SAFE public projects git trees. - safe/jmp/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 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 int ocfs2_update_entry(struct inode *dir, handle_t *handle,
 275                        struct buffer_head *de_bh, struct ocfs2_dir_entry *de,
 276                        struct inode *new_entry_inode)
 277 {
 278         int ret;
 279
 280         ret = ocfs2_journal_access(handle, dir, de_bh,
 281                                    OCFS2_JOURNAL_ACCESS_WRITE);
 282         if (ret) {
 283                 mlog_errno(ret);
 284                 goto out;
 285         }
 286
 287         de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
 288         ocfs2_set_de_type(de, new_entry_inode->i_mode);
 289
 290         ocfs2_journal_dirty(handle, de_bh);
 291
 292 out:
 293         return ret;
 294 }
 295
 296 /*
 297  * ocfs2_delete_entry deletes a directory entry by merging it with the
 298  * previous entry
 299  */
 300 int ocfs2_delete_entry(handle_t *handle,
 301                        struct inode *dir,
 302                        struct ocfs2_dir_entry *de_del,
 303                        struct buffer_head *bh)
 304 {
 305         struct ocfs2_dir_entry *de, *pde;
 306         int i, status = -ENOENT;
 307
 308         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
 309
 310         i = 0;
 311         pde = NULL;
 312         de = (struct ocfs2_dir_entry *) bh->b_data;
 313         while (i < bh->b_size) {
 314                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
 315                         status = -EIO;
 316                         mlog_errno(status);
 317                         goto bail;
 318                 }
 319                 if (de == de_del)  {
 320                         status = ocfs2_journal_access(handle, dir, bh,
 321                                                       OCFS2_JOURNAL_ACCESS_WRITE);
 322                         if (status < 0) {
 323                                 status = -EIO;
 324                                 mlog_errno(status);
 325                                 goto bail;
 326                         }
 327                         if (pde)
 328                                 pde->rec_len =
 329                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
 330                                                     le16_to_cpu(de->rec_len));
 331                         else
 332                                 de->inode = 0;
 333                         dir->i_version++;
 334                         status = ocfs2_journal_dirty(handle, bh);
 335                         goto bail;
 336                 }
 337                 i += le16_to_cpu(de->rec_len);
 338                 pde = de;
 339                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
 340         }
 341 bail:
 342         mlog_exit(status);
 343         return status;
 344 }
 345
 346 /* we don't always have a dentry for what we want to add, so people
 347  * like orphan dir can call this instead.
 348  *
 349  * If you pass me insert_bh, I'll skip the search of the other dir
 350  * blocks and put the record in there.
 351  */
 352 int __ocfs2_add_entry(handle_t *handle,
 353                       struct inode *dir,
 354                       const char *name, int namelen,
 355                       struct inode *inode, u64 blkno,
 356                       struct buffer_head *parent_fe_bh,
 357                       struct buffer_head *insert_bh)
 358 {
 359         unsigned long offset;
 360         unsigned short rec_len;
 361         struct ocfs2_dir_entry *de, *de1;
 362         struct super_block *sb;
 363         int retval, status;
 364
 365         mlog_entry_void();
 366
 367         sb = dir->i_sb;
 368
 369         if (!namelen)
 370                 return -EINVAL;
 371
 372         rec_len = OCFS2_DIR_REC_LEN(namelen);
 373         offset = 0;
 374         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
 375         while (1) {
 376                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
 377                 /* These checks should've already been passed by the
 378                  * prepare function, but I guess we can leave them
 379                  * here anyway. */
 380                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
 381                         retval = -ENOENT;
 382                         goto bail;
 383                 }
 384                 if (ocfs2_match(namelen, name, de)) {
 385                         retval = -EEXIST;
 386                         goto bail;
 387                 }
 388                 if (((le64_to_cpu(de->inode) == 0) &&
 389                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
 390                     (le16_to_cpu(de->rec_len) >=
 391                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
 392                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 393                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 394                         if (retval < 0) {
 395                                 mlog_errno(retval);
 396                                 goto bail;
 397                         }
 398
 399                         status = ocfs2_journal_access(handle, dir, insert_bh,
 400                                                       OCFS2_JOURNAL_ACCESS_WRITE);
 401                         /* By now the buffer is marked for journaling */
 402                         offset += le16_to_cpu(de->rec_len);
 403                         if (le64_to_cpu(de->inode)) {
 404                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
 405                                         OCFS2_DIR_REC_LEN(de->name_len));
 406                                 de1->rec_len =
 407                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
 408                                         OCFS2_DIR_REC_LEN(de->name_len));
