err.no Git - linux-2.6/blob - 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
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dir.h"
  51 #include "dlmglue.h"
  52 #include "extent_map.h"
  53 #include "file.h"
  54 #include "inode.h"
  55 #include "journal.h"
  56 #include "namei.h"
  57 #include "suballoc.h"
  58 #include "super.h"
  59 #include "uptodate.h"
  60
  61 #include "buffer_head_io.h"
  62
  63 #define NAMEI_RA_CHUNKS  2
  64 #define NAMEI_RA_BLOCKS  4
  65 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  66 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  67
  68 static unsigned char ocfs2_filetype_table[] = {
  69         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  70 };
  71
  72 static int ocfs2_extend_dir(struct ocfs2_super *osb,
  73                             struct inode *dir,
  74                             struct buffer_head *parent_fe_bh,
  75                             struct buffer_head **new_de_bh);
  76 static int ocfs2_do_extend_dir(struct super_block *sb,
  77                                handle_t *handle,
  78                                struct inode *dir,
  79                                struct buffer_head *parent_fe_bh,
  80                                struct ocfs2_alloc_context *data_ac,
  81                                struct ocfs2_alloc_context *meta_ac,
  82                                struct buffer_head **new_bh);
  83
  84 int ocfs2_check_dir_entry(struct inode * dir,
  85                           struct ocfs2_dir_entry * de,
  86                           struct buffer_head * bh,
  87                           unsigned long offset)
  88 {
  89         const char *error_msg = NULL;
  90         const int rlen = le16_to_cpu(de->rec_len);
  91
  92         if (rlen < OCFS2_DIR_REC_LEN(1))
  93                 error_msg = "rec_len is smaller than minimal";
  94         else if (rlen % 4 != 0)
  95                 error_msg = "rec_len % 4 != 0";
  96         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
  97                 error_msg = "rec_len is too small for name_len";
  98         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
  99                 error_msg = "directory entry across blocks";
 100
 101         if (error_msg != NULL)
 102                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
 103                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
 104                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
 105                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
 106                      de->name_len);
 107         return error_msg == NULL ? 1 : 0;
 108 }
 109
 110 static inline int ocfs2_match(int len,
 111                               const char * const name,
 112                               struct ocfs2_dir_entry *de)
 113 {
 114         if (len != de->name_len)
 115                 return 0;
 116         if (!de->inode)
 117                 return 0;
 118         return !memcmp(name, de->name, len);
 119 }
 120
 121 /*
 122  * Returns 0 if not found, -1 on failure, and 1 on success
 123  */
 124 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
 125                                         struct inode *dir,
 126                                         const char *name, int namelen,
 127                                         unsigned long offset,
 128                                         struct ocfs2_dir_entry **res_dir)
 129 {
 130         struct ocfs2_dir_entry *de;
 131         char *dlimit, *de_buf;
 132         int de_len;
 133         int ret = 0;
 134
 135         mlog_entry_void();
 136
 137         de_buf = bh->b_data;
 138         dlimit = de_buf + dir->i_sb->s_blocksize;
 139
 140         while (de_buf < dlimit) {
 141                 /* this code is executed quadratically often */
 142                 /* do minimal checking `by hand' */
 143
 144                 de = (struct ocfs2_dir_entry *) de_buf;
 145
 146                 if (de_buf + namelen <= dlimit &&
 147                     ocfs2_match(namelen, name, de)) {
 148                         /* found a match - just to be sure, do a full check */
 149                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
 150                                 ret = -1;
 151                                 goto bail;
 152                         }
 153                         *res_dir = de;
 154                         ret = 1;
 155                         goto bail;
 156                 }
 157
 158                 /* prevent looping on a bad block */
 159                 de_len = le16_to_cpu(de->rec_len);
 160                 if (de_len <= 0) {
 161                         ret = -1;
 162                         goto bail;
 163                 }
 164
 165                 de_buf += de_len;
 166                 offset += de_len;
 167         }
 168
 169 bail:
 170         mlog_exit(ret);
 171         return ret;
 172 }
 173
 174 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
 175                                      struct inode *dir,
 176                                      struct ocfs2_dir_entry **res_dir)
 177 {
 178         struct super_block *sb;
 179         struct buffer_head *bh_use[NAMEI_RA_SIZE];
 180         struct buffer_head *bh, *ret = NULL;
 181         unsigned long start, block, b;
 182         int ra_max = 0;         /* Number of bh's in the readahead
 183                                    buffer, bh_use[] */
 184         int ra_ptr = 0;         /* Current index into readahead
 185                                    buffer */
 186         int num = 0;
 187         int nblocks, i, err;
 188
 189         mlog_entry_void();
 190
 191         *res_dir = NULL;
 192         sb = dir->i_sb;
 193
 194         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 195         start = OCFS2_I(dir)->ip_dir_start_lookup;
 196         if (start >= nblocks)
 197                 start = 0;
 198         block = start;
 199
 200 restart:
 201         do {
 202                 /*
 203                  * We deal with the read-ahead logic here.
