git.oblomov.eu Git - linux-2.6/blob - fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   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 "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "symlink.h"
  60 #include "sysfile.h"
  61 #include "uptodate.h"
  62 #include "vote.h"
  63
  64 #include "buffer_head_io.h"
  65
  66 #define NAMEI_RA_CHUNKS  2
  67 #define NAMEI_RA_BLOCKS  4
  68 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  69 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  70
  71 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  72                                         struct inode *dir,
  73                                         const char *name, int namelen,
  74                                         unsigned long offset,
  75                                         struct ocfs2_dir_entry **res_dir);
  76
  77 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
  78                               struct inode *dir,
  79                               struct ocfs2_dir_entry *de_del,
  80                               struct buffer_head *bh);
  81
  82 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
  83                              struct inode *dir,
  84                              const char *name, int namelen,
  85                              struct inode *inode, u64 blkno,
  86                              struct buffer_head *parent_fe_bh,
  87                              struct buffer_head *insert_bh);
  88
  89 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  90                               struct inode *dir,
  91                               struct dentry *dentry, int mode,
  92                               dev_t dev,
  93                               struct buffer_head **new_fe_bh,
  94                               struct buffer_head *parent_fe_bh,
  95                               struct ocfs2_journal_handle *handle,
  96                               struct inode **ret_inode,
  97                               struct ocfs2_alloc_context *inode_ac);
  98
  99 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 100                               struct ocfs2_journal_handle *handle,
 101                               struct inode *parent,
 102                               struct inode *inode,
 103                               struct buffer_head *fe_bh,
 104                               struct ocfs2_alloc_context *data_ac);
 105
 106 static int ocfs2_double_lock(struct ocfs2_super *osb,
 107                              struct ocfs2_journal_handle *handle,
 108                              struct buffer_head **bh1,
 109                              struct inode *inode1,
 110                              struct buffer_head **bh2,
 111                              struct inode *inode2);
 112
 113 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 114                                     struct ocfs2_journal_handle *handle,
 115                                     struct inode *inode,
 116                                     char *name,
 117                                     struct buffer_head **de_bh);
 118
 119 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 120                             struct ocfs2_journal_handle *handle,
 121                             struct inode *inode,
 122                             struct ocfs2_dinode *fe,
 123                             char *name,
 124                             struct buffer_head *de_bh);
 125
 126 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 127                                      struct ocfs2_journal_handle *handle,
 128                                      struct inode *inode,
 129                                      const char *symname);
 130
 131 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
 132                                   struct dentry *dentry,
 133                                   struct inode *inode, u64 blkno,
 134                                   struct buffer_head *parent_fe_bh,
 135                                   struct buffer_head *insert_bh)
 136 {
 137         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 138                                  dentry->d_name.name, dentry->d_name.len,
 139                                  inode, blkno, parent_fe_bh, insert_bh);
 140 }
 141
 142 /* An orphan dir name is an 8 byte value, printed as a hex string */
 143 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 144
 145 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 146                                    struct nameidata *nd)
 147 {
 148         int status;
 149         u64 blkno;
 150         struct buffer_head *dirent_bh = NULL;
 151         struct inode *inode = NULL;
 152         struct dentry *ret;
 153         struct ocfs2_dir_entry *dirent;
 154         struct ocfs2_inode_info *oi;
 155
 156         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 157                    dentry->d_name.len, dentry->d_name.name);
 158
 159         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 160                 ret = ERR_PTR(-ENAMETOOLONG);
 161                 goto bail;
 162         }
 163
 164         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 165              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 166
 167         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
 168         if (status < 0) {
 169                 if (status != -ENOENT)
 170                         mlog_errno(status);
 171                 ret = ERR_PTR(status);
 172                 goto bail;
 173         }
 174
 175         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 176                                           dentry->d_name.len, &blkno,
 177                                           dir, &dirent_bh, &dirent);
 178         if (status < 0)
 179                 goto bail_add;
 180
 181         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
 182         if (IS_ERR(inode)) {
 183                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 184                      (unsigned long long)blkno);
 185                 ret = ERR_PTR(-EACCES);
 186                 goto bail_unlock;
 187         }
 188
 189         oi = OCFS2_I(inode);
 190         /* Clear any orphaned state... If we were able to look up the
 191          * inode from a directory, it certainly can't be orphaned. We
 192          * might have the bad state from a node which intended to
 193          * orphan this inode but crashed before it could commit the
 194          * unlink. */
 195         spin_lock(&oi->ip_lock);
 196         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 197         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 198         spin_unlock(&oi->ip_lock);
 199
 200 bail_add:
 201
 202         dentry->d_op = &ocfs2_dentry_ops;
 203         ret = d_splice_alias(inode, dentry);
 204
 205 bail_unlock:
 206         /* Don't drop the cluster lock until *after* the d_add --
 207          * unlink on another node will message us to remove that
 208          * dentry under this lock so otherwise we can race this with
 209          * the vote thread and have a stale dentry. */
 210         ocfs2_meta_unlock(dir, 0);
 211
 212 bail:
 213         if (dirent_bh)
 214                 brelse(dirent_bh);
 215
 216         mlog_exit_ptr(ret);
 217
 218         return ret;
 219 }
 220
 221 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 222                               struct ocfs2_journal_handle *handle,
 223                               struct inode *parent,
 224                               struct inode *inode,
 225                               struct buffer_head *fe_bh,
 226                               struct ocfs2_alloc_context *data_ac)
 227 {
 228         int status;
 229         struct buffer_head *new_bh = NULL;
 230         struct ocfs2_dir_entry *de = NULL;
 231
 232         mlog_entry_void();
 233
 234         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 235                                      data_ac, NULL, &new_bh);
 236         if (status < 0) {
 237                 mlog_errno(status);
 238                 goto bail;
 239         }
 240
 241         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 242
 243         status = ocfs2_journal_access(handle, inode, new_bh,
 244                                       OCFS2_JOURNAL_ACCESS_CREATE);
 245         if (status < 0) {
 246                 mlog_errno(status);
 247                 goto bail;
 248         }
 249         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 250
 251         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 252         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 253         de->name_len = 1;
 254         de->rec_len =
 255                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 256         strcpy(de->name, ".");
 257         ocfs2_set_de_type(de, S_IFDIR);
 258         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 259         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 260         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 261                                   OCFS2_DIR_REC_LEN(1));
 262         de->name_len = 2;
 263         strcpy(de->name, "..");
 264         ocfs2_set_de_type(de, S_IFDIR);
 265
 266         status = ocfs2_journal_dirty(handle, new_bh);
 267         if (status < 0) {
 268                 mlog_errno(status);
 269                 goto bail;
 270         }
 271
 272         i_size_write(inode, inode->i_sb->s_blocksize);
 273         inode->i_nlink = 2;
 274         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 275         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 276         if (status < 0) {
 277                 mlog_errno(status);
 278                 goto bail;
 279         }
 280
 281         status = 0;
 282 bail:
 283         if (new_bh)
 284                 brelse(new_bh);
 285
 286         mlog_exit(status);
 287         return status;
 288 }
 289
 290 static int ocfs2_mknod(struct inode *dir,
 291                        struct dentry *dentry,
 292                        int mode,
 293                        dev_t dev)
 294 {
 295         int status = 0;
 296         struct buffer_head *parent_fe_bh = NULL;
 297         struct ocfs2_journal_handle *handle = NULL;
 298         struct ocfs2_super *osb;
 299         struct ocfs2_dinode *dirfe;
 300         struct buffer_head *new_fe_bh = NULL;
 301         struct buffer_head *de_bh = NULL;
 302         struct inode *inode = NULL;
 303         struct ocfs2_alloc_context *inode_ac = NULL;
 304         struct ocfs2_alloc_context *data_ac = NULL;
 305
 306         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 307                    (unsigned long)dev, dentry->d_name.len,
 308                    dentry->d_name.name);
 309
 310         /* get our super block */
 311         osb = OCFS2_SB(dir->i_sb);
 312
 313         handle = ocfs2_alloc_handle(osb);
 314         if (handle == NULL) {
 315                 status = -ENOMEM;
 316                 mlog_errno(status);
 317                 goto leave;
 318         }
 319
 320         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 321         if (status < 0) {
