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

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "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         handle = ocfs2_alloc_handle(osb);
 337         if (handle == NULL) {
 338                 status = -ENOMEM;
 339                 mlog_errno(status);
 340                 goto leave;
 341         }
 342
 343         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 344         if (status < 0) {
 345                 if (status != -ENOENT)
 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         if (status == -ENOSPC)
 459                 mlog(0, "Disk is full\n");
 460
 461         if (new_fe_bh)
 462                 brelse(new_fe_bh);
 463
 464         if (de_bh)
 465                 brelse(de_bh);
 466
 467         if (parent_fe_bh)
 468                 brelse(parent_fe_bh);
 469
 470         if ((status < 0) && inode)
 471                 iput(inode);
 472
 473         if (inode_ac)
 474                 ocfs2_free_alloc_context(inode_ac);
 475
 476         if (data_ac)
 477                 ocfs2_free_alloc_context(data_ac);
 478
 479         mlog_exit(status);
 480
 481         return status;
 482 }
 483
 484 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 485                               struct inode *dir,
 486                               struct dentry *dentry, int mode,
 487                               dev_t dev,
 488                               struct buffer_head **new_fe_bh,
 489                               struct buffer_head *parent_fe_bh,
 490                               struct ocfs2_journal_handle *handle,
 491                               struct inode **ret_inode,
 492                               struct ocfs2_alloc_context *inode_ac)
 493 {
 494         int status = 0;
 495         struct ocfs2_dinode *fe = NULL;
 496         struct ocfs2_extent_list *fel;
 497         u64 fe_blkno = 0;
 498         u16 suballoc_bit;
 499         struct inode *inode = NULL;
 500
 501         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 502                    (unsigned long)dev, dentry->d_name.len,
 503                    dentry->d_name.name);
 504
 505         *new_fe_bh = NULL;
 506         *ret_inode = NULL;
 507
 508         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 509                                        &fe_blkno);
 510         if (status < 0) {
 511                 mlog_errno(status);
 512                 goto leave;
 513         }
 514
 515         inode = new_inode(dir->i_sb);
 516         if (IS_ERR(inode)) {
 517                 status = PTR_ERR(inode);
 518                 mlog(ML_ERROR, "new_inode failed!\n");
 519                 goto leave;
 520         }
 521
 522         /* populate as many fields early on as possible - many of
 523          * these are used by the support functions here and in
 524          * callers. */
 525         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 526         OCFS2_I(inode)->ip_blkno = fe_blkno;
 527         if (S_ISDIR(mode))
 528                 inode->i_nlink = 2;
 529         else
 530                 inode->i_nlink = 1;
 531         inode->i_mode = mode;
 532         spin_lock(&osb->osb_lock);
 533         inode->i_generation = osb->s_next_generation++;
 534         spin_unlock(&osb->osb_lock);
 535
 536         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 537         if (!*new_fe_bh) {
 538                 status = -EIO;
 539                 mlog_errno(status);
 540                 goto leave;
 541         }
 542         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 543
 544         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 545                                       OCFS2_JOURNAL_ACCESS_CREATE);
 546         if (status < 0) {
 547                 mlog_errno(status);
 548                 goto leave;
 549         }
 550
 551         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 552         memset(fe, 0, osb->sb->s_blocksize);
 553
 554         fe->i_generation = cpu_to_le32(inode->i_generation);
 555         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 556         fe->i_blkno = cpu_to_le64(fe_blkno);
 557         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 558         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 559         fe->i_uid = cpu_to_le32(current->fsuid);
 560         if (dir->i_mode & S_ISGID) {
 561                 fe->i_gid = cpu_to_le32(dir->i_gid);
 562                 if (S_ISDIR(mode))
 563                         mode |= S_ISGID;
 564         } else
 565                 fe->i_gid = cpu_to_le32(current->fsgid);
 566         fe->i_mode = cpu_to_le16(mode);
 567         if (S_ISCHR(mode) || S_ISBLK(mode))
 568                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 569
 570         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 571
 572         fe->i_last_eb_blk = 0;
 573         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 574         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 575         fe->i_atime = fe->i_ctime = fe->i_mtime =
 576                 cpu_to_le64(CURRENT_TIME.tv_sec);
 577         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 578                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 579         fe->i_dtime = 0;
 580
 581         fel = &fe->id2.i_list;
 582         fel->l_tree_depth = 0;
 583         fel->l_next_free_rec = 0;
 584         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 585
 586         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 587         if (status < 0) {
 588                 mlog_errno(status);
 589                 goto leave;
 590         }
 591
 592         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 593                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 594                      "i_blkno=%llu, i_ino=%lu\n",
 595                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 596                      (unsigned long long)fe->i_blkno, inode->i_ino);
 597                 BUG();
 598         }
 599
 600         ocfs2_inode_set_new(osb, inode);
 601         status = ocfs2_create_new_inode_locks(inode);
 602         if (status < 0)
 603                 mlog_errno(status);
 604
 605         status = 0; /* error in ocfs2_create_new_inode_locks is not
 606                      * critical */
 607
 608         *ret_inode = inode;
 609 leave:
 610         if (status < 0) {
 611                 if (*new_fe_bh) {
 612                         brelse(*new_fe_bh);
 613                         *new_fe_bh = NULL;
 614                 }
 615                 if (inode)
 616                         iput(inode);
 617         }
 618
 619         mlog_exit(status);
 620         return status;
 621 }
 622
 623 static int ocfs2_mkdir(struct inode *dir,
 624                        struct dentry *dentry,
 625                        int mode)
 626 {
 627         int ret;
 628
 629         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 630                    dentry->d_name.len, dentry->d_name.name);
 631         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 632         mlog_exit(ret);
 633
 634         return ret;
 635 }
 636
 637 static int ocfs2_create(struct inode *dir,
 638                         struct dentry *dentry,
 639                         int mode,
 640                         struct nameidata *nd)
 641 {
 642         int ret;
 643
 644         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 645                    dentry->d_name.len, dentry->d_name.name);
 646         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 647         mlog_exit(ret);
 648
 649         return ret;
 650 }
 651
 652 static int ocfs2_link(struct dentry *old_dentry,
 653                       struct inode *dir,
 654                       struct dentry *dentry)
 655 {
 656         struct ocfs2_journal_handle *handle = NULL;
 657         struct inode *inode = old_dentry->d_inode;
 658         int err;
 659         struct buffer_head *fe_bh = NULL;
 660         struct buffer_head *parent_fe_bh = NULL;
 661         struct buffer_head *de_bh = NULL;
 662         struct ocfs2_dinode *fe = NULL;
 663         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 664
 665         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 666                    old_dentry->d_name.len, old_dentry->d_name.name,
 667                    dentry->d_name.len, dentry->d_name.name);
 668
 669         if (S_ISDIR(inode->i_mode)) {
 670                 err = -EPERM;
 671                 goto bail;
 672         }
 673
 674         handle = ocfs2_alloc_handle(osb);
 675         if (handle == NULL) {
 676                 err = -ENOMEM;
 677                 goto bail;
 678         }
 679
 680         err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 681         if (err < 0) {
 682                 if (err != -ENOENT)
 683                         mlog_errno(err);
 684                 goto bail;
 685         }
 686
 687         if (!dir->i_nlink) {
 688                 err = -ENOENT;
 689                 goto bail;
 690         }
 691
 692         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 693                                         dentry->d_name.len);
 694         if (err)
 695                 goto bail;
 696
 697         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 698                                            dentry->d_name.name,
 699                                            dentry->d_name.len, &de_bh);
 700         if (err < 0) {
 701                 mlog_errno(err);
 702                 goto bail;