 409                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 410                                 de = de1;
 411                         }
 412                         de->file_type = OCFS2_FT_UNKNOWN;
 413                         if (blkno) {
 414                                 de->inode = cpu_to_le64(blkno);
 415                                 ocfs2_set_de_type(de, inode->i_mode);
 416                         } else
 417                                 de->inode = 0;
 418                         de->name_len = namelen;
 419                         memcpy(de->name, name, namelen);
 420
 421                         dir->i_version++;
 422                         status = ocfs2_journal_dirty(handle, insert_bh);
 423                         retval = 0;
 424                         goto bail;
 425                 }
 426                 offset += le16_to_cpu(de->rec_len);
 427                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
 428         }
 429
 430         /* when you think about it, the assert above should prevent us
 431          * from ever getting here. */
 432         retval = -ENOSPC;
 433 bail:
 434
 435         mlog_exit(retval);
 436         return retval;
 437 }
 438
 439 static int ocfs2_dir_foreach_blk(struct inode *inode, unsigned long *f_version,
 440                                  loff_t *f_pos, void *priv, filldir_t filldir)
 441 {
 442         int error = 0;
 443         unsigned long offset, blk, last_ra_blk = 0;
 444         int i, stored;
 445         struct buffer_head * bh, * tmp;
 446         struct ocfs2_dir_entry * de;
 447         int err;
 448         struct super_block * sb = inode->i_sb;
 449         unsigned int ra_sectors = 16;
 450
 451         stored = 0;
 452         bh = NULL;
 453
 454         offset = (*f_pos) & (sb->s_blocksize - 1);
 455
 456         while (!error && !stored && *f_pos < i_size_read(inode)) {
 457                 blk = (*f_pos) >> sb->s_blocksize_bits;
 458                 bh = ocfs2_bread(inode, blk, &err, 0);
 459                 if (!bh) {
 460                         mlog(ML_ERROR,
 461                              "directory #%llu contains a hole at offset %lld\n",
 462                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 463                              *f_pos);
 464                         *f_pos += sb->s_blocksize - offset;
 465                         continue;
 466                 }
 467
 468                 /* The idea here is to begin with 8k read-ahead and to stay
 469                  * 4k ahead of our current position.
 470                  *
 471                  * TODO: Use the pagecache for this. We just need to
 472                  * make sure it's cluster-safe... */
 473                 if (!last_ra_blk
 474                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
 475                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
 476                              i > 0; i--) {
 477                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
 478                                 if (tmp)
 479                                         brelse(tmp);
 480                         }
 481                         last_ra_blk = blk;
 482                         ra_sectors = 8;
 483                 }
 484
 485 revalidate:
 486                 /* If the dir block has changed since the last call to
 487                  * readdir(2), then we might be pointing to an invalid
 488                  * dirent right now.  Scan from the start of the block
 489                  * to make sure. */
 490                 if (*f_version != inode->i_version) {
 491                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 492                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 493                                 /* It's too expensive to do a full
 494                                  * dirent test each time round this
 495                                  * loop, but we do have to test at
 496                                  * least that it is non-zero.  A
 497                                  * failure will be detected in the
 498                                  * dirent test below. */
 499                                 if (le16_to_cpu(de->rec_len) <
 500                                     OCFS2_DIR_REC_LEN(1))
 501                                         break;
 502                                 i += le16_to_cpu(de->rec_len);
 503                         }
 504                         offset = i;
 505                         *f_pos = ((*f_pos) & ~(sb->s_blocksize - 1))
 506                                 | offset;
 507                         *f_version = inode->i_version;
 508                 }
 509
 510                 while (!error && *f_pos < i_size_read(inode)
 511                        && offset < sb->s_blocksize) {
 512                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 513                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 514                                 /* On error, skip the f_pos to the
 515                                    next block. */
 516                                 *f_pos = ((*f_pos) | (sb->s_blocksize - 1)) + 1;
 517                                 brelse(bh);
 518                                 goto out;
 519                         }
 520                         offset += le16_to_cpu(de->rec_len);
 521                         if (le64_to_cpu(de->inode)) {
 522                                 /* We might block in the next section
 523                                  * if the data destination is
 524                                  * currently swapped out.  So, use a
 525                                  * version stamp to detect whether or
 526                                  * not the directory has been modified
 527                                  * during the copy operation.
