
 549 } /* obdfs_create */
 550
 551 int obdfs_mknod (struct inode * dir, struct dentry *dentry, int mode, int rdev)
 552 {
 553         struct inode * inode;
 554         struct page *page;
 555         struct ext2_dir_entry_2 * de;
 556         int err;
 557
 558         ENTRY;
 559         inode = obdfs_new_inode(dir, mode);
 560         if ( IS_ERR(inode) ) {
 561                 EXIT;
 562                 return PTR_ERR(inode);
 563         }
 564
 565         inode->i_uid = current->fsuid;
 566         init_special_inode(inode, mode, rdev);
 567         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 568         if (!page)
 569                 goto out_no_entry;
 570         de->inode = cpu_to_le32(inode->i_ino);
 571         dir->i_version = ++event;
 572         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 573         mark_inode_dirty(inode);
 574
 575         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 576         UnlockPage(page);
 577
 578         d_instantiate(dentry, inode);
 579         page_cache_release(page);
 580         err = 0;
 581 out:
 582         return err;
 583
 584 out_no_entry:
 585         inode->i_nlink--;
 586         mark_inode_dirty(inode);
 587         iput(inode);
 588         goto out;
 589 } /* obdfs_mknod */
 590
 591 int obdfs_mkdir(struct inode * dir, struct dentry * dentry, int mode)
 592 {
 593         struct inode * inode;
 594         struct page *page, *inode_page;
 595         struct ext2_dir_entry_2 * de;
 596         int err;
 597
 598         ENTRY;
 599
 600         err = -EMLINK;
 601         if (dir->i_nlink >= EXT2_LINK_MAX)
 602                 goto out;
 603
 604         mode |= S_IFDIR;
 605         if (dir->i_mode & S_ISGID)
 606                 mode |= S_ISGID;
 607
 608         inode = obdfs_new_inode(dir, mode);
 609         if ( IS_ERR(inode) ) {
 610                 EXIT;
 611                 return PTR_ERR(inode);
 612         }
 613
 614         inode->i_op = &obdfs_dir_inode_operations;
 615         inode->i_blocks = 0;
 616         inode_page = obdfs_getpage(inode, 0, 1, LOCKED);
 617         if (!inode_page) {
 618                 inode->i_nlink--; /* is this nlink == 0? */
 619                 mark_inode_dirty(inode);
 620                 iput (inode);
 621                 return err;
 622         }
 623
 624         /* create . and .. */
 625         de = (struct ext2_dir_entry_2 *) page_address(inode_page);
 626         de->inode = cpu_to_le32(inode->i_ino);
 627         de->name_len = 1;
 628         de->rec_len = cpu_to_le16(EXT2_DIR_REC_LEN(de->name_len));
 629         strcpy (de->name, ".");
 630         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 631
 632         de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 633         de->inode = cpu_to_le32(dir->i_ino);
 634         de->rec_len = cpu_to_le16(PAGE_SIZE - EXT2_DIR_REC_LEN(1));
 635         de->name_len = 2;
 636         strcpy (de->name, "..");
 637         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 638
 639         /* XXX handle err */
 640         err = obdfs_do_writepage(inode, inode_page, IS_SYNC(inode));
 641         inode->i_blocks = PAGE_SIZE/inode->i_sb->s_blocksize;
 642         inode->i_size = PAGE_SIZE;
 643         UnlockPage(inode_page);
 644         page_cache_release(inode_page);
 645
 646         inode->i_nlink = 2;
 647         mark_inode_dirty(inode);
 648
 649         /* now deal with the parent */
 650         page = obdfs_add_entry(dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 651         if (!page) {
 652                 goto out_no_entry;
 653         }
 654
 655         de->inode = cpu_to_le32(inode->i_ino);
 656         ext2_set_de_type(dir->i_sb, de, S_IFDIR);
 657         dir->i_version = ++event;
 658
 659         dir->i_nlink++;
 660         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 661         mark_inode_dirty(dir);
 662         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 663
 664         UnlockPage(page);
 665
 666         page_cache_release(page);