 703         }
 704
 705         err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 706         if (err < 0) {
 707                 if (err != -ENOENT)
 708                         mlog_errno(err);
 709                 goto bail;
 710         }
 711
 712         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 713         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 714                 err = -EMLINK;
 715                 goto bail;
 716         }
 717
 718         handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
 719         if (IS_ERR(handle)) {
 720                 err = PTR_ERR(handle);
 721                 handle = NULL;
 722                 mlog_errno(err);
 723                 goto bail;
 724         }
 725
 726         err = ocfs2_journal_access(handle, inode, fe_bh,
 727                                    OCFS2_JOURNAL_ACCESS_WRITE);
 728         if (err < 0) {
 729                 mlog_errno(err);
 730                 goto bail;
 731         }
 732
 733         inc_nlink(inode);
 734         inode->i_ctime = CURRENT_TIME;
 735         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 736         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 737         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 738
 739         err = ocfs2_journal_dirty(handle, fe_bh);
 740         if (err < 0) {
 741                 le16_add_cpu(&fe->i_links_count, -1);
 742                 drop_nlink(inode);
 743                 mlog_errno(err);
 744                 goto bail;
 745         }
 746
 747         err = ocfs2_add_entry(handle, dentry, inode,
 748                               OCFS2_I(inode)->ip_blkno,
 749                               parent_fe_bh, de_bh);
 750         if (err) {
 751                 le16_add_cpu(&fe->i_links_count, -1);
 752                 drop_nlink(inode);
 753                 mlog_errno(err);
 754                 goto bail;
 755         }
 756
 757         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 758         if (err) {
 759                 mlog_errno(err);
 760                 goto bail;
 761         }
 762
 763         atomic_inc(&inode->i_count);
 764         dentry->d_op = &ocfs2_dentry_ops;
 765         d_instantiate(dentry, inode);
 766 bail:
 767         if (handle)
 768                 ocfs2_commit_trans(handle);
 769         if (de_bh)
 770                 brelse(de_bh);
 771         if (fe_bh)
 772                 brelse(fe_bh);
 773         if (parent_fe_bh)
 774                 brelse(parent_fe_bh);
 775
 776         mlog_exit(err);
 777
 778         return err;
 779 }
 780
 781 /*
 782  * Takes and drops an exclusive lock on the given dentry. This will
 783  * force other nodes to drop it.
 784  */
 785 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 786 {
 787         int ret;
 788
 789         ret = ocfs2_dentry_lock(dentry, 1);
 790         if (ret)
 791                 mlog_errno(ret);
 792         else
 793                 ocfs2_dentry_unlock(dentry, 1);
 794
 795         return ret;
 796 }
 797
 798 static inline int inode_is_unlinkable(struct inode *inode)
 799 {
 800         if (S_ISDIR(inode->i_mode)) {
 801                 if (inode->i_nlink == 2)
 802                         return 1;
 803                 return 0;
 804         }
 805
 806         if (inode->i_nlink == 1)
 807                 return 1;
 808         return 0;
 809 }
 810
 811 static int ocfs2_unlink(struct inode *dir,
 812                         struct dentry *dentry)
 813 {
 814         int status;
 815         struct inode *inode = dentry->d_inode;
 816         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 817         u64 blkno;
 818         struct ocfs2_dinode *fe = NULL;
 819         struct buffer_head *fe_bh = NULL;
 820         struct buffer_head *parent_node_bh = NULL;
 821         struct ocfs2_journal_handle *handle = NULL;
 822         struct ocfs2_dir_entry *dirent = NULL;
 823         struct buffer_head *dirent_bh = NULL;
 824         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 825         struct buffer_head *orphan_entry_bh = NULL;
 826
 827         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 828                    dentry->d_name.len, dentry->d_name.name);
 829
 830         BUG_ON(dentry->d_parent->d_inode != dir);
 831
 832         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 833
 834         if (inode == osb->root_inode) {
 835                 mlog(0, "Cannot delete the root directory\n");
 836                 status = -EPERM;
 837                 goto leave;
 838         }
 839
 840         handle = ocfs2_alloc_handle(osb);
 841         if (handle == NULL) {
 842                 status = -ENOMEM;
 843                 mlog_errno(status);
 844                 goto leave;
 845         }
 846
 847         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
 848         if (status < 0) {
 849                 if (status != -ENOENT)
 850                         mlog_errno(status);
 851                 goto leave;
 852         }
 853
 854         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 855                                           dentry->d_name.len, &blkno,
 856                                           dir, &dirent_bh, &dirent);
 857         if (status < 0) {
 858                 if (status != -ENOENT)
 859                         mlog_errno(status);
 860                 goto leave;
 861         }
 862
 863         if (OCFS2_I(inode)->ip_blkno != blkno) {
 864                 status = -ENOENT;
 865
 866                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 867                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 868                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 869                 goto leave;
 870         }
 871
 872         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 873         if (status < 0) {
 874                 if (status != -ENOENT)
 875                         mlog_errno(status);
 876                 goto leave;
 877         }
 878
 879         if (S_ISDIR(inode->i_mode)) {
 880                 if (!ocfs2_empty_dir(inode)) {
 881                         status = -ENOTEMPTY;
 882                         goto leave;
 883                 } else if (inode->i_nlink != 2) {
 884                         status = -ENOTEMPTY;
 885                         goto leave;
 886                 }
 887         }
 888
 889         status = ocfs2_remote_dentry_delete(dentry);
 890         if (status < 0) {
 891                 /* This vote should succeed under all normal
 892                  * circumstances. */
 893                 mlog_errno(status);
 894                 goto leave;
 895         }
 896
 897         if (inode_is_unlinkable(inode)) {
 898                 status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 899                                                   orphan_name,
 900                                                   &orphan_entry_bh);
 901                 if (status < 0) {
 902                         mlog_errno(status);
 903                         goto leave;
 904                 }
 905         }
 906
 907         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
 908         if (IS_ERR(handle)) {
 909                 status = PTR_ERR(handle);
 910                 handle = NULL;
 911                 mlog_errno(status);
 912                 goto leave;
 913         }
 914
 915         status = ocfs2_journal_access(handle, inode, fe_bh,
 916                                       OCFS2_JOURNAL_ACCESS_WRITE);
 917         if (status < 0) {
 918                 mlog_errno(status);
 919                 goto leave;
 920         }
 921
 922         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 923
 924         if (inode_is_unlinkable(inode)) {
 925                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 926                                           orphan_entry_bh);
 927                 if (status < 0) {
 928                         mlog_errno(status);
 929                         goto leave;
 930                 }
 931         }
 932
 933         /* delete the name from the parent dir */
 934         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 935         if (status < 0) {
 936                 mlog_errno(status);
 937                 goto leave;
 938         }
 939
 940         if (S_ISDIR(inode->i_mode))
 941                 drop_nlink(inode);
 942         drop_nlink(inode);
 943         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 944
 945         status = ocfs2_journal_dirty(handle, fe_bh);
 946         if (status < 0) {
 947                 mlog_errno(status);
 948                 goto leave;
 949         }
 950
 951         if (S_ISDIR(inode->i_mode)) {
 952                 drop_nlink(dir);
 953                 status = ocfs2_mark_inode_dirty(handle, dir,
 954                                                 parent_node_bh);
 955                 if (status < 0) {
 956                         mlog_errno(status);
 957                         inc_nlink(dir);
 958                 }
 959         }
 960
 961 leave:
 962         if (handle)
 963                 ocfs2_commit_trans(handle);
 964
 965         if (fe_bh)
 966                 brelse(fe_bh);
 967
 968         if (dirent_bh)
 969                 brelse(dirent_bh);
 970
 971         if (parent_node_bh)
 972                 brelse(parent_node_bh);
 973
 974         if (orphan_entry_bh)
 975                 brelse(orphan_entry_bh);
 976
 977         mlog_exit(status);
 978
 979         return status;
 980 }
 981
 982 /*
 983  * The only place this should be used is rename!