 322                 if (status != -ENOENT)
 323                         mlog_errno(status);
 324                 goto leave;
 325         }
 326
 327         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 328                 status = -EMLINK;
 329                 goto leave;
 330         }
 331
 332         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 333         if (!dirfe->i_links_count) {
 334                 /* can't make a file in a deleted directory. */
 335                 status = -ENOENT;
 336                 goto leave;
 337         }
 338
 339         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 340                                            dentry->d_name.len);
 341         if (status)
 342                 goto leave;
 343
 344         /* get a spot inside the dir. */
 345         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 346                                               dentry->d_name.name,
 347                                               dentry->d_name.len, &de_bh);
 348         if (status < 0) {
 349                 mlog_errno(status);
 350                 goto leave;
 351         }
 352
 353         /* reserve an inode spot */
 354         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
 355         if (status < 0) {
 356                 if (status != -ENOSPC)
 357                         mlog_errno(status);
 358                 goto leave;
 359         }
 360
 361         /* are we making a directory? If so, reserve a cluster for his
 362          * 1st extent. */
 363         if (S_ISDIR(mode)) {
 364                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 365                 if (status < 0) {
 366                         if (status != -ENOSPC)
 367                                 mlog_errno(status);
 368                         goto leave;
 369                 }
 370         }
 371
 372         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
 373         if (IS_ERR(handle)) {
 374                 status = PTR_ERR(handle);
 375                 handle = NULL;
 376                 mlog_errno(status);
 377                 goto leave;
 378         }
 379
 380         /* do the real work now. */
 381         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 382                                     &new_fe_bh, parent_fe_bh, handle,
 383                                     &inode, inode_ac);
 384         if (status < 0) {
 385                 mlog_errno(status);
 386                 goto leave;
 387         }
 388
 389         if (S_ISDIR(mode)) {
 390                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 391                                             new_fe_bh, data_ac);
 392                 if (status < 0) {
 393                         mlog_errno(status);
 394                         goto leave;
 395                 }
 396
 397                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 398                                               OCFS2_JOURNAL_ACCESS_WRITE);
 399                 if (status < 0) {
 400                         mlog_errno(status);
 401                         goto leave;
 402                 }
 403                 le16_add_cpu(&dirfe->i_links_count, 1);
 404                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 405                 if (status < 0) {
 406                         mlog_errno(status);
 407                         goto leave;
 408                 }
 409                 dir->i_nlink++;
 410         }
 411
 412         status = ocfs2_add_entry(handle, dentry, inode,
 413                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 414                                  de_bh);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto leave;
 418         }
 419
 420         insert_inode_hash(inode);
 421         dentry->d_op = &ocfs2_dentry_ops;
 422         d_instantiate(dentry, inode);
 423         status = 0;
 424 leave:
 425         if (handle)
 426                 ocfs2_commit_trans(handle);
 427
 428         if (status == -ENOSPC)
 429                 mlog(0, "Disk is full\n");
 430
 431         if (new_fe_bh)
 432                 brelse(new_fe_bh);
 433
 434         if (de_bh)
 435                 brelse(de_bh);
 436
 437         if (parent_fe_bh)
 438                 brelse(parent_fe_bh);
 439
 440         if ((status < 0) && inode)
 441                 iput(inode);
 442
 443         if (inode_ac)
 444                 ocfs2_free_alloc_context(inode_ac);
 445
 446         if (data_ac)
 447                 ocfs2_free_alloc_context(data_ac);
 448
 449         mlog_exit(status);
 450
 451         return status;
 452 }
 453
 454 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 455                               struct inode *dir,
 456                               struct dentry *dentry, int mode,
 457                               dev_t dev,
 458                               struct buffer_head **new_fe_bh,
 459                               struct buffer_head *parent_fe_bh,
 460                               struct ocfs2_journal_handle *handle,
 461                               struct inode **ret_inode,
 462                               struct ocfs2_alloc_context *inode_ac)
 463 {
 464         int status = 0;
 465         struct ocfs2_dinode *fe = NULL;
 466         struct ocfs2_extent_list *fel;
 467         u64 fe_blkno = 0;
 468         u16 suballoc_bit;
 469         struct inode *inode = NULL;
 470
 471         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 472                    (unsigned long)dev, dentry->d_name.len,
 473                    dentry->d_name.name);
 474
 475         *new_fe_bh = NULL;
 476         *ret_inode = NULL;
 477
 478         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 479                                        &fe_blkno);
 480         if (status < 0) {
 481                 mlog_errno(status);
 482                 goto leave;
 483         }
 484
 485         inode = new_inode(dir->i_sb);
 486         if (IS_ERR(inode)) {
 487                 status = PTR_ERR(inode);
 488                 mlog(ML_ERROR, "new_inode failed!\n");
 489                 goto leave;
 490         }
 491
 492         /* populate as many fields early on as possible - many of
 493          * these are used by the support functions here and in
 494          * callers. */
 495         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 496         OCFS2_I(inode)->ip_blkno = fe_blkno;
 497         if (S_ISDIR(mode))
 498                 inode->i_nlink = 2;
 499         else
 500                 inode->i_nlink = 1;
 501         inode->i_mode = mode;
 502         spin_lock(&osb->osb_lock);
 503         inode->i_generation = osb->s_next_generation++;
 504         spin_unlock(&osb->osb_lock);
 505
 506         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 507         if (!*new_fe_bh) {
 508                 status = -EIO;
 509                 mlog_errno(status);
 510                 goto leave;
 511         }
 512         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 513
 514         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 515                                       OCFS2_JOURNAL_ACCESS_CREATE);
 516         if (status < 0) {
 517                 mlog_errno(status);
 518                 goto leave;
 519         }
 520
 521         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 522         memset(fe, 0, osb->sb->s_blocksize);
 523
 524         fe->i_generation = cpu_to_le32(inode->i_generation);
 525         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 526         fe->i_blkno = cpu_to_le64(fe_blkno);
 527         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 528         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 529         fe->i_uid = cpu_to_le32(current->fsuid);
 530         if (dir->i_mode & S_ISGID) {
 531                 fe->i_gid = cpu_to_le32(dir->i_gid);
 532                 if (S_ISDIR(mode))
 533                         mode |= S_ISGID;
 534         } else
 535                 fe->i_gid = cpu_to_le32(current->fsgid);
 536         fe->i_mode = cpu_to_le16(mode);
 537         if (S_ISCHR(mode) || S_ISBLK(mode))
 538                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 539
 540         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 541
 542         fe->i_last_eb_blk = 0;
 543         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 544         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 545         fe->i_atime = fe->i_ctime = fe->i_mtime =
 546                 cpu_to_le64(CURRENT_TIME.tv_sec);
 547         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 548                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 549         fe->i_dtime = 0;
 550
 551         fel = &fe->id2.i_list;
 552         fel->l_tree_depth = 0;
 553         fel->l_next_free_rec = 0;
 554         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 555
 556         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 557         if (status < 0) {
 558                 mlog_errno(status);
 559                 goto leave;
 560         }
 561
 562         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 563                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 564                      "i_blkno=%llu, i_ino=%lu\n",
 565                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 566                      (unsigned long long)fe->i_blkno, inode->i_ino);
 567                 BUG();
 568         }
 569
 570         ocfs2_inode_set_new(osb, inode);
 571         status = ocfs2_create_new_inode_locks(inode);
 572         if (status < 0)
 573                 mlog_errno(status);
 574
 575         status = 0; /* error in ocfs2_create_new_inode_locks is not
 576                      * critical */
 577
 578         *ret_inode = inode;
 579 leave:
 580         if (status < 0) {
 581                 if (*new_fe_bh) {
 582                         brelse(*new_fe_bh);
 583                         *new_fe_bh = NULL;
 584                 }
 585                 if (inode)
 586                         iput(inode);
 587         }
 588
 589         mlog_exit(status);
 590         return status;
 591 }
 592
 593 static int ocfs2_mkdir(struct inode *dir,
 594                        struct dentry *dentry,
 595                        int mode)
 596 {
 597         int ret;
 598
 599         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 600                    dentry->d_name.len, dentry->d_name.name);
 601         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 602         mlog_exit(ret);
 603
 604         return ret;
 605 }
 606
 607 static int ocfs2_create(struct inode *dir,
 608                         struct dentry *dentry,
 609                         int mode,
 610                         struct nameidata *nd)
 611 {
 612         int ret;
 613
 614         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 615                    dentry->d_name.len, dentry->d_name.name);
 616         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 617         mlog_exit(ret);
 618
 619         return ret;
 620 }
 621
 622 static int ocfs2_link(struct dentry *old_dentry,
