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 static int ocfs2_dir_foreach_blk(struct inode *inode, unsigned long *f_version,
 418                                  loff_t *f_pos, void *priv, filldir_t filldir)
 419 {
 420         int error = 0;
 421         unsigned long offset, blk, last_ra_blk = 0;
 422         int i, stored;
 423         struct buffer_head * bh, * tmp;
 424         struct ocfs2_dir_entry * de;
 425         int err;
 426         struct super_block * sb = inode->i_sb;
 427         unsigned int ra_sectors = 16;
 428
 429         stored = 0;
 430         bh = NULL;
 431
 432         offset = (*f_pos) & (sb->s_blocksize - 1);
 433
 434         while (!error && !stored && *f_pos < i_size_read(inode)) {
 435                 blk = (*f_pos) >> sb->s_blocksize_bits;
 436                 bh = ocfs2_bread(inode, blk, &err, 0);
 437                 if (!bh) {
 438                         mlog(ML_ERROR,
 439                              "directory #%llu contains a hole at offset %lld\n",
 440                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 441                              *f_pos);
 442                         *f_pos += sb->s_blocksize - offset;
 443                         continue;
 444                 }
 445
 446                 /* The idea here is to begin with 8k read-ahead and to stay
 447                  * 4k ahead of our current position.
 448                  *
 449                  * TODO: Use the pagecache for this. We just need to
 450                  * make sure it's cluster-safe... */
 451                 if (!last_ra_blk
 452                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
 453                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
 454                              i > 0; i--) {
 455                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
 456                                 if (tmp)
 457                                         brelse(tmp);
 458                         }
 459                         last_ra_blk = blk;
 460                         ra_sectors = 8;
 461                 }
 462
 463 revalidate:
 464                 /* If the dir block has changed since the last call to
 465                  * readdir(2), then we might be pointing to an invalid
 466                  * dirent right now.  Scan from the start of the block
 467                  * to make sure. */
 468                 if (*f_version != inode->i_version) {
 469                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 470                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 471                                 /* It's too expensive to do a full
 472                                  * dirent test each time round this
 473                                  * loop, but we do have to test at
 474                                  * least that it is non-zero.  A
 475                                  * failure will be detected in the
 476                                  * dirent test below. */
 477                                 if (le16_to_cpu(de->rec_len) <
 478                                     OCFS2_DIR_REC_LEN(1))
 479                                         break;
 480                                 i += le16_to_cpu(de->rec_len);
 481                         }
 482                         offset = i;
 483                         *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
 484                                 | offset;
 485                         *f_version = inode->i_version;
 486                 }
 487
 488                 while (!error && *f_pos < i_size_read(inode)
 489                        && offset < sb->s_blocksize) {
 490                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 491                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 492                                 /* On error, skip the f_pos to the
 493                                    next block. */
 494                                 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
 495                                 brelse(bh);
 496                                 goto out;
 497                         }
 498                         offset += le16_to_cpu(de->rec_len);
 499                         if (le64_to_cpu(de->inode)) {
 500                                 /* We might block in the next section
 501                                  * if the data destination is
 502                                  * currently swapped out.  So, use a
 503                                  * version stamp to detect whether or
 504                                  * not the directory has been modified
 505                                  * during the copy operation.