 984  * if they have the same id, then the 1st one is the only one locked.
 985  */
 986 static int ocfs2_double_lock(struct ocfs2_super *osb,
 987                              struct ocfs2_journal_handle *handle,
 988                              struct buffer_head **bh1,
 989                              struct inode *inode1,
 990                              struct buffer_head **bh2,
 991                              struct inode *inode2)
 992 {
 993         int status;
 994         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 995         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 996         struct buffer_head **tmpbh;
 997         struct inode *tmpinode;
 998
 999         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1000                    (unsigned long long)oi1->ip_blkno,
1001                    (unsigned long long)oi2->ip_blkno);
1002
1003         BUG_ON(!handle);
1004
1005         if (*bh1)
1006                 *bh1 = NULL;
1007         if (*bh2)
1008                 *bh2 = NULL;
1009
1010         /* we always want to lock the one with the lower lockid first. */
1011         if (oi1->ip_blkno != oi2->ip_blkno) {
1012                 if (oi1->ip_blkno < oi2->ip_blkno) {
1013                         /* switch id1 and id2 around */
1014                         mlog(0, "switching them around...\n");
1015                         tmpbh = bh2;
1016                         bh2 = bh1;
1017                         bh1 = tmpbh;
1018
1019                         tmpinode = inode2;
1020                         inode2 = inode1;
1021                         inode1 = tmpinode;
1022                 }
1023                 /* lock id2 */
1024                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1025                 if (status < 0) {
1026                         if (status != -ENOENT)
1027                                 mlog_errno(status);
1028                         goto bail;
1029                 }
1030         }
1031         /* lock id1 */
1032         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1033         if (status < 0) {
1034                 if (status != -ENOENT)
1035                         mlog_errno(status);
1036                 goto bail;
1037         }
1038 bail:
1039         mlog_exit(status);
1040         return status;
1041 }
1042
1043 #define PARENT_INO(buffer) \
1044         ((struct ocfs2_dir_entry *) \
1045          ((char *)buffer + \
1046           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1047
1048 static int ocfs2_rename(struct inode *old_dir,
1049                         struct dentry *old_dentry,
1050                         struct inode *new_dir,
1051                         struct dentry *new_dentry)
1052 {
1053         int status = 0, rename_lock = 0;
1054         struct inode *old_inode = old_dentry->d_inode;
1055         struct inode *new_inode = new_dentry->d_inode;
1056         struct ocfs2_dinode *newfe = NULL;
1057         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1058         struct buffer_head *orphan_entry_bh = NULL;
1059         struct buffer_head *newfe_bh = NULL;
1060         struct buffer_head *insert_entry_bh = NULL;
1061         struct ocfs2_super *osb = NULL;
1062         u64 newfe_blkno;
1063         struct ocfs2_journal_handle *handle = NULL;
1064         struct buffer_head *old_dir_bh = NULL;
1065         struct buffer_head *new_dir_bh = NULL;
1066         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1067                                                                // and new_dentry
1068         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1069         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1070                                                     // this is the 1st dirent bh
1071         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1072
1073         /* At some point it might be nice to break this function up a
1074          * bit. */
1075
1076         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1077                    old_dir, old_dentry, new_dir, new_dentry,
1078                    old_dentry->d_name.len, old_dentry->d_name.name,
1079                    new_dentry->d_name.len, new_dentry->d_name.name);
1080
1081         osb = OCFS2_SB(old_dir->i_sb);
1082
1083         if (new_inode) {
1084                 if (!igrab(new_inode))
1085                         BUG();
1086         }
1087
1088         /* Assume a directory heirarchy thusly:
1089          * a/b/c
1090          * a/d
1091          * a,b,c, and d are all directories.
1092          *
1093          * from cwd of 'a' on both nodes:
1094          * node1: mv b/c d
1095          * node2: mv d   b/c
1096          *
1097          * And that's why, just like the VFS, we need a file system
1098          * rename lock. */
1099         if (old_dentry != new_dentry) {
1100                 status = ocfs2_rename_lock(osb);
1101                 if (status < 0) {
1102                         mlog_errno(status);
1103                         goto bail;
1104                 }
1105                 rename_lock = 1;
1106         }
1107
1108         handle = ocfs2_alloc_handle(osb);
1109         if (handle == NULL) {
1110                 status = -ENOMEM;
1111                 mlog_errno(status);
1112                 goto bail;
1113         }
1114
1115         /* if old and new are the same, this'll just do one lock. */
1116         status = ocfs2_double_lock(osb, handle,
1117                                   &old_dir_bh, old_dir,
1118                                   &new_dir_bh, new_dir);
1119         if (status < 0) {
1120                 mlog_errno(status);
1121                 goto bail;
1122         }
1123
1124         /* make sure both dirs have bhs
1125          * get an extra ref on old_dir_bh if old==new */
1126         if (!new_dir_bh) {
1127                 if (old_dir_bh) {
1128                         new_dir_bh = old_dir_bh;
1129                         get_bh(new_dir_bh);
1130                 } else {
1131                         mlog(ML_ERROR, "no old_dir_bh!\n");
1132                         status = -EIO;
1133                         goto bail;
1134                 }
1135         }
1136
1137         /*
1138          * Though we don't require an inode meta data update if
1139          * old_inode is not a directory, we lock anyway here to ensure
1140          * the vote thread on other nodes won't have to concurrently
1141          * downconvert the inode and the dentry locks.