 623                       struct inode *dir,
 624                       struct dentry *dentry)
 625 {
 626         struct ocfs2_journal_handle *handle = NULL;
 627         struct inode *inode = old_dentry->d_inode;
 628         int err;
 629         struct buffer_head *fe_bh = NULL;
 630         struct buffer_head *parent_fe_bh = NULL;
 631         struct buffer_head *de_bh = NULL;
 632         struct ocfs2_dinode *fe = NULL;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634
 635         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 636                    old_dentry->d_name.len, old_dentry->d_name.name,
 637                    dentry->d_name.len, dentry->d_name.name);
 638
 639         if (S_ISDIR(inode->i_mode)) {
 640                 err = -EPERM;
 641                 goto bail;
 642         }
 643
 644         if (inode->i_nlink >= OCFS2_LINK_MAX) {
 645                 err = -EMLINK;
 646                 goto bail;
 647         }
 648
 649         handle = ocfs2_alloc_handle(osb);
 650         if (handle == NULL) {
 651                 err = -ENOMEM;
 652                 goto bail;
 653         }
 654
 655         err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 656         if (err < 0) {
 657                 if (err != -ENOENT)
 658                         mlog_errno(err);
 659                 goto bail;
 660         }
 661
 662         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 663                                         dentry->d_name.len);
 664         if (err)
 665                 goto bail;
 666
 667         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 668                                            dentry->d_name.name,
 669                                            dentry->d_name.len, &de_bh);
 670         if (err < 0) {
 671                 mlog_errno(err);
 672                 goto bail;
 673         }
 674
 675         err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 676         if (err < 0) {
 677                 if (err != -ENOENT)
 678                         mlog_errno(err);
 679                 goto bail;
 680         }
 681
 682         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 683         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 684                 err = -EMLINK;
 685                 goto bail;
 686         }
 687
 688         handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
 689         if (IS_ERR(handle)) {
 690                 err = PTR_ERR(handle);
 691                 handle = NULL;
 692                 mlog_errno(err);
 693                 goto bail;
 694         }
 695
 696         err = ocfs2_journal_access(handle, inode, fe_bh,
 697                                    OCFS2_JOURNAL_ACCESS_WRITE);
 698         if (err < 0) {
 699                 mlog_errno(err);
 700                 goto bail;
 701         }
 702
 703         inode->i_nlink++;
 704         inode->i_ctime = CURRENT_TIME;
 705         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 706         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 707         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 708
 709         err = ocfs2_journal_dirty(handle, fe_bh);
 710         if (err < 0) {
 711                 le16_add_cpu(&fe->i_links_count, -1);
 712                 inode->i_nlink--;
 713                 mlog_errno(err);
 714                 goto bail;
 715         }
 716
 717         err = ocfs2_add_entry(handle, dentry, inode,
 718                               OCFS2_I(inode)->ip_blkno,
 719                               parent_fe_bh, de_bh);
 720         if (err) {
 721                 le16_add_cpu(&fe->i_links_count, -1);
 722                 inode->i_nlink--;
 723                 mlog_errno(err);
 724                 goto bail;
 725         }
 726
 727         atomic_inc(&inode->i_count);
 728         dentry->d_op = &ocfs2_dentry_ops;
 729         d_instantiate(dentry, inode);
 730 bail:
 731         if (handle)
 732                 ocfs2_commit_trans(handle);
 733         if (de_bh)
 734                 brelse(de_bh);
 735         if (fe_bh)
 736                 brelse(fe_bh);
 737         if (parent_fe_bh)
 738                 brelse(parent_fe_bh);
 739
 740         mlog_exit(err);
 741
 742         return err;
 743 }
 744
 745 static int ocfs2_unlink(struct inode *dir,
 746                         struct dentry *dentry)
 747 {
 748         int status;
 749         unsigned int saved_nlink = 0;
 750         struct inode *inode = dentry->d_inode;
 751         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 752         u64 blkno;
 753         struct ocfs2_dinode *fe = NULL;
 754         struct buffer_head *fe_bh = NULL;
 755         struct buffer_head *parent_node_bh = NULL;
 756         struct ocfs2_journal_handle *handle = NULL;
 757         struct ocfs2_dir_entry *dirent = NULL;
 758         struct buffer_head *dirent_bh = NULL;
 759         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 760         struct buffer_head *orphan_entry_bh = NULL;
 761
 762         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 763                    dentry->d_name.len, dentry->d_name.name);
 764
 765         BUG_ON(dentry->d_parent->d_inode != dir);
 766
 767         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 768
 769         if (inode == osb->root_inode) {
 770                 mlog(0, "Cannot delete the root directory\n");
 771                 status = -EPERM;
 772                 goto leave;
 773         }
 774
 775         handle = ocfs2_alloc_handle(osb);
 776         if (handle == NULL) {
 777                 status = -ENOMEM;
 778                 mlog_errno(status);
 779                 goto leave;
 780         }
 781
 782         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
 783         if (status < 0) {
 784                 if (status != -ENOENT)
 785                         mlog_errno(status);
 786                 goto leave;
 787         }
 788
 789         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 790                                           dentry->d_name.len, &blkno,
 791                                           dir, &dirent_bh, &dirent);
 792         if (status < 0) {
 793                 if (status != -ENOENT)
 794                         mlog_errno(status);
 795                 goto leave;
 796         }
 797
 798         if (OCFS2_I(inode)->ip_blkno != blkno) {
 799                 status = -ENOENT;
 800
 801                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 802                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 803                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 804                 goto leave;
 805         }
 806
 807         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 808         if (status < 0) {
 809                 if (status != -ENOENT)
 810                         mlog_errno(status);
 811                 goto leave;
 812         }
 813
 814         if (S_ISDIR(inode->i_mode)) {
 815                 if (!ocfs2_empty_dir(inode)) {
 816                         status = -ENOTEMPTY;
 817                         goto leave;
 818                 } else if (inode->i_nlink != 2) {
 819                         status = -ENOTEMPTY;
 820                         goto leave;
 821                 }
 822         }
 823
 824         /* There are still a few steps left until we can consider the
 825          * unlink to have succeeded. Save off nlink here before
 826          * modification so we can set it back in case we hit an issue
 827          * before commit. */
 828         saved_nlink = inode->i_nlink;
 829         if (S_ISDIR(inode->i_mode))
 830                 inode->i_nlink = 0;
 831         else
 832                 inode->i_nlink--;
 833
 834         status = ocfs2_request_unlink_vote(inode, dentry,
 835                                            (unsigned int) inode->i_nlink);
 836         if (status < 0) {
 837                 /* This vote should succeed under all normal
 838                  * circumstances. */
 839                 mlog_errno(status);
 840                 goto leave;
 841         }
 842
 843         if (!inode->i_nlink) {
 844                 status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 845                                                   orphan_name,
 846                                                   &orphan_entry_bh);
 847                 if (status < 0) {
 848                         mlog_errno(status);
 849                         goto leave;
 850                 }
 851         }
 852
 853         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
 854         if (IS_ERR(handle)) {
 855                 status = PTR_ERR(handle);
 856                 handle = NULL;
 857                 mlog_errno(status);
 858                 goto leave;
 859         }
 860
 861         status = ocfs2_journal_access(handle, inode, fe_bh,
 862                                       OCFS2_JOURNAL_ACCESS_WRITE);
 863         if (status < 0) {
 864                 mlog_errno(status);
 865                 goto leave;
 866         }
 867
 868         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 869
 870         if (!inode->i_nlink) {
 871                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 872                                           orphan_entry_bh);
 873                 if (status < 0) {
 874                         mlog_errno(status);
 875                         goto leave;
 876                 }
 877         }
 878
 879         /* delete the name from the parent dir */
 880         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 881         if (status < 0) {
 882                 mlog_errno(status);
 883                 goto leave;
 884         }
 885
 886         /* We can set nlink on the dinode now. clear the saved version
 887          * so that it doesn't get set later. */
 888         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 889         saved_nlink = 0;
 890
 891         status = ocfs2_journal_dirty(handle, fe_bh);
 892         if (status < 0) {
 893                 mlog_errno(status);
 894                 goto leave;
 895         }
 896
 897         if (S_ISDIR(inode->i_mode)) {
 898                 dir->i_nlink--;
 899                 status = ocfs2_mark_inode_dirty(handle, dir,
 900                                                 parent_node_bh);
 901                 if (status < 0) {
 902                         mlog_errno(status);
 903                         dir->i_nlink++;
 904                 }
 905         }
 906
 907 leave:
 908         if (status < 0 && saved_nlink)
 909                 inode->i_nlink = saved_nlink;
 910
 911         if (handle)
 912                 ocfs2_commit_trans(handle);
 913
 914         if (fe_bh)
 915                 brelse(fe_bh);
 916
 917         if (dirent_bh)
 918                 brelse(dirent_bh);
 919
 920         if (parent_node_bh)
 921                 brelse(parent_node_bh);
 922
 923         if (orphan_entry_bh)
 924                 brelse(orphan_entry_bh);
 925
 926         mlog_exit(status);
 927
 928         return status;
 929 }
 930
 931 /*
 932  * The only place this should be used is rename!
