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         if (atomic_read(&old_dentry->d_count) > 2) {
1089                 shrink_dcache_parent(old_dentry);
1090                 if (atomic_read(&old_dentry->d_count) > 2) {
1091                         status = -EBUSY;
1092                         goto bail;
1093                 }
1094         }
1095
1096         /* Assume a directory heirarchy thusly:
1097          * a/b/c
1098          * a/d
1099          * a,b,c, and d are all directories.
1100          *
1101          * from cwd of 'a' on both nodes:
1102          * node1: mv b/c d
1103          * node2: mv d   b/c
1104          *
1105          * And that's why, just like the VFS, we need a file system
1106          * rename lock. */
1107         if (old_dentry != new_dentry) {
1108                 status = ocfs2_rename_lock(osb);
1109                 if (status < 0) {
1110                         mlog_errno(status);
1111                         goto bail;
1112                 }
1113                 rename_lock = 1;
1114         }
1115
1116         handle = ocfs2_alloc_handle(osb);
1117         if (handle == NULL) {
1118                 status = -ENOMEM;
1119                 mlog_errno(status);
1120                 goto bail;
1121         }
1122
1123         /* if old and new are the same, this'll just do one lock. */
1124         status = ocfs2_double_lock(osb, handle,
1125                                   &old_dir_bh, old_dir,
1126                                   &new_dir_bh, new_dir);
1127         if (status < 0) {
1128                 mlog_errno(status);
1129                 goto bail;
1130         }
1131
1132         /* make sure both dirs have bhs
1133          * get an extra ref on old_dir_bh if old==new */
1134         if (!new_dir_bh) {
1135                 if (old_dir_bh) {
1136                         new_dir_bh = old_dir_bh;
1137                         get_bh(new_dir_bh);
1138                 } else {
1139                         mlog(ML_ERROR, "no old_dir_bh!\n");
1140                         status = -EIO;
1141                         goto bail;
1142                 }
1143         }
1144
1145         /*
1146          * Though we don't require an inode meta data update if
1147          * old_inode is not a directory, we lock anyway here to ensure
1148          * the vote thread on other nodes won't have to concurrently
1149          * downconvert the inode and the dentry locks.
1150          */
1151         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1152         if (status < 0) {
1153                 if (status != -ENOENT)
1154                         mlog_errno(status);
1155                 goto bail;
1156         }
1157
1158         status = ocfs2_remote_dentry_delete(old_dentry);
1159         if (status < 0) {
1160                 mlog_errno(status);
1161                 goto bail;
1162         }
1163
1164         if (S_ISDIR(old_inode->i_mode)) {
1165                 status = -EIO;
1166                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1167                 if (!old_inode_de_bh)
1168                         goto bail;
1169
1170                 status = -EIO;
1171                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1172                     OCFS2_I(old_dir)->ip_blkno)
1173                         goto bail;
1174                 status = -EMLINK;
1175                 if (!new_inode && new_dir!=old_dir &&
1176                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1177                         goto bail;
1178         }
1179
1180         status = -ENOENT;
1181         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1182                                      old_dentry->d_name.len,
1183                                      old_dir, &old_de);
1184         if (!old_de_bh)
1185                 goto bail;
1186
1187         /*
1188          *  Check for inode number is _not_ due to possible IO errors.