 506                                  */
 507                                 unsigned long version = *f_version;
 508                                 unsigned char d_type = DT_UNKNOWN;
 509
 510                                 if (de->file_type < OCFS2_FT_MAX)
 511                                         d_type = ocfs2_filetype_table[de->file_type];
 512                                 error = filldir(priv, de->name,
 513                                                 de->name_len,
 514                                                 *f_pos,
 515                                                 le64_to_cpu(de->inode),
 516                                                 d_type);
 517                                 if (error)
 518                                         break;
 519                                 if (version != *f_version)
 520                                         goto revalidate;
 521                                 stored ++;
 522                         }
 523                         *f_pos += le16_to_cpu(de->rec_len);
 524                 }
 525                 offset = 0;
 526                 brelse(bh);
 527         }
 528
 529         stored = 0;
 530 out:
 531         return stored;
 532 }
 533
 534 /*
 535  * ocfs2_readdir()
 536  *
 537  */
 538 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
 539 {
 540         int error = 0;
 541         struct inode *inode = filp->f_path.dentry->d_inode;
 542         int lock_level = 0;
 543
 544         mlog_entry("dirino=%llu\n",
 545                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 546
 547         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
 548         if (lock_level && error >= 0) {
 549                 /* We release EX lock which used to update atime
 550                  * and get PR lock again to reduce contention
 551                  * on commonly accessed directories. */
 552                 ocfs2_meta_unlock(inode, 1);
 553                 lock_level = 0;
 554                 error = ocfs2_meta_lock(inode, NULL, 0);
 555         }
 556         if (error < 0) {
 557                 if (error != -ENOENT)
 558                         mlog_errno(error);
 559                 /* we haven't got any yet, so propagate the error. */
 560                 goto bail_nolock;
 561         }
 562
 563         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
 564                                       dirent, filldir);
 565
 566         ocfs2_meta_unlock(inode, lock_level);
 567
 568 bail_nolock:
 569         mlog_exit(error);
 570
 571         return error;
 572 }
 573
 574 /*
 575  * NOTE: this should always be called with parent dir i_mutex taken.
 576  */
 577 int ocfs2_find_files_on_disk(const char *name,
 578                              int namelen,
 579                              u64 *blkno,
 580                              struct inode *inode,
 581                              struct buffer_head **dirent_bh,
 582                              struct ocfs2_dir_entry **dirent)
 583 {
 584         int status = -ENOENT;
 585
 586         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 587                    namelen, name, blkno, inode, dirent_bh, dirent);
 588
 589         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 590         if (!*dirent_bh || !*dirent) {
 591                 status = -ENOENT;
 592                 goto leave;
 593         }
 594
 595         *blkno = le64_to_cpu((*dirent)->inode);
 596
 597         status = 0;
 598 leave:
 599         if (status < 0) {
 600                 *dirent = NULL;
 601                 if (*dirent_bh) {
 602                         brelse(*dirent_bh);
 603                         *dirent_bh = NULL;
 604                 }
 605         }
 606
 607         mlog_exit(status);
 608         return status;
 609 }
 610
 611 /* Check for a name within a directory.
 612  *
 613  * Return 0 if the name does not exist
 614  * Return -EEXIST if the directory contains the name
 615  *
 616  * Callers should have i_mutex + a cluster lock on dir
 617  */
 618 int ocfs2_check_dir_for_entry(struct inode *dir,
 619                               const char *name,
 620                               int namelen)
 621 {
 622         int ret;
 623         struct buffer_head *dirent_bh = NULL;
 624         struct ocfs2_dir_entry *dirent = NULL;
 625
 626         mlog_entry("dir %llu, name '%.*s'\n",
 627                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 628
 629         ret = -EEXIST;
 630         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 631         if (dirent_bh)
 632                 goto bail;
 633
 634         ret = 0;
 635 bail:
 636         if (dirent_bh)
 637                 brelse(dirent_bh);
 638
 639         mlog_exit(ret);
 640         return ret;
 641 }
 642
 643 /*
 644  * routine to check that the specified directory is empty (for rmdir)
 645  */
 646 int ocfs2_empty_dir(struct inode *inode)
 647 {
 648         unsigned long offset;
 649         struct buffer_head * bh;
 650         struct ocfs2_dir_entry * de, * de1;
 651         struct super_block * sb;
 652         int err;
 653
 654         sb = inode->i_sb;
 655         if ((i_size_read(inode) <
 656              (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
 657             !(bh = ocfs2_bread(inode, 0, &err, 0))) {
 658                 mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
 659                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 660                 return 1;
 661         }
 662
 663         de = (struct ocfs2_dir_entry *) bh->b_data;
 664         de1 = (struct ocfs2_dir_entry *)