 667         d_instantiate(dentry, inode);
 668 out:
 669         EXIT;
 670         return err;
 671
 672 out_no_entry:
 673         inode->i_nlink = 0;
 674         mark_inode_dirty(inode);
 675         iput (inode);
 676         EXIT;
 677         goto out;
 678 } /* obdfs_mkdir */
 679
 680
 681 /*
 682  * routine to check that the specified directory is empty (for rmdir)
 683  */
 684 static int empty_dir (struct inode * inode)
 685 {
 686         unsigned long offset;
 687         struct page *page;
 688         struct ext2_dir_entry_2 * de, * de1;
 689         struct super_block * sb;
 690
 691         sb = inode->i_sb;
 692         if (inode->i_size < EXT2_DIR_REC_LEN(1) + EXT2_DIR_REC_LEN(2) ||
 693             !(page = obdfs_getpage (inode, 0, 0, LOCKED))) {
 694                 ext2_warning (inode->i_sb, "empty_dir",
 695                               "bad directory (dir #%lu) - no data block",
 696                               inode->i_ino);
 697                 return 1;
 698         }
 699         de = (struct ext2_dir_entry_2 *) page_address(page);
 700         de1 = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 701         if (le32_to_cpu(de->inode) != inode->i_ino || !le32_to_cpu(de1->inode) ||
 702             strcmp (".", de->name) || strcmp ("..", de1->name)) {
 703                 ext2_warning (inode->i_sb, "empty_dir",
 704                               "bad directory (dir #%lu) - no `.' or `..'",
 705                               inode->i_ino);
 706                 page_cache_release(page);
 707                 return 1;
 708         }
 709         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 710         de = (struct ext2_dir_entry_2 *) ((char *) de1 + le16_to_cpu(de1->rec_len));
 711         while (offset < inode->i_size ) {
 712                 if (!page || (void *) de >= (void *) (page_address(page) + PAGE_SIZE)) {
 713                         if (page) {
 714                                 UnlockPage(page);
 715                                 page_cache_release(page);
 716                         }
 717                         page = obdfs_getpage(inode, offset, 0, LOCKED);
 718                         if (!page) {
 719 #if 0
 720                                 ext2_error (sb, "empty_dir",
 721                                             "directory #%lu contains a hole at offset %lu",
 722                                             inode->i_ino, offset);
 723 #endif
 724                                 offset += sb->s_blocksize;
 725                                 continue;
 726                         }
 727                         de = (struct ext2_dir_entry_2 *) page_address(page);
 728                 }
 729                 if (!obdfs_check_dir_entry ("empty_dir", inode, de, page,
 730                                            offset)) {
 731                         UnlockPage(page);
 732                         page_cache_release(page);
 733                         return 1;
 734                 }
 735                 if (le32_to_cpu(de->inode)) {
 736                         UnlockPage(page);
 737                         page_cache_release(page);
 738                         return 0;
 739                 }
 740                 offset += le16_to_cpu(de->rec_len);
 741                 de = (struct ext2_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
 742         }
 743         UnlockPage(page);
 744         page_cache_release(page);
 745         return 1;
 746 } /* empty_dir */
 747
 748 int obdfs_rmdir (struct inode * dir, struct dentry *dentry)
 749 {
 750         int retval;
 751         struct inode * inode;
 752         struct page *page;
 753         struct ext2_dir_entry_2 * de;
 754         int err;
 755
 756         ENTRY;
 757
 758         retval = -ENOENT;
 759         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 760         if (!page)
 761                 goto end_rmdir;
 762
 763         inode = dentry->d_inode;
 764         DQUOT_INIT(inode);
 765
 766         retval = -EIO;
