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 "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "vote.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 #define NAMEI_RA_CHUNKS  2
  68 #define NAMEI_RA_BLOCKS  4
  69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  71
  72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  73                                         struct inode *dir,
  74                                         const char *name, int namelen,
  75                                         unsigned long offset,
  76                                         struct ocfs2_dir_entry **res_dir);
  77
  78 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
  84                              struct inode *dir,
  85                              const char *name, int namelen,
  86                              struct inode *inode, u64 blkno,
  87                              struct buffer_head *parent_fe_bh,
  88                              struct buffer_head *insert_bh);
  89
  90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  91                               struct inode *dir,
  92                               struct dentry *dentry, int mode,
  93                               dev_t dev,
  94                               struct buffer_head **new_fe_bh,
  95                               struct buffer_head *parent_fe_bh,
  96                               struct ocfs2_journal_handle *handle,
  97                               struct inode **ret_inode,
  98                               struct ocfs2_alloc_context *inode_ac);
  99
 100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 101                               struct ocfs2_journal_handle *handle,
 102                               struct inode *parent,
 103                               struct inode *inode,
 104                               struct buffer_head *fe_bh,
 105                               struct ocfs2_alloc_context *data_ac);
 106
 107 static int ocfs2_double_lock(struct ocfs2_super *osb,
 108                              struct ocfs2_journal_handle *handle,
 109                              struct buffer_head **bh1,
 110                              struct inode *inode1,
 111                              struct buffer_head **bh2,
 112                              struct inode *inode2);
 113
 114 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 115                                     struct ocfs2_journal_handle *handle,
 116                                     struct inode *inode,
 117                                     char *name,
 118                                     struct buffer_head **de_bh);
 119
 120 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 121                             struct ocfs2_journal_handle *handle,
 122                             struct inode *inode,
 123                             struct ocfs2_dinode *fe,
 124                             char *name,
 125                             struct buffer_head *de_bh);
 126
 127 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 128                                      struct ocfs2_journal_handle *handle,
 129                                      struct inode *inode,
 130                                      const char *symname);
 131
 132 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
 133                                   struct dentry *dentry,
 134                                   struct inode *inode, u64 blkno,
 135                                   struct buffer_head *parent_fe_bh,
 136                                   struct buffer_head *insert_bh)
 137 {
 138         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 139                                  dentry->d_name.name, dentry->d_name.len,
 140                                  inode, blkno, parent_fe_bh, insert_bh);
 141 }
 142
 143 /* An orphan dir name is an 8 byte value, printed as a hex string */
 144 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 145
 146 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 147                                    struct nameidata *nd)
 148 {
 149         int status;
 150         u64 blkno;
 151         struct buffer_head *dirent_bh = NULL;
 152         struct inode *inode = NULL;
 153         struct dentry *ret;
 154         struct ocfs2_dir_entry *dirent;
 155         struct ocfs2_inode_info *oi;
 156
 157         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 158                    dentry->d_name.len, dentry->d_name.name);
 159
 160         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 161                 ret = ERR_PTR(-ENAMETOOLONG);
 162                 goto bail;
 163         }
 164
 165         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 166              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 167
 168         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
 169         if (status < 0) {
 170                 if (status != -ENOENT)
 171                         mlog_errno(status);
 172                 ret = ERR_PTR(status);
 173                 goto bail;
 174         }
 175
 176         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 177                                           dentry->d_name.len, &blkno,
 178                                           dir, &dirent_bh, &dirent);
 179         if (status < 0)
 180                 goto bail_add;
 181
 182         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 183         if (IS_ERR(inode)) {
 184                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 185                      (unsigned long long)blkno);
 186                 ret = ERR_PTR(-EACCES);
 187                 goto bail_unlock;
 188         }
 189
 190         oi = OCFS2_I(inode);
 191         /* Clear any orphaned state... If we were able to look up the
 192          * inode from a directory, it certainly can't be orphaned. We
 193          * might have the bad state from a node which intended to
 194          * orphan this inode but crashed before it could commit the
 195          * unlink. */
 196         spin_lock(&oi->ip_lock);
 197         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 198         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 199         spin_unlock(&oi->ip_lock);
 200
 201 bail_add:
 202         dentry->d_op = &ocfs2_dentry_ops;
 203         ret = d_splice_alias(inode, dentry);
 204
 205         if (inode) {
 206                 /*
 207                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 208                  * dentry, it will d_move() it on top of ourse. The
 209                  * return value will indicate this however, so in
 210                  * those cases, we switch them around for the locking
 211                  * code.
 212                  *
 213                  * NOTE: This dentry already has ->d_op set from
 214                  * ocfs2_get_parent() and ocfs2_get_dentry()
 215                  */
 216                 if (ret)
 217                         dentry = ret;
 218
 219                 status = ocfs2_dentry_attach_lock(dentry, inode,
 220                                                   OCFS2_I(dir)->ip_blkno);
 221                 if (status) {
 222                         mlog_errno(status);
 223                         ret = ERR_PTR(status);
 224                         goto bail_unlock;
 225                 }
 226         }
 227
 228 bail_unlock:
 229         /* Don't drop the cluster lock until *after* the d_add --
 230          * unlink on another node will message us to remove that
 231          * dentry under this lock so otherwise we can race this with
 232          * the vote thread and have a stale dentry. */
 233         ocfs2_meta_unlock(dir, 0);
 234
 235 bail:
 236         if (dirent_bh)
 237                 brelse(dirent_bh);
 238
 239         mlog_exit_ptr(ret);
 240
 241         return ret;
 242 }
 243
 244 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 245                               struct ocfs2_journal_handle *handle,
 246                               struct inode *parent,
 247                               struct inode *inode,
 248                               struct buffer_head *fe_bh,
 249                               struct ocfs2_alloc_context *data_ac)
 250 {
 251         int status;
 252         struct buffer_head *new_bh = NULL;
 253         struct ocfs2_dir_entry *de = NULL;
 254
 255         mlog_entry_void();
 256
 257         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 258                                      data_ac, NULL, &new_bh);
 259         if (status < 0) {
 260                 mlog_errno(status);
 261                 goto bail;
 262         }
 263
 264         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 265
 266         status = ocfs2_journal_access(handle, inode, new_bh,
 267                                       OCFS2_JOURNAL_ACCESS_CREATE);
 268         if (status < 0) {
 269                 mlog_errno(status);
 270                 goto bail;
 271         }
 272         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 273
 274         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 275         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 276         de->name_len = 1;
 277         de->rec_len =
 278                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 279         strcpy(de->name, ".");
 280         ocfs2_set_de_type(de, S_IFDIR);
 281         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 282         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 283         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 284                                   OCFS2_DIR_REC_LEN(1));
 285         de->name_len = 2;
 286         strcpy(de->name, "..");
 287         ocfs2_set_de_type(de, S_IFDIR);
 288
 289         status = ocfs2_journal_dirty(handle, new_bh);
 290         if (status < 0) {
 291                 mlog_errno(status);
 292                 goto bail;
 293         }
 294
 295         i_size_write(inode, inode->i_sb->s_blocksize);
 296         inode->i_nlink = 2;
 297         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 298         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 299         if (status < 0) {
 300                 mlog_errno(status);
 301                 goto bail;
 302         }
 303
 304         status = 0;
 305 bail:
 306         if (new_bh)
 307                 brelse(new_bh);
 308
 309         mlog_exit(status);
 310         return status;
 311 }
 312
 313 static int ocfs2_mknod(struct inode *dir,
 314                        struct dentry *dentry,
 315                        int mode,
 316                        dev_t dev)
 317 {
 318         int status = 0;
 319         struct buffer_head *parent_fe_bh = NULL;
 320         struct ocfs2_journal_handle *handle = NULL;
 321         struct ocfs2_super *osb;
 322         struct ocfs2_dinode *dirfe;
 323         struct buffer_head *new_fe_bh = NULL;
 324         struct buffer_head *de_bh = NULL;
 325         struct inode *inode = NULL;
 326         struct ocfs2_alloc_context *inode_ac = NULL;
 327         struct ocfs2_alloc_context *data_ac = NULL;
 328
 329         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 330                    (unsigned long)dev, dentry->d_name.len,
 331                    dentry->d_name.name);
 332
 333         /* get our super block */
 334         osb = OCFS2_SB(dir->i_sb);
 335
 336         status = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
 337         if (status < 0) {
 338                 if (status != -ENOENT)