 528                                  */
 529                                 unsigned long version = *f_version;
 530                                 unsigned char d_type = DT_UNKNOWN;
 531
 532                                 if (de->file_type < OCFS2_FT_MAX)
 533                                         d_type = ocfs2_filetype_table[de->file_type];
 534                                 error = filldir(priv, de->name,
 535                                                 de->name_len,
 536                                                 *f_pos,
 537                                                 le64_to_cpu(de->inode),
 538                                                 d_type);
 539                                 if (error)
 540                                         break;
 541                                 if (version != *f_version)
 542                                         goto revalidate;
 543                                 stored ++;
 544                         }
 545                         *f_pos += le16_to_cpu(de->rec_len);
 546                 }
 547                 offset = 0;
 548                 brelse(bh);
 549         }
 550
 551         stored = 0;
 552 out:
 553         return stored;
 554 }
 555
 556 /*
 557  * This is intended to be called from inside other kernel functions,
 558  * so we fake some arguments.
 559  */
 560 int ocfs2_dir_foreach(struct inode *inode, loff_t *f_pos, void *priv,
 561                       filldir_t filldir)
 562 {
 563         int ret = 0;
 564         unsigned long version = inode->i_version;
 565
 566         while (*f_pos < i_size_read(inode)) {
 567                 ret = ocfs2_dir_foreach_blk(inode, &version, f_pos, priv,
 568                                             filldir);
 569                 if (ret)
 570                         break;
 571         }
 572
 573         return 0;
 574 }
 575
 576 /*
 577  * ocfs2_readdir()
 578  *
 579  */
 580 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
 581 {
 582         int error = 0;
 583         struct inode *inode = filp->f_path.dentry->d_inode;
 584         int lock_level = 0;
 585
 586         mlog_entry("dirino=%llu\n",
 587                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
 588
 589         error = ocfs2_meta_lock_atime(inode, filp->f_vfsmnt, &lock_level);
 590         if (lock_level && error >= 0) {
 591                 /* We release EX lock which used to update atime
 592                  * and get PR lock again to reduce contention
 593                  * on commonly accessed directories. */
 594                 ocfs2_meta_unlock(inode, 1);
 595                 lock_level = 0;
 596                 error = ocfs2_meta_lock(inode, NULL, 0);
 597         }
 598         if (error < 0) {
 599                 if (error != -ENOENT)
 600                         mlog_errno(error);
 601                 /* we haven't got any yet, so propagate the error. */
 602                 goto bail_nolock;
 603         }
 604
 605         error = ocfs2_dir_foreach_blk(inode, &filp->f_version, &filp->f_pos,
 606                                       dirent, filldir);
 607
 608         ocfs2_meta_unlock(inode, lock_level);
 609
 610 bail_nolock:
 611         mlog_exit(error);
 612
 613         return error;
 614 }
 615
 616 /*
 617  * NOTE: this should always be called with parent dir i_mutex taken.
 618  */
 619 int ocfs2_find_files_on_disk(const char *name,
 620                              int namelen,
 621                              u64 *blkno,
 622                              struct inode *inode,
 623                              struct buffer_head **dirent_bh,
 624                              struct ocfs2_dir_entry **dirent)
 625 {
 626         int status = -ENOENT;
 627
 628         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 629                    namelen, name, blkno, inode, dirent_bh, dirent);
 630
 631         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 632         if (!*dirent_bh || !*dirent) {
 633                 status = -ENOENT;
 634                 goto leave;
 635         }
 636
 637         *blkno = le64_to_cpu((*dirent)->inode);
 638
 639         status = 0;
 640 leave:
 641         if (status < 0) {
 642                 *dirent = NULL;
 643                 if (*dirent_bh) {
 644                         brelse(*dirent_bh);
 645                         *dirent_bh = NULL;
 646                 }
 647         }
 648
 649         mlog_exit(status);
 650         return status;
 651 }
 652
 653 /*
 654  * Convenience function for callers which just want the block number
 655  * mapped to a name and don't require the full dirent info, etc.
 656  */
 657 int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
 658                                int namelen, u64 *blkno)
 659 {
 660         int ret;
 661         struct buffer_head *bh = NULL;
 662         struct ocfs2_dir_entry *dirent = NULL;
 663
 664         ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &bh, &dirent);
 665         brelse(bh);
 666
 667         return ret;
 668 }
 669
 670 /* Check for a name within a directory.