 933  * if they have the same id, then the 1st one is the only one locked.
 934  */
 935 static int ocfs2_double_lock(struct ocfs2_super *osb,
 936                              struct ocfs2_journal_handle *handle,
 937                              struct buffer_head **bh1,
 938                              struct inode *inode1,
 939                              struct buffer_head **bh2,
 940                              struct inode *inode2)
 941 {
 942         int status;
 943         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 944         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 945         struct buffer_head **tmpbh;
 946         struct inode *tmpinode;
 947
 948         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 949                    (unsigned long long)oi1->ip_blkno,
 950                    (unsigned long long)oi2->ip_blkno);
 951
 952         BUG_ON(!handle);
 953
 954         if (*bh1)
 955                 *bh1 = NULL;
 956         if (*bh2)
 957                 *bh2 = NULL;
 958
 959         /* we always want to lock the one with the lower lockid first. */
 960         if (oi1->ip_blkno != oi2->ip_blkno) {
 961                 if (oi1->ip_blkno < oi2->ip_blkno) {
 962                         /* switch id1 and id2 around */
 963                         mlog(0, "switching them around...\n");
 964                         tmpbh = bh2;
 965                         bh2 = bh1;
 966                         bh1 = tmpbh;
 967
 968                         tmpinode = inode2;
 969                         inode2 = inode1;
 970                         inode1 = tmpinode;
 971                 }
 972                 /* lock id2 */
 973                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
 974                 if (status < 0) {
 975                         if (status != -ENOENT)
 976                                 mlog_errno(status);
 977                         goto bail;
 978                 }
 979         }
 980         /* lock id1 */
 981         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
 982         if (status < 0) {
 983                 if (status != -ENOENT)
 984                         mlog_errno(status);
 985                 goto bail;
 986         }
 987 bail:
 988         mlog_exit(status);
 989         return status;
 990 }
 991
 992 #define PARENT_INO(buffer) \
 993         ((struct ocfs2_dir_entry *) \
 994          ((char *)buffer + \
 995           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
 996
 997 static int ocfs2_rename(struct inode *old_dir,
 998                         struct dentry *old_dentry,
 999                         struct inode *new_dir,
1000                         struct dentry *new_dentry)
1001 {
1002         int status = 0, rename_lock = 0;
1003         struct inode *old_inode = old_dentry->d_inode;
1004         struct inode *new_inode = new_dentry->d_inode;
1005         struct ocfs2_dinode *newfe = NULL;
1006         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1007         struct buffer_head *orphan_entry_bh = NULL;
1008         struct buffer_head *newfe_bh = NULL;
1009         struct buffer_head *insert_entry_bh = NULL;
1010         struct ocfs2_super *osb = NULL;
1011         u64 newfe_blkno;
1012         struct ocfs2_journal_handle *handle = NULL;
1013         struct buffer_head *old_dir_bh = NULL;
1014         struct buffer_head *new_dir_bh = NULL;
1015         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1016                                                                // and new_dentry
1017         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1018         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1019                                                     // this is the 1st dirent bh
1020         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1021         unsigned int links_count;
1022
1023         /* At some point it might be nice to break this function up a
1024          * bit. */
1025
1026         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1027                    old_dir, old_dentry, new_dir, new_dentry,
1028                    old_dentry->d_name.len, old_dentry->d_name.name,
1029                    new_dentry->d_name.len, new_dentry->d_name.name);
1030
1031         osb = OCFS2_SB(old_dir->i_sb);
1032
1033         if (new_inode) {
1034                 if (!igrab(new_inode))
1035                         BUG();
1036         }
1037
1038         if (atomic_read(&old_dentry->d_count) > 2) {
1039                 shrink_dcache_parent(old_dentry);
1040                 if (atomic_read(&old_dentry->d_count) > 2) {
1041                         status = -EBUSY;
1042                         goto bail;
1043                 }
1044         }
1045
1046         /* Assume a directory heirarchy thusly:
1047          * a/b/c
1048          * a/d
1049          * a,b,c, and d are all directories.
1050          *
1051          * from cwd of 'a' on both nodes:
1052          * node1: mv b/c d
1053          * node2: mv d   b/c
1054          *
1055          * And that's why, just like the VFS, we need a file system
1056          * rename lock. */
1057         if (old_dentry != new_dentry) {
1058                 status = ocfs2_rename_lock(osb);
1059                 if (status < 0) {
1060                         mlog_errno(status);
1061                         goto bail;
1062                 }
1063                 rename_lock = 1;
1064         }
1065
1066         handle = ocfs2_alloc_handle(osb);
1067         if (handle == NULL) {
1068                 status = -ENOMEM;
1069                 mlog_errno(status);
1070                 goto bail;
1071         }
1072
1073         /* if old and new are the same, this'll just do one lock. */
1074         status = ocfs2_double_lock(osb, handle,
1075                                   &old_dir_bh, old_dir,
1076                                   &new_dir_bh, new_dir);
1077         if (status < 0) {
1078                 mlog_errno(status);
1079                 goto bail;
1080         }
1081
1082         /* make sure both dirs have bhs
1083          * get an extra ref on old_dir_bh if old==new */
1084         if (!new_dir_bh) {
1085                 if (old_dir_bh) {
1086                         new_dir_bh = old_dir_bh;
1087                         get_bh(new_dir_bh);
1088                 } else {
1089                         mlog(ML_ERROR, "no old_dir_bh!\n");
1090                         status = -EIO;
1091                         goto bail;
1092                 }
1093         }
1094
1095         if (S_ISDIR(old_inode->i_mode)) {
1096                 /* Directories actually require metadata updates to
1097                  * the directory info so we can't get away with not
1098                  * doing node locking on it. */
1099                 status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1100                 if (status < 0) {
1101                         if (status != -ENOENT)
1102                                 mlog_errno(status);
1103                         goto bail;
1104                 }
1105
1106                 status = ocfs2_request_rename_vote(old_inode, old_dentry);
1107                 if (status < 0) {
1108                         mlog_errno(status);
1109                         goto bail;
1110                 }
1111
1112                 status = -EIO;
1113                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1114                 if (!old_inode_de_bh)
1115                         goto bail;
1116
1117                 status = -EIO;
1118                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1119                     OCFS2_I(old_dir)->ip_blkno)
1120                         goto bail;
1121                 status = -EMLINK;
1122                 if (!new_inode && new_dir!=old_dir &&
1123                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1124                         goto bail;
1125         } else {
1126                 /* Ah, the simple case - we're a file so just send a
1127                  * message. */
1128                 status = ocfs2_request_rename_vote(old_inode, old_dentry);
1129                 if (status < 0) {
1130                         mlog_errno(status);
1131                         goto bail;
1132                 }
1133         }
1134
1135         status = -ENOENT;
1136         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1137                                      old_dentry->d_name.len,
1138                                      old_dir, &old_de);
1139         if (!old_de_bh)
1140                 goto bail;
1141
1142         /*
1143          *  Check for inode number is _not_ due to possible IO errors.
