fs/ocfs2/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dir.c
   5  *
   6  * Creates, reads, walks and deletes directory-nodes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "uptodate.h"
  61
  62 #include "buffer_head_io.h"
  63
  64 #define NAMEI_RA_CHUNKS  2
  65 #define NAMEI_RA_BLOCKS  4
  66 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  67 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  68
  69 static unsigned char ocfs2_filetype_table[] = {
  70         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  71 };
  72
  73 static int ocfs2_extend_dir(struct ocfs2_super *osb,
  74                             struct inode *dir,
  75                             struct buffer_head *parent_fe_bh,
  76                             unsigned int blocks_wanted,
  77                             struct buffer_head **new_de_bh);
  78 static int ocfs2_do_extend_dir(struct super_block *sb,
  79                                handle_t *handle,
  80                                struct inode *dir,
  81                                struct buffer_head *parent_fe_bh,
  82                                struct ocfs2_alloc_context *data_ac,
  83                                struct ocfs2_alloc_context *meta_ac,
  84                                struct buffer_head **new_bh);
  85
  86 /*
  87  * bh passed here can be an inode block or a dir data block, depending
  88  * on the inode inline data flag.
  89  */
  90 static int ocfs2_check_dir_entry(struct inode * dir,
  91                                  struct ocfs2_dir_entry * de,
  92                                  struct buffer_head * bh,
  93                                  unsigned long offset)
  94 {
  95         const char *error_msg = NULL;
  96         const int rlen = le16_to_cpu(de->rec_len);
  97
  98         if (rlen < OCFS2_DIR_REC_LEN(1))
  99                 error_msg = "rec_len is smaller than minimal";
 100         else if (rlen % 4 != 0)
 101                 error_msg = "rec_len % 4 != 0";
 102         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
 103                 error_msg = "rec_len is too small for name_len";
 104         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
 105                 error_msg = "directory entry across blocks";
 106
 107         if (error_msg != NULL)
 108                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
 109                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
 110                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
 111                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
 112                      de->name_len);
 113         return error_msg == NULL ? 1 : 0;
 114 }
 115
 116 static inline int ocfs2_match(int len,
 117                               const char * const name,
 118                               struct ocfs2_dir_entry *de)
 119 {
 120         if (len != de->name_len)
 121                 return 0;
 122         if (!de->inode)
 123                 return 0;
 124         return !memcmp(name, de->name, len);
 125 }
 126
 127 /*
 128  * Returns 0 if not found, -1 on failure, and 1 on success
 129  */
 130 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
 131                                         struct inode *dir,
 132                                         const char *name, int namelen,
 133                                         unsigned long offset,
 134                                         char *first_de,
 135                                         unsigned int bytes,
 136                                         struct ocfs2_dir_entry **res_dir)
 137 {
 138         struct ocfs2_dir_entry *de;
 139         char *dlimit, *de_buf;
 140         int de_len;
 141         int ret = 0;
 142
 143         mlog_entry_void();
 144
 145         de_buf = first_de;
 146         dlimit = de_buf + bytes;
 147
 148         while (de_buf < dlimit) {
 149                 /* this code is executed quadratically often */
 150                 /* do minimal checking `by hand' */
 151
 152                 de = (struct ocfs2_dir_entry *) de_buf;
 153
 154                 if (de_buf + namelen <= dlimit &&
 155                     ocfs2_match(namelen, name, de)) {
 156                         /* found a match - just to be sure, do a full check */
 157                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
 158                                 ret = -1;
 159                                 goto bail;
 160                         }
 161                         *res_dir = de;
 162                         ret = 1;
 163                         goto bail;
 164                 }
 165
 166                 /* prevent looping on a bad block */
 167                 de_len = le16_to_cpu(de->rec_len);
 168                 if (de_len <= 0) {
 169                         ret = -1;
 170                         goto bail;
 171                 }
 172
 173                 de_buf += de_len;
 174                 offset += de_len;
 175         }
 176
 177 bail:
 178         mlog_exit(ret);
 179         return ret;
 180 }
 181
 182 static struct buffer_head *ocfs2_find_entry_id(const char *name,
 183                                                int namelen,
 184                                                struct inode *dir,
 185                                                struct ocfs2_dir_entry **res_dir)
 186 {
 187         int ret, found;
 188         struct buffer_head *di_bh = NULL;
 189         struct ocfs2_dinode *di;
 190         struct ocfs2_inline_data *data;
 191
 192         ret = ocfs2_read_inode_block(dir, &di_bh);
 193         if (ret) {
 194                 mlog_errno(ret);
 195                 goto out;
 196         }
 197
 198         di = (struct ocfs2_dinode *)di_bh->b_data;
 199         data = &di->id2.i_data;
 200
 201         found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
 202                                       data->id_data, i_size_read(dir), res_dir);
 203         if (found == 1)
 204                 return di_bh;
 205
 206         brelse(di_bh);
 207 out:
 208         return NULL;
 209 }
 210
 211 static int ocfs2_validate_dir_block(struct super_block *sb,
 212                                     struct buffer_head *bh)
 213 {
 214         /*
 215          * Nothing yet.  We don't validate dirents here, that's handled
 216          * in-place when the code walks them.
 217          */
 218         mlog(0, "Validating dirblock %llu\n",
 219              (unsigned long long)bh->b_blocknr);
 220
 221         return 0;
 222 }
 223
 224 /*
 225  * This function forces all errors to -EIO for consistency with its
 226  * predecessor, ocfs2_bread().  We haven't audited what returning the
 227  * real error codes would do to callers.  We log the real codes with
 228  * mlog_errno() before we squash them.
 229  */
 230 static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
 231                                 struct buffer_head **bh, int flags)
 232 {
 233         int rc = 0;
 234         struct buffer_head *tmp = *bh;
 235
 236         rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
 237                                     ocfs2_validate_dir_block);
 238         if (rc)
 239                 mlog_errno(rc);
 240
 241         /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
 242         if (!rc && !*bh)
 243                 *bh = tmp;
 244
 245         return rc ? -EIO : 0;
 246 }
 247
 248 static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
 249                                                struct inode *dir,
 250                                                struct ocfs2_dir_entry **res_dir)
 251 {
 252         struct super_block *sb;
 253         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 254         struct buffer_head *bh, *ret = NULL;
 255         unsigned long start, block, b;
 256         int ra_max = 0;         /* Number of bh's in the readahead
 257                                    buffer, bh_use[] */
 258         int ra_ptr = 0;         /* Current index into readahead
 259                                    buffer */
 260         int num = 0;
 261         int nblocks, i, err;
 262
 263         mlog_entry_void();
 264
 265         sb = dir->i_sb;
 266
 267         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 268         start = OCFS2_I(dir)->ip_dir_start_lookup;
 269         if (start >= nblocks)
 270                 start = 0;
 271         block = start;
 272
 273 restart:
 274         do {
 275                 /*
 276                  * We deal with the read-ahead logic here.
 277                  */
 278                 if (ra_ptr >= ra_max) {
 279                         /* Refill the readahead buffer */
 280                         ra_ptr = 0;
 281                         b = block;
 282                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 283                                 /*
 284                                  * Terminate if we reach the end of the
 285                                  * directory and must wrap, or if our
 286                                  * search has finished at this block.