1142          */
1143         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1144         if (status < 0) {
1145                 if (status != -ENOENT)
1146                         mlog_errno(status);
1147                 goto bail;
1148         }
1149
1150         status = ocfs2_remote_dentry_delete(old_dentry);
1151         if (status < 0) {
1152                 mlog_errno(status);
1153                 goto bail;
1154         }
1155
1156         if (S_ISDIR(old_inode->i_mode)) {
1157                 status = -EIO;
1158                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1159                 if (!old_inode_de_bh)
1160                         goto bail;
1161
1162                 status = -EIO;
1163                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1164                     OCFS2_I(old_dir)->ip_blkno)
1165                         goto bail;
1166                 status = -EMLINK;
1167                 if (!new_inode && new_dir!=old_dir &&
1168                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1169                         goto bail;
1170         }
1171
1172         status = -ENOENT;
1173         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1174                                      old_dentry->d_name.len,
1175                                      old_dir, &old_de);
1176         if (!old_de_bh)
1177                 goto bail;
1178
1179         /*
1180          *  Check for inode number is _not_ due to possible IO errors.
1181          *  We might rmdir the source, keep it as pwd of some process
1182          *  and merrily kill the link to whatever was created under the
1183          *  same name. Goodbye sticky bit ;-<
1184          */
1185         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1186                 goto bail;
1187
1188         /* check if the target already exists (in which case we need
1189          * to delete it */
1190         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1191                                           new_dentry->d_name.len,
1192                                           &newfe_blkno, new_dir, &new_de_bh,
1193                                           &new_de);
1194         /* The only error we allow here is -ENOENT because the new
1195          * file not existing is perfectly valid. */
1196         if ((status < 0) && (status != -ENOENT)) {
1197                 /* If we cannot find the file specified we should just */
1198                 /* return the error... */
1199                 mlog_errno(status);
1200                 goto bail;
1201         }
1202
1203         if (!new_de && new_inode)
1204                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1205                      "directory!", new_inode->i_ino);
1206
1207         /* In case we need to overwrite an existing file, we blow it
1208          * away first */
1209         if (new_de) {
1210                 /* VFS didn't think there existed an inode here, but
1211                  * someone else in the cluster must have raced our
1212                  * rename to create one. Today we error cleanly, in
1213                  * the future we should consider calling iget to build
1214                  * a new struct inode for this entry. */
1215                 if (!new_inode) {
1216                         status = -EACCES;
1217
1218                         mlog(0, "We found an inode for name %.*s but VFS "
1219                              "didn't give us one.\n", new_dentry->d_name.len,
1220                              new_dentry->d_name.name);
1221                         goto bail;
1222                 }
1223
1224                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1225                         status = -EACCES;
1226
1227                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1228                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1229                              (unsigned long long)newfe_blkno,
1230                              OCFS2_I(new_inode)->ip_flags);
1231                         goto bail;
1232                 }
1233
1234                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1235                 if (status < 0) {
1236                         if (status != -ENOENT)
1237                                 mlog_errno(status);
1238                         goto bail;
1239                 }
1240
1241                 status = ocfs2_remote_dentry_delete(new_dentry);
1242                 if (status < 0) {
1243                         mlog_errno(status);
1244                         goto bail;
1245                 }
1246
1247                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1248
1249                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1250                      "newfebh=%p bhblocknr=%llu\n", new_de,
1251                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1252                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1253
1254                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1255                         status = ocfs2_prepare_orphan_dir(osb, handle,
1256                                                           new_inode,
1257                                                           orphan_name,
1258                                                           &orphan_entry_bh);
1259                         if (status < 0) {
1260                                 mlog_errno(status);
1261                                 goto bail;
1262                         }
1263                 }
1264         } else {
1265                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1266
1267                 status = ocfs2_check_dir_for_entry(new_dir,
1268                                                    new_dentry->d_name.name,
1269                                                    new_dentry->d_name.len);
1270                 if (status)
1271                         goto bail;
1272
1273                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1274                                                       new_dentry->d_name.name,
1275                                                       new_dentry->d_name.len,
1276                                                       &insert_entry_bh);
1277                 if (status < 0) {
1278                         mlog_errno(status);
1279                         goto bail;
1280                 }
1281         }
1282
1283         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1284         if (IS_ERR(handle)) {
1285                 status = PTR_ERR(handle);
1286                 handle = NULL;
1287                 mlog_errno(status);
1288                 goto bail;
1289         }
1290
1291         if (new_de) {
1292                 if (S_ISDIR(new_inode->i_mode)) {
1293                         if (!ocfs2_empty_dir(new_inode) ||
1294                             new_inode->i_nlink != 2) {
1295                                 status = -ENOTEMPTY;
1296                                 goto bail;
1297                         }
1298                 }
1299                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1300                                               OCFS2_JOURNAL_ACCESS_WRITE);
1301                 if (status < 0) {
1302                         mlog_errno(status);
1303                         goto bail;
1304                 }
1305
1306                 if (S_ISDIR(new_inode->i_mode) ||
1307                     (newfe->i_links_count == cpu_to_le16(1))){
1308                         status = ocfs2_orphan_add(osb, handle, new_inode,
1309                                                   newfe, orphan_name,
1310                                                   orphan_entry_bh);
1311                         if (status < 0) {
1312                                 mlog_errno(status);
1313                                 goto bail;
1314                         }
1315                 }
1316
1317                 /* change the dirent to point to the correct inode */
1318                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1319                                               OCFS2_JOURNAL_ACCESS_WRITE);
1320                 if (status < 0) {
1321                         mlog_errno(status);
1322                         goto bail;
1323                 }
1324                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1325                 new_de->file_type = old_de->file_type;
1326                 new_dir->i_version++;
1327                 status = ocfs2_journal_dirty(handle, new_de_bh);
1328                 if (status < 0) {
1329                         mlog_errno(status);
1330                         goto bail;
1331                 }