 671  *
 672  * Return 0 if the name does not exist
 673  * Return -EEXIST if the directory contains the name
 674  *
 675  * Callers should have i_mutex + a cluster lock on dir
 676  */
 677 int ocfs2_check_dir_for_entry(struct inode *dir,
 678                               const char *name,
 679                               int namelen)
 680 {
 681         int ret;
 682         struct buffer_head *dirent_bh = NULL;
 683         struct ocfs2_dir_entry *dirent = NULL;
 684
 685         mlog_entry("dir %llu, name '%.*s'\n",
 686                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 687
 688         ret = -EEXIST;
 689         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 690         if (dirent_bh)
 691                 goto bail;
 692
 693         ret = 0;
 694 bail:
 695         if (dirent_bh)
 696                 brelse(dirent_bh);
 697
 698         mlog_exit(ret);
 699         return ret;
 700 }
 701
 702 struct ocfs2_empty_dir_priv {
 703         unsigned seen_dot;
 704         unsigned seen_dot_dot;
 705         unsigned seen_other;
 706 };
 707 static int ocfs2_empty_dir_filldir(void *priv, const char *name, int name_len,
 708                                    loff_t pos, u64 ino, unsigned type)
 709 {
 710         struct ocfs2_empty_dir_priv *p = priv;
 711
 712         /*
 713          * Check the positions of "." and ".." records to be sure
 714          * they're in the correct place.
 715          */
 716         if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
 717                 p->seen_dot = 1;
 718                 return 0;
 719         }
 720
 721         if (name_len == 2 && !strncmp("..", name, 2) &&
 722             pos == OCFS2_DIR_REC_LEN(1)) {
 723                 p->seen_dot_dot = 1;
 724                 return 0;
 725         }
 726
 727         p->seen_other = 1;
 728         return 1;
 729 }
 730 /*
 731  * routine to check that the specified directory is empty (for rmdir)
 732  *
 733  * Returns 1 if dir is empty, zero otherwise.
 734  */
 735 int ocfs2_empty_dir(struct inode *inode)
 736 {
 737         int ret;
 738         loff_t start = 0;
 739         struct ocfs2_empty_dir_priv priv;
 740
 741         memset(&priv, 0, sizeof(priv));
 742
 743         ret = ocfs2_dir_foreach(inode, &start, &priv, ocfs2_empty_dir_filldir);
 744         if (ret)
 745                 mlog_errno(ret);
 746
 747         if (!priv.seen_dot || !priv.seen_dot_dot) {
 748                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 749                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 750                 /*
 751                  * XXX: Is it really safe to allow an unlink to continue?
 752                  */
 753                 return 1;
 754         }
 755
 756         return !priv.seen_other;
 757 }
 758
 759 int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 760                        handle_t *handle,
 761                        struct inode *parent,
 762                        struct inode *inode,
 763                        struct buffer_head *fe_bh,
 764                        struct ocfs2_alloc_context *data_ac)
 765 {
 766         int status;
 767         struct buffer_head *new_bh = NULL;
 768         struct ocfs2_dir_entry *de = NULL;
 769
 770         mlog_entry_void();
 771
 772         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 773                                      data_ac, NULL, &new_bh);
 774         if (status < 0) {
 775                 mlog_errno(status);
 776                 goto bail;
 777         }
 778
 779         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 780
 781         status = ocfs2_journal_access(handle, inode, new_bh,
 782                                       OCFS2_JOURNAL_ACCESS_CREATE);
 783         if (status < 0) {
 784                 mlog_errno(status);
 785                 goto bail;
 786         }
 787         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 788
 789         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 790         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 791         de->name_len = 1;
 792         de->rec_len =
 793                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 794         strcpy(de->name, ".");
 795         ocfs2_set_de_type(de, S_IFDIR);
 796         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 797         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 798         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 799                                   OCFS2_DIR_REC_LEN(1));
 800         de->name_len = 2;
 801         strcpy(de->name, "..");
 802         ocfs2_set_de_type(de, S_IFDIR);
 803
 804         status = ocfs2_journal_dirty(handle, new_bh);
 805         if (status < 0) {
 806                 mlog_errno(status);
 807                 goto bail;
 808         }
 809
 810         i_size_write(inode, inode->i_sb->s_blocksize);
 811         inode->i_nlink = 2;
 812         inode->i_blocks = ocfs2_inode_sector_count(inode);
 813         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 814         if (status < 0) {
 815                 mlog_errno(status);
 816                 goto bail;
 817         }
 818
 819         status = 0;
 820 bail:
 821         if (new_bh)
 822                 brelse(new_bh);
 823
 824         mlog_exit(status);
 825         return status;
 826 }
 827
 828 /* returns a bh of the 1st new block in the allocation. */
 829 static int ocfs2_do_extend_dir(struct super_block *sb,