 287                                  */
 288                                 if (b >= nblocks || (num && block == start)) {
 289                                         bh_use[ra_max] = NULL;
 290                                         break;
 291                                 }
 292                                 num++;
 293
 294                                 bh = NULL;
 295                                 err = ocfs2_read_dir_block(dir, b++, &bh,
 296                                                            OCFS2_BH_READAHEAD);
 297                                 bh_use[ra_max] = bh;
 298                         }
 299                 }
 300                 if ((bh = bh_use[ra_ptr++]) == NULL)
 301                         goto next;
 302                 if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
 303                         /* read error, skip block & hope for the best.
 304                          * ocfs2_read_dir_block() has released the bh. */
 305                         ocfs2_error(dir->i_sb, "reading directory %llu, "
 306                                     "offset %lu\n",
 307                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
 308                                     block);
 309                         goto next;
 310                 }
 311                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
 312                                           block << sb->s_blocksize_bits,
 313                                           bh->b_data, sb->s_blocksize,
 314                                           res_dir);
 315                 if (i == 1) {
 316                         OCFS2_I(dir)->ip_dir_start_lookup = block;
 317                         ret = bh;
 318                         goto cleanup_and_exit;
 319                 } else {
 320                         brelse(bh);
 321                         if (i < 0)
 322                                 goto cleanup_and_exit;
 323                 }
 324         next:
 325                 if (++block >= nblocks)
 326                         block = 0;
 327         } while (block != start);
 328
 329         /*
 330          * If the directory has grown while we were searching, then
 331          * search the last part of the directory before giving up.
 332          */
 333         block = nblocks;
 334         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 335         if (block < nblocks) {
 336                 start = 0;
 337                 goto restart;
 338         }
 339
 340 cleanup_and_exit:
 341         /* Clean up the read-ahead blocks */
 342         for (; ra_ptr < ra_max; ra_ptr++)
 343                 brelse(bh_use[ra_ptr]);
 344
 345         mlog_exit_ptr(ret);
 346         return ret;
 347 }
 348
 349 /*
 350  * Try to find an entry of the provided name within 'dir'.
 351  *
 352  * If nothing was found, NULL is returned. Otherwise, a buffer_head
 353  * and pointer to the dir entry are passed back.
 354  *
 355  * Caller can NOT assume anything about the contents of the
 356  * buffer_head - it is passed back only so that it can be passed into
 357  * any one of the manipulation functions (add entry, delete entry,
 358  * etc). As an example, bh in the extent directory case is a data
 359  * block, in the inline-data case it actually points to an inode.
 360  */
 361 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
 362                                      struct inode *dir,
 363                                      struct ocfs2_dir_entry **res_dir)
 364 {
 365         *res_dir = NULL;
 366
 367         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 368                 return ocfs2_find_entry_id(name, namelen, dir, res_dir);
 369
 370         return ocfs2_find_entry_el(name, namelen, dir, res_dir);
 371 }
 372
 373 /*
 374  * Update inode number and type of a previously found directory entry.
 375  */
 376 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
 377                        struct buffer_head *de_bh, struct ocfs2_dir_entry *de,
 378                        struct inode *new_entry_inode)
 379 {
 380         int ret;
 381         ocfs2_journal_access_func access = ocfs2_journal_access_db;
 382
 383         /*
 384          * The same code works fine for both inline-data and extent
 385          * based directories, so no need to split this up.  The only
 386          * difference is the journal_access function.
 387          */
 388
 389         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 390                 access = ocfs2_journal_access_di;
 391
 392         ret = access(handle, dir, de_bh, OCFS2_JOURNAL_ACCESS_WRITE);
 393         if (ret) {
 394                 mlog_errno(ret);
 395                 goto out;
 396         }
 397
 398         de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
 399         ocfs2_set_de_type(de, new_entry_inode->i_mode);
 400
 401         ocfs2_journal_dirty(handle, de_bh);
 402
 403 out:
 404         return ret;
 405 }
 406
 407 static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
 408                                 struct ocfs2_dir_entry *de_del,
 409                                 struct buffer_head *bh, char *first_de,
 410                                 unsigned int bytes)
 411 {
 412         struct ocfs2_dir_entry *de, *pde;
 413         int i, status = -ENOENT;
 414         ocfs2_journal_access_func access = ocfs2_journal_access_db;
 415
 416         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
 417
 418         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 419                 access = ocfs2_journal_access_di;
 420
 421         i = 0;
 422         pde = NULL;
 423         de = (struct ocfs2_dir_entry *) first_de;
 424         while (i < bytes) {
 425                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
 426                         status = -EIO;
 427                         mlog_errno(status);
 428                         goto bail;
 429                 }
 430                 if (de == de_del)  {
 431                         status = access(handle, dir, bh,
 432                                         OCFS2_JOURNAL_ACCESS_WRITE);
 433                         if (status < 0) {
 434                                 status = -EIO;
 435                                 mlog_errno(status);
 436                                 goto bail;
 437                         }
 438                         if (pde)
 439                                 le16_add_cpu(&pde->rec_len,
 440                                                 le16_to_cpu(de->rec_len));
 441                         else
 442                                 de->inode = 0;
 443                         dir->i_version++;
 444                         status = ocfs2_journal_dirty(handle, bh);
 445                         goto bail;
 446                 }
 447                 i += le16_to_cpu(de->rec_len);
 448                 pde = de;
 449                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
 450         }
 451 bail:
 452         mlog_exit(status);
 453         return status;
 454 }
 455
 456 static inline int ocfs2_delete_entry_id(handle_t *handle,
 457                                         struct inode *dir,
 458                                         struct ocfs2_dir_entry *de_del,
 459                                         struct buffer_head *bh)
 460 {
 461         int ret;
 462         struct buffer_head *di_bh = NULL;
 463         struct ocfs2_dinode *di;
 464         struct ocfs2_inline_data *data;
 465
 466         ret = ocfs2_read_inode_block(dir, &di_bh);
 467         if (ret) {
 468                 mlog_errno(ret);
 469                 goto out;
 470         }
 471
 472         di = (struct ocfs2_dinode *)di_bh->b_data;
 473         data = &di->id2.i_data;
 474
 475         ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
 476                                    i_size_read(dir));
 477
 478         brelse(di_bh);
 479 out:
 480         return ret;
 481 }
 482
 483 static inline int ocfs2_delete_entry_el(handle_t *handle,
 484                                         struct inode *dir,
 485                                         struct ocfs2_dir_entry *de_del,
 486                                         struct buffer_head *bh)
 487 {
 488         return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
 489                                     bh->b_size);
 490 }
 491
 492 /*
 493  * ocfs2_delete_entry deletes a directory entry by merging it with the
 494  * previous entry
 495  */
 496 int ocfs2_delete_entry(handle_t *handle,
 497                        struct inode *dir,
 498                        struct ocfs2_dir_entry *de_del,
 499                        struct buffer_head *bh)
 500 {
 501         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 502                 return ocfs2_delete_entry_id(handle, dir, de_del, bh);
 503
 504         return ocfs2_delete_entry_el(handle, dir, de_del, bh);
 505 }
 506
 507 /*
 508  * Check whether 'de' has enough room to hold an entry of
 509  * 'new_rec_len' bytes.