1144          *  We might rmdir the source, keep it as pwd of some process
1145          *  and merrily kill the link to whatever was created under the
1146          *  same name. Goodbye sticky bit ;-<
1147          */
1148         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1149                 goto bail;
1150
1151         /* check if the target already exists (in which case we need
1152          * to delete it */
1153         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1154                                           new_dentry->d_name.len,
1155                                           &newfe_blkno, new_dir, &new_de_bh,
1156                                           &new_de);
1157         /* The only error we allow here is -ENOENT because the new
1158          * file not existing is perfectly valid. */
1159         if ((status < 0) && (status != -ENOENT)) {
1160                 /* If we cannot find the file specified we should just */
1161                 /* return the error... */
1162                 mlog_errno(status);
1163                 goto bail;
1164         }
1165
1166         if (!new_de && new_inode)
1167                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1168                      "directory!", new_inode->i_ino);
1169
1170         /* In case we need to overwrite an existing file, we blow it
1171          * away first */
1172         if (new_de) {
1173                 /* VFS didn't think there existed an inode here, but
1174                  * someone else in the cluster must have raced our
1175                  * rename to create one. Today we error cleanly, in
1176                  * the future we should consider calling iget to build
1177                  * a new struct inode for this entry. */
1178                 if (!new_inode) {
1179                         status = -EACCES;
1180
1181                         mlog(0, "We found an inode for name %.*s but VFS "
1182                              "didn't give us one.\n", new_dentry->d_name.len,
1183                              new_dentry->d_name.name);
1184                         goto bail;
1185                 }
1186
1187                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1188                         status = -EACCES;
1189
1190                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1191                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1192                              (unsigned long long)newfe_blkno,
1193                              OCFS2_I(new_inode)->ip_flags);
1194                         goto bail;
1195                 }
1196
1197                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1198                 if (status < 0) {
1199                         if (status != -ENOENT)
1200                                 mlog_errno(status);
1201                         goto bail;
1202                 }
1203
1204                 if (S_ISDIR(new_inode->i_mode))
1205                         links_count = 0;
1206                 else
1207                         links_count = (unsigned int) (new_inode->i_nlink - 1);
1208
1209                 status = ocfs2_request_unlink_vote(new_inode, new_dentry,
1210                                                    links_count);
1211                 if (status < 0) {
1212                         mlog_errno(status);
1213                         goto bail;
1214                 }
1215
1216                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1217
1218                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1219                      "newfebh=%p bhblocknr=%llu\n", new_de,
1220                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1221                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1222
1223                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1224                         status = ocfs2_prepare_orphan_dir(osb, handle,
1225                                                           new_inode,
1226                                                           orphan_name,
1227                                                           &orphan_entry_bh);
1228                         if (status < 0) {
1229                                 mlog_errno(status);
1230                                 goto bail;
1231                         }
1232                 }
1233         } else {
1234                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1235
1236                 status = ocfs2_check_dir_for_entry(new_dir,
1237                                                    new_dentry->d_name.name,
1238                                                    new_dentry->d_name.len);
1239                 if (status)
1240                         goto bail;
1241
1242                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1243                                                       new_dentry->d_name.name,
1244                                                       new_dentry->d_name.len,
1245                                                       &insert_entry_bh);
1246                 if (status < 0) {
1247                         mlog_errno(status);
1248                         goto bail;
1249                 }
1250         }
1251
1252         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1253         if (IS_ERR(handle)) {
1254                 status = PTR_ERR(handle);
1255                 handle = NULL;
1256                 mlog_errno(status);
1257                 goto bail;
1258         }
1259
1260         if (new_de) {
1261                 if (S_ISDIR(new_inode->i_mode)) {
1262                         if (!ocfs2_empty_dir(new_inode) ||
1263                             new_inode->i_nlink != 2) {
1264                                 status = -ENOTEMPTY;
1265                                 goto bail;
1266                         }
1267                 }
1268                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1269                                               OCFS2_JOURNAL_ACCESS_WRITE);
1270                 if (status < 0) {
1271                         mlog_errno(status);
1272                         goto bail;
1273                 }
1274
1275                 if (S_ISDIR(new_inode->i_mode) ||
1276                     (newfe->i_links_count == cpu_to_le16(1))){
1277                         status = ocfs2_orphan_add(osb, handle, new_inode,
1278                                                   newfe, orphan_name,
1279                                                   orphan_entry_bh);
1280                         if (status < 0) {
1281                                 mlog_errno(status);
1282                                 goto bail;
1283                         }
1284                 }
1285
1286                 /* change the dirent to point to the correct inode */
1287                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1288                                               OCFS2_JOURNAL_ACCESS_WRITE);
1289                 if (status < 0) {
1290                         mlog_errno(status);
1291                         goto bail;
1292                 }
1293                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1294                 new_de->file_type = old_de->file_type;
1295                 new_dir->i_version++;
1296                 status = ocfs2_journal_dirty(handle, new_de_bh);
1297                 if (status < 0) {
1298                         mlog_errno(status);
1299                         goto bail;
1300                 }
1301
1302                 if (S_ISDIR(new_inode->i_mode))
1303                         newfe->i_links_count = 0;
1304                 else
1305                         le16_add_cpu(&newfe->i_links_count, -1);
1306
1307                 status = ocfs2_journal_dirty(handle, newfe_bh);
1308                 if (status < 0) {
1309                         mlog_errno(status);
1310                         goto bail;
1311                 }
1312         } else {
1313                 /* if the name was not found in new_dir, add it now */
1314                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1315                                          OCFS2_I(old_inode)->ip_blkno,
1316                                          new_dir_bh, insert_entry_bh);
1317         }
1318
1319         old_inode->i_ctime = CURRENT_TIME;
1320         mark_inode_dirty(old_inode);
1321
1322         /* now that the name has been added to new_dir, remove the old name */
1323         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1324         if (status < 0) {
1325                 mlog_errno(status);
1326                 goto bail;
1327         }
1328
1329         if (new_inode) {
1330                 new_inode->i_nlink--;
1331                 new_inode->i_ctime = CURRENT_TIME;
1332         }
1333         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1334         if (old_inode_de_bh) {
1335                 status = ocfs2_journal_access(handle, old_inode,
1336                                              old_inode_de_bh,
1337                                              OCFS2_JOURNAL_ACCESS_WRITE);
1338                 PARENT_INO(old_inode_de_bh->b_data) =
1339                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1340                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1341                 old_dir->i_nlink--;
1342                 if (new_inode) {
1343                         new_inode->i_nlink--;
1344                 } else {
1345                         new_dir->i_nlink++;
1346                         mark_inode_dirty(new_dir);
1347                 }
1348         }
1349         mark_inode_dirty(old_dir);
1350         if (new_inode)
1351                 mark_inode_dirty(new_inode);
1352
1353         if (old_dir != new_dir)
1354                 if (new_dir_nlink != new_dir->i_nlink) {
1355                         if (!new_dir_bh) {
1356                                 mlog(ML_ERROR, "need to change nlink for new "
1357                                      "dir %llu from %d to %d but bh is NULL\n",
1358                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1359                                      (int)new_dir_nlink, new_dir->i_nlink);
1360                         } else {
1361                                 struct ocfs2_dinode *fe;
1362                                 status = ocfs2_journal_access(handle,
1363                                                               new_dir,
1364                                                               new_dir_bh,
1365                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1366                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1367                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1368                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1369                         }
1370                 }
1371
1372         if (old_dir_nlink != old_dir->i_nlink) {
1373                 if (!old_dir_bh) {
1374                         mlog(ML_ERROR, "need to change nlink for old dir "
1375                              "%llu from %d to %d but bh is NULL!\n",
1376                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1377                              (int)old_dir_nlink, old_dir->i_nlink);
1378                 } else {
1379                         struct ocfs2_dinode *fe;
1380                         status = ocfs2_journal_access(handle, old_dir,
1381                                                       old_dir_bh,
1382                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1383                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1384                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1385                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1386                 }
1387         }
1388
1389         status = 0;
1390 bail:
1391         if (rename_lock)
1392                 ocfs2_rename_unlock(osb);
1393
1394         if (handle)
1395                 ocfs2_commit_trans(handle);
1396
1397         if (new_inode)
1398                 sync_mapping_buffers(old_inode->i_mapping);
1399
1400         if (new_inode)
1401                 iput(new_inode);
1402         if (newfe_bh)
1403                 brelse(newfe_bh);
1404         if (old_dir_bh)
1405                 brelse(old_dir_bh);
1406         if (new_dir_bh)
1407                 brelse(new_dir_bh);
1408         if (new_de_bh)
1409                 brelse(new_de_bh);
1410         if (old_de_bh)
1411                 brelse(old_de_bh);
1412         if (old_inode_de_bh)
1413                 brelse(old_inode_de_bh);
1414         if (orphan_entry_bh)
1415                 brelse(orphan_entry_bh);
1416         if (insert_entry_bh)
1417                 brelse(insert_entry_bh);
1418
1419         mlog_exit(status);
1420
1421         return status;
1422 }
1423
1424 /*
1425  * we expect i_size = strlen(symname). Copy symname into the file
1426  * data, including the null terminator.