1189          *  We might rmdir the source, keep it as pwd of some process
1190          *  and merrily kill the link to whatever was created under the
1191          *  same name. Goodbye sticky bit ;-<
1192          */
1193         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1194                 goto bail;
1195
1196         /* check if the target already exists (in which case we need
1197          * to delete it */
1198         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1199                                           new_dentry->d_name.len,
1200                                           &newfe_blkno, new_dir, &new_de_bh,
1201                                           &new_de);
1202         /* The only error we allow here is -ENOENT because the new
1203          * file not existing is perfectly valid. */
1204         if ((status < 0) && (status != -ENOENT)) {
1205                 /* If we cannot find the file specified we should just */
1206                 /* return the error... */
1207                 mlog_errno(status);
1208                 goto bail;
1209         }
1210
1211         if (!new_de && new_inode)
1212                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1213                      "directory!", new_inode->i_ino);
1214
1215         /* In case we need to overwrite an existing file, we blow it
1216          * away first */
1217         if (new_de) {
1218                 /* VFS didn't think there existed an inode here, but
1219                  * someone else in the cluster must have raced our
1220                  * rename to create one. Today we error cleanly, in
1221                  * the future we should consider calling iget to build
1222                  * a new struct inode for this entry. */
1223                 if (!new_inode) {
1224                         status = -EACCES;
1225
1226                         mlog(0, "We found an inode for name %.*s but VFS "
1227                              "didn't give us one.\n", new_dentry->d_name.len,
1228                              new_dentry->d_name.name);
1229                         goto bail;
1230                 }
1231
1232                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1233                         status = -EACCES;
1234
1235                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1236                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1237                              (unsigned long long)newfe_blkno,
1238                              OCFS2_I(new_inode)->ip_flags);
1239                         goto bail;
1240                 }
1241
1242                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1243                 if (status < 0) {
1244                         if (status != -ENOENT)
1245                                 mlog_errno(status);
1246                         goto bail;
1247                 }
1248
1249                 status = ocfs2_remote_dentry_delete(new_dentry);
1250                 if (status < 0) {
1251                         mlog_errno(status);
1252                         goto bail;
1253                 }
1254
1255                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1256
1257                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1258                      "newfebh=%p bhblocknr=%llu\n", new_de,
1259                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1260                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1261
1262                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1263                         status = ocfs2_prepare_orphan_dir(osb, handle,
1264                                                           new_inode,
1265                                                           orphan_name,
1266                                                           &orphan_entry_bh);
1267                         if (status < 0) {
1268                                 mlog_errno(status);
1269                                 goto bail;
1270                         }
1271                 }
1272         } else {
1273                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1274
1275                 status = ocfs2_check_dir_for_entry(new_dir,
1276                                                    new_dentry->d_name.name,
1277                                                    new_dentry->d_name.len);
1278                 if (status)
1279                         goto bail;
1280
1281                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1282                                                       new_dentry->d_name.name,
1283                                                       new_dentry->d_name.len,
1284                                                       &insert_entry_bh);
1285                 if (status < 0) {
1286                         mlog_errno(status);
1287                         goto bail;
1288                 }
1289         }
1290
1291         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1292         if (IS_ERR(handle)) {
1293                 status = PTR_ERR(handle);
1294                 handle = NULL;
1295                 mlog_errno(status);
1296                 goto bail;
1297         }
1298
1299         if (new_de) {
1300                 if (S_ISDIR(new_inode->i_mode)) {
1301                         if (!ocfs2_empty_dir(new_inode) ||
1302                             new_inode->i_nlink != 2) {
1303                                 status = -ENOTEMPTY;
1304                                 goto bail;
1305                         }
1306                 }
1307                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1308                                               OCFS2_JOURNAL_ACCESS_WRITE);
1309                 if (status < 0) {
1310                         mlog_errno(status);
1311                         goto bail;
1312                 }
1313
1314                 if (S_ISDIR(new_inode->i_mode) ||
1315                     (newfe->i_links_count == cpu_to_le16(1))){
1316                         status = ocfs2_orphan_add(osb, handle, new_inode,
1317                                                   newfe, orphan_name,
1318                                                   orphan_entry_bh);
1319                         if (status < 0) {
1320                                 mlog_errno(status);
1321                                 goto bail;
1322                         }
1323                 }
1324
1325                 /* change the dirent to point to the correct inode */
1326                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1327                                               OCFS2_JOURNAL_ACCESS_WRITE);
1328                 if (status < 0) {