 339                         mlog_errno(status);
 340                 return status;
 341         }
 342
 343         handle = ocfs2_alloc_handle(osb);
 344         if (handle == NULL) {
 345                 status = -ENOMEM;
 346                 mlog_errno(status);
 347                 goto leave;
 348         }
 349
 350         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 351                 status = -EMLINK;
 352                 goto leave;
 353         }
 354
 355         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 356         if (!dirfe->i_links_count) {
 357                 /* can't make a file in a deleted directory. */
 358                 status = -ENOENT;
 359                 goto leave;
 360         }
 361
 362         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 363                                            dentry->d_name.len);
 364         if (status)
 365                 goto leave;
 366
 367         /* get a spot inside the dir. */
 368         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 369                                               dentry->d_name.name,
 370                                               dentry->d_name.len, &de_bh);
 371         if (status < 0) {
 372                 mlog_errno(status);
 373                 goto leave;
 374         }
 375
 376         /* reserve an inode spot */
 377         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
 378         if (status < 0) {
 379                 if (status != -ENOSPC)
 380                         mlog_errno(status);
 381                 goto leave;
 382         }
 383
 384         /* are we making a directory? If so, reserve a cluster for his
 385          * 1st extent. */
 386         if (S_ISDIR(mode)) {
 387                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 388                 if (status < 0) {
 389                         if (status != -ENOSPC)
 390                                 mlog_errno(status);
 391                         goto leave;
 392                 }
 393         }
 394
 395         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
 396         if (IS_ERR(handle)) {
 397                 status = PTR_ERR(handle);
 398                 handle = NULL;
 399                 mlog_errno(status);
 400                 goto leave;
 401         }
 402
 403         /* do the real work now. */
 404         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 405                                     &new_fe_bh, parent_fe_bh, handle,
 406                                     &inode, inode_ac);
 407         if (status < 0) {
 408                 mlog_errno(status);
 409                 goto leave;
 410         }
 411
 412         if (S_ISDIR(mode)) {
 413                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 414                                             new_fe_bh, data_ac);
 415                 if (status < 0) {
 416                         mlog_errno(status);
 417                         goto leave;
 418                 }
 419
 420                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 421                                               OCFS2_JOURNAL_ACCESS_WRITE);
 422                 if (status < 0) {
 423                         mlog_errno(status);
 424                         goto leave;
 425                 }
 426                 le16_add_cpu(&dirfe->i_links_count, 1);
 427                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 428                 if (status < 0) {
 429                         mlog_errno(status);
 430                         goto leave;
 431                 }
 432                 inc_nlink(dir);
 433         }
 434
 435         status = ocfs2_add_entry(handle, dentry, inode,
 436                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 437                                  de_bh);
 438         if (status < 0) {
 439                 mlog_errno(status);
 440                 goto leave;
 441         }
 442
 443         status = ocfs2_dentry_attach_lock(dentry, inode,
 444                                           OCFS2_I(dir)->ip_blkno);
 445         if (status) {
 446                 mlog_errno(status);
 447                 goto leave;
 448         }
 449
 450         insert_inode_hash(inode);
 451         dentry->d_op = &ocfs2_dentry_ops;
 452         d_instantiate(dentry, inode);
 453         status = 0;
 454 leave:
 455         if (handle)
 456                 ocfs2_commit_trans(handle);
 457
 458         ocfs2_meta_unlock(dir, 1);
 459
 460         if (status == -ENOSPC)
 461                 mlog(0, "Disk is full\n");
 462
 463         if (new_fe_bh)
 464                 brelse(new_fe_bh);
 465
 466         if (de_bh)
 467                 brelse(de_bh);
 468
 469         if (parent_fe_bh)
 470                 brelse(parent_fe_bh);
 471
 472         if ((status < 0) && inode)
 473                 iput(inode);
 474
 475         if (inode_ac)
 476                 ocfs2_free_alloc_context(inode_ac);
 477
 478         if (data_ac)
 479                 ocfs2_free_alloc_context(data_ac);
 480
 481         mlog_exit(status);
 482
 483         return status;
 484 }
 485
 486 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 487                               struct inode *dir,
 488                               struct dentry *dentry, int mode,
 489                               dev_t dev,
 490                               struct buffer_head **new_fe_bh,
 491                               struct buffer_head *parent_fe_bh,
 492                               struct ocfs2_journal_handle *handle,
 493                               struct inode **ret_inode,
 494                               struct ocfs2_alloc_context *inode_ac)
 495 {
 496         int status = 0;
 497         struct ocfs2_dinode *fe = NULL;
 498         struct ocfs2_extent_list *fel;
 499         u64 fe_blkno = 0;
 500         u16 suballoc_bit;
 501         struct inode *inode = NULL;
 502
 503         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 504                    (unsigned long)dev, dentry->d_name.len,
 505                    dentry->d_name.name);
 506
 507         *new_fe_bh = NULL;
 508         *ret_inode = NULL;
 509
 510         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 511                                        &fe_blkno);
 512         if (status < 0) {
 513                 mlog_errno(status);
 514                 goto leave;
 515         }
 516
 517         inode = new_inode(dir->i_sb);
 518         if (IS_ERR(inode)) {
 519                 status = PTR_ERR(inode);
 520                 mlog(ML_ERROR, "new_inode failed!\n");
 521                 goto leave;
 522         }
 523
 524         /* populate as many fields early on as possible - many of
 525          * these are used by the support functions here and in
 526          * callers. */
 527         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 528         OCFS2_I(inode)->ip_blkno = fe_blkno;
 529         if (S_ISDIR(mode))
 530                 inode->i_nlink = 2;
 531         else
 532                 inode->i_nlink = 1;
 533         inode->i_mode = mode;
 534         spin_lock(&osb->osb_lock);
 535         inode->i_generation = osb->s_next_generation++;
 536         spin_unlock(&osb->osb_lock);
 537
 538         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 539         if (!*new_fe_bh) {
 540                 status = -EIO;
 541                 mlog_errno(status);
 542                 goto leave;
 543         }
 544         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 545
 546         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 547                                       OCFS2_JOURNAL_ACCESS_CREATE);
 548         if (status < 0) {
 549                 mlog_errno(status);
 550                 goto leave;
 551         }
 552
 553         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 554         memset(fe, 0, osb->sb->s_blocksize);
 555
 556         fe->i_generation = cpu_to_le32(inode->i_generation);
 557         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 558         fe->i_blkno = cpu_to_le64(fe_blkno);
 559         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 560         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 561         fe->i_uid = cpu_to_le32(current->fsuid);
 562         if (dir->i_mode & S_ISGID) {
 563                 fe->i_gid = cpu_to_le32(dir->i_gid);
 564                 if (S_ISDIR(mode))
 565                         mode |= S_ISGID;
 566         } else
 567                 fe->i_gid = cpu_to_le32(current->fsgid);
 568         fe->i_mode = cpu_to_le16(mode);
 569         if (S_ISCHR(mode) || S_ISBLK(mode))
 570                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 571
 572         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 573
 574         fe->i_last_eb_blk = 0;
 575         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 576         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 577         fe->i_atime = fe->i_ctime = fe->i_mtime =
 578                 cpu_to_le64(CURRENT_TIME.tv_sec);
 579         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 580                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 581         fe->i_dtime = 0;
 582
 583         fel = &fe->id2.i_list;
 584         fel->l_tree_depth = 0;
 585         fel->l_next_free_rec = 0;
 586         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 587
 588         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 589         if (status < 0) {
 590                 mlog_errno(status);
 591                 goto leave;
 592         }
 593
 594         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 595                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 596                      "i_blkno=%llu, i_ino=%lu\n",
 597                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 598                      (unsigned long long)fe->i_blkno, inode->i_ino);
 599                 BUG();
 600         }
 601
 602         ocfs2_inode_set_new(osb, inode);
 603         status = ocfs2_create_new_inode_locks(inode);
 604         if (status < 0)
 605                 mlog_errno(status);
 606
 607         status = 0; /* error in ocfs2_create_new_inode_locks is not
 608                      * critical */
 609
 610         *ret_inode = inode;
 611 leave:
 612         if (status < 0) {
 613                 if (*new_fe_bh) {
 614                         brelse(*new_fe_bh);
 615                         *new_fe_bh = NULL;
 616                 }
 617                 if (inode)
 618                         iput(inode);
 619         }
 620
 621         mlog_exit(status);
 622         return status;
 623 }
 624
 625 static int ocfs2_mkdir(struct inode *dir,
 626                        struct dentry *dentry,
 627                        int mode)
 628 {
 629         int ret;
 630
 631         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 632                    dentry->d_name.len, dentry->d_name.name);
 633         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 634         mlog_exit(ret);
 635
 636         return ret;
 637 }
 638
 639 static int ocfs2_create(struct inode *dir,
 640                         struct dentry *dentry,
 641                         int mode,
 642                         struct nameidata *nd)
 643 {
 644         int ret;
 645
 646         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 647                    dentry->d_name.len, dentry->d_name.name);
 648         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 649         mlog_exit(ret);