 665                         ((char *)de + le16_to_cpu(de->rec_len));
 666         if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
 667                         !le64_to_cpu(de1->inode) ||
 668                         strcmp(".", de->name) ||
 669                         strcmp("..", de1->name)) {
 670                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 671                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 672                 brelse(bh);
 673                 return 1;
 674         }
 675         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 676         de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
 677         while (offset < i_size_read(inode) ) {
 678                 if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
 679                         brelse(bh);
 680                         bh = ocfs2_bread(inode,
 681                                          offset >> sb->s_blocksize_bits, &err, 0);
 682                         if (!bh) {
 683                                 mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
 684                                      (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
 685                                 offset += sb->s_blocksize;
 686                                 continue;
 687                         }
 688                         de = (struct ocfs2_dir_entry *) bh->b_data;
 689                 }
 690                 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 691                         brelse(bh);
 692                         return 1;
 693                 }
 694                 if (le64_to_cpu(de->inode)) {
 695                         brelse(bh);
 696                         return 0;
 697                 }
 698                 offset += le16_to_cpu(de->rec_len);
 699                 de = (struct ocfs2_dir_entry *)
 700                         ((char *)de + le16_to_cpu(de->rec_len));
 701         }
 702         brelse(bh);
 703         return 1;
 704 }
 705
 706 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 707                        handle_t *handle,
 708                        struct inode *parent,
 709                        struct inode *inode,
 710                        struct buffer_head *fe_bh,
 711                        struct ocfs2_alloc_context *data_ac)
 712 {
 713         int status;
 714         struct buffer_head *new_bh = NULL;
 715         struct ocfs2_dir_entry *de = NULL;
 716
 717         mlog_entry_void();
 718
 719         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 720                                      data_ac, NULL, &new_bh);
 721         if (status < 0) {
 722                 mlog_errno(status);
 723                 goto bail;
 724         }
 725
 726         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 727
 728         status = ocfs2_journal_access(handle, inode, new_bh,
 729                                       OCFS2_JOURNAL_ACCESS_CREATE);
 730         if (status < 0) {
 731                 mlog_errno(status);
 732                 goto bail;
 733         }
 734         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 735
 736         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 737         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 738         de->name_len = 1;
 739         de->rec_len =
 740                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 741         strcpy(de->name, ".");
 742         ocfs2_set_de_type(de, S_IFDIR);
 743         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 744         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 745         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 746                                   OCFS2_DIR_REC_LEN(1));
 747         de->name_len = 2;
 748         strcpy(de->name, "..");
 749         ocfs2_set_de_type(de, S_IFDIR);
 750
 751         status = ocfs2_journal_dirty(handle, new_bh);
 752         if (status < 0) {
 753                 mlog_errno(status);
 754                 goto bail;
 755         }
 756
 757         i_size_write(inode, inode->i_sb->s_blocksize);
 758         inode->i_nlink = 2;
 759         inode->i_blocks = ocfs2_inode_sector_count(inode);
 760         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 761         if (status < 0) {
 762                 mlog_errno(status);
 763                 goto bail;
 764         }
 765
 766         status = 0;
 767 bail:
 768         if (new_bh)
 769                 brelse(new_bh);
 770
 771         mlog_exit(status);
 772         return status;
 773 }
 774
 775 /* returns a bh of the 1st new block in the allocation. */
 776 static int ocfs2_do_extend_dir(struct super_block *sb,
 777                                handle_t *handle,
 778                                struct inode *dir,
 779                                struct buffer_head *parent_fe_bh,
 780                                struct ocfs2_alloc_context *data_ac,
 781                                struct ocfs2_alloc_context *meta_ac,
 782                                struct buffer_head **new_bh)
 783 {
 784         int status;
 785         int extend;
 786         u64 p_blkno, v_blkno;
 787
 788         spin_lock(&OCFS2_I(dir)->ip_lock);
 789         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 790         spin_unlock(&OCFS2_I(dir)->ip_lock);
 791
 792         if (extend) {
 793                 u32 offset = OCFS2_I(dir)->ip_clusters;
 794
 795                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
 796                                                     1, 0, parent_fe_bh, handle,
 797                                                     data_ac, meta_ac, NULL);