1329                         mlog_errno(status);
1330                         goto bail;
1331                 }
1332                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1333                 new_de->file_type = old_de->file_type;
1334                 new_dir->i_version++;
1335                 status = ocfs2_journal_dirty(handle, new_de_bh);
1336                 if (status < 0) {
1337                         mlog_errno(status);
1338                         goto bail;
1339                 }
1340
1341                 if (S_ISDIR(new_inode->i_mode))
1342                         newfe->i_links_count = 0;
1343                 else
1344                         le16_add_cpu(&newfe->i_links_count, -1);
1345
1346                 status = ocfs2_journal_dirty(handle, newfe_bh);
1347                 if (status < 0) {
1348                         mlog_errno(status);
1349                         goto bail;
1350                 }
1351         } else {
1352                 /* if the name was not found in new_dir, add it now */
1353                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1354                                          OCFS2_I(old_inode)->ip_blkno,
1355                                          new_dir_bh, insert_entry_bh);
1356         }
1357
1358         old_inode->i_ctime = CURRENT_TIME;
1359         mark_inode_dirty(old_inode);
1360
1361         /* now that the name has been added to new_dir, remove the old name */
1362         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1363         if (status < 0) {
1364                 mlog_errno(status);
1365                 goto bail;
1366         }
1367
1368         if (new_inode) {
1369                 new_inode->i_nlink--;
1370                 new_inode->i_ctime = CURRENT_TIME;
1371         }
1372         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1373         if (old_inode_de_bh) {
1374                 status = ocfs2_journal_access(handle, old_inode,
1375                                              old_inode_de_bh,
1376                                              OCFS2_JOURNAL_ACCESS_WRITE);
1377                 PARENT_INO(old_inode_de_bh->b_data) =
1378                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1379                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1380                 old_dir->i_nlink--;
1381                 if (new_inode) {
1382                         new_inode->i_nlink--;
1383                 } else {
1384                         inc_nlink(new_dir);
1385                         mark_inode_dirty(new_dir);
1386                 }
1387         }
1388         mark_inode_dirty(old_dir);
1389         if (new_inode)
1390                 mark_inode_dirty(new_inode);
1391
1392         if (old_dir != new_dir)
1393                 if (new_dir_nlink != new_dir->i_nlink) {
1394                         if (!new_dir_bh) {
1395                                 mlog(ML_ERROR, "need to change nlink for new "
1396                                      "dir %llu from %d to %d but bh is NULL\n",
1397                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1398                                      (int)new_dir_nlink, new_dir->i_nlink);
1399                         } else {
1400                                 struct ocfs2_dinode *fe;
1401                                 status = ocfs2_journal_access(handle,
1402                                                               new_dir,
1403                                                               new_dir_bh,
1404                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1405                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1406                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1407                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1408                         }
1409                 }
1410
1411         if (old_dir_nlink != old_dir->i_nlink) {
1412                 if (!old_dir_bh) {
1413                         mlog(ML_ERROR, "need to change nlink for old dir "
1414                              "%llu from %d to %d but bh is NULL!\n",
1415                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1416                              (int)old_dir_nlink, old_dir->i_nlink);
1417                 } else {
1418                         struct ocfs2_dinode *fe;
1419                         status = ocfs2_journal_access(handle, old_dir,
1420                                                       old_dir_bh,
1421                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1422                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1423                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1424                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1425                 }
1426         }
1427
1428         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1429         status = 0;
1430 bail:
1431         if (rename_lock)
1432                 ocfs2_rename_unlock(osb);
1433
1434         if (handle)
1435                 ocfs2_commit_trans(handle);
1436
1437         if (new_inode)
1438                 sync_mapping_buffers(old_inode->i_mapping);
1439
1440         if (new_inode)
1441                 iput(new_inode);
1442         if (newfe_bh)
1443                 brelse(newfe_bh);
1444         if (old_dir_bh)
1445                 brelse(old_dir_bh);
1446         if (new_dir_bh)
1447                 brelse(new_dir_bh);
1448         if (new_de_bh)
1449                 brelse(new_de_bh);
1450         if (old_de_bh)
1451                 brelse(old_de_bh);
1452         if (old_inode_de_bh)
1453                 brelse(old_inode_de_bh);
1454         if (orphan_entry_bh)
1455                 brelse(orphan_entry_bh);
1456         if (insert_entry_bh)
1457                 brelse(insert_entry_bh);
1458
1459         mlog_exit(status);
1460
1461         return status;
1462 }
1463
1464 /*
1465  * we expect i_size = strlen(symname). Copy symname into the file
1466  * data, including the null terminator.
1467  */
1468 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1469                                      struct ocfs2_journal_handle *handle,
1470                                      struct inode *inode,
1471                                      const char *symname)
1472 {
1473         struct buffer_head **bhs = NULL;
1474         const char *c;
1475         struct super_block *sb = osb->sb;
1476         u64 p_blkno;
1477         int p_blocks;
1478         int virtual, blocks, status, i, bytes_left;
1479
1480         bytes_left = i_size_read(inode) + 1;
1481         /* we can't trust i_blocks because we're actually going to
1482          * write i_size + 1 bytes. */