1427  */
1428 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1429                                      struct ocfs2_journal_handle *handle,
1430                                      struct inode *inode,
1431                                      const char *symname)
1432 {
1433         struct buffer_head **bhs = NULL;
1434         const char *c;
1435         struct super_block *sb = osb->sb;
1436         u64 p_blkno;
1437         int p_blocks;
1438         int virtual, blocks, status, i, bytes_left;
1439
1440         bytes_left = i_size_read(inode) + 1;
1441         /* we can't trust i_blocks because we're actually going to
1442          * write i_size + 1 bytes. */
1443         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1444
1445         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1446                         (unsigned long long)inode->i_blocks,
1447                         i_size_read(inode), blocks);
1448
1449         /* Sanity check -- make sure we're going to fit. */
1450         if (bytes_left >
1451             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1452                 status = -EIO;
1453                 mlog_errno(status);
1454                 goto bail;
1455         }
1456
1457         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1458         if (!bhs) {
1459                 status = -ENOMEM;
1460                 mlog_errno(status);
1461                 goto bail;
1462         }
1463
1464         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1465                                              &p_blocks);
1466         if (status < 0) {
1467                 mlog_errno(status);
1468                 goto bail;
1469         }
1470
1471         /* links can never be larger than one cluster so we know this
1472          * is all going to be contiguous, but do a sanity check
1473          * anyway. */
1474         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1475                 status = -EIO;
1476                 mlog_errno(status);
1477                 goto bail;
1478         }
1479
1480         virtual = 0;
1481         while(bytes_left > 0) {
1482                 c = &symname[virtual * sb->s_blocksize];
1483
1484                 bhs[virtual] = sb_getblk(sb, p_blkno);
1485                 if (!bhs[virtual]) {
1486                         status = -ENOMEM;
1487                         mlog_errno(status);
1488                         goto bail;
1489                 }
1490                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1491
1492                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1493                                               OCFS2_JOURNAL_ACCESS_CREATE);
1494                 if (status < 0) {
1495                         mlog_errno(status);
1496                         goto bail;
1497                 }
1498
1499                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1500
1501                 memcpy(bhs[virtual]->b_data, c,
1502                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1503                        bytes_left);
1504
1505                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1506                 if (status < 0) {
1507                         mlog_errno(status);
1508                         goto bail;
1509                 }
1510
1511                 virtual++;
1512                 p_blkno++;
1513                 bytes_left -= sb->s_blocksize;
1514         }
1515
1516         status = 0;
1517 bail:
1518
1519         if (bhs) {
1520                 for(i = 0; i < blocks; i++)
1521                         if (bhs[i])
1522                                 brelse(bhs[i]);
1523                 kfree(bhs);
1524         }
1525
1526         mlog_exit(status);
1527         return status;
1528 }
1529
1530 static int ocfs2_symlink(struct inode *dir,
1531                          struct dentry *dentry,
1532                          const char *symname)
1533 {
1534         int status, l, credits;
1535         u64 newsize;
1536         struct ocfs2_super *osb = NULL;
1537         struct inode *inode = NULL;
1538         struct super_block *sb;
1539         struct buffer_head *new_fe_bh = NULL;
1540         struct buffer_head *de_bh = NULL;
1541         struct buffer_head *parent_fe_bh = NULL;
1542         struct ocfs2_dinode *fe = NULL;
1543         struct ocfs2_dinode *dirfe;
1544         struct ocfs2_journal_handle *handle = NULL;
1545         struct ocfs2_alloc_context *inode_ac = NULL;
1546         struct ocfs2_alloc_context *data_ac = NULL;
1547
1548         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1549                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1550
1551         sb = dir->i_sb;
1552         osb = OCFS2_SB(sb);
1553
1554         l = strlen(symname) + 1;
1555
1556         credits = ocfs2_calc_symlink_credits(sb);
1557
1558         handle = ocfs2_alloc_handle(osb);
1559         if (handle == NULL) {
1560                 status = -ENOMEM;
1561                 mlog_errno(status);
1562                 goto bail;
1563         }
1564
1565         /* lock the parent directory */
1566         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1567         if (status < 0) {
1568                 if (status != -ENOENT)
1569                         mlog_errno(status);
1570                 goto bail;
1571         }
1572
1573         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1574         if (!dirfe->i_links_count) {
1575                 /* can't make a file in a deleted directory. */
1576                 status = -ENOENT;
1577                 goto bail;
1578         }
1579
1580         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1581                                            dentry->d_name.len);
1582         if (status)
1583                 goto bail;
1584
1585         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1586                                               dentry->d_name.name,
1587                                               dentry->d_name.len, &de_bh);
1588         if (status < 0) {
1589                 mlog_errno(status);
1590                 goto bail;
1591         }
1592
1593         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1594         if (status < 0) {
1595                 if (status != -ENOSPC)
1596                         mlog_errno(status);
1597                 goto bail;
1598         }
1599
1600         /* don't reserve bitmap space for fast symlinks. */
1601         if (l > ocfs2_fast_symlink_chars(sb)) {
1602                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1603                 if (status < 0) {
1604                         if (status != -ENOSPC)
1605                                 mlog_errno(status);
1606                         goto bail;
1607                 }
1608         }
1609
1610         handle = ocfs2_start_trans(osb, handle, credits);
1611         if (IS_ERR(handle)) {
1612                 status = PTR_ERR(handle);
1613                 handle = NULL;
1614                 mlog_errno(status);
1615                 goto bail;
1616         }
1617
1618         status = ocfs2_mknod_locked(osb, dir, dentry,
1619                                     S_IFLNK | S_IRWXUGO, 0,
1620                                     &new_fe_bh, parent_fe_bh, handle,
1621                                     &inode, inode_ac);
1622         if (status < 0) {
1623                 mlog_errno(status);
1624                 goto bail;
1625         }
1626
1627         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1628         inode->i_rdev = 0;
1629         newsize = l - 1;
1630         if (l > ocfs2_fast_symlink_chars(sb)) {
1631                 inode->i_op = &ocfs2_symlink_inode_operations;
1632                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1633                                                     handle, data_ac, NULL,
1634                                                     NULL);
1635                 if (status < 0) {
1636                         if (status != -ENOSPC && status != -EINTR) {
1637                                 mlog(ML_ERROR,
1638                                      "Failed to extend file to %llu\n",
1639                                      (unsigned long long)newsize);
1640                                 mlog_errno(status);
1641                                 status = -ENOSPC;
1642                         }
1643                         goto bail;
1644                 }
1645                 i_size_write(inode, newsize);
1646                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1647         } else {
1648                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1649                 memcpy((char *) fe->id2.i_symlink, symname, l);
1650                 i_size_write(inode, newsize);
1651                 inode->i_blocks = 0;
1652         }
1653
1654         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1655         if (status < 0) {
1656                 mlog_errno(status);
1657                 goto bail;
1658         }
1659
1660         if (!ocfs2_inode_is_fast_symlink(inode)) {
1661                 status = ocfs2_create_symlink_data(osb, handle, inode,
1662                                                    symname);
1663                 if (status < 0) {
1664                         mlog_errno(status);
1665                         goto bail;
1666                 }
1667         }
1668
1669         status = ocfs2_add_entry(handle, dentry, inode,
1670                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1671                                  de_bh);
1672         if (status < 0) {
1673                 mlog_errno(status);
1674                 goto bail;
1675         }
1676
1677         insert_inode_hash(inode);
1678         dentry->d_op = &ocfs2_dentry_ops;
1679         d_instantiate(dentry, inode);
1680 bail:
1681         if (handle)
1682                 ocfs2_commit_trans(handle);
1683         if (new_fe_bh)
1684                 brelse(new_fe_bh);
1685         if (parent_fe_bh)
1686                 brelse(parent_fe_bh);
1687         if (de_bh)
1688                 brelse(de_bh);
1689         if (inode_ac)
1690                 ocfs2_free_alloc_context(inode_ac);
1691         if (data_ac)
1692                 ocfs2_free_alloc_context(data_ac);
1693         if ((status < 0) && inode)
1694                 iput(inode);
1695
1696         mlog_exit(status);
1697
1698         return status;
1699 }
1700
1701 int ocfs2_check_dir_entry(struct inode * dir,
1702                           struct ocfs2_dir_entry * de,
1703                           struct buffer_head * bh,
1704                           unsigned long offset)
1705 {
1706         const char *error_msg = NULL;
1707         const int rlen = le16_to_cpu(de->rec_len);
1708
1709         if (rlen < OCFS2_DIR_REC_LEN(1))
1710                 error_msg = "rec_len is smaller than minimal";
1711         else if (rlen % 4 != 0)
1712                 error_msg = "rec_len % 4 != 0";
1713         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1714                 error_msg = "rec_len is too small for name_len";
1715         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1716                 error_msg = "directory entry across blocks";
1717
1718         if (error_msg != NULL)
1719                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1720                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1721                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1722                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1723                      de->name_len);
1724         return error_msg == NULL ? 1 : 0;
1725 }
1726
1727 /* we don't always have a dentry for what we want to add, so people
1728  * like orphan dir can call this instead.