1332
1333                 if (S_ISDIR(new_inode->i_mode))
1334                         newfe->i_links_count = 0;
1335                 else
1336                         le16_add_cpu(&newfe->i_links_count, -1);
1337
1338                 status = ocfs2_journal_dirty(handle, newfe_bh);
1339                 if (status < 0) {
1340                         mlog_errno(status);
1341                         goto bail;
1342                 }
1343         } else {
1344                 /* if the name was not found in new_dir, add it now */
1345                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1346                                          OCFS2_I(old_inode)->ip_blkno,
1347                                          new_dir_bh, insert_entry_bh);
1348         }
1349
1350         old_inode->i_ctime = CURRENT_TIME;
1351         mark_inode_dirty(old_inode);
1352
1353         /* now that the name has been added to new_dir, remove the old name */
1354         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1355         if (status < 0) {
1356                 mlog_errno(status);
1357                 goto bail;
1358         }
1359
1360         if (new_inode) {
1361                 new_inode->i_nlink--;
1362                 new_inode->i_ctime = CURRENT_TIME;
1363         }
1364         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1365         if (old_inode_de_bh) {
1366                 status = ocfs2_journal_access(handle, old_inode,
1367                                              old_inode_de_bh,
1368                                              OCFS2_JOURNAL_ACCESS_WRITE);
1369                 PARENT_INO(old_inode_de_bh->b_data) =
1370                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1371                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1372                 old_dir->i_nlink--;
1373                 if (new_inode) {
1374                         new_inode->i_nlink--;
1375                 } else {
1376                         inc_nlink(new_dir);
1377                         mark_inode_dirty(new_dir);
1378                 }
1379         }
1380         mark_inode_dirty(old_dir);
1381         if (new_inode)
1382                 mark_inode_dirty(new_inode);
1383
1384         if (old_dir != new_dir)
1385                 if (new_dir_nlink != new_dir->i_nlink) {
1386                         if (!new_dir_bh) {
1387                                 mlog(ML_ERROR, "need to change nlink for new "
1388                                      "dir %llu from %d to %d but bh is NULL\n",
1389                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1390                                      (int)new_dir_nlink, new_dir->i_nlink);
1391                         } else {
1392                                 struct ocfs2_dinode *fe;
1393                                 status = ocfs2_journal_access(handle,
1394                                                               new_dir,
1395                                                               new_dir_bh,
1396                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1397                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1398                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1399                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1400                         }
1401                 }
1402
1403         if (old_dir_nlink != old_dir->i_nlink) {
1404                 if (!old_dir_bh) {
1405                         mlog(ML_ERROR, "need to change nlink for old dir "
1406                              "%llu from %d to %d but bh is NULL!\n",
1407                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1408                              (int)old_dir_nlink, old_dir->i_nlink);
1409                 } else {
1410                         struct ocfs2_dinode *fe;
1411                         status = ocfs2_journal_access(handle, old_dir,
1412                                                       old_dir_bh,
1413                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1414                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1415                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1416                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1417                 }
1418         }
1419
1420         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1421         status = 0;
1422 bail:
1423         if (rename_lock)
1424                 ocfs2_rename_unlock(osb);
1425
1426         if (handle)
1427                 ocfs2_commit_trans(handle);
1428
1429         if (new_inode)
1430                 sync_mapping_buffers(old_inode->i_mapping);
1431
1432         if (new_inode)
1433                 iput(new_inode);
1434         if (newfe_bh)
1435                 brelse(newfe_bh);
1436         if (old_dir_bh)
1437                 brelse(old_dir_bh);
1438         if (new_dir_bh)
1439                 brelse(new_dir_bh);
1440         if (new_de_bh)
1441                 brelse(new_de_bh);
1442         if (old_de_bh)
1443                 brelse(old_de_bh);
1444         if (old_inode_de_bh)
1445                 brelse(old_inode_de_bh);
1446         if (orphan_entry_bh)
1447                 brelse(orphan_entry_bh);
1448         if (insert_entry_bh)
1449                 brelse(insert_entry_bh);
1450
1451         mlog_exit(status);
1452
1453         return status;
1454 }
1455
1456 /*
1457  * we expect i_size = strlen(symname). Copy symname into the file
1458  * data, including the null terminator.
1459  */
1460 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1461                                      struct ocfs2_journal_handle *handle,
1462                                      struct inode *inode,
1463                                      const char *symname)
1464 {
1465         struct buffer_head **bhs = NULL;
1466         const char *c;
1467         struct super_block *sb = osb->sb;
1468         u64 p_blkno;
1469         int p_blocks;
1470         int virtual, blocks, status, i, bytes_left;
1471
1472         bytes_left = i_size_read(inode) + 1;
1473         /* we can't trust i_blocks because we're actually going to
1474          * write i_size + 1 bytes. */
1475         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1476
1477         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1478                         (unsigned long long)inode->i_blocks,
1479                         i_size_read(inode), blocks);
1480
1481         /* Sanity check -- make sure we're going to fit. */
1482         if (bytes_left >
1483             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1484                 status = -EIO;
1485                 mlog_errno(status);
1486                 goto bail;
1487         }
1488
1489         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1490         if (!bhs) {
1491                 status = -ENOMEM;
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495
1496         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1497                                              &p_blocks);
1498         if (status < 0) {
1499                 mlog_errno(status);
1500                 goto bail;
1501         }
1502
1503         /* links can never be larger than one cluster so we know this
1504          * is all going to be contiguous, but do a sanity check
1505          * anyway. */
1506         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1507                 status = -EIO;
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         virtual = 0;
1513         while(bytes_left > 0) {
1514                 c = &symname[virtual * sb->s_blocksize];
1515
1516                 bhs[virtual] = sb_getblk(sb, p_blkno);
1517                 if (!bhs[virtual]) {
1518                         status = -ENOMEM;
1519                         mlog_errno(status);
1520                         goto bail;
1521                 }
1522                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1523
1524                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1525                                               OCFS2_JOURNAL_ACCESS_CREATE);
1526                 if (status < 0) {
1527                         mlog_errno(status);
1528                         goto bail;
1529                 }
1530
1531                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1532
1533                 memcpy(bhs[virtual]->b_data, c,
1534                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1535                        bytes_left);
1536
1537                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1538                 if (status < 0) {
1539                         mlog_errno(status);