 510  */
 511 static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
 512                                          unsigned int new_rec_len)
 513 {
 514         unsigned int de_really_used;
 515
 516         /* Check whether this is an empty record with enough space */
 517         if (le64_to_cpu(de->inode) == 0 &&
 518             le16_to_cpu(de->rec_len) >= new_rec_len)
 519                 return 1;
 520
 521         /*
 522          * Record might have free space at the end which we can
 523          * use.
 524          */
 525         de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
 526         if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
 527             return 1;
 528
 529         return 0;
 530 }
 531
 532 /* we don't always have a dentry for what we want to add, so people
 533  * like orphan dir can call this instead.
 534  *
 535  * If you pass me insert_bh, I'll skip the search of the other dir
 536  * blocks and put the record in there.
 537  */
 538 int __ocfs2_add_entry(handle_t *handle,
 539                       struct inode *dir,
 540                       const char *name, int namelen,
 541                       struct inode *inode, u64 blkno,
 542                       struct buffer_head *parent_fe_bh,
 543                       struct buffer_head *insert_bh)
 544 {
 545         unsigned long offset;
 546         unsigned short rec_len;
 547         struct ocfs2_dir_entry *de, *de1;
 548         struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
 549         struct super_block *sb = dir->i_sb;
 550         int retval, status;
 551         unsigned int size = sb->s_blocksize;
 552         char *data_start = insert_bh->b_data;
 553
 554         mlog_entry_void();
 555
 556         if (!namelen)
 557                 return -EINVAL;
 558
 559         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
 560                 data_start = di->id2.i_data.id_data;
 561                 size = i_size_read(dir);
 562
 563                 BUG_ON(insert_bh != parent_fe_bh);
 564         }
 565
 566         rec_len = OCFS2_DIR_REC_LEN(namelen);
 567         offset = 0;
 568         de = (struct ocfs2_dir_entry *) data_start;
 569         while (1) {
 570                 BUG_ON((char *)de >= (size + data_start));
 571
 572                 /* These checks should've already been passed by the
 573                  * prepare function, but I guess we can leave them
 574                  * here anyway. */
 575                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
 576                         retval = -ENOENT;
 577                         goto bail;
 578                 }
 579                 if (ocfs2_match(namelen, name, de)) {
 580                         retval = -EEXIST;
 581                         goto bail;
 582                 }
 583
 584                 if (ocfs2_dirent_would_fit(de, rec_len)) {
 585                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 586                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 587                         if (retval < 0) {
 588                                 mlog_errno(retval);
 589                                 goto bail;
 590                         }
 591
 592                         if (insert_bh == parent_fe_bh)
 593                                 status = ocfs2_journal_access_di(handle, dir,
 594                                                                  insert_bh,
 595                                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 596                         else
 597                                 status = ocfs2_journal_access_db(handle, dir,
 598                                                                  insert_bh,
 599                                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 600                         /* By now the buffer is marked for journaling */
 601                         offset += le16_to_cpu(de->rec_len);
 602                         if (le64_to_cpu(de->inode)) {
 603                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
 604                                         OCFS2_DIR_REC_LEN(de->name_len));
 605                                 de1->rec_len =
 606                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
 607                                         OCFS2_DIR_REC_LEN(de->name_len));
 608                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 609                                 de = de1;
 610                         }
 611                         de->file_type = OCFS2_FT_UNKNOWN;
 612                         if (blkno) {
 613                                 de->inode = cpu_to_le64(blkno);
 614                                 ocfs2_set_de_type(de, inode->i_mode);
 615                         } else
 616                                 de->inode = 0;
 617                         de->name_len = namelen;
 618                         memcpy(de->name, name, namelen);
 619
 620                         dir->i_version++;
 621                         status = ocfs2_journal_dirty(handle, insert_bh);
 622                         retval = 0;
 623                         goto bail;
 624                 }
 625                 offset += le16_to_cpu(de->rec_len);
 626                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
 627         }
 628
 629         /* when you think about it, the assert above should prevent us
 630          * from ever getting here. */
 631         retval = -ENOSPC;
 632 bail:
 633
 634         mlog_exit(retval);
 635         return retval;
 636 }
 637
 638 static int ocfs2_dir_foreach_blk_id(struct inode *inode,
 639                                     u64 *f_version,
 640                                     loff_t *f_pos, void *priv,
 641                                     filldir_t filldir, int *filldir_err)
 642 {
 643         int ret, i, filldir_ret;
 644         unsigned long offset = *f_pos;
 645         struct buffer_head *di_bh = NULL;
 646         struct ocfs2_dinode *di;
 647         struct ocfs2_inline_data *data;
 648         struct ocfs2_dir_entry *de;
 649
 650         ret = ocfs2_read_inode_block(inode, &di_bh);
 651         if (ret) {
 652                 mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
 653                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 654                 goto out;
 655         }
 656
 657         di = (struct ocfs2_dinode *)di_bh->b_data;
 658         data = &di->id2.i_data;
 659
 660         while (*f_pos < i_size_read(inode)) {
 661 revalidate:
 662                 /* If the dir block has changed since the last call to
 663                  * readdir(2), then we might be pointing to an invalid
 664                  * dirent right now.  Scan from the start of the block
 665                  * to make sure. */
 666                 if (*f_version != inode->i_version) {
 667                         for (i = 0; i < i_size_read(inode) && i < offset; ) {
 668                                 de = (struct ocfs2_dir_entry *)
 669                                         (data->id_data + i);
 670                                 /* It's too expensive to do a full
 671                                  * dirent test each time round this
 672                                  * loop, but we do have to test at
 673                                  * least that it is non-zero.  A
 674                                  * failure will be detected in the
 675                                  * dirent test below. */
 676                                 if (le16_to_cpu(de->rec_len) <
 677                                     OCFS2_DIR_REC_LEN(1))
 678                                         break;
 679                                 i += le16_to_cpu(de->rec_len);
 680                         }
 681                         *f_pos = offset = i;
 682                         *f_version = inode->i_version;
 683                 }
 684
 685                 de = (struct ocfs2_dir_entry *) (data->id_data + *f_pos);
 686                 if (!ocfs2_check_dir_entry(inode, de, di_bh, *f_pos)) {
 687                         /* On error, skip the f_pos to the end. */
 688                         *f_pos = i_size_read(inode);
 689                         goto out;
 690                 }
 691                 offset += le16_to_cpu(de->rec_len);
 692                 if (le64_to_cpu(de->inode)) {
 693                         /* We might block in the next section
 694                          * if the data destination is
 695                          * currently swapped out.  So, use a
 696                          * version stamp to detect whether or
 697                          * not the directory has been modified
 698                          * during the copy operation.