1729  *
1730  * If you pass me insert_bh, I'll skip the search of the other dir
1731  * blocks and put the record in there.
1732  */
1733 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1734                              struct inode *dir,
1735                              const char *name, int namelen,
1736                              struct inode *inode, u64 blkno,
1737                              struct buffer_head *parent_fe_bh,
1738                              struct buffer_head *insert_bh)
1739 {
1740         unsigned long offset;
1741         unsigned short rec_len;
1742         struct ocfs2_dir_entry *de, *de1;
1743         struct super_block *sb;
1744         int retval, status;
1745
1746         mlog_entry_void();
1747
1748         sb = dir->i_sb;
1749
1750         if (!namelen)
1751                 return -EINVAL;
1752
1753         rec_len = OCFS2_DIR_REC_LEN(namelen);
1754         offset = 0;
1755         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1756         while (1) {
1757                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1758                 /* These checks should've already been passed by the
1759                  * prepare function, but I guess we can leave them
1760                  * here anyway. */
1761                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1762                         retval = -ENOENT;
1763                         goto bail;
1764                 }
1765                 if (ocfs2_match(namelen, name, de)) {
1766                         retval = -EEXIST;
1767                         goto bail;
1768                 }
1769                 if (((le64_to_cpu(de->inode) == 0) &&
1770                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1771                     (le16_to_cpu(de->rec_len) >=
1772                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1773                         status = ocfs2_journal_access(handle, dir, insert_bh,
1774                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1775                         /* By now the buffer is marked for journaling */
1776                         offset += le16_to_cpu(de->rec_len);
1777                         if (le64_to_cpu(de->inode)) {
1778                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1779                                         OCFS2_DIR_REC_LEN(de->name_len));
1780                                 de1->rec_len =
1781                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1782                                         OCFS2_DIR_REC_LEN(de->name_len));
1783                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1784                                 de = de1;
1785                         }
1786                         de->file_type = OCFS2_FT_UNKNOWN;
1787                         if (blkno) {
1788                                 de->inode = cpu_to_le64(blkno);
1789                                 ocfs2_set_de_type(de, inode->i_mode);
1790                         } else
1791                                 de->inode = 0;
1792                         de->name_len = namelen;
1793                         memcpy(de->name, name, namelen);
1794
1795                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1796                         dir->i_version++;
1797                         status = ocfs2_journal_dirty(handle, insert_bh);
1798                         retval = 0;
1799                         goto bail;
1800                 }
1801                 offset += le16_to_cpu(de->rec_len);
1802                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1803         }
1804
1805         /* when you think about it, the assert above should prevent us
1806          * from ever getting here. */
1807         retval = -ENOSPC;
1808 bail:
1809
1810         mlog_exit(retval);
1811         return retval;
1812 }
1813
1814
1815 /*
1816  * ocfs2_delete_entry deletes a directory entry by merging it with the
1817  * previous entry
1818  */
1819 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1820                               struct inode *dir,
1821                               struct ocfs2_dir_entry *de_del,
1822                               struct buffer_head *bh)
1823 {
1824         struct ocfs2_dir_entry *de, *pde;
1825         int i, status = -ENOENT;
1826
1827         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1828
1829         i = 0;
1830         pde = NULL;
1831         de = (struct ocfs2_dir_entry *) bh->b_data;
1832         while (i < bh->b_size) {
1833                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1834                         status = -EIO;
1835                         mlog_errno(status);
1836                         goto bail;
1837                 }
1838                 if (de == de_del)  {
1839                         status = ocfs2_journal_access(handle, dir, bh,
1840                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1841                         if (status < 0) {
1842                                 status = -EIO;
1843                                 mlog_errno(status);
1844                                 goto bail;
1845                         }
1846                         if (pde)
1847                                 pde->rec_len =
1848                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1849                                                     le16_to_cpu(de->rec_len));
1850                         else
1851                                 de->inode = 0;
1852                         dir->i_version++;
1853                         status = ocfs2_journal_dirty(handle, bh);
1854                         goto bail;
1855                 }
1856                 i += le16_to_cpu(de->rec_len);
1857                 pde = de;
1858                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1859         }
1860 bail:
1861         mlog_exit(status);
1862         return status;
1863 }
1864
1865 /*
1866  * Returns 0 if not found, -1 on failure, and 1 on success
1867  */
1868 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1869                                         struct inode *dir,
1870                                         const char *name, int namelen,
1871                                         unsigned long offset,
1872                                         struct ocfs2_dir_entry **res_dir)
1873 {
1874         struct ocfs2_dir_entry *de;
1875         char *dlimit, *de_buf;
1876         int de_len;
1877         int ret = 0;
1878
1879         mlog_entry_void();
1880
1881         de_buf = bh->b_data;
1882         dlimit = de_buf + dir->i_sb->s_blocksize;
1883
1884         while (de_buf < dlimit) {
1885                 /* this code is executed quadratically often */
1886                 /* do minimal checking `by hand' */
1887
1888                 de = (struct ocfs2_dir_entry *) de_buf;
1889
1890                 if (de_buf + namelen <= dlimit &&
1891                     ocfs2_match(namelen, name, de)) {
1892                         /* found a match - just to be sure, do a full check */
1893                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1894                                 ret = -1;
1895                                 goto bail;
1896                         }
1897                         *res_dir = de;
1898                         ret = 1;
1899                         goto bail;
1900                 }
1901
1902                 /* prevent looping on a bad block */
1903                 de_len = le16_to_cpu(de->rec_len);
1904                 if (de_len <= 0) {
1905                         ret = -1;
1906                         goto bail;
1907                 }
1908
1909                 de_buf += de_len;
1910                 offset += de_len;
1911         }
1912
1913 bail:
1914         mlog_exit(ret);
1915         return ret;
1916 }
1917
1918 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1919                                      struct inode *dir,
1920                                      struct ocfs2_dir_entry **res_dir)
1921 {
1922         struct super_block *sb;
1923         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1924         struct buffer_head *bh, *ret = NULL;
1925         unsigned long start, block, b;
1926         int ra_max = 0;         /* Number of bh's in the readahead
1927                                    buffer, bh_use[] */
1928         int ra_ptr = 0;         /* Current index into readahead
1929                                    buffer */
1930         int num = 0;
1931         int nblocks, i, err;
1932
1933         mlog_entry_void();
1934
1935         *res_dir = NULL;
1936         sb = dir->i_sb;
1937
1938         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1939         start = OCFS2_I(dir)->ip_dir_start_lookup;
1940         if (start >= nblocks)
1941                 start = 0;
1942         block = start;
1943
1944 restart:
1945         do {
1946                 /*
1947                  * We deal with the read-ahead logic here.
1948                  */
1949                 if (ra_ptr >= ra_max) {
1950                         /* Refill the readahead buffer */
1951                         ra_ptr = 0;
1952                         b = block;
1953                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1954                                 /*
1955                                  * Terminate if we reach the end of the
1956                                  * directory and must wrap, or if our
1957                                  * search has finished at this block.
1958                                  */
1959                                 if (b >= nblocks || (num && block == start)) {
1960                                         bh_use[ra_max] = NULL;
1961                                         break;
1962                                 }
1963                                 num++;
1964
1965                                 /* XXX: questionable readahead stuff here */
1966                                 bh = ocfs2_bread(dir, b++, &err, 1);
1967                                 bh_use[ra_max] = bh;
1968 #if 0           // ???