1483         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1484
1485         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1486                         (unsigned long long)inode->i_blocks,
1487                         i_size_read(inode), blocks);
1488
1489         /* Sanity check -- make sure we're going to fit. */
1490         if (bytes_left >
1491             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1492                 status = -EIO;
1493                 mlog_errno(status);
1494                 goto bail;
1495         }
1496
1497         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1498         if (!bhs) {
1499                 status = -ENOMEM;
1500                 mlog_errno(status);
1501                 goto bail;
1502         }
1503
1504         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1505                                              &p_blocks);
1506         if (status < 0) {
1507                 mlog_errno(status);
1508                 goto bail;
1509         }
1510
1511         /* links can never be larger than one cluster so we know this
1512          * is all going to be contiguous, but do a sanity check
1513          * anyway. */
1514         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1515                 status = -EIO;
1516                 mlog_errno(status);
1517                 goto bail;
1518         }
1519
1520         virtual = 0;
1521         while(bytes_left > 0) {
1522                 c = &symname[virtual * sb->s_blocksize];
1523
1524                 bhs[virtual] = sb_getblk(sb, p_blkno);
1525                 if (!bhs[virtual]) {
1526                         status = -ENOMEM;
1527                         mlog_errno(status);
1528                         goto bail;
1529                 }
1530                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1531
1532                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1533                                               OCFS2_JOURNAL_ACCESS_CREATE);
1534                 if (status < 0) {
1535                         mlog_errno(status);
1536                         goto bail;
1537                 }
1538
1539                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1540
1541                 memcpy(bhs[virtual]->b_data, c,
1542                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1543                        bytes_left);
1544
1545                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1546                 if (status < 0) {
1547                         mlog_errno(status);
1548                         goto bail;
1549                 }
1550
1551                 virtual++;
1552                 p_blkno++;
1553                 bytes_left -= sb->s_blocksize;
1554         }
1555
1556         status = 0;
1557 bail:
1558
1559         if (bhs) {
1560                 for(i = 0; i < blocks; i++)
1561                         if (bhs[i])
1562                                 brelse(bhs[i]);
1563                 kfree(bhs);
1564         }
1565
1566         mlog_exit(status);
1567         return status;
1568 }
1569
1570 static int ocfs2_symlink(struct inode *dir,
1571                          struct dentry *dentry,
1572                          const char *symname)
1573 {
1574         int status, l, credits;
1575         u64 newsize;
1576         struct ocfs2_super *osb = NULL;
1577         struct inode *inode = NULL;
1578         struct super_block *sb;
1579         struct buffer_head *new_fe_bh = NULL;
1580         struct buffer_head *de_bh = NULL;
1581         struct buffer_head *parent_fe_bh = NULL;
1582         struct ocfs2_dinode *fe = NULL;
1583         struct ocfs2_dinode *dirfe;
1584         struct ocfs2_journal_handle *handle = NULL;
1585         struct ocfs2_alloc_context *inode_ac = NULL;
1586         struct ocfs2_alloc_context *data_ac = NULL;
1587
1588         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1589                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1590
1591         sb = dir->i_sb;
1592         osb = OCFS2_SB(sb);
1593
1594         l = strlen(symname) + 1;
1595
1596         credits = ocfs2_calc_symlink_credits(sb);
1597
1598         handle = ocfs2_alloc_handle(osb);
1599         if (handle == NULL) {
1600                 status = -ENOMEM;
1601                 mlog_errno(status);
1602                 goto bail;
1603         }
1604
1605         /* lock the parent directory */
1606         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1607         if (status < 0) {
1608                 if (status != -ENOENT)
1609                         mlog_errno(status);
1610                 goto bail;
1611         }
1612
1613         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1614         if (!dirfe->i_links_count) {
1615                 /* can't make a file in a deleted directory. */
1616                 status = -ENOENT;
1617                 goto bail;
1618         }
1619
1620         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1621                                            dentry->d_name.len);
1622         if (status)
1623                 goto bail;
1624
1625         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1626                                               dentry->d_name.name,
1627                                               dentry->d_name.len, &de_bh);
1628         if (status < 0) {
1629                 mlog_errno(status);
1630                 goto bail;
1631         }
1632
1633         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1634         if (status < 0) {
1635                 if (status != -ENOSPC)
1636                         mlog_errno(status);
1637                 goto bail;
1638         }
1639
1640         /* don't reserve bitmap space for fast symlinks. */
1641         if (l > ocfs2_fast_symlink_chars(sb)) {
1642                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1643                 if (status < 0) {
1644                         if (status != -ENOSPC)
1645                                 mlog_errno(status);
1646                         goto bail;
1647                 }
1648         }
1649
1650         handle = ocfs2_start_trans(osb, handle, credits);
1651         if (IS_ERR(handle)) {
1652                 status = PTR_ERR(handle);
1653                 handle = NULL;
1654                 mlog_errno(status);
1655                 goto bail;
1656         }
1657
1658         status = ocfs2_mknod_locked(osb, dir, dentry,
1659                                     S_IFLNK | S_IRWXUGO, 0,
1660                                     &new_fe_bh, parent_fe_bh, handle,
1661                                     &inode, inode_ac);
1662         if (status < 0) {
1663                 mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1668         inode->i_rdev = 0;