 767         if (le32_to_cpu(de->inode) != inode->i_ino)
 768                 goto end_rmdir;
 769
 770         retval = -ENOTEMPTY;
 771         if (!empty_dir (inode))
 772                 goto end_rmdir;
 773
 774         retval = obdfs_delete_entry (de, page);
 775         dir->i_version = ++event;
 776         if (retval)
 777                 goto end_rmdir;
 778         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 779         /* XXX handle err? */
 780         UnlockPage(page);
 781
 782         if (inode->i_nlink != 2)
 783                 ext2_warning (inode->i_sb, "ext2_rmdir",
 784                               "empty directory has nlink!=2 (%d)",
 785                               inode->i_nlink);
 786         inode->i_version = ++event;
 787         inode->i_nlink = 0;
 788         inode->i_size = 0;
 789         mark_inode_dirty(inode);
 790         dir->i_nlink--;
 791         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 792         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 793         mark_inode_dirty(dir);
 794         d_delete(dentry);
 795
 796 end_rmdir:
 797         if ( page )
 798                 page_cache_release(page);
 799         EXIT;
 800         return retval;
 801 } /* obdfs_rmdir */
 802
 803 int obdfs_unlink(struct inode * dir, struct dentry *dentry)
 804 {
 805         int retval;
 806         struct inode * inode;
 807         struct page *page;
 808         struct ext2_dir_entry_2 * de;
 809         int err;
 810
 811         ENTRY;
 812
 813         retval = -ENOENT;
 814         page = obdfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, LOCKED);
 815         if (!page)
 816                 goto end_unlink;
 817
 818         inode = dentry->d_inode;
 819         DQUOT_INIT(inode);
 820
 821         retval = -EIO;
 822         if (le32_to_cpu(de->inode) != inode->i_ino)
 823                 goto end_unlink;
 824
 825         if (!inode->i_nlink) {
 826                 ext2_warning (inode->i_sb, "ext2_unlink",
 827                               "Deleting nonexistent file (%lu), %d",
 828                               inode->i_ino, inode->i_nlink);
 829                 inode->i_nlink = 1;
 830         }
 831         retval = obdfs_delete_entry (de, page);
 832         if (retval)
 833                 goto end_unlink;
 834         dir->i_version = ++event;
 835         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 836         /* XXX handle err? */
 837         UnlockPage(page);
 838
 839         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 840         dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
 841         mark_inode_dirty(dir);
 842         inode->i_nlink--;
 843         mark_inode_dirty(inode);
 844         inode->i_ctime = dir->i_ctime;
 845         retval = 0;
 846         d_delete(dentry);       /* This also frees the inode */
 847
 848 end_unlink:
 849         if (page)
 850                 page_cache_release(page);
 851         EXIT;
 852         return retval;
 853 } /* obdfs_unlink */
 854
 855 int obdfs_symlink (struct inode * dir, struct dentry *dentry,
 856                    const char * symname)
 857 {
 858         struct ext2_dir_entry_2 * de;
 859         struct inode * inode;
 860         struct obdfs_inode_info *oinfo;
 861         struct page* page = NULL, * name_page = NULL;
 862         char * link;
 863         int i, l, err = -EIO;
 864         char c;
 865
 866         ENTRY;
 867         inode = obdfs_new_inode(dir, S_IFLNK | S_IRWXUGO);
 868         if ( IS_ERR(inode) ) {
 869                 EXIT;
 870                 return PTR_ERR(inode);
 871         }
 872
 873         inode->i_op = &obdfs_symlink_inode_operations;
 874         for (l = 0; l < inode->i_sb->s_blocksize - 1 && symname [l]; l++)
 875                 ;
 876
 877         oinfo = OBDFS_INFO(inode);
 878
 879         if (l >= sizeof(oinfo->oi_inline)) {
 880                 CDEBUG(D_INODE, "l=%d, normal symlink\n", l);
 881
 882                 name_page = obdfs_getpage(inode, 0, 1, LOCKED);