 204                  */
 205                 if (ra_ptr >= ra_max) {
 206                         /* Refill the readahead buffer */
 207                         ra_ptr = 0;
 208                         b = block;
 209                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
 210                                 /*
 211                                  * Terminate if we reach the end of the
 212                                  * directory and must wrap, or if our
 213                                  * search has finished at this block.
 214                                  */
 215                                 if (b >= nblocks || (num && block == start)) {
 216                                         bh_use[ra_max] = NULL;
 217                                         break;
 218                                 }
 219                                 num++;
 220
 221                                 bh = ocfs2_bread(dir, b++, &err, 1);
 222                                 bh_use[ra_max] = bh;
 223                         }
 224                 }
 225                 if ((bh = bh_use[ra_ptr++]) == NULL)
 226                         goto next;
 227                 wait_on_buffer(bh);
 228                 if (!buffer_uptodate(bh)) {
 229                         /* read error, skip block & hope for the best */
 230                         ocfs2_error(dir->i_sb, "reading directory %llu, "
 231                                     "offset %lu\n",
 232                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
 233                                     block);
 234                         brelse(bh);
 235                         goto next;
 236                 }
 237                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
 238                                           block << sb->s_blocksize_bits,
 239                                           res_dir);
 240                 if (i == 1) {
 241                         OCFS2_I(dir)->ip_dir_start_lookup = block;
 242                         ret = bh;
 243                         goto cleanup_and_exit;
 244                 } else {
 245                         brelse(bh);
 246                         if (i < 0)
 247                                 goto cleanup_and_exit;
 248                 }
 249         next:
 250                 if (++block >= nblocks)
 251                         block = 0;
 252         } while (block != start);
 253
 254         /*
 255          * If the directory has grown while we were searching, then
 256          * search the last part of the directory before giving up.
 257          */
 258         block = nblocks;
 259         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
 260         if (block < nblocks) {
 261                 start = 0;
 262                 goto restart;
 263         }
 264
 265 cleanup_and_exit:
 266         /* Clean up the read-ahead blocks */
 267         for (; ra_ptr < ra_max; ra_ptr++)
 268                 brelse(bh_use[ra_ptr]);
 269
 270         mlog_exit_ptr(ret);
 271         return ret;
 272 }
 273
 274 /*
 275  * ocfs2_delete_entry deletes a directory entry by merging it with the
 276  * previous entry
 277  */
 278 int ocfs2_delete_entry(handle_t *handle,
 279                        struct inode *dir,
 280                        struct ocfs2_dir_entry *de_del,
 281                        struct buffer_head *bh)
 282 {
 283         struct ocfs2_dir_entry *de, *pde;
 284         int i, status = -ENOENT;
 285
 286         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
 287
 288         i = 0;
 289         pde = NULL;
 290         de = (struct ocfs2_dir_entry *) bh->b_data;
 291         while (i < bh->b_size) {
 292                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
 293                         status = -EIO;
 294                         mlog_errno(status);
 295                         goto bail;
 296                 }
 297                 if (de == de_del)  {
 298                         status = ocfs2_journal_access(handle, dir, bh,
 299                                                       OCFS2_JOURNAL_ACCESS_WRITE);
 300                         if (status < 0) {
 301                                 status = -EIO;
 302                                 mlog_errno(status);
 303                                 goto bail;
 304                         }
 305                         if (pde)
 306                                 pde->rec_len =
 307                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 308                                                     le16_to_cpu(de->rec_len));
 309                         else
 310                                 de->inode = 0;
 311                         dir->i_version++;
 312                         status = ocfs2_journal_dirty(handle, bh);
 313                         goto bail;
 314                 }
 315                 i += le16_to_cpu(de->rec_len);
 316                 pde = de;
 317                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
 318         }
 319 bail:
 320         mlog_exit(status);
 321         return status;
 322 }
 323
 324 /* we don't always have a dentry for what we want to add, so people
 325  * like orphan dir can call this instead.