1669         newsize = l - 1;
1670         if (l > ocfs2_fast_symlink_chars(sb)) {
1671                 inode->i_op = &ocfs2_symlink_inode_operations;
1672                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1673                                                     handle, data_ac, NULL,
1674                                                     NULL);
1675                 if (status < 0) {
1676                         if (status != -ENOSPC && status != -EINTR) {
1677                                 mlog(ML_ERROR,
1678                                      "Failed to extend file to %llu\n",
1679                                      (unsigned long long)newsize);
1680                                 mlog_errno(status);
1681                                 status = -ENOSPC;
1682                         }
1683                         goto bail;
1684                 }
1685                 i_size_write(inode, newsize);
1686                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1687         } else {
1688                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1689                 memcpy((char *) fe->id2.i_symlink, symname, l);
1690                 i_size_write(inode, newsize);
1691                 inode->i_blocks = 0;
1692         }
1693
1694         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1695         if (status < 0) {
1696                 mlog_errno(status);
1697                 goto bail;
1698         }
1699
1700         if (!ocfs2_inode_is_fast_symlink(inode)) {
1701                 status = ocfs2_create_symlink_data(osb, handle, inode,
1702                                                    symname);
1703                 if (status < 0) {
1704                         mlog_errno(status);
1705                         goto bail;
1706                 }
1707         }
1708
1709         status = ocfs2_add_entry(handle, dentry, inode,
1710                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1711                                  de_bh);
1712         if (status < 0) {
1713                 mlog_errno(status);
1714                 goto bail;
1715         }
1716
1717         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1718         if (status) {
1719                 mlog_errno(status);
1720                 goto bail;
1721         }
1722
1723         insert_inode_hash(inode);
1724         dentry->d_op = &ocfs2_dentry_ops;
1725         d_instantiate(dentry, inode);
1726 bail:
1727         if (handle)
1728                 ocfs2_commit_trans(handle);
1729         if (new_fe_bh)
1730                 brelse(new_fe_bh);
1731         if (parent_fe_bh)
1732                 brelse(parent_fe_bh);
1733         if (de_bh)
1734                 brelse(de_bh);
1735         if (inode_ac)
1736                 ocfs2_free_alloc_context(inode_ac);
1737         if (data_ac)
1738                 ocfs2_free_alloc_context(data_ac);
1739         if ((status < 0) && inode)
1740                 iput(inode);
1741
1742         mlog_exit(status);
1743
1744         return status;
1745 }
1746
1747 int ocfs2_check_dir_entry(struct inode * dir,
1748                           struct ocfs2_dir_entry * de,
1749                           struct buffer_head * bh,
1750                           unsigned long offset)
1751 {
1752         const char *error_msg = NULL;
1753         const int rlen = le16_to_cpu(de->rec_len);
1754
1755         if (rlen < OCFS2_DIR_REC_LEN(1))
1756                 error_msg = "rec_len is smaller than minimal";
1757         else if (rlen % 4 != 0)
1758                 error_msg = "rec_len % 4 != 0";
1759         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1760                 error_msg = "rec_len is too small for name_len";
1761         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1762                 error_msg = "directory entry across blocks";
1763
1764         if (error_msg != NULL)
1765                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1766                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1767                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1768                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1769                      de->name_len);
1770         return error_msg == NULL ? 1 : 0;
1771 }
1772
1773 /* we don't always have a dentry for what we want to add, so people
1774  * like orphan dir can call this instead.
1775  *
1776  * If you pass me insert_bh, I'll skip the search of the other dir
1777  * blocks and put the record in there.
1778  */
1779 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1780                              struct inode *dir,
1781                              const char *name, int namelen,
1782                              struct inode *inode, u64 blkno,
1783                              struct buffer_head *parent_fe_bh,
1784                              struct buffer_head *insert_bh)
1785 {
1786         unsigned long offset;
1787         unsigned short rec_len;
1788         struct ocfs2_dir_entry *de, *de1;
1789         struct super_block *sb;
1790         int retval, status;
1791
1792         mlog_entry_void();
1793
1794         sb = dir->i_sb;
1795
1796         if (!namelen)
1797                 return -EINVAL;
1798
1799         rec_len = OCFS2_DIR_REC_LEN(namelen);
1800         offset = 0;
1801         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1802         while (1) {
1803                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1804                 /* These checks should've already been passed by the
1805                  * prepare function, but I guess we can leave them
1806                  * here anyway. */
1807                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1808                         retval = -ENOENT;
1809                         goto bail;
1810                 }
1811                 if (ocfs2_match(namelen, name, de)) {
1812                         retval = -EEXIST;
1813                         goto bail;
1814                 }
1815                 if (((le64_to_cpu(de->inode) == 0) &&
1816                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1817                     (le16_to_cpu(de->rec_len) >=
1818                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1819                         status = ocfs2_journal_access(handle, dir, insert_bh,
1820                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1821                         /* By now the buffer is marked for journaling */
1822                         offset += le16_to_cpu(de->rec_len);
1823                         if (le64_to_cpu(de->inode)) {
