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 static 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  * This is intended to be called from inside other kernel functions,
 536  * so we fake some arguments.
 537  */
 538 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
 539                       filldir_t filldir)
 540 {
 541         int ret = 0;
 542         unsigned long version = inode->i_version;
 543
 544         while (*f_pos < i_size_read(inode)) {
 545                 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
 546                                             filldir);
 547                 if (ret)
 548                         break;
 549         }
 550
 551         return 0;
 552 }
 553
 554 /*
 555  * ocfs2_readdir()
 556  *
 557  */
 558 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
 559 {
 560         int error = 0;
 561         struct inode *inode = filp->f_path.dentry->d_inode;
 562         int lock_level = 0;
 563
 564         mlog_entry("dirino=%llu\n",
 565                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 566
 567         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
 568         if (lock_level && error >= 0) {
 569                 /* We release EX lock which used to update atime
 570                  * and get PR lock again to reduce contention
 571                  * on commonly accessed directories. */
 572                 ocfs2_meta_unlock(inode, 1);
 573                 lock_level = 0;
 574                 error = ocfs2_meta_lock(inode, NULL, 0);
 575         }
 576         if (error < 0) {
 577                 if (error != -ENOENT)
 578                         mlog_errno(error);
 579                 /* we haven't got any yet, so propagate the error. */
 580                 goto bail_nolock;
 581         }
 582
 583         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
 584                                       dirent, filldir);
 585
 586         ocfs2_meta_unlock(inode, lock_level);
 587
 588 bail_nolock:
 589         mlog_exit(error);
 590
 591         return error;
 592 }
 593
 594 /*
 595  * NOTE: this should always be called with parent dir i_mutex taken.
 596  */
 597 int ocfs2_find_files_on_disk(const char *name,
 598                              int namelen,
 599                              u64 *blkno,
 600                              struct inode *inode,
 601                              struct buffer_head **dirent_bh,
 602                              struct ocfs2_dir_entry **dirent)
 603 {
 604         int status = -ENOENT;
 605
 606         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 607                    namelen, name, blkno, inode, dirent_bh, dirent);
 608
 609         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 610         if (!*dirent_bh || !*dirent) {
 611                 status = -ENOENT;
 612                 goto leave;
 613         }
 614
 615         *blkno = le64_to_cpu((*dirent)->inode);
 616
 617         status = 0;
 618 leave:
 619         if (status < 0) {
 620                 *dirent = NULL;
 621                 if (*dirent_bh) {
 622                         brelse(*dirent_bh);
 623                         *dirent_bh = NULL;
 624                 }
 625         }
 626
 627         mlog_exit(status);
 628         return status;
 629 }
 630
 631 /*
 632  * Convenience function for callers which just want the block number
 633  * mapped to a name and don't require the full dirent info, etc.
 634  */
 635 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
 636                                int namelen, u64 *blkno)
 637 {
 638         int ret;
 639         struct buffer_head *bh = NULL;
 640         struct ocfs2_dir_entry *dirent = NULL;
 641
 642         ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent);
 643         brelse(bh);
 644
 645         return ret;
 646 }
 647
 648 /* Check for a name within a directory.
 649  *
 650  * Return 0 if the name does not exist
 651  * Return -EEXIST if the directory contains the name
 652  *
 653  * Callers should have i_mutex + a cluster lock on dir
 654  */
 655 int ocfs2_check_dir_for_entry(struct inode *dir,
 656                               const char *name,
 657                               int namelen)
 658 {
 659         int ret;
 660         struct buffer_head *dirent_bh = NULL;
 661         struct ocfs2_dir_entry *dirent = NULL;
 662
 663         mlog_entry("dir %llu, name '%.*s'\n",
 664                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 665
 666         ret = -EEXIST;
 667         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 668         if (dirent_bh)
 669                 goto bail;
 670
 671         ret = 0;
 672 bail:
 673         if (dirent_bh)
 674                 brelse(dirent_bh);
 675
 676         mlog_exit(ret);
 677         return ret;
 678 }
 679
 680 struct ocfs2_empty_dir_priv {
 681         unsigned seen_dot;
 682         unsigned seen_dot_dot;
 683         unsigned seen_other;
 684 };
 685 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
 686                                    loff_t pos, u64 ino, unsigned type)
 687 {
 688         struct ocfs2_empty_dir_priv *p = priv;
 689
 690         /*
 691          * Check the positions of "." and ".." records to be sure
 692          * they're in the correct place.