 326  *
 327  * If you pass me insert_bh, I'll skip the search of the other dir
 328  * blocks and put the record in there.
 329  */
 330 int __ocfs2_add_entry(handle_t *handle,
 331                       struct inode *dir,
 332                       const char *name, int namelen,
 333                       struct inode *inode, u64 blkno,
 334                       struct buffer_head *parent_fe_bh,
 335                       struct buffer_head *insert_bh)
 336 {
 337         unsigned long offset;
 338         unsigned short rec_len;
 339         struct ocfs2_dir_entry *de, *de1;
 340         struct super_block *sb;
 341         int retval, status;
 342
 343         mlog_entry_void();
 344
 345         sb = dir->i_sb;
 346
 347         if (!namelen)
 348                 return -EINVAL;
 349
 350         rec_len = OCFS2_DIR_REC_LEN(namelen);
 351         offset = 0;
 352         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
 353         while (1) {
 354                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
 355                 /* These checks should've already been passed by the
 356                  * prepare function, but I guess we can leave them
 357                  * here anyway. */
 358                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
 359                         retval = -ENOENT;
 360                         goto bail;
 361                 }
 362                 if (ocfs2_match(namelen, name, de)) {
 363                         retval = -EEXIST;
 364                         goto bail;
 365                 }
 366                 if (((le64_to_cpu(de->inode) == 0) &&
 367                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
 368                     (le16_to_cpu(de->rec_len) >=
 369                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
 370                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 371                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 372                         if (retval < 0) {
 373                                 mlog_errno(retval);
 374                                 goto bail;
 375                         }
 376
 377                         status = ocfs2_journal_access(handle, dir, insert_bh,
 378                                                       OCFS2_JOURNAL_ACCESS_WRITE);
 379                         /* By now the buffer is marked for journaling */
 380                         offset += le16_to_cpu(de->rec_len);
 381                         if (le64_to_cpu(de->inode)) {
 382                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
 383                                         OCFS2_DIR_REC_LEN(de->name_len));
 384                                 de1->rec_len =
 385                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
 386                                         OCFS2_DIR_REC_LEN(de->name_len));
 387                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 388                                 de = de1;
 389                         }
 390                         de->file_type = OCFS2_FT_UNKNOWN;
 391                         if (blkno) {
 392                                 de->inode = cpu_to_le64(blkno);
 393                                 ocfs2_set_de_type(de, inode->i_mode);
 394                         } else
 395                                 de->inode = 0;
 396                         de->name_len = namelen;
 397                         memcpy(de->name, name, namelen);
 398
 399                         dir->i_version++;
 400                         status = ocfs2_journal_dirty(handle, insert_bh);
 401                         retval = 0;
 402                         goto bail;
 403                 }
 404                 offset += le16_to_cpu(de->rec_len);
 405                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
 406         }
 407
 408         /* when you think about it, the assert above should prevent us
 409          * from ever getting here. */
 410         retval = -ENOSPC;
 411 bail:
 412
 413         mlog_exit(retval);
 414         return retval;
 415 }
 416
 417 /*
 418  * ocfs2_readdir()
 419  *
 420  */
 421 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
 422 {
 423         int error = 0;
 424         unsigned long offset, blk, last_ra_blk = 0;
 425         int i, stored;
 426         struct buffer_head * bh, * tmp;
 427         struct ocfs2_dir_entry * de;
 428         int err;
 429         struct inode *inode = filp->f_path.dentry->d_inode;
 430         struct super_block * sb = inode->i_sb;
 431         unsigned int ra_sectors = 16;
 432         int lock_level = 0;
 433
 434         mlog_entry("dirino=%llu\n",
 435                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 436
 437         stored = 0;
 438         bh = NULL;
 439
 440         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