1824                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1825                                         OCFS2_DIR_REC_LEN(de->name_len));
1826                                 de1->rec_len =
1827                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1828                                         OCFS2_DIR_REC_LEN(de->name_len));
1829                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1830                                 de = de1;
1831                         }
1832                         de->file_type = OCFS2_FT_UNKNOWN;
1833                         if (blkno) {
1834                                 de->inode = cpu_to_le64(blkno);
1835                                 ocfs2_set_de_type(de, inode->i_mode);
1836                         } else
1837                                 de->inode = 0;
1838                         de->name_len = namelen;
1839                         memcpy(de->name, name, namelen);
1840
1841                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1842                         dir->i_version++;
1843                         status = ocfs2_journal_dirty(handle, insert_bh);
1844                         retval = 0;
1845                         goto bail;
1846                 }
1847                 offset += le16_to_cpu(de->rec_len);
1848                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1849         }
1850
1851         /* when you think about it, the assert above should prevent us
1852          * from ever getting here. */
1853         retval = -ENOSPC;
1854 bail:
1855
1856         mlog_exit(retval);
1857         return retval;
1858 }
1859
1860
1861 /*
1862  * ocfs2_delete_entry deletes a directory entry by merging it with the
1863  * previous entry
1864  */
1865 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1866                               struct inode *dir,
1867                               struct ocfs2_dir_entry *de_del,
1868                               struct buffer_head *bh)
1869 {
1870         struct ocfs2_dir_entry *de, *pde;
1871         int i, status = -ENOENT;
1872
1873         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1874
1875         i = 0;
1876         pde = NULL;
1877         de = (struct ocfs2_dir_entry *) bh->b_data;
1878         while (i < bh->b_size) {
1879                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1880                         status = -EIO;
1881                         mlog_errno(status);
1882                         goto bail;
1883                 }
1884                 if (de == de_del)  {
1885                         status = ocfs2_journal_access(handle, dir, bh,
1886                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1887                         if (status < 0) {
1888                                 status = -EIO;
1889                                 mlog_errno(status);
1890                                 goto bail;
1891                         }
1892                         if (pde)
1893                                 pde->rec_len =
1894                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1895                                                     le16_to_cpu(de->rec_len));
1896                         else
1897                                 de->inode = 0;
1898                         dir->i_version++;
1899                         status = ocfs2_journal_dirty(handle, bh);
1900                         goto bail;
1901                 }
1902                 i += le16_to_cpu(de->rec_len);
1903                 pde = de;
1904                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1905         }
1906 bail:
1907         mlog_exit(status);
1908         return status;
1909 }
1910
1911 /*
1912  * Returns 0 if not found, -1 on failure, and 1 on success
1913  */
1914 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1915                                         struct inode *dir,
1916                                         const char *name, int namelen,
1917                                         unsigned long offset,
1918                                         struct ocfs2_dir_entry **res_dir)
1919 {
1920         struct ocfs2_dir_entry *de;
1921         char *dlimit, *de_buf;
1922         int de_len;
1923         int ret = 0;
1924
1925         mlog_entry_void();
1926
1927         de_buf = bh->b_data;
1928         dlimit = de_buf + dir->i_sb->s_blocksize;
1929
1930         while (de_buf < dlimit) {
1931                 /* this code is executed quadratically often */
1932                 /* do minimal checking `by hand' */
1933
1934                 de = (struct ocfs2_dir_entry *) de_buf;
1935
1936                 if (de_buf + namelen <= dlimit &&
1937                     ocfs2_match(namelen, name, de)) {
1938                         /* found a match - just to be sure, do a full check */
1939                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1940                                 ret = -1;
1941                                 goto bail;
1942                         }
1943                         *res_dir = de;
1944                         ret = 1;
1945                         goto bail;
1946                 }
1947
1948                 /* prevent looping on a bad block */
1949                 de_len = le16_to_cpu(de->rec_len);
1950                 if (de_len <= 0) {
1951                         ret = -1;
1952                         goto bail;
1953                 }
1954
1955                 de_buf += de_len;
1956                 offset += de_len;
1957         }
1958
1959 bail:
1960         mlog_exit(ret);
1961         return ret;
1962 }
1963
1964 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1965                                      struct inode *dir,
1966                                      struct ocfs2_dir_entry **res_dir)
1967 {
1968         struct super_block *sb;
1969         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1970         struct buffer_head *bh, *ret = NULL;
1971         unsigned long start, block, b;
1972         int ra_max = 0;         /* Number of bh's in the readahead
1973                                    buffer, bh_use[] */
1974         int ra_ptr = 0;         /* Current index into readahead
1975                                    buffer */
1976         int num = 0;
1977         int nblocks, i, err;
1978
1979         mlog_entry_void();
1980
1981         *res_dir = NULL;
1982         sb = dir->i_sb;
1983
1984         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1985         start = OCFS2_I(dir)->ip_dir_start_lookup;
1986         if (start >= nblocks)
1987                 start = 0;
1988         block = start;
1989
1990 restart:
1991         do {
1992                 /*
1993                  * We deal with the read-ahead logic here.