 693          */
 694         if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
 695                 p->seen_dot = 1;
 696                 return 0;
 697         }
 698
 699         if (name_len == 2 && !strncmp("..", name, 2) &&
 700             pos == OCFS2_DIR_REC_LEN(1)) {
 701                 p->seen_dot_dot = 1;
 702                 return 0;
 703         }
 704
 705         p->seen_other = 1;
 706         return 1;
 707 }
 708 /*
 709  * routine to check that the specified directory is empty (for rmdir)
 710  *
 711  * Returns 1 if dir is empty, zero otherwise.
 712  */
 713 int ocfs2_empty_dir(struct inode *inode)
 714 {
 715         int ret;
 716         loff_t start = 0;
 717         struct ocfs2_empty_dir_priv priv;
 718
 719         memset(&priv, 0, sizeof(priv));
 720
 721         ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
 722         if (ret)
 723                 mlog_errno(ret);
 724
 725         if (!priv.seen_dot || !priv.seen_dot_dot) {
 726                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 727                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 728                 /*
 729                  * XXX: Is it really safe to allow an unlink to continue?
 730                  */
 731                 return 1;
 732         }
 733
 734         return !priv.seen_other;
 735 }
 736
 737 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 738                        handle_t *handle,
 739                        struct inode *parent,
 740                        struct inode *inode,
 741                        struct buffer_head *fe_bh,
 742                        struct ocfs2_alloc_context *data_ac)
 743 {
 744         int status;
 745         struct buffer_head *new_bh = NULL;
 746         struct ocfs2_dir_entry *de = NULL;
 747
 748         mlog_entry_void();
 749
 750         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 751                                      data_ac, NULL, &new_bh);
 752         if (status < 0) {
 753                 mlog_errno(status);
 754                 goto bail;
 755         }
 756
 757         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 758
 759         status = ocfs2_journal_access(handle, inode, new_bh,
 760                                       OCFS2_JOURNAL_ACCESS_CREATE);
 761         if (status < 0) {
 762                 mlog_errno(status);
 763                 goto bail;
 764         }
 765         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 766
 767         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 768         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 769         de->name_len = 1;
 770         de->rec_len =
 771                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 772         strcpy(de->name, ".");
 773         ocfs2_set_de_type(de, S_IFDIR);
 774         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 775         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 776         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 777                                   OCFS2_DIR_REC_LEN(1));
 778         de->name_len = 2;
 779         strcpy(de->name, "..");
 780         ocfs2_set_de_type(de, S_IFDIR);
 781
 782         status = ocfs2_journal_dirty(handle, new_bh);
 783         if (status < 0) {
 784                 mlog_errno(status);
 785                 goto bail;
 786         }
 787
 788         i_size_write(inode, inode->i_sb->s_blocksize);
 789         inode->i_nlink = 2;
 790         inode->i_blocks = ocfs2_inode_sector_count(inode);
 791         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 792         if (status < 0) {
 793                 mlog_errno(status);
 794                 goto bail;
 795         }
 796
 797         status = 0;
 798 bail:
 799         if (new_bh)
 800                 brelse(new_bh);
 801
 802         mlog_exit(status);
 803         return status;
 804 }
 805
 806 /* returns a bh of the 1st new block in the allocation. */
 807 static int ocfs2_do_extend_dir(struct super_block *sb,
 808                                handle_t *handle,
 809                                struct inode *dir,
 810                                struct buffer_head *parent_fe_bh,
 811                                struct ocfs2_alloc_context *data_ac,
 812                                struct ocfs2_alloc_context *meta_ac,
 813                                struct buffer_head **new_bh)
 814 {
 815         int status;
 816         int extend;
 817         u64 p_blkno, v_blkno;
 818
 819         spin_lock(&OCFS2_I(dir)->ip_lock);
 820         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 821         spin_unlock(&OCFS2_I(dir)->ip_lock);
 822
 823         if (extend) {
 824                 u32 offset = OCFS2_I(dir)->ip_clusters;
 825
 826                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
 827                                                     1, 0, parent_fe_bh, handle,
 828                                                     data_ac, meta_ac, NULL);
 829                 BUG_ON(status == -EAGAIN);
 830                 if (status < 0) {
 831                         mlog_errno(status);
 832                         goto bail;
 833                 }
 834         }
 835
 836         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
 837         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