 699                          */
 700                         u64 version = *f_version;
 701                         unsigned char d_type = DT_UNKNOWN;
 702
 703                         if (de->file_type < OCFS2_FT_MAX)
 704                                 d_type = ocfs2_filetype_table[de->file_type];
 705
 706                         filldir_ret = filldir(priv, de->name,
 707                                               de->name_len,
 708                                               *f_pos,
 709                                               le64_to_cpu(de->inode),
 710                                               d_type);
 711                         if (filldir_ret) {
 712                                 if (filldir_err)
 713                                         *filldir_err = filldir_ret;
 714                                 break;
 715                         }
 716                         if (version != *f_version)
 717                                 goto revalidate;
 718                 }
 719                 *f_pos += le16_to_cpu(de->rec_len);
 720         }
 721
 722 out:
 723         brelse(di_bh);
 724
 725         return 0;
 726 }
 727
 728 static int ocfs2_dir_foreach_blk_el(struct inode *inode,
 729                                     u64 *f_version,
 730                                     loff_t *f_pos, void *priv,
 731                                     filldir_t filldir, int *filldir_err)
 732 {
 733         int error = 0;
 734         unsigned long offset, blk, last_ra_blk = 0;
 735         int i, stored;
 736         struct buffer_head * bh, * tmp;
 737         struct ocfs2_dir_entry * de;
 738         struct super_block * sb = inode->i_sb;
 739         unsigned int ra_sectors = 16;
 740
 741         stored = 0;
 742         bh = NULL;
 743
 744         offset = (*f_pos) & (sb->s_blocksize - 1);
 745
 746         while (!error && !stored && *f_pos < i_size_read(inode)) {
 747                 blk = (*f_pos) >> sb->s_blocksize_bits;
 748                 if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
 749                         /* Skip the corrupt dirblock and keep trying */
 750                         *f_pos += sb->s_blocksize - offset;
 751                         continue;
 752                 }
 753
 754                 /* The idea here is to begin with 8k read-ahead and to stay
 755                  * 4k ahead of our current position.
 756                  *
 757                  * TODO: Use the pagecache for this. We just need to
 758                  * make sure it's cluster-safe... */
 759                 if (!last_ra_blk
 760                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
 761                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
 762                              i > 0; i--) {
 763                                 tmp = NULL;
 764                                 if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
 765                                                           OCFS2_BH_READAHEAD))
 766                                         brelse(tmp);
 767                         }
 768                         last_ra_blk = blk;
 769                         ra_sectors = 8;
 770                 }
 771
 772 revalidate:
 773                 /* If the dir block has changed since the last call to
 774                  * readdir(2), then we might be pointing to an invalid
 775                  * dirent right now.  Scan from the start of the block
 776                  * to make sure. */
 777                 if (*f_version != inode->i_version) {
 778                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 779                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 780                                 /* It's too expensive to do a full
 781                                  * dirent test each time round this
 782                                  * loop, but we do have to test at
 783                                  * least that it is non-zero.  A
 784                                  * failure will be detected in the
 785                                  * dirent test below. */
 786                                 if (le16_to_cpu(de->rec_len) <
 787                                     OCFS2_DIR_REC_LEN(1))
 788                                         break;
 789                                 i += le16_to_cpu(de->rec_len);
 790                         }
 791                         offset = i;
 792                         *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
 793                                 | offset;
 794                         *f_version = inode->i_version;
 795                 }
 796
 797                 while (!error && *f_pos < i_size_read(inode)
 798                        && offset < sb->s_blocksize) {
 799                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 800                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 801                                 /* On error, skip the f_pos to the
 802                                    next block. */
 803                                 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
 804                                 brelse(bh);
 805                                 goto out;
 806                         }
 807                         offset += le16_to_cpu(de->rec_len);
 808                         if (le64_to_cpu(de->inode)) {
 809                                 /* We might block in the next section
 810                                  * if the data destination is
 811                                  * currently swapped out.  So, use a
 812                                  * version stamp to detect whether or
 813                                  * not the directory has been modified
 814                                  * during the copy operation.
 815                                  */
 816                                 unsigned long version = *f_version;
 817                                 unsigned char d_type = DT_UNKNOWN;
 818
 819                                 if (de->file_type < OCFS2_FT_MAX)
 820                                         d_type = ocfs2_filetype_table[de->file_type];
 821                                 error = filldir(priv, de->name,
 822                                                 de->name_len,
 823                                                 *f_pos,
 824                                                 le64_to_cpu(de->inode),
 825                                                 d_type);
 826                                 if (error) {
 827                                         if (filldir_err)
 828                                                 *filldir_err = error;
 829                                         break;
 830                                 }
 831                                 if (version != *f_version)
 832                                         goto revalidate;
 833                                 stored ++;
 834                         }
 835                         *f_pos += le16_to_cpu(de->rec_len);
 836                 }
 837                 offset = 0;
 838                 brelse(bh);
 839                 bh = NULL;
 840         }
 841
 842         stored = 0;
 843 out:
 844         return stored;
 845 }
 846
 847 static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
 848                                  loff_t *f_pos, void *priv, filldir_t filldir,
 849                                  int *filldir_err)
 850 {
 851         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
 852                 return ocfs2_dir_foreach_blk_id(inode, f_version, f_pos, priv,
 853                                                 filldir, filldir_err);
 854
 855         return ocfs2_dir_foreach_blk_el(inode, f_version, f_pos, priv, filldir,
 856                                         filldir_err);
 857 }
 858
 859 /*
 860  * This is intended to be called from inside other kernel functions,
 861  * so we fake some arguments.
 862  */
 863 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
 864                       filldir_t filldir)
 865 {
 866         int ret = 0, filldir_err = 0;
 867         u64 version = inode->i_version;
 868
 869         while (*f_pos < i_size_read(inode)) {
 870                 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
 871                                             filldir, &filldir_err);
 872                 if (ret || filldir_err)
 873                         break;
 874         }
 875
 876         if (ret > 0)
 877                 ret = -EIO;
 878
 879         return 0;
 880 }
 881
 882 /*
 883  * ocfs2_readdir()
 884  *
 885  */
 886 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
 887 {
 888         int error = 0;
 889         struct inode *inode = filp->f_path.dentry->d_inode;
 890         int lock_level = 0;
 891
 892         mlog_entry("dirino=%llu\n",
 893                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 894
 895         error = ocfs2_inode_lock_atime(inode, filp->f_vfsmnt, &lock_level);
 896         if (lock_level && error >= 0) {
 897                 /* We release EX lock which used to update atime
 898                  * and get PR lock again to reduce contention
 899                  * on commonly accessed directories. */
 900                 ocfs2_inode_unlock(inode, 1);
 901                 lock_level = 0;
 902                 error = ocfs2_inode_lock(inode, NULL, 0);
 903         }
 904         if (error < 0) {
 905                 if (error != -ENOENT)
 906                         mlog_errno(error);
 907                 /* we haven't got any yet, so propagate the error. */
 908                 goto bail_nolock;
 909         }
 910
 911         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
 912                                       dirent, filldir, NULL);
 913
 914         ocfs2_inode_unlock(inode, lock_level);
 915
 916 bail_nolock:
 917         mlog_exit(error);
 918
 919         return error;
 920 }
 921
 922 /*
 923  * NOTE: this should always be called with parent dir i_mutex taken.