1540                         goto bail;
1541                 }
1542
1543                 virtual++;
1544                 p_blkno++;
1545                 bytes_left -= sb->s_blocksize;
1546         }
1547
1548         status = 0;
1549 bail:
1550
1551         if (bhs) {
1552                 for(i = 0; i < blocks; i++)
1553                         if (bhs[i])
1554                                 brelse(bhs[i]);
1555                 kfree(bhs);
1556         }
1557
1558         mlog_exit(status);
1559         return status;
1560 }
1561
1562 static int ocfs2_symlink(struct inode *dir,
1563                          struct dentry *dentry,
1564                          const char *symname)
1565 {
1566         int status, l, credits;
1567         u64 newsize;
1568         struct ocfs2_super *osb = NULL;
1569         struct inode *inode = NULL;
1570         struct super_block *sb;
1571         struct buffer_head *new_fe_bh = NULL;
1572         struct buffer_head *de_bh = NULL;
1573         struct buffer_head *parent_fe_bh = NULL;
1574         struct ocfs2_dinode *fe = NULL;
1575         struct ocfs2_dinode *dirfe;
1576         struct ocfs2_journal_handle *handle = NULL;
1577         struct ocfs2_alloc_context *inode_ac = NULL;
1578         struct ocfs2_alloc_context *data_ac = NULL;
1579
1580         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1581                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1582
1583         sb = dir->i_sb;
1584         osb = OCFS2_SB(sb);
1585
1586         l = strlen(symname) + 1;
1587
1588         credits = ocfs2_calc_symlink_credits(sb);
1589
1590         handle = ocfs2_alloc_handle(osb);
1591         if (handle == NULL) {
1592                 status = -ENOMEM;
1593                 mlog_errno(status);
1594                 goto bail;
1595         }
1596
1597         /* lock the parent directory */
1598         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1599         if (status < 0) {
1600                 if (status != -ENOENT)
1601                         mlog_errno(status);
1602                 goto bail;
1603         }
1604
1605         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1606         if (!dirfe->i_links_count) {
1607                 /* can't make a file in a deleted directory. */
1608                 status = -ENOENT;
1609                 goto bail;
1610         }
1611
1612         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1613                                            dentry->d_name.len);
1614         if (status)
1615                 goto bail;
1616
1617         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1618                                               dentry->d_name.name,
1619                                               dentry->d_name.len, &de_bh);
1620         if (status < 0) {
1621                 mlog_errno(status);
1622                 goto bail;
1623         }
1624
1625         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1626         if (status < 0) {
1627                 if (status != -ENOSPC)
1628                         mlog_errno(status);
1629                 goto bail;
1630         }
1631
1632         /* don't reserve bitmap space for fast symlinks. */
1633         if (l > ocfs2_fast_symlink_chars(sb)) {
1634                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1635                 if (status < 0) {
1636                         if (status != -ENOSPC)
1637                                 mlog_errno(status);
1638                         goto bail;
1639                 }
1640         }
1641
1642         handle = ocfs2_start_trans(osb, handle, credits);
1643         if (IS_ERR(handle)) {
1644                 status = PTR_ERR(handle);
1645                 handle = NULL;
1646                 mlog_errno(status);
1647                 goto bail;
1648         }
1649
1650         status = ocfs2_mknod_locked(osb, dir, dentry,
1651                                     S_IFLNK | S_IRWXUGO, 0,
1652                                     &new_fe_bh, parent_fe_bh, handle,
1653                                     &inode, inode_ac);
1654         if (status < 0) {
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1660         inode->i_rdev = 0;
1661         newsize = l - 1;
1662         if (l > ocfs2_fast_symlink_chars(sb)) {
1663                 inode->i_op = &ocfs2_symlink_inode_operations;
1664                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1665                                                     handle, data_ac, NULL,
1666                                                     NULL);
1667                 if (status < 0) {
1668                         if (status != -ENOSPC && status != -EINTR) {
1669                                 mlog(ML_ERROR,
1670                                      "Failed to extend file to %llu\n",
1671                                      (unsigned long long)newsize);
1672                                 mlog_errno(status);
1673                                 status = -ENOSPC;
1674                         }
1675                         goto bail;
1676                 }
1677                 i_size_write(inode, newsize);
1678                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1679         } else {
1680                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1681                 memcpy((char *) fe->id2.i_symlink, symname, l);
1682                 i_size_write(inode, newsize);
1683                 inode->i_blocks = 0;
1684         }
1685
1686         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1687         if (status < 0) {
1688                 mlog_errno(status);
1689                 goto bail;
1690         }
1691
1692         if (!ocfs2_inode_is_fast_symlink(inode)) {
1693                 status = ocfs2_create_symlink_data(osb, handle, inode,
1694                                                    symname);
1695                 if (status < 0) {
1696                         mlog_errno(status);
1697                         goto bail;
1698                 }
1699         }
1700
1701         status = ocfs2_add_entry(handle, dentry, inode,
1702                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1703                                  de_bh);
1704         if (status < 0) {
1705                 mlog_errno(status);
1706                 goto bail;
1707         }
1708
1709         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1710         if (status) {
1711                 mlog_errno(status);
1712                 goto bail;
1713         }
1714
1715         insert_inode_hash(inode);
1716         dentry->d_op = &ocfs2_dentry_ops;
1717         d_instantiate(dentry, inode);
1718 bail:
1719         if (handle)
1720                 ocfs2_commit_trans(handle);
1721         if (new_fe_bh)
1722                 brelse(new_fe_bh);
1723         if (parent_fe_bh)
1724                 brelse(parent_fe_bh);
1725         if (de_bh)
1726                 brelse(de_bh);
1727         if (inode_ac)
1728                 ocfs2_free_alloc_context(inode_ac);
1729         if (data_ac)
1730                 ocfs2_free_alloc_context(data_ac);
1731         if ((status < 0) && inode)
1732                 iput(inode);
1733
1734         mlog_exit(status);
1735
1736         return status;
1737 }
1738
1739 int ocfs2_check_dir_entry(struct inode * dir,
1740                           struct ocfs2_dir_entry * de,
1741                           struct buffer_head * bh,
1742                           unsigned long offset)
1743 {
1744         const char *error_msg = NULL;
1745         const int rlen = le16_to_cpu(de->rec_len);
1746
1747         if (rlen < OCFS2_DIR_REC_LEN(1))
1748                 error_msg = "rec_len is smaller than minimal";
1749         else if (rlen % 4 != 0)
1750                 error_msg = "rec_len % 4 != 0";
1751         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1752                 error_msg = "rec_len is too small for name_len";
1753         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1754                 error_msg = "directory entry across blocks";
1755
1756         if (error_msg != NULL)
1757                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1758                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1759                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1760                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1761                      de->name_len);
1762         return error_msg == NULL ? 1 : 0;
1763 }
1764
1765 /* we don't always have a dentry for what we want to add, so people
1766  * like orphan dir can call this instead.
1767  *
1768  * If you pass me insert_bh, I'll skip the search of the other dir
1769  * blocks and put the record in there.