 838         if (status < 0) {
 839                 mlog_errno(status);
 840                 goto bail;
 841         }
 842
 843         *new_bh = sb_getblk(sb, p_blkno);
 844         if (!*new_bh) {
 845                 status = -EIO;
 846                 mlog_errno(status);
 847                 goto bail;
 848         }
 849         status = 0;
 850 bail:
 851         mlog_exit(status);
 852         return status;
 853 }
 854
 855 /* assumes you already have a cluster lock on the directory. */
 856 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 857                             struct inode *dir,
 858                             struct buffer_head *parent_fe_bh,
 859                             struct buffer_head **new_de_bh)
 860 {
 861         int status = 0;
 862         int credits, num_free_extents, drop_alloc_sem = 0;
 863         loff_t dir_i_size;
 864         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 865         struct ocfs2_alloc_context *data_ac = NULL;
 866         struct ocfs2_alloc_context *meta_ac = NULL;
 867         handle_t *handle = NULL;
 868         struct buffer_head *new_bh = NULL;
 869         struct ocfs2_dir_entry * de;
 870         struct super_block *sb = osb->sb;
 871
 872         mlog_entry_void();
 873
 874         dir_i_size = i_size_read(dir);
 875         mlog(0, "extending dir %llu (i_size = %lld)\n",
 876              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 877
 878         /* dir->i_size is always block aligned. */
 879         spin_lock(&OCFS2_I(dir)->ip_lock);
 880         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 881                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 882                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 883                 if (num_free_extents < 0) {
 884                         status = num_free_extents;
 885                         mlog_errno(status);
 886                         goto bail;
 887                 }
 888
 889                 if (!num_free_extents) {
 890                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
 891                         if (status < 0) {
 892                                 if (status != -ENOSPC)
 893                                         mlog_errno(status);
 894                                 goto bail;
 895                         }
 896                 }
 897
 898                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 899                 if (status < 0) {
 900                         if (status != -ENOSPC)
 901                                 mlog_errno(status);
 902                         goto bail;
 903                 }
 904
 905                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 906         } else {
 907                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 908                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 909         }
 910
 911         down_write(&OCFS2_I(dir)->ip_alloc_sem);
 912         drop_alloc_sem = 1;
 913
 914         handle = ocfs2_start_trans(osb, credits);
 915         if (IS_ERR(handle)) {
 916                 status = PTR_ERR(handle);
 917                 handle = NULL;
 918                 mlog_errno(status);
 919                 goto bail;
 920         }
 921
 922         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 923                                      data_ac, meta_ac, &new_bh);
 924         if (status < 0) {
 925                 mlog_errno(status);
 926                 goto bail;
 927         }
 928
 929         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 930
 931         status = ocfs2_journal_access(handle, dir, new_bh,
 932                                       OCFS2_JOURNAL_ACCESS_CREATE);
 933         if (status < 0) {
 934                 mlog_errno(status);
 935                 goto bail;
 936         }
 937         memset(new_bh->b_data, 0, sb->s_blocksize);
 938         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 939         de->inode = 0;
 940         de->rec_len = cpu_to_le16(sb->s_blocksize);
 941         status = ocfs2_journal_dirty(handle, new_bh);
 942         if (status < 0) {
 943                 mlog_errno(status);
 944                 goto bail;
 945         }
 946
 947         dir_i_size += dir->i_sb->s_blocksize;
 948         i_size_write(dir, dir_i_size);
 949         dir->i_blocks = ocfs2_inode_sector_count(dir);
 950         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 951         if (status < 0) {
 952                 mlog_errno(status);
 953                 goto bail;
 954         }
 955
 956         *new_de_bh = new_bh;
 957         get_bh(*new_de_bh);
 958 bail:
 959         if (drop_alloc_sem)
 960                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
 961         if (handle)
 962                 ocfs2_commit_trans(osb, handle);
 963
 964         if (data_ac)
 965                 ocfs2_free_alloc_context(data_ac);
 966         if (meta_ac)
 967                 ocfs2_free_alloc_context(meta_ac);
 968
 969         if (new_bh)
 970                 brelse(new_bh);
 971
 972         mlog_exit(status);
 973         return status;
 974 }
 975
 976 /*
 977  * Search the dir for a good spot, extending it if necessary. The
 978  * block containing an appropriate record is returned in ret_de_bh.