1969                                 if (bh)
1970                                         ll_rw_block(READ, 1, &bh);
1971 #endif
1972                         }
1973                 }
1974                 if ((bh = bh_use[ra_ptr++]) == NULL)
1975                         goto next;
1976                 wait_on_buffer(bh);
1977                 if (!buffer_uptodate(bh)) {
1978                         /* read error, skip block & hope for the best */
1979                         brelse(bh);
1980                         goto next;
1981                 }
1982                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
1983                                           block << sb->s_blocksize_bits,
1984                                           res_dir);
1985                 if (i == 1) {
1986                         OCFS2_I(dir)->ip_dir_start_lookup = block;
1987                         ret = bh;
1988                         goto cleanup_and_exit;
1989                 } else {
1990                         brelse(bh);
1991                         if (i < 0)
1992                                 goto cleanup_and_exit;
1993                 }
1994         next:
1995                 if (++block >= nblocks)
1996                         block = 0;
1997         } while (block != start);
1998
1999         /*
2000          * If the directory has grown while we were searching, then
2001          * search the last part of the directory before giving up.
2002          */
2003         block = nblocks;
2004         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2005         if (block < nblocks) {
2006                 start = 0;
2007                 goto restart;
2008         }
2009
2010 cleanup_and_exit:
2011         /* Clean up the read-ahead blocks */
2012         for (; ra_ptr < ra_max; ra_ptr++)
2013                 brelse(bh_use[ra_ptr]);
2014
2015         mlog_exit_ptr(ret);
2016         return ret;
2017 }
2018
2019 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2020 {
2021         int status, namelen;
2022
2023         mlog_entry_void();
2024
2025         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2026                            (long long)blkno);
2027         if (namelen <= 0) {
2028                 if (namelen)
2029                         status = namelen;
2030                 else
2031                         status = -EINVAL;
2032                 mlog_errno(status);
2033                 goto bail;
2034         }
2035         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2036                 status = -EINVAL;
2037                 mlog_errno(status);
2038                 goto bail;
2039         }
2040
2041         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2042              namelen);
2043
2044         status = 0;
2045 bail:
2046         mlog_exit(status);
2047         return status;
2048 }
2049
2050 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2051                                     struct ocfs2_journal_handle *handle,
2052                                     struct inode *inode,
2053                                     char *name,
2054                                     struct buffer_head **de_bh)
2055 {
2056         struct inode *orphan_dir_inode = NULL;
2057         struct buffer_head *orphan_dir_bh = NULL;
2058         int status = 0;
2059
2060         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2061         if (status < 0) {
2062                 mlog_errno(status);
2063                 goto leave;
2064         }
2065
2066         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2067                                                        ORPHAN_DIR_SYSTEM_INODE,
2068                                                        osb->slot_num);
2069         if (!orphan_dir_inode) {
2070                 status = -ENOENT;
2071                 mlog_errno(status);
2072                 goto leave;
2073         }
2074
2075         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2076         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2077         if (status < 0) {
2078                 mlog_errno(status);
2079                 goto leave;
2080         }
2081
2082         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2083                                               orphan_dir_bh, name,
2084                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2085         if (status < 0) {
2086                 mlog_errno(status);
2087                 goto leave;
2088         }
2089
2090 leave:
2091         if (orphan_dir_inode)
2092                 iput(orphan_dir_inode);
2093
2094         if (orphan_dir_bh)
2095                 brelse(orphan_dir_bh);
2096
2097         mlog_exit(status);
2098         return status;
2099 }
2100
2101 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2102                             struct ocfs2_journal_handle *handle,
2103                             struct inode *inode,
2104                             struct ocfs2_dinode *fe,
2105                             char *name,
2106                             struct buffer_head *de_bh)
2107 {
2108         struct inode *orphan_dir_inode = NULL;
2109         struct buffer_head *orphan_dir_bh = NULL;
2110         int status = 0;
2111         struct ocfs2_dinode *orphan_fe;
2112
2113         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2114
2115         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2116                                                        ORPHAN_DIR_SYSTEM_INODE,
2117                                                        osb->slot_num);
2118         if (!orphan_dir_inode) {
2119                 status = -ENOENT;
2120                 mlog_errno(status);
2121                 goto leave;
2122         }
2123
2124         status = ocfs2_read_block(osb,
2125                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2126                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2127                                   orphan_dir_inode);
2128         if (status < 0) {
2129                 mlog_errno(status);
2130                 goto leave;
2131         }
2132
2133         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2134                                       OCFS2_JOURNAL_ACCESS_WRITE);
2135         if (status < 0) {
2136                 mlog_errno(status);
2137                 goto leave;
2138         }
2139
2140         /* we're a cluster, and nlink can change on disk from
2141          * underneath us... */
2142         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2143         if (S_ISDIR(inode->i_mode))
2144                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2145         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2146
2147         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2148         if (status < 0) {
2149                 mlog_errno(status);
2150                 goto leave;
2151         }
2152
2153         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2154                                    OCFS2_ORPHAN_NAMELEN, inode,
2155                                    OCFS2_I(inode)->ip_blkno,
2156                                    orphan_dir_bh, de_bh);
2157         if (status < 0) {
2158                 mlog_errno(status);
2159                 goto leave;
2160         }
2161
2162         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2163
2164         /* Record which orphan dir our inode now resides
2165          * in. delete_inode will use this to determine which orphan
2166          * dir to lock. */
2167         spin_lock(&OCFS2_I(inode)->ip_lock);
2168         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2169         spin_unlock(&OCFS2_I(inode)->ip_lock);
2170
2171         mlog(0, "Inode %llu orphaned in slot %d\n",
2172              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2173
2174 leave:
2175         if (orphan_dir_inode)
2176                 iput(orphan_dir_inode);
2177
2178         if (orphan_dir_bh)
2179                 brelse(orphan_dir_bh);
2180
2181         mlog_exit(status);
2182         return status;
2183 }
2184
2185 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2186 int ocfs2_orphan_del(struct ocfs2_super *osb,
2187                      struct ocfs2_journal_handle *handle,
2188                      struct inode *orphan_dir_inode,
2189                      struct inode *inode,
2190                      struct buffer_head *orphan_dir_bh)
2191 {
2192         char name[OCFS2_ORPHAN_NAMELEN + 1];
2193         struct ocfs2_dinode *orphan_fe;
2194         int status = 0;
2195         struct buffer_head *target_de_bh = NULL;
2196         struct ocfs2_dir_entry *target_de = NULL;
2197
2198         mlog_entry_void();
2199
2200         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2201         if (status < 0) {
2202                 mlog_errno(status);
2203                 goto leave;
2204         }
2205
2206         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2207              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2208              OCFS2_ORPHAN_NAMELEN);
2209
2210         /* find it's spot in the orphan directory */
2211         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2212                                         orphan_dir_inode, &target_de);
2213         if (!target_de_bh) {
2214                 status = -ENOENT;
2215                 mlog_errno(status);
2216                 goto leave;
2217         }
2218
2219         /* remove it from the orphan directory */
2220         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2221                                     target_de_bh);
2222         if (status < 0) {
2223                 mlog_errno(status);
2224                 goto leave;
2225         }
2226
2227         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2228                                       OCFS2_JOURNAL_ACCESS_WRITE);
2229         if (status < 0) {
2230                 mlog_errno(status);
2231                 goto leave;
2232         }
2233
2234         /* do the i_nlink dance! :) */
2235         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2236         if (S_ISDIR(inode->i_mode))
2237                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2238         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2239
2240         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2241         if (status < 0) {
2242                 mlog_errno(status);
2243                 goto leave;
2244         }
2245
2246 leave:
2247         if (target_de_bh)
2248                 brelse(target_de_bh);
2249
2250         mlog_exit(status);
2251         return status;
2252 }
2253
2254 struct inode_operations ocfs2_dir_iops = {
2255         .create         = ocfs2_create,
2256         .lookup         = ocfs2_lookup,
2257         .link           = ocfs2_link,
2258         .unlink         = ocfs2_unlink,
2259         .rmdir          = ocfs2_unlink,
2260         .symlink        = ocfs2_symlink,
2261         .mkdir          = ocfs2_mkdir,
2262         .mknod          = ocfs2_mknod,
2263         .rename         = ocfs2_rename,
2264         .setattr        = ocfs2_setattr,
2265         .getattr        = ocfs2_getattr,
2266 };