1770  */
1771 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1772                              struct inode *dir,
1773                              const char *name, int namelen,
1774                              struct inode *inode, u64 blkno,
1775                              struct buffer_head *parent_fe_bh,
1776                              struct buffer_head *insert_bh)
1777 {
1778         unsigned long offset;
1779         unsigned short rec_len;
1780         struct ocfs2_dir_entry *de, *de1;
1781         struct super_block *sb;
1782         int retval, status;
1783
1784         mlog_entry_void();
1785
1786         sb = dir->i_sb;
1787
1788         if (!namelen)
1789                 return -EINVAL;
1790
1791         rec_len = OCFS2_DIR_REC_LEN(namelen);
1792         offset = 0;
1793         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1794         while (1) {
1795                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1796                 /* These checks should've already been passed by the
1797                  * prepare function, but I guess we can leave them
1798                  * here anyway. */
1799                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1800                         retval = -ENOENT;
1801                         goto bail;
1802                 }
1803                 if (ocfs2_match(namelen, name, de)) {
1804                         retval = -EEXIST;
1805                         goto bail;
1806                 }
1807                 if (((le64_to_cpu(de->inode) == 0) &&
1808                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1809                     (le16_to_cpu(de->rec_len) >=
1810                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1811                         status = ocfs2_journal_access(handle, dir, insert_bh,
1812                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1813                         /* By now the buffer is marked for journaling */
1814                         offset += le16_to_cpu(de->rec_len);
1815                         if (le64_to_cpu(de->inode)) {
1816                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1817                                         OCFS2_DIR_REC_LEN(de->name_len));
1818                                 de1->rec_len =
1819                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1820                                         OCFS2_DIR_REC_LEN(de->name_len));
1821                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1822                                 de = de1;
1823                         }
1824                         de->file_type = OCFS2_FT_UNKNOWN;
1825                         if (blkno) {
1826                                 de->inode = cpu_to_le64(blkno);
1827                                 ocfs2_set_de_type(de, inode->i_mode);
1828                         } else
1829                                 de->inode = 0;
1830                         de->name_len = namelen;
1831                         memcpy(de->name, name, namelen);
1832
1833                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1834                         dir->i_version++;
1835                         status = ocfs2_journal_dirty(handle, insert_bh);
1836                         retval = 0;
1837                         goto bail;
1838                 }
1839                 offset += le16_to_cpu(de->rec_len);
1840                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1841         }
1842
1843         /* when you think about it, the assert above should prevent us
1844          * from ever getting here. */
1845         retval = -ENOSPC;
1846 bail:
1847
1848         mlog_exit(retval);
1849         return retval;
1850 }
1851
1852
1853 /*
1854  * ocfs2_delete_entry deletes a directory entry by merging it with the
1855  * previous entry
1856  */
1857 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1858                               struct inode *dir,
1859                               struct ocfs2_dir_entry *de_del,
1860                               struct buffer_head *bh)
1861 {
1862         struct ocfs2_dir_entry *de, *pde;
1863         int i, status = -ENOENT;
1864
1865         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1866
1867         i = 0;
1868         pde = NULL;
1869         de = (struct ocfs2_dir_entry *) bh->b_data;
1870         while (i < bh->b_size) {
1871                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1872                         status = -EIO;
1873                         mlog_errno(status);
1874                         goto bail;
1875                 }
1876                 if (de == de_del)  {
1877                         status = ocfs2_journal_access(handle, dir, bh,
1878                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1879                         if (status < 0) {
1880                                 status = -EIO;
1881                                 mlog_errno(status);
1882                                 goto bail;
1883                         }
1884                         if (pde)
1885                                 pde->rec_len =
1886                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1887                                                     le16_to_cpu(de->rec_len));
1888                         else
1889                                 de->inode = 0;
1890                         dir->i_version++;
1891                         status = ocfs2_journal_dirty(handle, bh);
1892                         goto bail;
1893                 }
1894                 i += le16_to_cpu(de->rec_len);
1895                 pde = de;
1896                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1897         }
1898 bail:
1899         mlog_exit(status);
1900         return status;
1901 }
1902
1903 /*
1904  * Returns 0 if not found, -1 on failure, and 1 on success
1905  */
1906 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1907                                         struct inode *dir,
1908                                         const char *name, int namelen,
1909                                         unsigned long offset,
1910                                         struct ocfs2_dir_entry **res_dir)
1911 {
1912         struct ocfs2_dir_entry *de;
1913         char *dlimit, *de_buf;
1914         int de_len;
1915         int ret = 0;
1916
1917         mlog_entry_void();
1918
1919         de_buf = bh->b_data;
1920         dlimit = de_buf + dir->i_sb->s_blocksize;
1921
1922         while (de_buf < dlimit) {
1923                 /* this code is executed quadratically often */
1924                 /* do minimal checking `by hand' */
1925
1926                 de = (struct ocfs2_dir_entry *) de_buf;
1927
1928                 if (de_buf + namelen <= dlimit &&
1929                     ocfs2_match(namelen, name, de)) {
1930                         /* found a match - just to be sure, do a full check */
1931                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1932                                 ret = -1;
1933                                 goto bail;
1934                         }
1935                         *res_dir = de;
1936                         ret = 1;
1937                         goto bail;
1938                 }
1939
1940                 /* prevent looping on a bad block */
1941                 de_len = le16_to_cpu(de->rec_len);
1942                 if (de_len <= 0) {
1943                         ret = -1;
1944                         goto bail;
1945                 }
1946
1947                 de_buf += de_len;
1948                 offset += de_len;
1949         }
1950
1951 bail:
1952         mlog_exit(ret);
1953         return ret;
1954 }
1955
1956 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1957                                      struct inode *dir,
1958                                      struct ocfs2_dir_entry **res_dir)
1959 {
1960         struct super_block *sb;
1961         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1962         struct buffer_head *bh, *ret = NULL;
1963         unsigned long start, block, b;
1964         int ra_max = 0;         /* Number of bh's in the readahead
1965                                    buffer, bh_use[] */
1966         int ra_ptr = 0;         /* Current index into readahead
1967                                    buffer */
1968         int num = 0;
1969         int nblocks, i, err;
1970
1971         mlog_entry_void();
1972
1973         *res_dir = NULL;
1974         sb = dir->i_sb;
1975
1976         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1977         start = OCFS2_I(dir)->ip_dir_start_lookup;
1978         if (start >= nblocks)
1979                 start = 0;
1980         block = start;
1981
1982 restart:
1983         do {
1984                 /*
1985                  * We deal with the read-ahead logic here.
1986                  */
1987                 if (ra_ptr >= ra_max) {
1988                         /* Refill the readahead buffer */
1989                         ra_ptr = 0;
1990                         b = block;
1991                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1992                                 /*
1993                                  * Terminate if we reach the end of the
1994                                  * directory and must wrap, or if our
1995                                  * search has finished at this block.
1996                                  */
1997                                 if (b >= nblocks || (num && block == start)) {
1998                                         bh_use[ra_max] = NULL;
1999                                         break;
2000                                 }
2001                                 num++;
2002
2003                                 bh = ocfs2_bread(dir, b++, &err, 1);
2004                                 bh_use[ra_max] = bh;
2005                         }
2006                 }
2007                 if ((bh = bh_use[ra_ptr++]) == NULL)
2008                         goto next;
2009                 wait_on_buffer(bh);
2010                 if (!buffer_uptodate(bh)) {
2011                         /* read error, skip block & hope for the best */
2012                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2013                                     "offset %lu\n",
2014                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2015                                     block);
2016                         brelse(bh);
2017                         goto next;
2018                 }
2019                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2020                                           block << sb->s_blocksize_bits,
2021                                           res_dir);
2022                 if (i == 1) {
2023                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2024                         ret = bh;
2025                         goto cleanup_and_exit;
2026                 } else {
2027                         brelse(bh);
2028                         if (i < 0)
2029                                 goto cleanup_and_exit;
2030                 }
2031         next:
2032                 if (++block >= nblocks)
2033                         block = 0;
2034         } while (block != start);
2035
2036         /*
2037          * If the directory has grown while we were searching, then
2038          * search the last part of the directory before giving up.