 798                 BUG_ON(status == -EAGAIN);
 799                 if (status < 0) {
 800                         mlog_errno(status);
 801                         goto bail;
 802                 }
 803         }
 804
 805         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
 806         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
 807         if (status < 0) {
 808                 mlog_errno(status);
 809                 goto bail;
 810         }
 811
 812         *new_bh = sb_getblk(sb, p_blkno);
 813         if (!*new_bh) {
 814                 status = -EIO;
 815                 mlog_errno(status);
 816                 goto bail;
 817         }
 818         status = 0;
 819 bail:
 820         mlog_exit(status);
 821         return status;
 822 }
 823
 824 /* assumes you already have a cluster lock on the directory. */
 825 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 826                             struct inode *dir,
 827                             struct buffer_head *parent_fe_bh,
 828                             struct buffer_head **new_de_bh)
 829 {
 830         int status = 0;
 831         int credits, num_free_extents, drop_alloc_sem = 0;
 832         loff_t dir_i_size;
 833         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 834         struct ocfs2_alloc_context *data_ac = NULL;
 835         struct ocfs2_alloc_context *meta_ac = NULL;
 836         handle_t *handle = NULL;
 837         struct buffer_head *new_bh = NULL;
 838         struct ocfs2_dir_entry * de;
 839         struct super_block *sb = osb->sb;
 840
 841         mlog_entry_void();
 842
 843         dir_i_size = i_size_read(dir);
 844         mlog(0, "extending dir %llu (i_size = %lld)\n",
 845              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 846
 847         /* dir->i_size is always block aligned. */
 848         spin_lock(&OCFS2_I(dir)->ip_lock);
 849         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 850                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 851                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 852                 if (num_free_extents < 0) {
 853                         status = num_free_extents;
 854                         mlog_errno(status);
 855                         goto bail;
 856                 }
 857
 858                 if (!num_free_extents) {
 859                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
 860                         if (status < 0) {
 861                                 if (status != -ENOSPC)
 862                                         mlog_errno(status);
 863                                 goto bail;
 864                         }
 865                 }
 866
 867                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 868                 if (status < 0) {
 869                         if (status != -ENOSPC)
 870                                 mlog_errno(status);
 871                         goto bail;
 872                 }
 873
 874                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 875         } else {
 876                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 877                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 878         }
 879
 880         down_write(&OCFS2_I(dir)->ip_alloc_sem);
 881         drop_alloc_sem = 1;
 882
 883         handle = ocfs2_start_trans(osb, credits);
 884         if (IS_ERR(handle)) {
 885                 status = PTR_ERR(handle);
 886                 handle = NULL;
 887                 mlog_errno(status);
 888                 goto bail;
 889         }
 890
 891         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 892                                      data_ac, meta_ac, &new_bh);
 893         if (status < 0) {
 894                 mlog_errno(status);
 895                 goto bail;
 896         }
 897
 898         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 899
 900         status = ocfs2_journal_access(handle, dir, new_bh,
 901                                       OCFS2_JOURNAL_ACCESS_CREATE);
 902         if (status < 0) {
 903                 mlog_errno(status);
 904                 goto bail;
 905         }
 906         memset(new_bh->b_data, 0, sb->s_blocksize);
 907         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 908         de->inode = 0;
 909         de->rec_len = cpu_to_le16(sb->s_blocksize);
 910         status = ocfs2_journal_dirty(handle, new_bh);
 911         if (status < 0) {
 912                 mlog_errno(status);
 913                 goto bail;
 914         }
 915
 916         dir_i_size += dir->i_sb->s_blocksize;
 917         i_size_write(dir, dir_i_size);
 918         dir->i_blocks = ocfs2_inode_sector_count(dir);
 919         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 920         if (status < 0) {
 921                 mlog_errno(status);
 922                 goto bail;
 923         }
 924
 925         *new_de_bh = new_bh;
 926         get_bh(*new_de_bh);
 927 bail:
 928         if (drop_alloc_sem)
 929                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
 930         if (handle)
 931                 ocfs2_commit_trans(osb, handle);
 932
 933         if (data_ac)
 934                 ocfs2_free_alloc_context(data_ac);
 935         if (meta_ac)
 936                 ocfs2_free_alloc_context(meta_ac);
 937
 938         if (new_bh)
 939                 brelse(new_bh);
 940
 941         mlog_exit(status);
 942         return status;
 943 }
 944
 945 /*
 946  * Search the dir for a good spot, extending it if necessary. The
 947  * block containing an appropriate record is returned in ret_de_bh.