 650
 651         return ret;
 652 }
 653
 654 static int ocfs2_link(struct dentry *old_dentry,
 655                       struct inode *dir,
 656                       struct dentry *dentry)
 657 {
 658         struct ocfs2_journal_handle *handle = NULL;
 659         struct inode *inode = old_dentry->d_inode;
 660         int err;
 661         struct buffer_head *fe_bh = NULL;
 662         struct buffer_head *parent_fe_bh = NULL;
 663         struct buffer_head *de_bh = NULL;
 664         struct ocfs2_dinode *fe = NULL;
 665         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 666
 667         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 668                    old_dentry->d_name.len, old_dentry->d_name.name,
 669                    dentry->d_name.len, dentry->d_name.name);
 670
 671         if (S_ISDIR(inode->i_mode)) {
 672                 err = -EPERM;
 673                 goto bail;
 674         }
 675
 676         handle = ocfs2_alloc_handle(osb);
 677         if (handle == NULL) {
 678                 err = -ENOMEM;
 679                 goto bail;
 680         }
 681
 682         err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 683         if (err < 0) {
 684                 if (err != -ENOENT)
 685                         mlog_errno(err);
 686                 goto bail;
 687         }
 688
 689         if (!dir->i_nlink) {
 690                 err = -ENOENT;
 691                 goto bail;
 692         }
 693
 694         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 695                                         dentry->d_name.len);
 696         if (err)
 697                 goto bail;
 698
 699         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 700                                            dentry->d_name.name,
 701                                            dentry->d_name.len, &de_bh);
 702         if (err < 0) {
 703                 mlog_errno(err);
 704                 goto bail;
 705         }
 706
 707         err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 708         if (err < 0) {
 709                 if (err != -ENOENT)
 710                         mlog_errno(err);
 711                 goto bail;
 712         }
 713
 714         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 715         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 716                 err = -EMLINK;
 717                 goto bail;
 718         }
 719
 720         handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
 721         if (IS_ERR(handle)) {
 722                 err = PTR_ERR(handle);
 723                 handle = NULL;
 724                 mlog_errno(err);
 725                 goto bail;
 726         }
 727
 728         err = ocfs2_journal_access(handle, inode, fe_bh,
 729                                    OCFS2_JOURNAL_ACCESS_WRITE);
 730         if (err < 0) {
 731                 mlog_errno(err);
 732                 goto bail;
 733         }
 734
 735         inc_nlink(inode);
 736         inode->i_ctime = CURRENT_TIME;
 737         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 738         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 739         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 740
 741         err = ocfs2_journal_dirty(handle, fe_bh);
 742         if (err < 0) {
 743                 le16_add_cpu(&fe->i_links_count, -1);
 744                 drop_nlink(inode);
 745                 mlog_errno(err);
 746                 goto bail;
 747         }
 748
 749         err = ocfs2_add_entry(handle, dentry, inode,
 750                               OCFS2_I(inode)->ip_blkno,
 751                               parent_fe_bh, de_bh);
 752         if (err) {
 753                 le16_add_cpu(&fe->i_links_count, -1);
 754                 drop_nlink(inode);
 755                 mlog_errno(err);
 756                 goto bail;
 757         }
 758
 759         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 760         if (err) {
 761                 mlog_errno(err);
 762                 goto bail;
 763         }
 764
 765         atomic_inc(&inode->i_count);
 766         dentry->d_op = &ocfs2_dentry_ops;
 767         d_instantiate(dentry, inode);
 768 bail:
 769         if (handle)
 770                 ocfs2_commit_trans(handle);
 771         if (de_bh)
 772                 brelse(de_bh);
 773         if (fe_bh)
 774                 brelse(fe_bh);
 775         if (parent_fe_bh)
 776                 brelse(parent_fe_bh);
 777
 778         mlog_exit(err);
 779
 780         return err;
 781 }
 782
 783 /*
 784  * Takes and drops an exclusive lock on the given dentry. This will
 785  * force other nodes to drop it.
 786  */
 787 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 788 {
 789         int ret;
 790
 791         ret = ocfs2_dentry_lock(dentry, 1);
 792         if (ret)
 793                 mlog_errno(ret);
 794         else
 795                 ocfs2_dentry_unlock(dentry, 1);
 796
 797         return ret;
 798 }
 799
 800 static inline int inode_is_unlinkable(struct inode *inode)
 801 {
 802         if (S_ISDIR(inode->i_mode)) {
 803                 if (inode->i_nlink == 2)
 804                         return 1;
 805                 return 0;
 806         }
 807
 808         if (inode->i_nlink == 1)
 809                 return 1;
 810         return 0;
 811 }
 812
 813 static int ocfs2_unlink(struct inode *dir,
 814                         struct dentry *dentry)
 815 {
 816         int status;
 817         struct inode *inode = dentry->d_inode;
 818         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 819         u64 blkno;
 820         struct ocfs2_dinode *fe = NULL;
 821         struct buffer_head *fe_bh = NULL;
 822         struct buffer_head *parent_node_bh = NULL;
 823         struct ocfs2_journal_handle *handle = NULL;
 824         struct ocfs2_dir_entry *dirent = NULL;
 825         struct buffer_head *dirent_bh = NULL;
 826         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 827         struct buffer_head *orphan_entry_bh = NULL;
 828
 829         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 830                    dentry->d_name.len, dentry->d_name.name);
 831
 832         BUG_ON(dentry->d_parent->d_inode != dir);
 833
 834         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 835
 836         if (inode == osb->root_inode) {
 837                 mlog(0, "Cannot delete the root directory\n");
 838                 status = -EPERM;
 839                 goto leave;
 840         }
 841
 842         handle = ocfs2_alloc_handle(osb);
 843         if (handle == NULL) {
 844                 status = -ENOMEM;
 845                 mlog_errno(status);
 846                 goto leave;
 847         }
 848
 849         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
 850         if (status < 0) {
 851                 if (status != -ENOENT)
 852                         mlog_errno(status);
 853                 goto leave;
 854         }
 855
 856         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 857                                           dentry->d_name.len, &blkno,
 858                                           dir, &dirent_bh, &dirent);
 859         if (status < 0) {
 860                 if (status != -ENOENT)
 861                         mlog_errno(status);
 862                 goto leave;
 863         }
 864
 865         if (OCFS2_I(inode)->ip_blkno != blkno) {
 866                 status = -ENOENT;
 867
 868                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 869                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 870                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 871                 goto leave;
 872         }
 873
 874         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 875         if (status < 0) {
 876                 if (status != -ENOENT)
 877                         mlog_errno(status);
 878                 goto leave;
 879         }
 880
 881         if (S_ISDIR(inode->i_mode)) {
 882                 if (!ocfs2_empty_dir(inode)) {
 883                         status = -ENOTEMPTY;
 884                         goto leave;
 885                 } else if (inode->i_nlink != 2) {
 886                         status = -ENOTEMPTY;
 887                         goto leave;
 888                 }
 889         }
 890
 891         status = ocfs2_remote_dentry_delete(dentry);
 892         if (status < 0) {
 893                 /* This vote should succeed under all normal
 894                  * circumstances. */
 895                 mlog_errno(status);
 896                 goto leave;
 897         }
 898
 899         if (inode_is_unlinkable(inode)) {
 900                 status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 901                                                   orphan_name,
 902                                                   &orphan_entry_bh);
 903                 if (status < 0) {
 904                         mlog_errno(status);
 905                         goto leave;
 906                 }
 907         }
 908
 909         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
 910         if (IS_ERR(handle)) {
 911                 status = PTR_ERR(handle);
 912                 handle = NULL;
 913                 mlog_errno(status);
 914                 goto leave;
 915         }
 916
 917         status = ocfs2_journal_access(handle, inode, fe_bh,
 918                                       OCFS2_JOURNAL_ACCESS_WRITE);
 919         if (status < 0) {
 920                 mlog_errno(status);
 921                 goto leave;
 922         }
 923
 924         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 925
 926         if (inode_is_unlinkable(inode)) {
 927                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 928                                           orphan_entry_bh);
 929                 if (status < 0) {
 930                         mlog_errno(status);
 931                         goto leave;
 932                 }
 933         }
 934
 935         /* delete the name from the parent dir */
 936         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 937         if (status < 0) {
 938                 mlog_errno(status);
 939                 goto leave;
 940         }
 941
 942         if (S_ISDIR(inode->i_mode))
 943                 drop_nlink(inode);
 944         drop_nlink(inode);
 945         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 946
 947         status = ocfs2_journal_dirty(handle, fe_bh);
 948         if (status < 0) {
 949                 mlog_errno(status);
 950                 goto leave;
 951         }
 952
 953         if (S_ISDIR(inode->i_mode)) {
 954                 drop_nlink(dir);
 955                 status = ocfs2_mark_inode_dirty(handle, dir,
 956                                                 parent_node_bh);
 957                 if (status < 0) {
 958                         mlog_errno(status);
 959                         inc_nlink(dir);
 960                 }
 961         }
 962
 963 leave:
 964         if (handle)
 965                 ocfs2_commit_trans(handle);
 966
 967         if (fe_bh)
 968                 brelse(fe_bh);
 969
 970         if (dirent_bh)
 971                 brelse(dirent_bh);
 972
 973         if (parent_node_bh)
 974                 brelse(parent_node_bh);
 975
 976         if (orphan_entry_bh)
 977                 brelse(orphan_entry_bh);
 978
 979         mlog_exit(status);
 980
 981         return status;
 982 }
 983
 984 /*
 985  * The only place this should be used is rename!