 924  */
 925 int ocfs2_find_files_on_disk(const char *name,
 926                              int namelen,
 927                              u64 *blkno,
 928                              struct inode *inode,
 929                              struct buffer_head **dirent_bh,
 930                              struct ocfs2_dir_entry **dirent)
 931 {
 932         int status = -ENOENT;
 933
 934         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 935                    namelen, name, blkno, inode, dirent_bh, dirent);
 936
 937         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 938         if (!*dirent_bh || !*dirent) {
 939                 status = -ENOENT;
 940                 goto leave;
 941         }
 942
 943         *blkno = le64_to_cpu((*dirent)->inode);
 944
 945         status = 0;
 946 leave:
 947         if (status < 0) {
 948                 *dirent = NULL;
 949                 brelse(*dirent_bh);
 950                 *dirent_bh = NULL;
 951         }
 952
 953         mlog_exit(status);
 954         return status;
 955 }
 956
 957 /*
 958  * Convenience function for callers which just want the block number
 959  * mapped to a name and don't require the full dirent info, etc.
 960  */
 961 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
 962                                int namelen, u64 *blkno)
 963 {
 964         int ret;
 965         struct buffer_head *bh = NULL;
 966         struct ocfs2_dir_entry *dirent = NULL;
 967
 968         ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent);
 969         brelse(bh);
 970
 971         return ret;
 972 }
 973
 974 /* Check for a name within a directory.
 975  *
 976  * Return 0 if the name does not exist
 977  * Return -EEXIST if the directory contains the name
 978  *
 979  * Callers should have i_mutex + a cluster lock on dir
 980  */
 981 int ocfs2_check_dir_for_entry(struct inode *dir,
 982                               const char *name,
 983                               int namelen)
 984 {
 985         int ret;
 986         struct buffer_head *dirent_bh = NULL;
 987         struct ocfs2_dir_entry *dirent = NULL;
 988
 989         mlog_entry("dir %llu, name '%.*s'\n",
 990                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 991
 992         ret = -EEXIST;
 993         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 994         if (dirent_bh)
 995                 goto bail;
 996
 997         ret = 0;
 998 bail:
 999         brelse(dirent_bh);
1000
1001         mlog_exit(ret);
1002         return ret;
1003 }
1004
1005 struct ocfs2_empty_dir_priv {
1006         unsigned seen_dot;
1007         unsigned seen_dot_dot;
1008         unsigned seen_other;
1009 };
1010 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
1011                                    loff_t pos, u64 ino, unsigned type)
1012 {
1013         struct ocfs2_empty_dir_priv *p = priv;
1014
1015         /*
1016          * Check the positions of "." and ".." records to be sure
1017          * they're in the correct place.
1018          */
1019         if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
1020                 p->seen_dot = 1;
1021                 return 0;
1022         }
1023
1024         if (name_len == 2 && !strncmp("..", name, 2) &&
1025             pos == OCFS2_DIR_REC_LEN(1)) {
1026                 p->seen_dot_dot = 1;
1027                 return 0;
1028         }
1029
1030         p->seen_other = 1;
1031         return 1;
1032 }
1033 /*
1034  * routine to check that the specified directory is empty (for rmdir)
1035  *
1036  * Returns 1 if dir is empty, zero otherwise.
1037  */
1038 int ocfs2_empty_dir(struct inode *inode)
1039 {
1040         int ret;
1041         loff_t start = 0;
1042         struct ocfs2_empty_dir_priv priv;
1043
1044         memset(&priv, 0, sizeof(priv));
1045
1046         ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
1047         if (ret)
1048                 mlog_errno(ret);
1049
1050         if (!priv.seen_dot || !priv.seen_dot_dot) {
1051                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
1052                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
1053                 /*
1054                  * XXX: Is it really safe to allow an unlink to continue?
1055                  */
1056                 return 1;
1057         }
1058
1059         return !priv.seen_other;
1060 }
1061
1062 static void ocfs2_fill_initial_dirents(struct inode *inode,
1063                                        struct inode *parent,
1064                                        char *start, unsigned int size)
1065 {
1066         struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
1067
1068         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
1069         de->name_len = 1;
1070         de->rec_len =
1071                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1072         strcpy(de->name, ".");
1073         ocfs2_set_de_type(de, S_IFDIR);
1074
1075         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
1076         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
1077         de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
1078         de->name_len = 2;
1079         strcpy(de->name, "..");
1080         ocfs2_set_de_type(de, S_IFDIR);
1081 }
1082
1083 /*
1084  * This works together with code in ocfs2_mknod_locked() which sets
1085  * the inline-data flag and initializes the inline-data section.
1086  */
1087 static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
1088                                  handle_t *handle,
1089                                  struct inode *parent,
1090                                  struct inode *inode,
1091                                  struct buffer_head *di_bh)
1092 {
1093         int ret;
1094         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1095         struct ocfs2_inline_data *data = &di->id2.i_data;
1096         unsigned int size = le16_to_cpu(data->id_count);
1097
1098         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1099                                       OCFS2_JOURNAL_ACCESS_WRITE);
1100         if (ret) {
1101                 mlog_errno(ret);
1102                 goto out;
1103         }
1104
1105         ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
1106
1107         ocfs2_journal_dirty(handle, di_bh);
1108         if (ret) {
1109                 mlog_errno(ret);
1110                 goto out;
1111         }
1112
1113         i_size_write(inode, size);
1114         inode->i_nlink = 2;
1115         inode->i_blocks = ocfs2_inode_sector_count(inode);
1116
1117         ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
1118         if (ret < 0)
1119                 mlog_errno(ret);
1120
1121 out:
1122         return ret;
1123 }
1124
1125 static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
1126                                  handle_t *handle,
1127                                  struct inode *parent,
1128                                  struct inode *inode,
1129                                  struct buffer_head *fe_bh,
1130                                  struct ocfs2_alloc_context *data_ac)
1131 {
1132         int status;
1133         struct buffer_head *new_bh = NULL;
1134
1135         mlog_entry_void();
1136
1137         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
1138                                      data_ac, NULL, &new_bh);
1139         if (status < 0) {
1140                 mlog_errno(status);
1141                 goto bail;
1142         }
1143
1144         ocfs2_set_new_buffer_uptodate(inode, new_bh);
1145
1146         status = ocfs2_journal_access_db(handle, inode, new_bh,
1147                                          OCFS2_JOURNAL_ACCESS_CREATE);
1148         if (status < 0) {
1149                 mlog_errno(status);
1150                 goto bail;
1151         }
1152         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
1153
1154         ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data,
1155                                    osb->sb->s_blocksize);
1156
1157         status = ocfs2_journal_dirty(handle, new_bh);
1158         if (status < 0) {
1159                 mlog_errno(status);
1160                 goto bail;
1161         }
1162
1163         i_size_write(inode, inode->i_sb->s_blocksize);
1164         inode->i_nlink = 2;
1165         inode->i_blocks = ocfs2_inode_sector_count(inode);
1166         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
1167         if (status < 0) {
1168                 mlog_errno(status);
1169                 goto bail;
1170         }
1171
1172         status = 0;
1173 bail:
1174         brelse(new_bh);
1175
1176         mlog_exit(status);
1177         return status;
1178 }
1179
1180 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
1181                        handle_t *handle,
1182                        struct inode *parent,
1183                        struct inode *inode,
1184                        struct buffer_head *fe_bh,
1185                        struct ocfs2_alloc_context *data_ac)
1186 {
1187         BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
1188
1189         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
1190                 return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
1191
1192         return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
1193                                      data_ac);
1194 }
1195
1196 static void ocfs2_expand_last_dirent(char *start, unsigned int old_size,
1197                                      unsigned int new_size)
1198 {
1199         struct ocfs2_dir_entry *de;
1200         struct ocfs2_dir_entry *prev_de;
1201         char *de_buf, *limit;
1202         unsigned int bytes = new_size - old_size;
1203
1204         limit = start + old_size;
1205         de_buf = start;
1206         de = (struct ocfs2_dir_entry *)de_buf;
1207         do {
1208                 prev_de = de;
1209                 de_buf += le16_to_cpu(de->rec_len);
1210                 de = (struct ocfs2_dir_entry *)de_buf;
1211         } while (de_buf < limit);
1212
1213         le16_add_cpu(&prev_de->rec_len, bytes);
1214 }
1215
1216 /*
1217  * We allocate enough clusters to fulfill "blocks_wanted", but set
1218  * i_size to exactly one block. Ocfs2_extend_dir() will handle the
1219  * rest automatically for us.