 979  */
 980 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
 981                                  struct inode *dir,
 982                                  struct buffer_head *parent_fe_bh,
 983                                  const char *name,
 984                                  int namelen,
 985                                  struct buffer_head **ret_de_bh)
 986 {
 987         unsigned long offset;
 988         struct buffer_head * bh = NULL;
 989         unsigned short rec_len;
 990         struct ocfs2_dinode *fe;
 991         struct ocfs2_dir_entry *de;
 992         struct super_block *sb;
 993         int status;
 994
 995         mlog_entry_void();
 996
 997         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
 998              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 999
1000         BUG_ON(!S_ISDIR(dir->i_mode));
1001         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1002         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
1003
1004         sb = dir->i_sb;
1005
1006         if (!namelen) {
1007                 status = -EINVAL;
1008                 mlog_errno(status);
1009                 goto bail;
1010         }
1011
1012         bh = ocfs2_bread(dir, 0, &status, 0);
1013         if (!bh) {
1014                 mlog_errno(status);
1015                 goto bail;
1016         }
1017
1018         rec_len = OCFS2_DIR_REC_LEN(namelen);
1019         offset = 0;
1020         de = (struct ocfs2_dir_entry *) bh->b_data;
1021         while (1) {
1022                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
1023                         brelse(bh);
1024                         bh = NULL;
1025
1026                         if (i_size_read(dir) <= offset) {
1027                                 status = ocfs2_extend_dir(osb,
1028                                                           dir,
1029                                                           parent_fe_bh,
1030                                                           &bh);
1031                                 if (status < 0) {
1032                                         mlog_errno(status);
1033                                         goto bail;
1034                                 }
1035                                 BUG_ON(!bh);
1036                                 *ret_de_bh = bh;
1037                                 get_bh(*ret_de_bh);
1038                                 goto bail;
1039                         }
1040                         bh = ocfs2_bread(dir,
1041                                          offset >> sb->s_blocksize_bits,
1042                                          &status,
1043                                          0);
1044                         if (!bh) {
1045                                 mlog_errno(status);
1046                                 goto bail;
1047                         }
1048                         /* move to next block */
1049                         de = (struct ocfs2_dir_entry *) bh->b_data;
1050                 }
1051                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1052                         status = -ENOENT;
1053                         goto bail;
1054                 }
1055                 if (ocfs2_match(namelen, name, de)) {
1056                         status = -EEXIST;
1057                         goto bail;
1058                 }
1059                 if (((le64_to_cpu(de->inode) == 0) &&
1060                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1061                     (le16_to_cpu(de->rec_len) >=
1062                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1063                         /* Ok, we found a spot. Return this bh and let
1064                          * the caller actually fill it in. */
1065                         *ret_de_bh = bh;
1066                         get_bh(*ret_de_bh);
1067                         status = 0;
1068                         goto bail;
1069                 }
1070                 offset += le16_to_cpu(de->rec_len);
1071                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1072         }
1073
1074         status = 0;
1075 bail:
1076         if (bh)
1077                 brelse(bh);
1078
1079         mlog_exit(status);
1080         return status;
1081 }