 986  * if they have the same id, then the 1st one is the only one locked.
 987  */
 988 static int ocfs2_double_lock(struct ocfs2_super *osb,
 989                              struct ocfs2_journal_handle *handle,
 990                              struct buffer_head **bh1,
 991                              struct inode *inode1,
 992                              struct buffer_head **bh2,
 993                              struct inode *inode2)
 994 {
 995         int status;
 996         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 997         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 998         struct buffer_head **tmpbh;
 999         struct inode *tmpinode;
1000
1001         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1002                    (unsigned long long)oi1->ip_blkno,
1003                    (unsigned long long)oi2->ip_blkno);
1004
1005         BUG_ON(!handle);
1006
1007         if (*bh1)
1008                 *bh1 = NULL;
1009         if (*bh2)
1010                 *bh2 = NULL;
1011
1012         /* we always want to lock the one with the lower lockid first. */
1013         if (oi1->ip_blkno != oi2->ip_blkno) {
1014                 if (oi1->ip_blkno < oi2->ip_blkno) {
1015                         /* switch id1 and id2 around */
1016                         mlog(0, "switching them around...\n");
1017                         tmpbh = bh2;
1018                         bh2 = bh1;
1019                         bh1 = tmpbh;
1020
1021                         tmpinode = inode2;
1022                         inode2 = inode1;
1023                         inode1 = tmpinode;
1024                 }
1025                 /* lock id2 */
1026                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1027                 if (status < 0) {
1028                         if (status != -ENOENT)
1029                                 mlog_errno(status);
1030                         goto bail;
1031                 }
1032         }
1033         /* lock id1 */
1034         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1035         if (status < 0) {
1036                 if (status != -ENOENT)
1037                         mlog_errno(status);
1038                 goto bail;
1039         }
1040 bail:
1041         mlog_exit(status);
1042         return status;
1043 }
1044
1045 #define PARENT_INO(buffer) \
1046         ((struct ocfs2_dir_entry *) \
1047          ((char *)buffer + \
1048           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1049
1050 static int ocfs2_rename(struct inode *old_dir,
1051                         struct dentry *old_dentry,
1052                         struct inode *new_dir,
1053                         struct dentry *new_dentry)
1054 {
1055         int status = 0, rename_lock = 0;
1056         struct inode *old_inode = old_dentry->d_inode;
1057         struct inode *new_inode = new_dentry->d_inode;
1058         struct ocfs2_dinode *newfe = NULL;
1059         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1060         struct buffer_head *orphan_entry_bh = NULL;
1061         struct buffer_head *newfe_bh = NULL;
1062         struct buffer_head *insert_entry_bh = NULL;
1063         struct ocfs2_super *osb = NULL;
1064         u64 newfe_blkno;
1065         struct ocfs2_journal_handle *handle = NULL;
1066         struct buffer_head *old_dir_bh = NULL;
1067         struct buffer_head *new_dir_bh = NULL;
1068         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1069                                                                // and new_dentry
1070         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1071         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1072                                                     // this is the 1st dirent bh
1073         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1074
1075         /* At some point it might be nice to break this function up a
1076          * bit. */
1077
1078         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1079                    old_dir, old_dentry, new_dir, new_dentry,
1080                    old_dentry->d_name.len, old_dentry->d_name.name,
1081                    new_dentry->d_name.len, new_dentry->d_name.name);
1082
1083         osb = OCFS2_SB(old_dir->i_sb);
1084
1085         if (new_inode) {
1086                 if (!igrab(new_inode))
1087                         BUG();
1088         }
1089
1090         /* Assume a directory heirarchy thusly:
1091          * a/b/c
1092          * a/d
1093          * a,b,c, and d are all directories.
1094          *
1095          * from cwd of 'a' on both nodes:
1096          * node1: mv b/c d
1097          * node2: mv d   b/c
1098          *
1099          * And that's why, just like the VFS, we need a file system
1100          * rename lock. */
1101         if (old_dentry != new_dentry) {
1102                 status = ocfs2_rename_lock(osb);
1103                 if (status < 0) {
1104                         mlog_errno(status);
1105                         goto bail;
1106                 }
1107                 rename_lock = 1;
1108         }
1109
1110         handle = ocfs2_alloc_handle(osb);
1111         if (handle == NULL) {
1112                 status = -ENOMEM;
1113                 mlog_errno(status);
1114                 goto bail;
1115         }
1116
1117         /* if old and new are the same, this'll just do one lock. */
1118         status = ocfs2_double_lock(osb, handle,
1119                                   &old_dir_bh, old_dir,
1120                                   &new_dir_bh, new_dir);
1121         if (status < 0) {
1122                 mlog_errno(status);
1123                 goto bail;
1124         }
1125
1126         /* make sure both dirs have bhs
1127          * get an extra ref on old_dir_bh if old==new */
1128         if (!new_dir_bh) {
1129                 if (old_dir_bh) {
1130                         new_dir_bh = old_dir_bh;
1131                         get_bh(new_dir_bh);
1132                 } else {
1133                         mlog(ML_ERROR, "no old_dir_bh!\n");
1134                         status = -EIO;
1135                         goto bail;
1136                 }
1137         }
1138
1139         /*
1140          * Though we don't require an inode meta data update if
1141          * old_inode is not a directory, we lock anyway here to ensure
1142          * the vote thread on other nodes won't have to concurrently
1143          * downconvert the inode and the dentry locks.
1144          */
1145         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1146         if (status < 0) {
1147                 if (status != -ENOENT)
1148                         mlog_errno(status);
1149                 goto bail;
1150         }
1151
1152         status = ocfs2_remote_dentry_delete(old_dentry);
1153         if (status < 0) {
1154                 mlog_errno(status);
1155                 goto bail;
1156         }
1157
1158         if (S_ISDIR(old_inode->i_mode)) {
1159                 status = -EIO;
1160                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1161                 if (!old_inode_de_bh)
1162                         goto bail;
1163
1164                 status = -EIO;
1165                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1166                     OCFS2_I(old_dir)->ip_blkno)
1167                         goto bail;
1168                 status = -EMLINK;
1169                 if (!new_inode && new_dir!=old_dir &&
1170                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1171                         goto bail;
1172         }
1173
1174         status = -ENOENT;
1175         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1176                                      old_dentry->d_name.len,
1177                                      old_dir, &old_de);
1178         if (!old_de_bh)
1179                 goto bail;
1180
1181         /*
1182          *  Check for inode number is _not_ due to possible IO errors.