1220  *
1221  * *first_block_bh is a pointer to the 1st data block allocated to the
1222  *  directory.
1223  */
1224 static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
1225                                    unsigned int blocks_wanted,
1226                                    struct buffer_head **first_block_bh)
1227 {
1228         u32 alloc, bit_off, len;
1229         struct super_block *sb = dir->i_sb;
1230         int ret, credits = ocfs2_inline_to_extents_credits(sb);
1231         u64 blkno, bytes = blocks_wanted << sb->s_blocksize_bits;
1232         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
1233         struct ocfs2_inode_info *oi = OCFS2_I(dir);
1234         struct ocfs2_alloc_context *data_ac;
1235         struct buffer_head *dirdata_bh = NULL;
1236         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1237         handle_t *handle;
1238         struct ocfs2_extent_tree et;
1239         int did_quota = 0;
1240
1241         ocfs2_init_dinode_extent_tree(&et, dir, di_bh);
1242
1243         alloc = ocfs2_clusters_for_bytes(sb, bytes);
1244
1245         /*
1246          * We should never need more than 2 clusters for this -
1247          * maximum dirent size is far less than one block. In fact,
1248          * the only time we'd need more than one cluster is if
1249          * blocksize == clustersize and the dirent won't fit in the
1250          * extra space that the expansion to a single block gives. As
1251          * of today, that only happens on 4k/4k file systems.
1252          */
1253         BUG_ON(alloc > 2);
1254
1255         ret = ocfs2_reserve_clusters(osb, alloc, &data_ac);
1256         if (ret) {
1257                 mlog_errno(ret);
1258                 goto out;
1259         }
1260
1261         down_write(&oi->ip_alloc_sem);
1262
1263         /*
1264          * Prepare for worst case allocation scenario of two separate
1265          * extents.
1266          */
1267         if (alloc == 2)
1268                 credits += OCFS2_SUBALLOC_ALLOC;
1269
1270         handle = ocfs2_start_trans(osb, credits);
1271         if (IS_ERR(handle)) {
1272                 ret = PTR_ERR(handle);
1273                 mlog_errno(ret);
1274                 goto out_sem;
1275         }
1276
1277         if (vfs_dq_alloc_space_nodirty(dir,
1278                                 ocfs2_clusters_to_bytes(osb->sb, alloc))) {
1279                 ret = -EDQUOT;
1280                 goto out_commit;
1281         }
1282         did_quota = 1;
1283         /*
1284          * Try to claim as many clusters as the bitmap can give though
1285          * if we only get one now, that's enough to continue. The rest
1286          * will be claimed after the conversion to extents.
1287          */
1288         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
1289         if (ret) {
1290                 mlog_errno(ret);
1291                 goto out_commit;
1292         }
1293
1294         /*
1295          * Operations are carefully ordered so that we set up the new
1296          * data block first. The conversion from inline data to
1297          * extents follows.
1298          */
1299         blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
1300         dirdata_bh = sb_getblk(sb, blkno);
1301         if (!dirdata_bh) {
1302                 ret = -EIO;
1303                 mlog_errno(ret);
1304                 goto out_commit;
1305         }
1306
1307         ocfs2_set_new_buffer_uptodate(dir, dirdata_bh);
1308
1309         ret = ocfs2_journal_access_db(handle, dir, dirdata_bh,
1310                                       OCFS2_JOURNAL_ACCESS_CREATE);
1311         if (ret) {
1312                 mlog_errno(ret);
1313                 goto out_commit;
1314         }
1315
1316         memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
1317         memset(dirdata_bh->b_data + i_size_read(dir), 0,
1318                sb->s_blocksize - i_size_read(dir));
1319         ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir),
1320                                  sb->s_blocksize);
1321
1322         ret = ocfs2_journal_dirty(handle, dirdata_bh);
1323         if (ret) {
1324                 mlog_errno(ret);
1325                 goto out_commit;
1326         }
1327
1328         /*
1329          * Set extent, i_size, etc on the directory. After this, the
1330          * inode should contain the same exact dirents as before and
1331          * be fully accessible from system calls.
1332          *
1333          * We let the later dirent insert modify c/mtime - to the user
1334          * the data hasn't changed.
1335          */
1336         ret = ocfs2_journal_access_di(handle, dir, di_bh,
1337                                       OCFS2_JOURNAL_ACCESS_CREATE);
1338         if (ret) {
1339                 mlog_errno(ret);
1340                 goto out_commit;
1341         }
1342
1343         spin_lock(&oi->ip_lock);
1344         oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
1345         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1346         spin_unlock(&oi->ip_lock);
1347
1348         ocfs2_dinode_new_extent_list(dir, di);
1349
1350         i_size_write(dir, sb->s_blocksize);
1351         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1352
1353         di->i_size = cpu_to_le64(sb->s_blocksize);
1354         di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
1355         di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
1356
1357         /*
1358          * This should never fail as our extent list is empty and all
1359          * related blocks have been journaled already.
1360          */
1361         ret = ocfs2_insert_extent(osb, handle, dir, &et, 0, blkno, len,
1362                                   0, NULL);
1363         if (ret) {
1364                 mlog_errno(ret);
1365                 goto out_commit;
1366         }
1367
1368         /*
1369          * Set i_blocks after the extent insert for the most up to
1370          * date ip_clusters value.
1371          */
1372         dir->i_blocks = ocfs2_inode_sector_count(dir);
1373
1374         ret = ocfs2_journal_dirty(handle, di_bh);
1375         if (ret) {
1376                 mlog_errno(ret);
1377                 goto out_commit;
1378         }
1379
1380         /*
1381          * We asked for two clusters, but only got one in the 1st
1382          * pass. Claim the 2nd cluster as a separate extent.