 830                                handle_t *handle,
 831                                struct inode *dir,
 832                                struct buffer_head *parent_fe_bh,
 833                                struct ocfs2_alloc_context *data_ac,
 834                                struct ocfs2_alloc_context *meta_ac,
 835                                struct buffer_head **new_bh)
 836 {
 837         int status;
 838         int extend;
 839         u64 p_blkno, v_blkno;
 840
 841         spin_lock(&OCFS2_I(dir)->ip_lock);
 842         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 843         spin_unlock(&OCFS2_I(dir)->ip_lock);
 844
 845         if (extend) {
 846                 u32 offset = OCFS2_I(dir)->ip_clusters;
 847
 848                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, &offset,
 849                                                     1, 0, parent_fe_bh, handle,
 850                                                     data_ac, meta_ac, NULL);
 851                 BUG_ON(status == -EAGAIN);
 852                 if (status < 0) {
 853                         mlog_errno(status);
 854                         goto bail;
 855                 }
 856         }
 857
 858         v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
 859         status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
 860         if (status < 0) {
 861                 mlog_errno(status);
 862                 goto bail;
 863         }
 864
 865         *new_bh = sb_getblk(sb, p_blkno);
 866         if (!*new_bh) {
 867                 status = -EIO;
 868                 mlog_errno(status);
 869                 goto bail;
 870         }
 871         status = 0;
 872 bail:
 873         mlog_exit(status);
 874         return status;
 875 }
 876
 877 /* assumes you already have a cluster lock on the directory. */
 878 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 879                             struct inode *dir,
 880                             struct buffer_head *parent_fe_bh,
 881                             struct buffer_head **new_de_bh)
 882 {
 883         int status = 0;
 884         int credits, num_free_extents, drop_alloc_sem = 0;
 885         loff_t dir_i_size;
 886         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 887         struct ocfs2_alloc_context *data_ac = NULL;
 888         struct ocfs2_alloc_context *meta_ac = NULL;
 889         handle_t *handle = NULL;
 890         struct buffer_head *new_bh = NULL;
 891         struct ocfs2_dir_entry * de;
 892         struct super_block *sb = osb->sb;
 893
 894         mlog_entry_void();
 895
 896         dir_i_size = i_size_read(dir);
 897         mlog(0, "extending dir %llu (i_size = %lld)\n",
 898              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 899
 900         /* dir->i_size is always block aligned. */
 901         spin_lock(&OCFS2_I(dir)->ip_lock);
 902         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 903                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 904                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 905                 if (num_free_extents < 0) {
 906                         status = num_free_extents;
 907                         mlog_errno(status);
 908                         goto bail;
 909                 }
 910
 911                 if (!num_free_extents) {
 912                         status = ocfs2_reserve_new_metadata(osb, fe, &meta_ac);
 913                         if (status < 0) {
 914                                 if (status != -ENOSPC)
 915                                         mlog_errno(status);
 916                                 goto bail;
 917                         }
 918                 }
 919
 920                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 921                 if (status < 0) {
 922                         if (status != -ENOSPC)
 923                                 mlog_errno(status);
 924                         goto bail;
 925                 }
 926
 927                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 928         } else {
 929                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 930                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 931         }
 932
 933         down_write(&OCFS2_I(dir)->ip_alloc_sem);
 934         drop_alloc_sem = 1;
 935
 936         handle = ocfs2_start_trans(osb, credits);
 937         if (IS_ERR(handle)) {
 938                 status = PTR_ERR(handle);
 939                 handle = NULL;
 940                 mlog_errno(status);
 941                 goto bail;
 942         }
 943
 944         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 945                                      data_ac, meta_ac, &new_bh);
 946         if (status < 0) {
 947                 mlog_errno(status);
 948                 goto bail;
 949         }
 950
 951         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 952
 953         status = ocfs2_journal_access(handle, dir, new_bh,
 954                                       OCFS2_JOURNAL_ACCESS_CREATE);
 955         if (status < 0) {
 956                 mlog_errno(status);
 957                 goto bail;
 958         }
 959         memset(new_bh->b_data, 0, sb->s_blocksize);
 960         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 961         de->inode = 0;
 962         de->rec_len = cpu_to_le16(sb->s_blocksize);
 963         status = ocfs2_journal_dirty(handle, new_bh);