1183          *  We might rmdir the source, keep it as pwd of some process
1184          *  and merrily kill the link to whatever was created under the
1185          *  same name. Goodbye sticky bit ;-<
1186          */
1187         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1188                 goto bail;
1189
1190         /* check if the target already exists (in which case we need
1191          * to delete it */
1192         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1193                                           new_dentry->d_name.len,
1194                                           &newfe_blkno, new_dir, &new_de_bh,
1195                                           &new_de);
1196         /* The only error we allow here is -ENOENT because the new
1197          * file not existing is perfectly valid. */
1198         if ((status < 0) && (status != -ENOENT)) {
1199                 /* If we cannot find the file specified we should just */
1200                 /* return the error... */
1201                 mlog_errno(status);
1202                 goto bail;
1203         }
1204
1205         if (!new_de && new_inode)
1206                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1207                      "directory!", new_inode->i_ino);
1208
1209         /* In case we need to overwrite an existing file, we blow it
1210          * away first */
1211         if (new_de) {
1212                 /* VFS didn't think there existed an inode here, but
1213                  * someone else in the cluster must have raced our
1214                  * rename to create one. Today we error cleanly, in
1215                  * the future we should consider calling iget to build
1216                  * a new struct inode for this entry. */
1217                 if (!new_inode) {
1218                         status = -EACCES;
1219
1220                         mlog(0, "We found an inode for name %.*s but VFS "
1221                              "didn't give us one.\n", new_dentry->d_name.len,
1222                              new_dentry->d_name.name);
1223                         goto bail;
1224                 }
1225
1226                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1227                         status = -EACCES;
1228
1229                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1230                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1231                              (unsigned long long)newfe_blkno,
1232                              OCFS2_I(new_inode)->ip_flags);
1233                         goto bail;
1234                 }
1235
1236                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1237                 if (status < 0) {
1238                         if (status != -ENOENT)
1239                                 mlog_errno(status);
1240                         goto bail;
1241                 }
1242
1243                 status = ocfs2_remote_dentry_delete(new_dentry);
1244                 if (status < 0) {
1245                         mlog_errno(status);
1246                         goto bail;
1247                 }
1248
1249                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1250
1251                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1252                      "newfebh=%p bhblocknr=%llu\n", new_de,
1253                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1254                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1255
1256                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1257                         status = ocfs2_prepare_orphan_dir(osb, handle,
1258                                                           new_inode,
1259                                                           orphan_name,
1260                                                           &orphan_entry_bh);
1261                         if (status < 0) {
1262                                 mlog_errno(status);
1263                                 goto bail;
1264                         }
1265                 }
1266         } else {
1267                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1268
1269                 status = ocfs2_check_dir_for_entry(new_dir,
1270                                                    new_dentry->d_name.name,
1271                                                    new_dentry->d_name.len);
1272                 if (status)
1273                         goto bail;
1274
1275                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1276                                                       new_dentry->d_name.name,
1277                                                       new_dentry->d_name.len,
1278                                                       &insert_entry_bh);
1279                 if (status < 0) {
1280                         mlog_errno(status);
1281                         goto bail;
1282                 }
1283         }
1284
1285         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1286         if (IS_ERR(handle)) {
1287                 status = PTR_ERR(handle);
1288                 handle = NULL;
1289                 mlog_errno(status);
1290                 goto bail;
1291         }
1292
1293         if (new_de) {
1294                 if (S_ISDIR(new_inode->i_mode)) {
1295                         if (!ocfs2_empty_dir(new_inode) ||
1296                             new_inode->i_nlink != 2) {
1297                                 status = -ENOTEMPTY;
1298                                 goto bail;
1299                         }
1300                 }
1301                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1302                                               OCFS2_JOURNAL_ACCESS_WRITE);
1303                 if (status < 0) {
1304                         mlog_errno(status);
1305                         goto bail;
1306                 }
1307
1308                 if (S_ISDIR(new_inode->i_mode) ||
1309                     (newfe->i_links_count == cpu_to_le16(1))){
1310                         status = ocfs2_orphan_add(osb, handle, new_inode,
1311                                                   newfe, orphan_name,
1312                                                   orphan_entry_bh);
1313                         if (status < 0) {
1314                                 mlog_errno(status);
1315                                 goto bail;
1316                         }
1317                 }
1318
1319                 /* change the dirent to point to the correct inode */
1320                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1321                                               OCFS2_JOURNAL_ACCESS_WRITE);
1322                 if (status < 0) {
1323                         mlog_errno(status);
1324                         goto bail;
1325                 }
1326                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1327                 new_de->file_type = old_de->file_type;
1328                 new_dir->i_version++;
1329                 status = ocfs2_journal_dirty(handle, new_de_bh);
1330                 if (status < 0) {
1331                         mlog_errno(status);
1332                         goto bail;
1333                 }
1334
1335                 if (S_ISDIR(new_inode->i_mode))
1336                         newfe->i_links_count = 0;
1337                 else
1338                         le16_add_cpu(&newfe->i_links_count, -1);
1339
1340                 status = ocfs2_journal_dirty(handle, newfe_bh);
1341                 if (status < 0) {
1342                         mlog_errno(status);
1343                         goto bail;
1344                 }
1345         } else {
1346                 /* if the name was not found in new_dir, add it now */
1347                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1348                                          OCFS2_I(old_inode)->ip_blkno,
1349                                          new_dir_bh, insert_entry_bh);
1350         }
1351
1352         old_inode->i_ctime = CURRENT_TIME;
1353         mark_inode_dirty(old_inode);
1354
1355         /* now that the name has been added to new_dir, remove the old name */
1356         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1357         if (status < 0) {
1358                 mlog_errno(status);
1359                 goto bail;
1360         }
1361
1362         if (new_inode) {
1363                 new_inode->i_nlink--;
1364                 new_inode->i_ctime = CURRENT_TIME;
1365         }
1366         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1367         if (old_inode_de_bh) {
1368                 status = ocfs2_journal_access(handle, old_inode,
1369                                              old_inode_de_bh,
1370                                              OCFS2_JOURNAL_ACCESS_WRITE);
1371                 PARENT_INO(old_inode_de_bh->b_data) =
1372                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1373                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1374                 old_dir->i_nlink--;
1375                 if (new_inode) {
1376                         new_inode->i_nlink--;
1377                 } else {
1378                         inc_nlink(new_dir);
1379                         mark_inode_dirty(new_dir);
1380                 }
1381         }
1382         mark_inode_dirty(old_dir);
1383         if (new_inode)
1384                 mark_inode_dirty(new_inode);
1385
1386         if (old_dir != new_dir)
1387                 if (new_dir_nlink != new_dir->i_nlink) {
1388                         if (!new_dir_bh) {
1389                                 mlog(ML_ERROR, "need to change nlink for new "
1390                                      "dir %llu from %d to %d but bh is NULL\n",
1391                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1392                                      (int)new_dir_nlink, new_dir->i_nlink);
1393                         } else {
1394                                 struct ocfs2_dinode *fe;
1395                                 status = ocfs2_journal_access(handle,
1396                                                               new_dir,
1397                                                               new_dir_bh,
1398                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1399                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1400                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1401                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1402                         }
1403                 }
1404
1405         if (old_dir_nlink != old_dir->i_nlink) {
1406                 if (!old_dir_bh) {
1407                         mlog(ML_ERROR, "need to change nlink for old dir "
1408                              "%llu from %d to %d but bh is NULL!\n",
1409                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1410                              (int)old_dir_nlink, old_dir->i_nlink);
1411                 } else {
1412                         struct ocfs2_dinode *fe;
1413                         status = ocfs2_journal_access(handle, old_dir,
1414                                                       old_dir_bh,
1415                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1416                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1417                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1418                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1419                 }
1420         }
1421
1422         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1423         status = 0;
1424 bail:
1425         if (rename_lock)
1426                 ocfs2_rename_unlock(osb);
1427
1428         if (handle)
1429                 ocfs2_commit_trans(handle);
1430
1431         if (new_inode)
1432                 sync_mapping_buffers(old_inode->i_mapping);
1433
1434         if (new_inode)
1435                 iput(new_inode);
1436         if (newfe_bh)
1437                 brelse(newfe_bh);
1438         if (old_dir_bh)
1439                 brelse(old_dir_bh);
1440         if (new_dir_bh)
1441                 brelse(new_dir_bh);
1442         if (new_de_bh)
1443                 brelse(new_de_bh);
1444         if (old_de_bh)
1445                 brelse(old_de_bh);
1446         if (old_inode_de_bh)
1447                 brelse(old_inode_de_bh);
1448         if (orphan_entry_bh)
1449                 brelse(orphan_entry_bh);
1450         if (insert_entry_bh)
1451                 brelse(insert_entry_bh);
1452
1453         mlog_exit(status);
1454
1455         return status;
1456 }
1457
1458 /*
1459  * we expect i_size = strlen(symname). Copy symname into the file
1460  * data, including the null terminator.