1383          */
1384         if (alloc > len) {
1385                 ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off,
1386                                            &len);
1387                 if (ret) {
1388                         mlog_errno(ret);
1389                         goto out_commit;
1390                 }
1391                 blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
1392
1393                 ret = ocfs2_insert_extent(osb, handle, dir, &et, 1,
1394                                           blkno, len, 0, NULL);
1395                 if (ret) {
1396                         mlog_errno(ret);
1397                         goto out_commit;
1398                 }
1399         }
1400
1401         *first_block_bh = dirdata_bh;
1402         dirdata_bh = NULL;
1403
1404 out_commit:
1405         if (ret < 0 && did_quota)
1406                 vfs_dq_free_space_nodirty(dir,
1407                         ocfs2_clusters_to_bytes(osb->sb, 2));
1408         ocfs2_commit_trans(osb, handle);
1409
1410 out_sem:
1411         up_write(&oi->ip_alloc_sem);
1412
1413 out:
1414         if (data_ac)
1415                 ocfs2_free_alloc_context(data_ac);
1416
1417         brelse(dirdata_bh);
1418
1419         return ret;
1420 }
1421
1422 /* returns a bh of the 1st new block in the allocation. */
1423 static int ocfs2_do_extend_dir(struct super_block *sb,
1424                                handle_t *handle,
1425                                struct inode *dir,
1426                                struct buffer_head *parent_fe_bh,
1427                                struct ocfs2_alloc_context *data_ac,
1428                                struct ocfs2_alloc_context *meta_ac,
1429                                struct buffer_head **new_bh)
1430 {
1431         int status;
1432         int extend, did_quota = 0;
1433         u64 p_blkno, v_blkno;
1434
1435         spin_lock(&OCFS2_I(dir)->ip_lock);
1436         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
1437         spin_unlock(&OCFS2_I(dir)->ip_lock);
1438
1439         if (extend) {
1440                 u32 offset = OCFS2_I(dir)->ip_clusters;
1441
1442                 if (vfs_dq_alloc_space_nodirty(dir,
1443                                         ocfs2_clusters_to_bytes(sb, 1))) {
1444                         status = -EDQUOT;
1445                         goto bail;
1446                 }
1447                 did_quota = 1;
1448
1449                 status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
1450                                               1, 0, parent_fe_bh, handle,
1451                                               data_ac, meta_ac, NULL);
1452                 BUG_ON(status == -EAGAIN);
1453                 if (status < 0) {
1454                         mlog_errno(status);
1455                         goto bail;
1456                 }
1457         }
1458
1459         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
1460         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
1461         if (status < 0) {
1462                 mlog_errno(status);
1463                 goto bail;
1464         }
1465
1466         *new_bh = sb_getblk(sb, p_blkno);
1467         if (!*new_bh) {
1468                 status = -EIO;
1469                 mlog_errno(status);
1470                 goto bail;
1471         }
1472         status = 0;
1473 bail:
1474         if (did_quota && status < 0)
1475                 vfs_dq_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
1476         mlog_exit(status);
1477         return status;
1478 }
1479
1480 /*
1481  * Assumes you already have a cluster lock on the directory.
1482  *
1483  * 'blocks_wanted' is only used if we have an inline directory which
1484  * is to be turned into an extent based one. The size of the dirent to
1485  * insert might be larger than the space gained by growing to just one
1486  * block, so we may have to grow the inode by two blocks in that case.
1487  */
1488 static int ocfs2_extend_dir(struct ocfs2_super *osb,
1489                             struct inode *dir,
1490                             struct buffer_head *parent_fe_bh,
1491                             unsigned int blocks_wanted,
1492                             struct buffer_head **new_de_bh)
1493 {
1494         int status = 0;
1495         int credits, num_free_extents, drop_alloc_sem = 0;
1496         loff_t dir_i_size;
1497         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1498         struct ocfs2_extent_list *el = &fe->id2.i_list;
1499         struct ocfs2_alloc_context *data_ac = NULL;
1500         struct ocfs2_alloc_context *meta_ac = NULL;
1501         handle_t *handle = NULL;
1502         struct buffer_head *new_bh = NULL;
1503         struct ocfs2_dir_entry * de;
1504         struct super_block *sb = osb->sb;
1505         struct ocfs2_extent_tree et;
1506
1507         mlog_entry_void();
1508
1509         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1510                 status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
1511                                                  blocks_wanted, &new_bh);
1512                 if (status) {
1513                         mlog_errno(status);
1514                         goto bail;
1515                 }
1516
1517                 if (blocks_wanted == 1) {
1518                         /*
1519                          * If the new dirent will fit inside the space
1520                          * created by pushing out to one block, then
1521                          * we can complete the operation
1522                          * here. Otherwise we have to expand i_size
1523                          * and format the 2nd block below.
1524                          */
1525                         BUG_ON(new_bh == NULL);
1526                         goto bail_bh;
1527                 }
1528
1529                 /*
1530                  * Get rid of 'new_bh' - we want to format the 2nd
1531                  * data block and return that instead.
1532                  */
1533                 brelse(new_bh);
1534                 new_bh = NULL;
1535
1536                 dir_i_size = i_size_read(dir);
1537                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
1538                 goto do_extend;
1539         }
1540
1541         dir_i_size = i_size_read(dir);
1542         mlog(0, "extending dir %llu (i_size = %lld)\n",
1543              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
1544
1545         /* dir->i_size is always block aligned. */
1546         spin_lock(&OCFS2_I(dir)->ip_lock);
1547         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
1548                 spin_unlock(&OCFS2_I(dir)->ip_lock);
1549                 ocfs2_init_dinode_extent_tree(&et, dir, parent_fe_bh);
1550                 num_free_extents = ocfs2_num_free_extents(osb, dir, &et);
1551                 if (num_free_extents < 0) {
1552                         status = num_free_extents;
1553                         mlog_errno(status);
1554                         goto bail;
1555                 }
1556
1557                 if (!num_free_extents) {
1558                         status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
1559                         if (status < 0) {
1560                                 if (status != -ENOSPC)
1561                                         mlog_errno(status);
1562                                 goto bail;
1563                         }
1564                 }
1565
1566                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1567                 if (status < 0) {
1568                         if (status != -ENOSPC)
1569                                 mlog_errno(status);
1570                         goto bail;
1571                 }
1572
1573                 credits = ocfs2_calc_extend_credits(sb, el, 1);
1574         } else {
1575                 spin_unlock(&OCFS2_I(dir)->ip_lock);
1576                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
1577         }
1578
1579 do_extend:
1580         down_write(&OCFS2_I(dir)->ip_alloc_sem);
1581         drop_alloc_sem = 1;
1582
1583         handle = ocfs2_start_trans(osb, credits);
1584         if (IS_ERR(handle)) {
1585                 status = PTR_ERR(handle);
1586                 handle = NULL;
1587                 mlog_errno(status);
1588                 goto bail;
1589         }
1590
1591         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
1592                                      data_ac, meta_ac, &new_bh);
1593         if (status < 0) {
1594                 mlog_errno(status);
1595                 goto bail;
1596         }
1597