 441         if (lock_level && error >= 0) {
 442                 /* We release EX lock which used to update atime
 443                  * and get PR lock again to reduce contention
 444                  * on commonly accessed directories. */
 445                 ocfs2_meta_unlock(inode, 1);
 446                 lock_level = 0;
 447                 error = ocfs2_meta_lock(inode, NULL, 0);
 448         }
 449         if (error < 0) {
 450                 if (error != -ENOENT)
 451                         mlog_errno(error);
 452                 /* we haven't got any yet, so propagate the error. */
 453                 stored = error;
 454                 goto bail_nolock;
 455         }
 456
 457         offset = filp->f_pos & (sb->s_blocksize - 1);
 458
 459         while (!error && !stored && filp->f_pos < i_size_read(inode)) {
 460                 blk = (filp->f_pos) >> sb->s_blocksize_bits;
 461                 bh = ocfs2_bread(inode, blk, &err, 0);
 462                 if (!bh) {
 463                         mlog(ML_ERROR,
 464                              "directory #%llu contains a hole at offset %lld\n",
 465                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 466                              filp->f_pos);
 467                         filp->f_pos += sb->s_blocksize - offset;
 468                         continue;
 469                 }
 470
 471                 /* The idea here is to begin with 8k read-ahead and to stay
 472                  * 4k ahead of our current position.
 473                  *
 474                  * TODO: Use the pagecache for this. We just need to
 475                  * make sure it's cluster-safe... */
 476                 if (!last_ra_blk
 477                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
 478                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
 479                              i > 0; i--) {
 480                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
 481                                 if (tmp)
 482                                         brelse(tmp);
 483                         }
 484                         last_ra_blk = blk;
 485                         ra_sectors = 8;
 486                 }
 487
 488 revalidate:
 489                 /* If the dir block has changed since the last call to
 490                  * readdir(2), then we might be pointing to an invalid
 491                  * dirent right now.  Scan from the start of the block
 492                  * to make sure. */
 493                 if (filp->f_version != inode->i_version) {
 494                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 495                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 496                                 /* It's too expensive to do a full
 497                                  * dirent test each time round this
 498                                  * loop, but we do have to test at
 499                                  * least that it is non-zero.  A
 500                                  * failure will be detected in the
 501                                  * dirent test below. */
 502                                 if (le16_to_cpu(de->rec_len) <
 503                                     OCFS2_DIR_REC_LEN(1))
 504                                         break;
 505                                 i += le16_to_cpu(de->rec_len);
 506                         }
 507                         offset = i;
 508                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 509                                 | offset;
 510                         filp->f_version = inode->i_version;
 511                 }
 512
 513                 while (!error && filp->f_pos < i_size_read(inode)
 514                        && offset < sb->s_blocksize) {
 515                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 516                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 517                                 /* On error, skip the f_pos to the
 518                                    next block. */
 519                                 filp->f_pos = (filp->f_pos |
 520                                                (sb->s_blocksize - 1)) + 1;
 521                                 brelse(bh);
 522                                 goto bail;
 523                         }
 524                         offset += le16_to_cpu(de->rec_len);
 525                         if (le64_to_cpu(de->inode)) {
 526                                 /* We might block in the next section
 527                                  * if the data destination is
 528                                  * currently swapped out.  So, use a
 529                                  * version stamp to detect whether or
 530                                  * not the directory has been modified
 531                                  * during the copy operation.