1461  */
1462 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1463                                      struct ocfs2_journal_handle *handle,
1464                                      struct inode *inode,
1465                                      const char *symname)
1466 {
1467         struct buffer_head **bhs = NULL;
1468         const char *c;
1469         struct super_block *sb = osb->sb;
1470         u64 p_blkno;
1471         int p_blocks;
1472         int virtual, blocks, status, i, bytes_left;
1473
1474         bytes_left = i_size_read(inode) + 1;
1475         /* we can't trust i_blocks because we're actually going to
1476          * write i_size + 1 bytes. */
1477         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1478
1479         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1480                         (unsigned long long)inode->i_blocks,
1481                         i_size_read(inode), blocks);
1482
1483         /* Sanity check -- make sure we're going to fit. */
1484         if (bytes_left >
1485             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1486                 status = -EIO;
1487                 mlog_errno(status);
1488                 goto bail;
1489         }
1490
1491         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1492         if (!bhs) {
1493                 status = -ENOMEM;
1494                 mlog_errno(status);
1495                 goto bail;
1496         }
1497
1498         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1499                                              &p_blocks);
1500         if (status < 0) {
1501                 mlog_errno(status);
1502                 goto bail;
1503         }
1504
1505         /* links can never be larger than one cluster so we know this
1506          * is all going to be contiguous, but do a sanity check
1507          * anyway. */
1508         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1509                 status = -EIO;
1510                 mlog_errno(status);
1511                 goto bail;
1512         }
1513
1514         virtual = 0;
1515         while(bytes_left > 0) {
1516                 c = &symname[virtual * sb->s_blocksize];
1517
1518                 bhs[virtual] = sb_getblk(sb, p_blkno);
1519                 if (!bhs[virtual]) {
1520                         status = -ENOMEM;
1521                         mlog_errno(status);
1522                         goto bail;
1523                 }
1524                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1525
1526                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1527                                               OCFS2_JOURNAL_ACCESS_CREATE);
1528                 if (status < 0) {
1529                         mlog_errno(status);
1530                         goto bail;
1531                 }
1532
1533                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1534
1535                 memcpy(bhs[virtual]->b_data, c,
1536                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1537                        bytes_left);
1538
1539                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1540                 if (status < 0) {
1541                         mlog_errno(status);
1542                         goto bail;
1543                 }
1544
1545                 virtual++;
1546                 p_blkno++;
1547                 bytes_left -= sb->s_blocksize;
1548         }
1549
1550         status = 0;
1551 bail:
1552
1553         if (bhs) {
1554                 for(i = 0; i < blocks; i++)
1555                         if (bhs[i])
1556                                 brelse(bhs[i]);
1557                 kfree(bhs);
1558         }
1559
1560         mlog_exit(status);
1561         return status;
1562 }
1563
1564 static int ocfs2_symlink(struct inode *dir,
1565                          struct dentry *dentry,
1566                          const char *symname)
1567 {
1568         int status, l, credits;
1569         u64 newsize;
1570         struct ocfs2_super *osb = NULL;
1571         struct inode *inode = NULL;
1572         struct super_block *sb;
1573         struct buffer_head *new_fe_bh = NULL;
1574         struct buffer_head *de_bh = NULL;
1575         struct buffer_head *parent_fe_bh = NULL;
1576         struct ocfs2_dinode *fe = NULL;
1577         struct ocfs2_dinode *dirfe;
1578         struct ocfs2_journal_handle *handle = NULL;
1579         struct ocfs2_alloc_context *inode_ac = NULL;
1580         struct ocfs2_alloc_context *data_ac = NULL;
1581
1582         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1583                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1584
1585         sb = dir->i_sb;
1586         osb = OCFS2_SB(sb);
1587
1588         l = strlen(symname) + 1;
1589
1590         credits = ocfs2_calc_symlink_credits(sb);
1591
1592         handle = ocfs2_alloc_handle(osb);
1593         if (handle == NULL) {
1594                 status = -ENOMEM;
1595                 mlog_errno(status);
1596                 goto bail;
1597         }
1598
1599         /* lock the parent directory */
1600         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1601         if (status < 0) {
1602                 if (status != -ENOENT)
1603                         mlog_errno(status);
1604                 goto bail;
1605         }
1606
1607         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1608         if (!dirfe->i_links_count) {
1609                 /* can't make a file in a deleted directory. */
1610                 status = -ENOENT;
1611                 goto bail;
1612         }
1613
1614         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1615                                            dentry->d_name.len);
1616         if (status)
1617                 goto bail;
1618
1619         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1620                                               dentry->d_name.name,
1621                                               dentry->d_name.len, &de_bh);
1622         if (status < 0) {
1623                 mlog_errno(status);
1624                 goto bail;
1625         }
1626
1627         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1628         if (status < 0) {
1629                 if (status != -ENOSPC)
1630                         mlog_errno(status);
1631                 goto bail;
1632         }
1633
1634         /* don't reserve bitmap space for fast symlinks. */
1635         if (l > ocfs2_fast_symlink_chars(sb)) {
1636                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1637                 if (status < 0) {
1638                         if (status != -ENOSPC)
1639                                 mlog_errno(status);
1640                         goto bail;
1641                 }
1642         }
1643
1644         handle = ocfs2_start_trans(osb, handle, credits);
1645         if (IS_ERR(handle)) {
1646                 status = PTR_ERR(handle);
1647                 handle = NULL;
1648                 mlog_errno(status);
1649                 goto bail;
1650         }
1651
1652         status = ocfs2_mknod_locked(osb, dir, dentry,
1653                                     S_IFLNK | S_IRWXUGO, 0,
1654                                     &new_fe_bh, parent_fe_bh, handle,
1655                                     &inode, inode_ac);
1656         if (status < 0) {
1657                 mlog_errno(status);
1658                 goto bail;
1659         }
1660
1661         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1662         inode->i_rdev = 0;
1663         newsize = l - 1;
1664         if (l > ocfs2_fast_symlink_chars(sb)) {
1665                 inode->i_op = &ocfs2_symlink_inode_operations;
1666                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1667                                                     handle, data_ac, NULL,
1668                                                     NULL);
1669                 if (status < 0) {
1670                         if (status != -ENOSPC && status != -EINTR) {
1671                                 mlog(ML_ERROR,
1672                                      "Failed to extend file to %llu\n",
1673                                      (unsigned long long)newsize);
1674                                 mlog_errno(status);
1675                                 status = -ENOSPC;
1676                         }
1677                         goto bail;
1678                 }
1679                 i_size_write(inode, newsize);
1680                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1681         } else {
1682                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1683                 memcpy((char *) fe->id2.i_symlink, symname, l);
1684                 i_size_write(inode, newsize);
1685                 inode->i_blocks = 0;
1686         }
1687
1688         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1689         if (status < 0) {
1690                 mlog_errno(status);
1691                 goto bail;
1692         }
1693
1694         if (!ocfs2_inode_is_fast_symlink(inode)) {
1695                 status = ocfs2_create_symlink_data(osb, handle, inode,
1696                                                    symname);
1697                 if (status < 0) {
1698                         mlog_errno(status);
1699                         goto bail;
1700                 }
1701         }
1702
1703         status = ocfs2_add_entry(handle, dentry, inode,
1704                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1705                                  de_bh);
1706         if (status < 0) {
1707                 mlog_errno(status);
1708                 goto bail;
1709         }
1710
1711         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1712         if (status) {
1713                 mlog_errno(status);
1714                 goto bail;
1715         }
1716
1717         insert_inode_hash(inode);
1718         dentry->d_op = &ocfs2_dentry_ops;
1719         d_instantiate(dentry, inode);
1720 bail:
1721         if (handle)
1722                 ocfs2_commit_trans(handle);
1723         if (new_fe_bh)
1724                 brelse(new_fe_bh);
1725         if (parent_fe_bh)
1726                 brelse(parent_fe_bh);
1727         if (de_bh)
1728                 brelse(de_bh);
1729         if (inode_ac)
1730                 ocfs2_free_alloc_context(inode_ac);
1731         if (data_ac)
1732                 ocfs2_free_alloc_context(data_ac);
1733         if ((status < 0) && inode)
1734                 iput(inode);
1735
1736         mlog_exit(status);
1737
1738         return status;
1739 }
1740
1741 int ocfs2_check_dir_entry(struct inode * dir,
1742                           struct ocfs2_dir_entry * de,
1743                           struct buffer_head * bh,
1744                           unsigned long offset)
1745 {
1746         const char *error_msg = NULL;
1747         const int rlen = le16_to_cpu(de->rec_len);
1748
1749         if (rlen < OCFS2_DIR_REC_LEN(1))
1750                 error_msg = "rec_len is smaller than minimal";
1751         else if (rlen % 4 != 0)
1752                 error_msg = "rec_len % 4 != 0";
1753         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1754                 error_msg = "rec_len is too small for name_len";
1755         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1756                 error_msg = "directory entry across blocks";
1757
1758         if (error_msg != NULL)
1759                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1760                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1761                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1762                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1763                      de->name_len);
1764         return error_msg == NULL ? 1 : 0;
1765 }
1766
1767 /* we don't always have a dentry for what we want to add, so people
1768  * like orphan dir can call this instead.