 883                 if (!name_page) {
 884                         inode->i_nlink--;
 885                         mark_inode_dirty(inode);
 886                         iput (inode);
 887                         EXIT;
 888                         return err;
 889                 }
 890                 link = (char *)page_address(name_page);
 891         } else {
 892                 link = oinfo->oi_inline;
 893                 oinfo->oi_flags |= OBD_FL_INLINEDATA;
 894
 895                 CDEBUG(D_INODE, "l=%d, fast symlink\n", l);
 896         }
 897         i = 0;
 898         while (i < inode->i_sb->s_blocksize - 1 && (c = *(symname++)))
 899                 link[i++] = c;
 900         link[i] = 0;
 901         if (name_page) {
 902                 err = obdfs_do_writepage(inode, name_page, IS_SYNC(inode));
 903                 /* XXX handle err */
 904                 PDEBUG(name_page, "symlink");
 905                 UnlockPage(name_page);
 906                 page_cache_release(name_page);
 907         }
 908         inode->i_size = i;
 909         mark_inode_dirty(inode);
 910
 911         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len,
 912                                 &de, &err);
 913         if (!page)
 914                 goto out_no_entry;
 915         de->inode = cpu_to_le32(inode->i_ino);
 916         ext2_set_de_type(dir->i_sb, de, S_IFLNK);
 917         dir->i_version = ++event;
 918         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 919         UnlockPage(page);
 920
 921         d_instantiate(dentry, inode);
 922 out:
 923         EXIT;
 924         return err;
 925
 926 out_no_entry:
 927         inode->i_nlink--;
 928         mark_inode_dirty(inode);
 929         iput (inode);
 930         goto out;
 931 } /* obdfs_symlink */
 932
 933 int obdfs_link (struct dentry * old_dentry,
 934                 struct inode * dir, struct dentry *dentry)
 935 {
 936         struct inode *inode = old_dentry->d_inode;
 937         struct ext2_dir_entry_2 * de;
 938         struct page *page;
 939         int err;
 940
 941         ENTRY;
 942
 943         if (S_ISDIR(inode->i_mode))
 944                 return -EPERM;
 945
 946         if (inode->i_nlink >= EXT2_LINK_MAX)
 947                 return -EMLINK;
 948
 949         page = obdfs_add_entry (dir, dentry->d_name.name, dentry->d_name.len, &de, &err);
 950         if (!page)
 951                 return err;
 952
 953         de->inode = cpu_to_le32(inode->i_ino);
 954         ext2_set_de_type(dir->i_sb, de, inode->i_mode);
 955         dir->i_version = ++event;
 956
 957         err = obdfs_do_writepage(dir, page, IS_SYNC(dir));
 958         UnlockPage(page);
 959
 960         page_cache_release(page);
 961         inode->i_nlink++;
 962         inode->i_ctime = CURRENT_TIME;
 963         mark_inode_dirty(inode);
 964         inode->i_count++;
 965         d_instantiate(dentry, inode);
 966         return err;
 967 } /* obdfs_link */
 968
 969 #define PARENT_INO(buffer) \
 970         ((struct ext2_dir_entry_2 *) ((char *) buffer + \
 971         le16_to_cpu(((struct ext2_dir_entry_2 *) buffer)->rec_len)))->inode
 972
 973 /*
 974  * Anybody can rename anything with this: the permission checks are left to the
 975  * higher-level routines.
 976  */
 977 int obdfs_rename (struct inode * old_dir, struct dentry *old_dentry,
 978                            struct inode * new_dir, struct dentry *new_dentry)
 979 {
 980         struct inode * old_inode, * new_inode;
 981         struct page * old_page, * new_page, * dir_page;
 982         struct ext2_dir_entry_2 * old_de, * new_de;
 983         int err;
 984
 985         ENTRY;
 986
 987         new_page = dir_page = NULL;
 988
 989         /* does the old entry exist? - if not get out */
 990         old_page = obdfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len, &old_de, NOLOCK);
 991         PDEBUG(old_page, "rename - old page");
 992         /*
 993          *  Check for inode number is _not_ due to possible IO errors.