 532                                  */
 533                                 unsigned long version = filp->f_version;
 534                                 unsigned char d_type = DT_UNKNOWN;
 535
 536                                 if (de->file_type < OCFS2_FT_MAX)
 537                                         d_type = ocfs2_filetype_table[de->file_type];
 538                                 error = filldir(dirent, de->name,
 539                                                 de->name_len,
 540                                                 filp->f_pos,
 541                                                 ino_from_blkno(sb, le64_to_cpu(de->inode)),
 542                                                 d_type);
 543                                 if (error)
 544                                         break;
 545                                 if (version != filp->f_version)
 546                                         goto revalidate;
 547                                 stored ++;
 548                         }
 549                         filp->f_pos += le16_to_cpu(de->rec_len);
 550                 }
 551                 offset = 0;
 552                 brelse(bh);
 553         }
 554
 555         stored = 0;
 556 bail:
 557         ocfs2_meta_unlock(inode, lock_level);
 558
 559 bail_nolock:
 560         mlog_exit(stored);
 561
 562         return stored;
 563 }
 564
 565 /*
 566  * NOTE: this should always be called with parent dir i_mutex taken.
 567  */
 568 int ocfs2_find_files_on_disk(const char *name,
 569                              int namelen,
 570                              u64 *blkno,
 571                              struct inode *inode,
 572                              struct buffer_head **dirent_bh,
 573                              struct ocfs2_dir_entry **dirent)
 574 {
 575         int status = -ENOENT;
 576
 577         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 578                    namelen, name, blkno, inode, dirent_bh, dirent);
 579
 580         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 581         if (!*dirent_bh || !*dirent) {
 582                 status = -ENOENT;
 583                 goto leave;
 584         }
 585
 586         *blkno = le64_to_cpu((*dirent)->inode);
 587
 588         status = 0;
 589 leave:
 590         if (status < 0) {
 591                 *dirent = NULL;
 592                 if (*dirent_bh) {
 593                         brelse(*dirent_bh);
 594                         *dirent_bh = NULL;
 595                 }
 596         }
 597
 598         mlog_exit(status);
 599         return status;
 600 }
 601
 602 /* Check for a name within a directory.
 603  *
 604  * Return 0 if the name does not exist
 605  * Return -EEXIST if the directory contains the name
 606  *
 607  * Callers should have i_mutex + a cluster lock on dir
 608  */
 609 int ocfs2_check_dir_for_entry(struct inode *dir,
 610                               const char *name,
 611                               int namelen)
 612 {
 613         int ret;
 614         struct buffer_head *dirent_bh = NULL;
 615         struct ocfs2_dir_entry *dirent = NULL;
 616
 617         mlog_entry("dir %llu, name '%.*s'\n",
 618                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 619
 620         ret = -EEXIST;
 621         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 622         if (dirent_bh)
 623                 goto bail;
 624
 625         ret = 0;
 626 bail:
 627         if (dirent_bh)
 628                 brelse(dirent_bh);
 629
 630         mlog_exit(ret);
 631         return ret;
 632 }
 633
 634 /*
 635  * routine to check that the specified directory is empty (for rmdir)
 636  */
 637 int ocfs2_empty_dir(struct inode *inode)
 638 {
 639         unsigned long offset;
 640         struct buffer_head * bh;
 641         struct ocfs2_dir_entry * de, * de1;
 642         struct super_block * sb;
 643         int err;
 644
 645         sb = inode->i_sb;
 646         if ((i_size_read(inode) <
 647              (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
 648             !(bh = ocfs2_bread(inode, 0, &err, 0))) {
 649                 mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
 650                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 651                 return 1;
 652         }
 653
 654         de = (struct ocfs2_dir_entry *) bh->b_data;
 655         de1 = (struct ocfs2_dir_entry *)
 656                         ((char *)de + le16_to_cpu(de->rec_len));
 657         if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
 658                         !le64_to_cpu(de1->inode) ||
 659                         strcmp(".", de->name) ||
 660                         strcmp("..", de1->name)) {
 661                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 662                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 663                 brelse(bh);
 664                 return 1;
 665         }
 666         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 667         de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
 668         while (offset < i_size_read(inode) ) {
 669                 if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
 670                         brelse(bh);