1994                  */
1995                 if (ra_ptr >= ra_max) {
1996                         /* Refill the readahead buffer */
1997                         ra_ptr = 0;
1998                         b = block;
1999                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2000                                 /*
2001                                  * Terminate if we reach the end of the
2002                                  * directory and must wrap, or if our
2003                                  * search has finished at this block.
2004                                  */
2005                                 if (b >= nblocks || (num && block == start)) {
2006                                         bh_use[ra_max] = NULL;
2007                                         break;
2008                                 }
2009                                 num++;
2010
2011                                 bh = ocfs2_bread(dir, b++, &err, 1);
2012                                 bh_use[ra_max] = bh;
2013                         }
2014                 }
2015                 if ((bh = bh_use[ra_ptr++]) == NULL)
2016                         goto next;
2017                 wait_on_buffer(bh);
2018                 if (!buffer_uptodate(bh)) {
2019                         /* read error, skip block & hope for the best */
2020                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2021                                     "offset %lu\n",
2022                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2023                                     block);
2024                         brelse(bh);
2025                         goto next;
2026                 }
2027                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2028                                           block << sb->s_blocksize_bits,
2029                                           res_dir);
2030                 if (i == 1) {
2031                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2032                         ret = bh;
2033                         goto cleanup_and_exit;
2034                 } else {
2035                         brelse(bh);
2036                         if (i < 0)
2037                                 goto cleanup_and_exit;
2038                 }
2039         next:
2040                 if (++block >= nblocks)
2041                         block = 0;
2042         } while (block != start);
2043
2044         /*
2045          * If the directory has grown while we were searching, then
2046          * search the last part of the directory before giving up.
2047          */
2048         block = nblocks;
2049         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2050         if (block < nblocks) {
2051                 start = 0;
2052                 goto restart;
2053         }
2054
2055 cleanup_and_exit:
2056         /* Clean up the read-ahead blocks */
2057         for (; ra_ptr < ra_max; ra_ptr++)
2058                 brelse(bh_use[ra_ptr]);
2059
2060         mlog_exit_ptr(ret);
2061         return ret;
2062 }
2063
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2065 {
2066         int status, namelen;
2067
2068         mlog_entry_void();
2069
2070         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2071                            (long long)blkno);
2072         if (namelen <= 0) {
2073                 if (namelen)
2074                         status = namelen;
2075                 else
2076                         status = -EINVAL;
2077                 mlog_errno(status);
2078                 goto bail;
2079         }
2080         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2081                 status = -EINVAL;
2082                 mlog_errno(status);
2083                 goto bail;
2084         }
2085
2086         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2087              namelen);
2088
2089         status = 0;
2090 bail:
2091         mlog_exit(status);
2092         return status;
2093 }
2094
2095 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2096                                     struct ocfs2_journal_handle *handle,
2097                                     struct inode *inode,
2098                                     char *name,
2099                                     struct buffer_head **de_bh)
2100 {
2101         struct inode *orphan_dir_inode = NULL;
2102         struct buffer_head *orphan_dir_bh = NULL;
2103         int status = 0;
2104
2105         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2106         if (status < 0) {
2107                 mlog_errno(status);
2108                 goto leave;
2109         }
2110
2111         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2112                                                        ORPHAN_DIR_SYSTEM_INODE,
2113                                                        osb->slot_num);
2114         if (!orphan_dir_inode) {
2115                 status = -ENOENT;
2116                 mlog_errno(status);
2117                 goto leave;
2118         }
2119
2120         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2121         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2122         if (status < 0) {
2123                 mlog_errno(status);
2124                 goto leave;
2125         }
2126
2127         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2128                                               orphan_dir_bh, name,
2129                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2130         if (status < 0) {
2131                 mlog_errno(status);
2132                 goto leave;
2133         }
2134
2135 leave:
2136         if (orphan_dir_inode)
2137                 iput(orphan_dir_inode);
2138
2139         if (orphan_dir_bh)
2140                 brelse(orphan_dir_bh);
2141
2142         mlog_exit(status);
2143         return status;
2144 }
2145
2146 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2147                             struct ocfs2_journal_handle *handle,
2148                             struct inode *inode,
2149                             struct ocfs2_dinode *fe,