1769  *
1770  * If you pass me insert_bh, I'll skip the search of the other dir
1771  * blocks and put the record in there.
1772  */
1773 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1774                              struct inode *dir,
1775                              const char *name, int namelen,
1776                              struct inode *inode, u64 blkno,
1777                              struct buffer_head *parent_fe_bh,
1778                              struct buffer_head *insert_bh)
1779 {
1780         unsigned long offset;
1781         unsigned short rec_len;
1782         struct ocfs2_dir_entry *de, *de1;
1783         struct super_block *sb;
1784         int retval, status;
1785
1786         mlog_entry_void();
1787
1788         sb = dir->i_sb;
1789
1790         if (!namelen)
1791                 return -EINVAL;
1792
1793         rec_len = OCFS2_DIR_REC_LEN(namelen);
1794         offset = 0;
1795         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1796         while (1) {
1797                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1798                 /* These checks should've already been passed by the
1799                  * prepare function, but I guess we can leave them
1800                  * here anyway. */
1801                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1802                         retval = -ENOENT;
1803                         goto bail;
1804                 }
1805                 if (ocfs2_match(namelen, name, de)) {
1806                         retval = -EEXIST;
1807                         goto bail;
1808                 }
1809                 if (((le64_to_cpu(de->inode) == 0) &&
1810                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1811                     (le16_to_cpu(de->rec_len) >=
1812                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1813                         status = ocfs2_journal_access(handle, dir, insert_bh,
1814                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1815                         /* By now the buffer is marked for journaling */
1816                         offset += le16_to_cpu(de->rec_len);
1817                         if (le64_to_cpu(de->inode)) {
1818                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1819                                         OCFS2_DIR_REC_LEN(de->name_len));
1820                                 de1->rec_len =
1821                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1822                                         OCFS2_DIR_REC_LEN(de->name_len));
1823                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1824                                 de = de1;
1825                         }
1826                         de->file_type = OCFS2_FT_UNKNOWN;
1827                         if (blkno) {
1828                                 de->inode = cpu_to_le64(blkno);
1829                                 ocfs2_set_de_type(de, inode->i_mode);
1830                         } else
1831                                 de->inode = 0;
1832                         de->name_len = namelen;
1833                         memcpy(de->name, name, namelen);
1834
1835                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1836                         dir->i_version++;
1837                         status = ocfs2_journal_dirty(handle, insert_bh);
1838                         retval = 0;
1839                         goto bail;
1840                 }
1841                 offset += le16_to_cpu(de->rec_len);
1842                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1843         }
1844
1845         /* when you think about it, the assert above should prevent us
1846          * from ever getting here. */
1847         retval = -ENOSPC;
1848 bail:
1849
1850         mlog_exit(retval);
1851         return retval;
1852 }
1853
1854
1855 /*
1856  * ocfs2_delete_entry deletes a directory entry by merging it with the
1857  * previous entry
1858  */
1859 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1860                               struct inode *dir,
1861                               struct ocfs2_dir_entry *de_del,
1862                               struct buffer_head *bh)
1863 {
1864         struct ocfs2_dir_entry *de, *pde;
1865         int i, status = -ENOENT;
1866
1867         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1868
1869         i = 0;
1870         pde = NULL;
1871         de = (struct ocfs2_dir_entry *) bh->b_data;
1872         while (i < bh->b_size) {
1873                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1874                         status = -EIO;
1875                         mlog_errno(status);
1876                         goto bail;
1877                 }
1878                 if (de == de_del)  {
1879                         status = ocfs2_journal_access(handle, dir, bh,
1880                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1881                         if (status < 0) {
1882                                 status = -EIO;
1883                                 mlog_errno(status);
1884                                 goto bail;
1885                         }
1886                         if (pde)
1887                                 pde->rec_len =
1888                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1889                                                     le16_to_cpu(de->rec_len));
1890                         else
1891                                 de->inode = 0;
1892                         dir->i_version++;
1893                         status = ocfs2_journal_dirty(handle, bh);
1894                         goto bail;
1895                 }
1896                 i += le16_to_cpu(de->rec_len);
1897                 pde = de;
1898                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1899         }
1900 bail:
1901         mlog_exit(status);
1902         return status;
1903 }
1904
1905 /*
1906  * Returns 0 if not found, -1 on failure, and 1 on success
1907  */
1908 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1909                                         struct inode *dir,
1910                                         const char *name, int namelen,
1911                                         unsigned long offset,
1912                                         struct ocfs2_dir_entry **res_dir)
1913 {
1914         struct ocfs2_dir_entry *de;
1915         char *dlimit, *de_buf;
1916         int de_len;
1917         int ret = 0;
1918
1919         mlog_entry_void();
1920
1921         de_buf = bh->b_data;
1922         dlimit = de_buf + dir->i_sb->s_blocksize;
1923
1924         while (de_buf < dlimit) {
1925                 /* this code is executed quadratically often */
1926                 /* do minimal checking `by hand' */
1927
1928                 de = (struct ocfs2_dir_entry *) de_buf;
1929
1930                 if (de_buf + namelen <= dlimit &&
1931                     ocfs2_match(namelen, name, de)) {
1932                         /* found a match - just to be sure, do a full check */
1933                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1934                                 ret = -1;
1935                                 goto bail;
1936                         }
1937                         *res_dir = de;
1938                         ret = 1;
1939                         goto bail;
1940                 }
1941
1942                 /* prevent looping on a bad block */
1943                 de_len = le16_to_cpu(de->rec_len);
1944                 if (de_len <= 0) {
1945                         ret = -1;
1946                         goto bail;
1947                 }
1948
1949                 de_buf += de_len;
1950                 offset += de_len;
1951         }
1952
1953 bail:
1954         mlog_exit(ret);
1955         return ret;
1956 }
1957
1958 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1959                                      struct inode *dir,
1960                                      struct ocfs2_dir_entry **res_dir)
1961 {
1962         struct super_block *sb;
1963         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1964         struct buffer_head *bh, *ret = NULL;
1965         unsigned long start, block, b;
1966         int ra_max = 0;         /* Number of bh's in the readahead
1967                                    buffer, bh_use[] */
1968         int ra_ptr = 0;         /* Current index into readahead
1969                                    buffer */
1970         int num = 0;
1971         int nblocks, i, err;
1972
1973         mlog_entry_void();
1974
1975         *res_dir = NULL;
1976         sb = dir->i_sb;
1977
1978         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1979         start = OCFS2_I(dir)->ip_dir_start_lookup;
1980         if (start >= nblocks)
1981                 start = 0;
1982         block = start;
1983
1984 restart:
1985         do {
1986                 /*
1987                  * We deal with the read-ahead logic here.
1988                  */
1989                 if (ra_ptr >= ra_max) {
1990                         /* Refill the readahead buffer */
1991                         ra_ptr = 0;
1992                         b = block;
1993                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1994                                 /*
1995                                  * Terminate if we reach the end of the
1996                                  * directory and must wrap, or if our
1997                                  * search has finished at this block.