 671                         bh = ocfs2_bread(inode,
 672                                          offset >> sb->s_blocksize_bits, &err, 0);
 673                         if (!bh) {
 674                                 mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
 675                                      (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
 676                                 offset += sb->s_blocksize;
 677                                 continue;
 678                         }
 679                         de = (struct ocfs2_dir_entry *) bh->b_data;
 680                 }
 681                 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 682                         brelse(bh);
 683                         return 1;
 684                 }
 685                 if (le64_to_cpu(de->inode)) {
 686                         brelse(bh);
 687                         return 0;
 688                 }
 689                 offset += le16_to_cpu(de->rec_len);
 690                 de = (struct ocfs2_dir_entry *)
 691                         ((char *)de + le16_to_cpu(de->rec_len));
 692         }
 693         brelse(bh);
 694         return 1;
 695 }
 696
 697 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 698                        handle_t *handle,
 699                        struct inode *parent,
 700                        struct inode *inode,
 701                        struct buffer_head *fe_bh,
 702                        struct ocfs2_alloc_context *data_ac)
 703 {
 704         int status;
 705         struct buffer_head *new_bh = NULL;
 706         struct ocfs2_dir_entry *de = NULL;
 707
 708         mlog_entry_void();
 709
 710         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 711                                      data_ac, NULL, &new_bh);
 712         if (status < 0) {
 713                 mlog_errno(status);
 714                 goto bail;
 715         }
 716
 717         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 718
 719         status = ocfs2_journal_access(handle, inode, new_bh,
 720                                       OCFS2_JOURNAL_ACCESS_CREATE);
 721         if (status < 0) {
 722                 mlog_errno(status);
 723                 goto bail;
 724         }
 725         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 726
 727         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 728         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 729         de->name_len = 1;
 730         de->rec_len =
 731                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 732         strcpy(de->name, ".");
 733         ocfs2_set_de_type(de, S_IFDIR);
 734         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 735         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 736         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 737                                   OCFS2_DIR_REC_LEN(1));
 738         de->name_len = 2;
 739         strcpy(de->name, "..");
 740         ocfs2_set_de_type(de, S_IFDIR);
 741
 742         status = ocfs2_journal_dirty(handle, new_bh);
 743         if (status < 0) {
 744                 mlog_errno(status);
 745                 goto bail;
 746         }
 747
 748         i_size_write(inode, inode->i_sb->s_blocksize);
 749         inode->i_nlink = 2;
 750         inode->i_blocks = ocfs2_inode_sector_count(inode);
 751         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 752         if (status < 0) {
 753                 mlog_errno(status);
 754                 goto bail;
 755         }
 756
 757         status = 0;
 758 bail:
 759         if (new_bh)
 760                 brelse(new_bh);
 761
 762         mlog_exit(status);
 763         return status;
 764 }
 765
 766 /* returns a bh of the 1st new block in the allocation. */
 767 static int ocfs2_do_extend_dir(struct super_block *sb,
 768                                handle_t *handle,
 769                                struct inode *dir,
 770                                struct buffer_head *parent_fe_bh,
 771                                struct ocfs2_alloc_context *data_ac,
 772                                struct ocfs2_alloc_context *meta_ac,
 773                                struct buffer_head **new_bh)
 774 {
 775         int status;
 776         int extend;
 777         u64 p_blkno, v_blkno;
 778
 779         spin_lock(&OCFS2_I(dir)->ip_lock);
 780         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 781         spin_unlock(&OCFS2_I(dir)->ip_lock);
 782
 783         if (extend) {
 784                 u32 offset = OCFS2_I(dir)->ip_clusters;
 785
 786                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
 787                                                     1, 0, parent_fe_bh, handle,
 788                                                     data_ac, meta_ac, NULL);
 789                 BUG_ON(status == -EAGAIN);
 790                 if (status < 0) {
 791                         mlog_errno(status);
 792                         goto bail;
 793                 }
 794         }
 795
 796         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
 797         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
 798         if (status < 0) {
 799                 mlog_errno(status);