 964         if (status < 0) {
 965                 mlog_errno(status);
 966                 goto bail;
 967         }
 968
 969         dir_i_size += dir->i_sb->s_blocksize;
 970         i_size_write(dir, dir_i_size);
 971         dir->i_blocks = ocfs2_inode_sector_count(dir);
 972         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 973         if (status < 0) {
 974                 mlog_errno(status);
 975                 goto bail;
 976         }
 977
 978         *new_de_bh = new_bh;
 979         get_bh(*new_de_bh);
 980 bail:
 981         if (drop_alloc_sem)
 982                 up_write(&OCFS2_I(dir)->ip_alloc_sem);
 983         if (handle)
 984                 ocfs2_commit_trans(osb, handle);
 985
 986         if (data_ac)
 987                 ocfs2_free_alloc_context(data_ac);
 988         if (meta_ac)
 989                 ocfs2_free_alloc_context(meta_ac);
 990
 991         if (new_bh)
 992                 brelse(new_bh);
 993
 994         mlog_exit(status);
 995         return status;
 996 }
 997
 998 /*
 999  * Search the dir for a good spot, extending it if necessary. The
1000  * block containing an appropriate record is returned in ret_de_bh.
1001  */
1002 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
1003                                  struct inode *dir,
1004                                  struct buffer_head *parent_fe_bh,
1005                                  const char *name,
1006                                  int namelen,
1007                                  struct buffer_head **ret_de_bh)
1008 {
1009         unsigned long offset;
1010         struct buffer_head * bh = NULL;
1011         unsigned short rec_len;
1012         struct ocfs2_dinode *fe;
1013         struct ocfs2_dir_entry *de;
1014         struct super_block *sb;
1015         int status;
1016
1017         mlog_entry_void();
1018
1019         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
1020              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
1021
1022         BUG_ON(!S_ISDIR(dir->i_mode));
1023         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1024         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
1025
1026         sb = dir->i_sb;
1027
1028         if (!namelen) {
1029                 status = -EINVAL;
1030                 mlog_errno(status);
1031                 goto bail;
1032         }
1033
1034         bh = ocfs2_bread(dir, 0, &status, 0);
1035         if (!bh) {
1036                 mlog_errno(status);
1037                 goto bail;
1038         }
1039
1040         rec_len = OCFS2_DIR_REC_LEN(namelen);
1041         offset = 0;
1042         de = (struct ocfs2_dir_entry *) bh->b_data;
1043         while (1) {
1044                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
1045                         brelse(bh);
1046                         bh = NULL;
1047
1048                         if (i_size_read(dir) <= offset) {
1049                                 status = ocfs2_extend_dir(osb,
1050                                                           dir,
1051                                                           parent_fe_bh,
1052                                                           &bh);
1053                                 if (status < 0) {
1054                                         mlog_errno(status);
1055                                         goto bail;
1056                                 }
1057                                 BUG_ON(!bh);
1058                                 *ret_de_bh = bh;
1059                                 get_bh(*ret_de_bh);
1060                                 goto bail;
1061                         }
1062                         bh = ocfs2_bread(dir,
1063                                          offset >> sb->s_blocksize_bits,
1064                                          &status,
1065                                          0);
1066                         if (!bh) {
1067                                 mlog_errno(status);
1068                                 goto bail;
1069                         }
1070                         /* move to next block */
1071                         de = (struct ocfs2_dir_entry *) bh->b_data;
1072                 }
1073                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1074                         status = -ENOENT;
1075                         goto bail;
1076                 }
1077                 if (ocfs2_match(namelen, name, de)) {
1078                         status = -EEXIST;
1079                         goto bail;
1080                 }
1081                 if (((le64_to_cpu(de->inode) == 0) &&
1082                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1083                     (le16_to_cpu(de->rec_len) >=
1084                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1085                         /* Ok, we found a spot. Return this bh and let
1086                          * the caller actually fill it in. */
1087                         *ret_de_bh = bh;
1088                         get_bh(*ret_de_bh);
1089                         status = 0;
1090                         goto bail;
1091                 }
1092                 offset += le16_to_cpu(de->rec_len);
1093                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
1094         }
1095
1096         status = 0;
1097 bail:
1098         if (bh)
1099                 brelse(bh);
1100
1101         mlog_exit(status);
1102         return status;
1103 }