1998                                  */
1999                                 if (b >= nblocks || (num && block == start)) {
2000                                         bh_use[ra_max] = NULL;
2001                                         break;
2002                                 }
2003                                 num++;
2004
2005                                 bh = ocfs2_bread(dir, b++, &err, 1);
2006                                 bh_use[ra_max] = bh;
2007                         }
2008                 }
2009                 if ((bh = bh_use[ra_ptr++]) == NULL)
2010                         goto next;
2011                 wait_on_buffer(bh);
2012                 if (!buffer_uptodate(bh)) {
2013                         /* read error, skip block & hope for the best */
2014                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2015                                     "offset %lu\n",
2016                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2017                                     block);
2018                         brelse(bh);
2019                         goto next;
2020                 }
2021                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2022                                           block << sb->s_blocksize_bits,
2023                                           res_dir);
2024                 if (i == 1) {
2025                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2026                         ret = bh;
2027                         goto cleanup_and_exit;
2028                 } else {
2029                         brelse(bh);
2030                         if (i < 0)
2031                                 goto cleanup_and_exit;
2032                 }
2033         next:
2034                 if (++block >= nblocks)
2035                         block = 0;
2036         } while (block != start);
2037
2038         /*
2039          * If the directory has grown while we were searching, then
2040          * search the last part of the directory before giving up.
2041          */
2042         block = nblocks;
2043         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2044         if (block < nblocks) {
2045                 start = 0;
2046                 goto restart;
2047         }
2048
2049 cleanup_and_exit:
2050         /* Clean up the read-ahead blocks */
2051         for (; ra_ptr < ra_max; ra_ptr++)
2052                 brelse(bh_use[ra_ptr]);
2053
2054         mlog_exit_ptr(ret);
2055         return ret;
2056 }
2057
2058 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2059 {
2060         int status, namelen;
2061
2062         mlog_entry_void();
2063
2064         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2065                            (long long)blkno);
2066         if (namelen <= 0) {
2067                 if (namelen)
2068                         status = namelen;
2069                 else
2070                         status = -EINVAL;
2071                 mlog_errno(status);
2072                 goto bail;
2073         }
2074         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2075                 status = -EINVAL;
2076                 mlog_errno(status);
2077                 goto bail;
2078         }
2079
2080         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2081              namelen);
2082
2083         status = 0;
2084 bail:
2085         mlog_exit(status);
2086         return status;
2087 }
2088
2089 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2090                                     struct ocfs2_journal_handle *handle,
2091                                     struct inode *inode,
2092                                     char *name,
2093                                     struct buffer_head **de_bh)
2094 {
2095         struct inode *orphan_dir_inode = NULL;
2096         struct buffer_head *orphan_dir_bh = NULL;
2097         int status = 0;
2098
2099         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2100         if (status < 0) {
2101                 mlog_errno(status);
2102                 goto leave;
2103         }
2104
2105         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2106                                                        ORPHAN_DIR_SYSTEM_INODE,
2107                                                        osb->slot_num);
2108         if (!orphan_dir_inode) {
2109                 status = -ENOENT;
2110                 mlog_errno(status);
2111                 goto leave;
2112         }
2113
2114         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2115         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2116         if (status < 0) {
2117                 mlog_errno(status);
2118                 goto leave;
2119         }
2120
2121         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2122                                               orphan_dir_bh, name,
2123                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2124         if (status < 0) {
2125                 mlog_errno(status);
2126                 goto leave;
2127         }
2128
2129 leave:
2130         if (orphan_dir_inode)
2131                 iput(orphan_dir_inode);
2132
2133         if (orphan_dir_bh)
2134                 brelse(orphan_dir_bh);
2135
2136         mlog_exit(status);
2137         return status;
2138 }
2139
2140 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2141                             struct ocfs2_journal_handle *handle,
2142                             struct inode *inode,
2143                             struct ocfs2_dinode *fe,
2144                             char *name,
2145                             struct buffer_head *de_bh)
2146 {
2147         struct inode *orphan_dir_inode = NULL;
2148         struct buffer_head *orphan_dir_bh = NULL;
2149         int status = 0;
2150         struct ocfs2_dinode *orphan_fe;
2151
2152         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2153
2154         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2155                                                        ORPHAN_DIR_SYSTEM_INODE,
2156                                                        osb->slot_num);
2157         if (!orphan_dir_inode) {
2158                 status = -ENOENT;
2159                 mlog_errno(status);
2160                 goto leave;
2161         }
2162
2163         status = ocfs2_read_block(osb,
2164                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2165                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2166                                   orphan_dir_inode);
2167         if (status < 0) {
2168                 mlog_errno(status);
2169                 goto leave;
2170         }
2171
2172         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2173                                       OCFS2_JOURNAL_ACCESS_WRITE);
2174         if (status < 0) {
2175                 mlog_errno(status);
2176                 goto leave;
2177         }
2178
2179         /* we're a cluster, and nlink can change on disk from
2180          * underneath us... */
2181         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2182         if (S_ISDIR(inode->i_mode))
2183                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2184         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2185
2186         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2187         if (status < 0) {
2188                 mlog_errno(status);
2189                 goto leave;
2190         }
2191
2192         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2193                                    OCFS2_ORPHAN_NAMELEN, inode,
2194                                    OCFS2_I(inode)->ip_blkno,
2195                                    orphan_dir_bh, de_bh);
2196         if (status < 0) {
2197                 mlog_errno(status);
2198                 goto leave;
2199         }
2200
2201         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2202
2203         /* Record which orphan dir our inode now resides
2204          * in. delete_inode will use this to determine which orphan
2205          * dir to lock. */
2206         spin_lock(&OCFS2_I(inode)->ip_lock);
2207         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2208         spin_unlock(&OCFS2_I(inode)->ip_lock);
2209
2210         mlog(0, "Inode %llu orphaned in slot %d\n",
2211              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2212
2213 leave:
2214         if (orphan_dir_inode)
2215                 iput(orphan_dir_inode);
2216
2217         if (orphan_dir_bh)
2218                 brelse(orphan_dir_bh);
2219
2220         mlog_exit(status);
2221         return status;
2222 }
2223
2224 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2225 int ocfs2_orphan_del(struct ocfs2_super *osb,
2226                      struct ocfs2_journal_handle *handle,
2227                      struct inode *orphan_dir_inode,
2228                      struct inode *inode,
2229                      struct buffer_head *orphan_dir_bh)
2230 {
2231         char name[OCFS2_ORPHAN_NAMELEN + 1];
2232         struct ocfs2_dinode *orphan_fe;
2233         int status = 0;
2234         struct buffer_head *target_de_bh = NULL;
2235         struct ocfs2_dir_entry *target_de = NULL;
2236
2237         mlog_entry_void();
2238
2239         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2240         if (status < 0) {
2241                 mlog_errno(status);
2242                 goto leave;
2243         }
2244
2245         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2246              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2247              OCFS2_ORPHAN_NAMELEN);
2248
2249         /* find it's spot in the orphan directory */
2250         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2251                                         orphan_dir_inode, &target_de);
2252         if (!target_de_bh) {
2253                 status = -ENOENT;
2254                 mlog_errno(status);
2255                 goto leave;
2256         }
2257
2258         /* remove it from the orphan directory */
2259         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2260                                     target_de_bh);
2261         if (status < 0) {
2262                 mlog_errno(status);
2263                 goto leave;
2264         }
2265
2266         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2267                                       OCFS2_JOURNAL_ACCESS_WRITE);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         /* do the i_nlink dance! :) */
2274         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2275         if (S_ISDIR(inode->i_mode))
2276                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2277         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2278
2279         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2280         if (status < 0) {
2281                 mlog_errno(status);
2282                 goto leave;
2283         }
2284
2285 leave:
2286         if (target_de_bh)
2287                 brelse(target_de_bh);
2288
2289         mlog_exit(status);
2290         return status;
2291 }
2292
2293 struct inode_operations ocfs2_dir_iops = {
2294         .create         = ocfs2_create,
2295         .lookup         = ocfs2_lookup,
2296         .link           = ocfs2_link,
2297         .unlink         = ocfs2_unlink,
2298         .rmdir          = ocfs2_unlink,
2299         .symlink        = ocfs2_symlink,
2300         .mkdir          = ocfs2_mkdir,
2301         .mknod          = ocfs2_mknod,
2302         .rename         = ocfs2_rename,
2303         .setattr        = ocfs2_setattr,
2304         .getattr        = ocfs2_getattr,
2305 };