 800                 goto bail;
 801         }
 802
 803         *new_bh = sb_getblk(sb, p_blkno);
 804         if (!*new_bh) {
 805                 status = -EIO;
 806                 mlog_errno(status);
 807                 goto bail;
 808         }
 809         status = 0;
 810 bail:
 811         mlog_exit(status);
 812         return status;
 813 }
 814
 815 /* assumes you already have a cluster lock on the directory. */
 816 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 817                             struct inode *dir,
 818                             struct buffer_head *parent_fe_bh,
 819                             struct buffer_head **new_de_bh)
 820 {
 821         int status = 0;
 822         int credits, num_free_extents, drop_alloc_sem = 0;
 823         loff_t dir_i_size;
 824         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 825         struct ocfs2_alloc_context *data_ac = NULL;
 826         struct ocfs2_alloc_context *meta_ac = NULL;
 827         handle_t *handle = NULL;
 828         struct buffer_head *new_bh = NULL;
 829         struct ocfs2_dir_entry * de;
 830         struct super_block *sb = osb->sb;
 831
 832         mlog_entry_void();
 833
 834         dir_i_size = i_size_read(dir);
 835         mlog(0, "extending dir %llu (i_size = %lld)\n",
 836              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 837
 838         /* dir->i_size is always block aligned. */
 839         spin_lock(&OCFS2_I(dir)->ip_lock);
 840         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 841                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 842                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 843                 if (num_free_extents < 0) {
 844                         status = num_free_extents;
 845                         mlog_errno(status);
 846                         goto bail;
 847                 }
 848
 849                 if (!num_free_extents) {
 850                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
 851                         if (status < 0) {
 852                                 if (status != -ENOSPC)
 853                                         mlog_errno(status);
 854                                 goto bail;
 855                         }
 856                 }
 857
 858                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 859                 if (status < 0) {
 860                         if (status != -ENOSPC)
 861                                 mlog_errno(status);
 862                         goto bail;
 863                 }
 864
 865                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 866         } else {
 867                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 868                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 869         }
 870
 871         down_write(&OCFS2_I(dir)->ip_alloc_sem);
 872         drop_alloc_sem = 1;
 873
 874         handle = ocfs2_start_trans(osb, credits);
 875         if (IS_ERR(handle)) {
 876                 status = PTR_ERR(handle);
 877                 handle = NULL;
 878                 mlog_errno(status);
 879                 goto bail;
 880         }
 881
 882         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 883                                      data_ac, meta_ac, &new_bh);
 884         if (status < 0) {
 885                 mlog_errno(status);
 886                 goto bail;
 887         }
 888
 889         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 890
 891         status = ocfs2_journal_access(handle, dir, new_bh,
 892                                       OCFS2_JOURNAL_ACCESS_CREATE);
 893         if (status < 0) {
 894                 mlog_errno(status);
 895                 goto bail;
 896         }
 897         memset(new_bh->b_data, 0, sb->s_blocksize);
 898         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 899         de->inode = 0;
 900         de->rec_len = cpu_to_le16(sb->s_blocksize);
 901         status = ocfs2_journal_dirty(handle, new_bh);
 902         if (status < 0) {
 903                 mlog_errno(status);
 904                 goto bail;
 905         }
 906
 907         dir_i_size += dir->i_sb->s_blocksize;
 908         i_size_write(dir, dir_i_size);
 909         dir->i_blocks = ocfs2_inode_sector_count(dir);
 910         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 911         if (status < 0) {
 912                 mlog_errno(status);
 913                 goto bail;
 914         }
 915
 916         *new_de_bh = new_bh;
 917         get_bh(*new_de_bh);
 918 bail:
 919         if (drop_alloc_sem)
 920                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
 921         if (handle)
 922                 ocfs2_commit_trans(osb, handle);
 923
 924         if (data_ac)
 925                 ocfs2_free_alloc_context(data_ac);
 926         if (meta_ac)
 927                 ocfs2_free_alloc_context(meta_ac);
 928
 929         if (new_bh)
 930                 brelse(new_bh);
 931
 932         mlog_exit(status);
 933         return status;
 934 }
 935
 936 /*
 937  * Search the dir for a good spot, extending it if necessary. The
 938  * block containing an appropriate record is returned in ret_de_bh.