1598         ocfs2_set_new_buffer_uptodate(dir, new_bh);
1599
1600         status = ocfs2_journal_access_db(handle, dir, new_bh,
1601                                          OCFS2_JOURNAL_ACCESS_CREATE);
1602         if (status < 0) {
1603                 mlog_errno(status);
1604                 goto bail;
1605         }
1606         memset(new_bh->b_data, 0, sb->s_blocksize);
1607         de = (struct ocfs2_dir_entry *) new_bh->b_data;
1608         de->inode = 0;
1609         de->rec_len = cpu_to_le16(sb->s_blocksize);
1610         status = ocfs2_journal_dirty(handle, new_bh);
1611         if (status < 0) {
1612                 mlog_errno(status);
1613                 goto bail;
1614         }
1615
1616         dir_i_size += dir->i_sb->s_blocksize;
1617         i_size_write(dir, dir_i_size);
1618         dir->i_blocks = ocfs2_inode_sector_count(dir);
1619         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1620         if (status < 0) {
1621                 mlog_errno(status);
1622                 goto bail;
1623         }
1624
1625 bail_bh:
1626         *new_de_bh = new_bh;
1627         get_bh(*new_de_bh);
1628 bail:
1629         if (drop_alloc_sem)
1630                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
1631         if (handle)
1632                 ocfs2_commit_trans(osb, handle);
1633
1634         if (data_ac)
1635                 ocfs2_free_alloc_context(data_ac);
1636         if (meta_ac)
1637                 ocfs2_free_alloc_context(meta_ac);
1638
1639         brelse(new_bh);
1640
1641         mlog_exit(status);
1642         return status;
1643 }
1644
1645 static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
1646                                    const char *name, int namelen,
1647                                    struct buffer_head **ret_de_bh,
1648                                    unsigned int *blocks_wanted)
1649 {
1650         int ret;
1651         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1652         struct ocfs2_dir_entry *de, *last_de = NULL;
1653         char *de_buf, *limit;
1654         unsigned long offset = 0;
1655         unsigned int rec_len, new_rec_len;
1656
1657         de_buf = di->id2.i_data.id_data;
1658         limit = de_buf + i_size_read(dir);
1659         rec_len = OCFS2_DIR_REC_LEN(namelen);
1660
1661         while (de_buf < limit) {
1662                 de = (struct ocfs2_dir_entry *)de_buf;
1663
1664                 if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
1665                         ret = -ENOENT;
1666                         goto out;
1667                 }
1668                 if (ocfs2_match(namelen, name, de)) {
1669                         ret = -EEXIST;
1670                         goto out;
1671                 }
1672                 if (ocfs2_dirent_would_fit(de, rec_len)) {
1673                         /* Ok, we found a spot. Return this bh and let
1674                          * the caller actually fill it in. */
1675                         *ret_de_bh = di_bh;
1676                         get_bh(*ret_de_bh);
1677                         ret = 0;
1678                         goto out;
1679                 }
1680
1681                 last_de = de;
1682                 de_buf += le16_to_cpu(de->rec_len);
1683                 offset += le16_to_cpu(de->rec_len);
1684         }
1685
1686         /*
1687          * We're going to require expansion of the directory - figure
1688          * out how many blocks we'll need so that a place for the
1689          * dirent can be found.
1690          */
1691         *blocks_wanted = 1;
1692         new_rec_len = le16_to_cpu(last_de->rec_len) + (dir->i_sb->s_blocksize - i_size_read(dir));
1693         if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
1694                 *blocks_wanted = 2;
1695
1696         ret = -ENOSPC;
1697 out:
1698         return ret;
1699 }
1700
1701 static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
1702                                    int namelen, struct buffer_head **ret_de_bh)
1703 {
1704         unsigned long offset;
1705         struct buffer_head *bh = NULL;
1706         unsigned short rec_len;
1707         struct ocfs2_dir_entry *de;
1708         struct super_block *sb = dir->i_sb;
1709         int status;
1710
1711         status = ocfs2_read_dir_block(dir, 0, &bh, 0);
1712         if (status) {
1713                 mlog_errno(status);
1714                 goto bail;
1715         }
1716
1717         rec_len = OCFS2_DIR_REC_LEN(namelen);
1718         offset = 0;
1719         de = (struct ocfs2_dir_entry *) bh->b_data;
1720         while (1) {
1721                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
1722                         brelse(bh);
1723                         bh = NULL;
1724
1725                         if (i_size_read(dir) <= offset) {
1726                                 /*
1727                                  * Caller will have to expand this
1728                                  * directory.
1729                                  */
1730                                 status = -ENOSPC;
1731                                 goto bail;
1732                         }
1733                         status = ocfs2_read_dir_block(dir,
1734                                              offset >> sb->s_blocksize_bits,
1735                                              &bh, 0);
1736                         if (status) {
1737                                 mlog_errno(status);
1738                                 goto bail;
1739                         }
1740                         /* move to next block */
1741                         de = (struct ocfs2_dir_entry *) bh->b_data;
1742                 }
1743                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1744                         status = -ENOENT;
1745                         goto bail;
1746                 }
1747                 if (ocfs2_match(namelen, name, de)) {
1748                         status = -EEXIST;
1749                         goto bail;
1750                 }
1751                 if (ocfs2_dirent_would_fit(de, rec_len)) {
1752                         /* Ok, we found a spot. Return this bh and let
1753                          * the caller actually fill it in. */
1754                         *ret_de_bh = bh;
1755                         get_bh(*ret_de_bh);
1756                         status = 0;
1757                         goto bail;
1758                 }
1759                 offset += le16_to_cpu(de->rec_len);
1760                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1761         }
1762
1763         status = 0;
1764 bail:
1765         brelse(bh);
1766
1767         mlog_exit(status);
1768         return status;
1769 }
1770
1771 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
1772                                  struct inode *dir,
1773                                  struct buffer_head *parent_fe_bh,
1774                                  const char *name,
1775                                  int namelen,
1776                                  struct buffer_head **ret_de_bh)
1777 {
1778         int ret;
1779         unsigned int blocks_wanted = 1;
1780         struct buffer_head *bh = NULL;
1781
1782         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
1783              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
1784
1785         *ret_de_bh = NULL;
1786
1787         if (!namelen) {
1788                 ret = -EINVAL;
1789                 mlog_errno(ret);
1790                 goto out;
1791         }
1792
1793         if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1794                 ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
1795                                               namelen, &bh, &blocks_wanted);
1796         } else
1797                 ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
1798
1799         if (ret && ret != -ENOSPC) {
1800                 mlog_errno(ret);
1801                 goto out;
1802         }
1803
1804         if (ret == -ENOSPC) {
1805                 /*
1806                  * We have to expand the directory to add this name.
1807                  */
1808                 BUG_ON(bh);
1809
1810                 ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
1811                                        &bh);
1812                 if (ret) {
1813                         if (ret != -ENOSPC)
1814                                 mlog_errno(ret);
1815                         goto out;
1816                 }
1817
1818                 BUG_ON(!bh);
1819         }
1820
1821         *ret_de_bh = bh;
1822         bh = NULL;
1823 out:
1824         brelse(bh);
1825         return ret;
1826 }