 948  */
 949 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
 950                                  struct inode *dir,
 951                                  struct buffer_head *parent_fe_bh,
 952                                  const char *name,
 953                                  int namelen,
 954                                  struct buffer_head **ret_de_bh)
 955 {
 956         unsigned long offset;
 957         struct buffer_head * bh = NULL;
 958         unsigned short rec_len;
 959         struct ocfs2_dinode *fe;
 960         struct ocfs2_dir_entry *de;
 961         struct super_block *sb;
 962         int status;
 963
 964         mlog_entry_void();
 965
 966         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
 967              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 968
 969         BUG_ON(!S_ISDIR(dir->i_mode));
 970         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 971         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
 972
 973         sb = dir->i_sb;
 974
 975         if (!namelen) {
 976                 status = -EINVAL;
 977                 mlog_errno(status);
 978                 goto bail;
 979         }
 980
 981         bh = ocfs2_bread(dir, 0, &status, 0);
 982         if (!bh) {
 983                 mlog_errno(status);
 984                 goto bail;
 985         }
 986
 987         rec_len = OCFS2_DIR_REC_LEN(namelen);
 988         offset = 0;
 989         de = (struct ocfs2_dir_entry *) bh->b_data;
 990         while (1) {
 991                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
 992                         brelse(bh);
 993                         bh = NULL;
 994
 995                         if (i_size_read(dir) <= offset) {
 996                                 status = ocfs2_extend_dir(osb,
 997                                                           dir,
 998                                                           parent_fe_bh,
 999                                                           &bh);
1000                                 if (status < 0) {
1001                                         mlog_errno(status);
1002                                         goto bail;
1003                                 }
1004                                 BUG_ON(!bh);
1005                                 *ret_de_bh = bh;
1006                                 get_bh(*ret_de_bh);
1007                                 goto bail;
1008                         }
1009                         bh = ocfs2_bread(dir,
1010                                          offset >> sb->s_blocksize_bits,
1011                                          &status,
1012                                          0);
1013                         if (!bh) {
1014                                 mlog_errno(status);
1015                                 goto bail;
1016                         }
1017                         /* move to next block */
1018                         de = (struct ocfs2_dir_entry *) bh->b_data;
1019                 }
1020                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1021                         status = -ENOENT;
1022                         goto bail;
1023                 }
1024                 if (ocfs2_match(namelen, name, de)) {
1025                         status = -EEXIST;
1026                         goto bail;
1027                 }
1028                 if (((le64_to_cpu(de->inode) == 0) &&
1029                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1030                     (le16_to_cpu(de->rec_len) >=
1031                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1032                         /* Ok, we found a spot. Return this bh and let
1033                          * the caller actually fill it in. */
1034                         *ret_de_bh = bh;
1035                         get_bh(*ret_de_bh);
1036                         status = 0;
1037                         goto bail;
1038                 }
1039                 offset += le16_to_cpu(de->rec_len);
1040                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1041         }
1042
1043         status = 0;
1044 bail:
1045         if (bh)
1046                 brelse(bh);
1047
1048         mlog_exit(status);
1049         return status;
1050 }