2039          */
2040         block = nblocks;
2041         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2042         if (block < nblocks) {
2043                 start = 0;
2044                 goto restart;
2045         }
2046
2047 cleanup_and_exit:
2048         /* Clean up the read-ahead blocks */
2049         for (; ra_ptr < ra_max; ra_ptr++)
2050                 brelse(bh_use[ra_ptr]);
2051
2052         mlog_exit_ptr(ret);
2053         return ret;
2054 }
2055
2056 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2057 {
2058         int status, namelen;
2059
2060         mlog_entry_void();
2061
2062         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2063                            (long long)blkno);
2064         if (namelen <= 0) {
2065                 if (namelen)
2066                         status = namelen;
2067                 else
2068                         status = -EINVAL;
2069                 mlog_errno(status);
2070                 goto bail;
2071         }
2072         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2073                 status = -EINVAL;
2074                 mlog_errno(status);
2075                 goto bail;
2076         }
2077
2078         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2079              namelen);
2080
2081         status = 0;
2082 bail:
2083         mlog_exit(status);
2084         return status;
2085 }
2086
2087 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2088                                     struct ocfs2_journal_handle *handle,
2089                                     struct inode *inode,
2090                                     char *name,
2091                                     struct buffer_head **de_bh)
2092 {
2093         struct inode *orphan_dir_inode = NULL;
2094         struct buffer_head *orphan_dir_bh = NULL;
2095         int status = 0;
2096
2097         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2098         if (status < 0) {
2099                 mlog_errno(status);
2100                 goto leave;
2101         }
2102
2103         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2104                                                        ORPHAN_DIR_SYSTEM_INODE,
2105                                                        osb->slot_num);
2106         if (!orphan_dir_inode) {
2107                 status = -ENOENT;
2108                 mlog_errno(status);
2109                 goto leave;
2110         }
2111
2112         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2113         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2114         if (status < 0) {
2115                 mlog_errno(status);
2116                 goto leave;
2117         }
2118
2119         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2120                                               orphan_dir_bh, name,
2121                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2122         if (status < 0) {
2123                 mlog_errno(status);
2124                 goto leave;
2125         }
2126
2127 leave:
2128         if (orphan_dir_inode)
2129                 iput(orphan_dir_inode);
2130
2131         if (orphan_dir_bh)
2132                 brelse(orphan_dir_bh);
2133
2134         mlog_exit(status);
2135         return status;
2136 }
2137
2138 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2139                             struct ocfs2_journal_handle *handle,
2140                             struct inode *inode,
2141                             struct ocfs2_dinode *fe,
2142                             char *name,
2143                             struct buffer_head *de_bh)
2144 {
2145         struct inode *orphan_dir_inode = NULL;
2146         struct buffer_head *orphan_dir_bh = NULL;
2147         int status = 0;
2148         struct ocfs2_dinode *orphan_fe;
2149
2150         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2151
2152         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2153                                                        ORPHAN_DIR_SYSTEM_INODE,
2154                                                        osb->slot_num);
2155         if (!orphan_dir_inode) {
2156                 status = -ENOENT;
2157                 mlog_errno(status);
2158                 goto leave;
2159         }
2160
2161         status = ocfs2_read_block(osb,
2162                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2163                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2164                                   orphan_dir_inode);
2165         if (status < 0) {
2166                 mlog_errno(status);
2167                 goto leave;
2168         }
2169
2170         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2171                                       OCFS2_JOURNAL_ACCESS_WRITE);
2172         if (status < 0) {
2173                 mlog_errno(status);
2174                 goto leave;
2175         }
2176
2177         /* we're a cluster, and nlink can change on disk from
2178          * underneath us... */
2179         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2180         if (S_ISDIR(inode->i_mode))
2181                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2182         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2183
2184         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2185         if (status < 0) {
2186                 mlog_errno(status);
2187                 goto leave;
2188         }
2189
2190         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2191                                    OCFS2_ORPHAN_NAMELEN, inode,
2192                                    OCFS2_I(inode)->ip_blkno,
2193                                    orphan_dir_bh, de_bh);
2194         if (status < 0) {
2195                 mlog_errno(status);
2196                 goto leave;
2197         }
2198
2199         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2200
2201         /* Record which orphan dir our inode now resides
2202          * in. delete_inode will use this to determine which orphan
2203          * dir to lock. */
2204         spin_lock(&OCFS2_I(inode)->ip_lock);
2205         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2206         spin_unlock(&OCFS2_I(inode)->ip_lock);
2207
2208         mlog(0, "Inode %llu orphaned in slot %d\n",
2209              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2210
2211 leave:
2212         if (orphan_dir_inode)
2213                 iput(orphan_dir_inode);
2214
2215         if (orphan_dir_bh)
2216                 brelse(orphan_dir_bh);
2217
2218         mlog_exit(status);
2219         return status;
2220 }
2221
2222 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2223 int ocfs2_orphan_del(struct ocfs2_super *osb,
2224                      struct ocfs2_journal_handle *handle,
2225                      struct inode *orphan_dir_inode,
2226                      struct inode *inode,
2227                      struct buffer_head *orphan_dir_bh)
2228 {
2229         char name[OCFS2_ORPHAN_NAMELEN + 1];
2230         struct ocfs2_dinode *orphan_fe;
2231         int status = 0;
2232         struct buffer_head *target_de_bh = NULL;
2233         struct ocfs2_dir_entry *target_de = NULL;
2234
2235         mlog_entry_void();
2236
2237         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2238         if (status < 0) {
2239                 mlog_errno(status);
2240                 goto leave;
2241         }
2242
2243         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2244              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2245              OCFS2_ORPHAN_NAMELEN);
2246
2247         /* find it's spot in the orphan directory */
2248         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2249                                         orphan_dir_inode, &target_de);
2250         if (!target_de_bh) {
2251                 status = -ENOENT;
2252                 mlog_errno(status);
2253                 goto leave;
2254         }
2255
2256         /* remove it from the orphan directory */
2257         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2258                                     target_de_bh);
2259         if (status < 0) {
2260                 mlog_errno(status);
2261                 goto leave;
2262         }
2263
2264         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2265                                       OCFS2_JOURNAL_ACCESS_WRITE);
2266         if (status < 0) {
2267                 mlog_errno(status);
2268                 goto leave;
2269         }
2270
2271         /* do the i_nlink dance! :) */
2272         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2273         if (S_ISDIR(inode->i_mode))
2274                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2275         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2276
2277         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2278         if (status < 0) {
2279                 mlog_errno(status);
2280                 goto leave;
2281         }
2282
2283 leave:
2284         if (target_de_bh)
2285                 brelse(target_de_bh);
2286
2287         mlog_exit(status);
2288         return status;
2289 }
2290
2291 struct inode_operations ocfs2_dir_iops = {
2292         .create         = ocfs2_create,
2293         .lookup         = ocfs2_lookup,
2294         .link           = ocfs2_link,
2295         .unlink         = ocfs2_unlink,
2296         .rmdir          = ocfs2_unlink,
2297         .symlink        = ocfs2_symlink,
2298         .mkdir          = ocfs2_mkdir,
2299         .mknod          = ocfs2_mknod,
2300         .rename         = ocfs2_rename,
2301         .setattr        = ocfs2_setattr,
2302         .getattr        = ocfs2_getattr,
2303 };