 994          *  We might rmdir the source, keep it as pwd of some process
 995          *  and merrily kill the link to whatever was created under the
 996          *  same name. Goodbye sticky bit ;-<
 997          */
 998         old_inode = old_dentry->d_inode;
 999         err = -ENOENT;
1000         if (!old_page || le32_to_cpu(old_de->inode) != old_inode->i_ino)
1001                 goto end_rename;
1002
1003         /* find new inode */
1004         new_inode = new_dentry->d_inode;
1005         new_page = obdfs_find_entry (new_dir, new_dentry->d_name.name,
1006                                 new_dentry->d_name.len, &new_de, NOLOCK);
1007         PDEBUG(new_page, "rename - new page ");
1008         if (new_page) {
1009                 if (!new_inode) {
1010                         page_cache_release(new_page);
1011                         new_page = NULL;
1012                 } else {
1013                         DQUOT_INIT(new_inode);
1014                 }
1015         }
1016         /* in this case we to check more ... */
1017         if (S_ISDIR(old_inode->i_mode)) {
1018                 /* can only rename into empty new directory */
1019                 if (new_inode) {
1020                         err = -ENOTEMPTY;
1021                         if (!empty_dir (new_inode))
1022                                 goto end_rename;
1023                 }
1024                 err = -EIO;
1025                 dir_page= obdfs_getpage (old_inode, 0, 0, LOCKED);
1026                 PDEBUG(dir_page, "rename dir page");
1027
1028                 if (!dir_page)
1029                         goto end_rename;
1030                 if (le32_to_cpu(PARENT_INO(page_address(dir_page))) != old_dir->i_ino)
1031                         goto end_rename;
1032                 err = -EMLINK;
1033                 if (!new_inode && new_dir!=old_dir &&
1034                                 new_dir->i_nlink >= EXT2_LINK_MAX)
1035                         goto end_rename;
1036         }
1037         /* create the target dir entry */
1038         if (!new_page) {
1039                 new_page = obdfs_add_entry (new_dir, new_dentry->d_name.name,
1040                                         new_dentry->d_name.len, &new_de,
1041                                         &err);
1042                 PDEBUG(new_page, "rename new page");
1043                 if (!new_page)
1044                         goto end_rename;
1045         }
1046         new_dir->i_version = ++event;
1047
1048         /*
1049          * remove the old entry
1050          */
1051         new_de->inode = le32_to_cpu(old_inode->i_ino);
1052         if (EXT2_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
1053                                       EXT2_FEATURE_INCOMPAT_FILETYPE))
1054                 new_de->file_type = old_de->file_type;
1055
1056         obdfs_delete_entry (old_de, old_page);
1057
1058         old_dir->i_version = ++event;
1059         if (new_inode) {
1060                 new_inode->i_nlink--;
1061                 new_inode->i_ctime = CURRENT_TIME;
1062                 mark_inode_dirty(new_inode);
1063         }
1064         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1065         old_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1066         mark_inode_dirty(old_dir);
1067         if (dir_page) {
1068                 PARENT_INO(page_address(dir_page)) = le32_to_cpu(new_dir->i_ino);
1069                 /* XXX handle err */
1070                 err = obdfs_do_writepage(old_inode, dir_page, IS_SYNC(old_inode));
1071                 old_dir->i_nlink--;
1072                 mark_inode_dirty(old_dir);
1073                 if (new_inode) {
1074                         new_inode->i_nlink--;
1075                         mark_inode_dirty(new_inode);
1076                 } else {
1077                         new_dir->i_nlink++;
1078                         new_dir->u.ext2_i.i_flags &= ~EXT2_BTREE_FL;
1079                         mark_inode_dirty(new_dir);
1080                 }
1081         }
1082         if ( old_page != new_page ) {
1083                 unsigned long index = old_page->index;
1084                 /* lock the old_page and release unlocked copy */
1085                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1086                 page_cache_release(old_page);
1087                 old_page = obdfs_getpage(old_dir, index >> PAGE_SHIFT, 0, LOCKED);
1088                 CDEBUG(D_INODE, "old_page at %p\n", old_page);
1089                 /* XXX handle err */
1090                 err = obdfs_do_writepage(old_dir, old_page, IS_SYNC(old_dir));
1091         }
1092
1093         err = obdfs_do_writepage(new_dir, new_page, IS_SYNC(new_dir));
1094
1095 end_rename:
1096         if (old_page && PageLocked(old_page) )
1097                 UnlockPage(old_page);
1098         if (old_page)
1099                 page_cache_release(old_page);
1100         if (new_page && PageLocked(new_page) )
1101                 UnlockPage(new_page);
1102         if (new_page)
1103                 page_cache_release(new_page);
1104         if (dir_page && PageLocked(dir_page) )
1105                 UnlockPage(dir_page);
1106         if (dir_page)
1107                 page_cache_release(dir_page);
1108
1109
1110         return err;
1111 } /* obdfs_rename */