2150                             char *name,
2151                             struct buffer_head *de_bh)
2152 {
2153         struct inode *orphan_dir_inode = NULL;
2154         struct buffer_head *orphan_dir_bh = NULL;
2155         int status = 0;
2156         struct ocfs2_dinode *orphan_fe;
2157
2158         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2159
2160         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2161                                                        ORPHAN_DIR_SYSTEM_INODE,
2162                                                        osb->slot_num);
2163         if (!orphan_dir_inode) {
2164                 status = -ENOENT;
2165                 mlog_errno(status);
2166                 goto leave;
2167         }
2168
2169         status = ocfs2_read_block(osb,
2170                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2171                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2172                                   orphan_dir_inode);
2173         if (status < 0) {
2174                 mlog_errno(status);
2175                 goto leave;
2176         }
2177
2178         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2179                                       OCFS2_JOURNAL_ACCESS_WRITE);
2180         if (status < 0) {
2181                 mlog_errno(status);
2182                 goto leave;
2183         }
2184
2185         /* we're a cluster, and nlink can change on disk from
2186          * underneath us... */
2187         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2188         if (S_ISDIR(inode->i_mode))
2189                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2190         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2191
2192         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2193         if (status < 0) {
2194                 mlog_errno(status);
2195                 goto leave;
2196         }
2197
2198         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2199                                    OCFS2_ORPHAN_NAMELEN, inode,
2200                                    OCFS2_I(inode)->ip_blkno,
2201                                    orphan_dir_bh, de_bh);
2202         if (status < 0) {
2203                 mlog_errno(status);
2204                 goto leave;
2205         }
2206
2207         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2208
2209         /* Record which orphan dir our inode now resides
2210          * in. delete_inode will use this to determine which orphan
2211          * dir to lock. */
2212         spin_lock(&OCFS2_I(inode)->ip_lock);
2213         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2214         spin_unlock(&OCFS2_I(inode)->ip_lock);
2215
2216         mlog(0, "Inode %llu orphaned in slot %d\n",
2217              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2218
2219 leave:
2220         if (orphan_dir_inode)
2221                 iput(orphan_dir_inode);
2222
2223         if (orphan_dir_bh)
2224                 brelse(orphan_dir_bh);
2225
2226         mlog_exit(status);
2227         return status;
2228 }
2229
2230 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2231 int ocfs2_orphan_del(struct ocfs2_super *osb,
2232                      struct ocfs2_journal_handle *handle,
2233                      struct inode *orphan_dir_inode,
2234                      struct inode *inode,
2235                      struct buffer_head *orphan_dir_bh)
2236 {
2237         char name[OCFS2_ORPHAN_NAMELEN + 1];
2238         struct ocfs2_dinode *orphan_fe;
2239         int status = 0;
2240         struct buffer_head *target_de_bh = NULL;
2241         struct ocfs2_dir_entry *target_de = NULL;
2242
2243         mlog_entry_void();
2244
2245         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2246         if (status < 0) {
2247                 mlog_errno(status);
2248                 goto leave;
2249         }
2250
2251         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2252              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2253              OCFS2_ORPHAN_NAMELEN);
2254
2255         /* find it's spot in the orphan directory */
2256         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2257                                         orphan_dir_inode, &target_de);
2258         if (!target_de_bh) {
2259                 status = -ENOENT;
2260                 mlog_errno(status);
2261                 goto leave;
2262         }
2263
2264         /* remove it from the orphan directory */
2265         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2266                                     target_de_bh);
2267         if (status < 0) {
2268                 mlog_errno(status);
2269                 goto leave;
2270         }
2271
2272         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2273                                       OCFS2_JOURNAL_ACCESS_WRITE);
2274         if (status < 0) {
2275                 mlog_errno(status);
2276                 goto leave;
2277         }
2278
2279         /* do the i_nlink dance! :) */
2280         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2281         if (S_ISDIR(inode->i_mode))
2282                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2283         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2284
2285         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto leave;
2289         }
2290
2291 leave:
2292         if (target_de_bh)
2293                 brelse(target_de_bh);
2294
2295         mlog_exit(status);
2296         return status;
2297 }
2298
2299 struct inode_operations ocfs2_dir_iops = {
2300         .create         = ocfs2_create,
2301         .lookup         = ocfs2_lookup,
2302         .link           = ocfs2_link,
2303         .unlink         = ocfs2_unlink,
2304         .rmdir          = ocfs2_unlink,
2305         .symlink        = ocfs2_symlink,
2306         .mkdir          = ocfs2_mkdir,
2307         .mknod          = ocfs2_mknod,
2308         .rename         = ocfs2_rename,
2309         .setattr        = ocfs2_setattr,
2310         .getattr        = ocfs2_getattr,
2311 };