 939  */
 940 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
 941                                  struct inode *dir,
 942                                  struct buffer_head *parent_fe_bh,
 943                                  const char *name,
 944                                  int namelen,
 945                                  struct buffer_head **ret_de_bh)
 946 {
 947         unsigned long offset;
 948         struct buffer_head * bh = NULL;
 949         unsigned short rec_len;
 950         struct ocfs2_dinode *fe;
 951         struct ocfs2_dir_entry *de;
 952         struct super_block *sb;
 953         int status;
 954
 955         mlog_entry_void();
 956
 957         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
 958              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 959
 960         BUG_ON(!S_ISDIR(dir->i_mode));
 961         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 962         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
 963
 964         sb = dir->i_sb;
 965
 966         if (!namelen) {
 967                 status = -EINVAL;
 968                 mlog_errno(status);
 969                 goto bail;
 970         }
 971
 972         bh = ocfs2_bread(dir, 0, &status, 0);
 973         if (!bh) {
 974                 mlog_errno(status);
 975                 goto bail;
 976         }
 977
 978         rec_len = OCFS2_DIR_REC_LEN(namelen);
 979         offset = 0;
 980         de = (struct ocfs2_dir_entry *) bh->b_data;
 981         while (1) {
 982                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
 983                         brelse(bh);
 984                         bh = NULL;
 985
 986                         if (i_size_read(dir) <= offset) {
 987                                 status = ocfs2_extend_dir(osb,
 988                                                           dir,
 989                                                           parent_fe_bh,
 990                                                           &bh);
 991                                 if (status < 0) {
 992                                         mlog_errno(status);
 993                                         goto bail;
 994                                 }
 995                                 BUG_ON(!bh);
 996                                 *ret_de_bh = bh;
 997                                 get_bh(*ret_de_bh);
 998                                 goto bail;
 999                         }
1000                         bh = ocfs2_bread(dir,
1001                                          offset >> sb->s_blocksize_bits,
1002                                          &status,
1003                                          0);
1004                         if (!bh) {
1005                                 mlog_errno(status);
1006                                 goto bail;
1007                         }
1008                         /* move to next block */
1009                         de = (struct ocfs2_dir_entry *) bh->b_data;
1010                 }
1011                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1012                         status = -ENOENT;
1013                         goto bail;
1014                 }
1015                 if (ocfs2_match(namelen, name, de)) {
1016                         status = -EEXIST;
1017                         goto bail;
1018                 }
1019                 if (((le64_to_cpu(de->inode) == 0) &&
1020                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1021                     (le16_to_cpu(de->rec_len) >=
1022                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1023                         /* Ok, we found a spot. Return this bh and let
1024                          * the caller actually fill it in. */
1025                         *ret_de_bh = bh;
1026                         get_bh(*ret_de_bh);
1027                         status = 0;
1028                         goto bail;
1029                 }
1030                 offset += le16_to_cpu(de->rec_len);
1031                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1032         }
1033
1034         status = 0;
1035 bail:
1036         if (bh)
1037                 brelse(bh);
1038
1039         mlog_exit(status);
1040         return status;
1041 }