err.no 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(handle_t *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(handle_t *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                               handle_t *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                               handle_t *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_prepare_orphan_dir(struct ocfs2_super *osb,
 108                                     struct inode **ret_orphan_dir,
 109                                     struct inode *inode,
 110                                     char *name,
 111                                     struct buffer_head **de_bh);
 112
 113 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 114                             handle_t *handle,
 115                             struct inode *inode,
 116                             struct ocfs2_dinode *fe,
 117                             char *name,
 118                             struct buffer_head *de_bh,
 119                             struct inode *orphan_dir_inode);
 120
 121 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 122                                      handle_t *handle,
 123                                      struct inode *inode,
 124                                      const char *symname);
 125
 126 static inline int ocfs2_add_entry(handle_t *handle,
 127                                   struct dentry *dentry,
 128                                   struct inode *inode, u64 blkno,
 129                                   struct buffer_head *parent_fe_bh,
 130                                   struct buffer_head *insert_bh)
 131 {
 132         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 133                                  dentry->d_name.name, dentry->d_name.len,
 134                                  inode, blkno, parent_fe_bh, insert_bh);
 135 }
 136
 137 /* An orphan dir name is an 8 byte value, printed as a hex string */
 138 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 139
 140 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 141                                    struct nameidata *nd)
 142 {
 143         int status;
 144         u64 blkno;
 145         struct buffer_head *dirent_bh = NULL;
 146         struct inode *inode = NULL;
 147         struct dentry *ret;
 148         struct ocfs2_dir_entry *dirent;
 149         struct ocfs2_inode_info *oi;
 150
 151         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 152                    dentry->d_name.len, dentry->d_name.name);
 153
 154         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 155                 ret = ERR_PTR(-ENAMETOOLONG);
 156                 goto bail;
 157         }
 158
 159         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 160              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 161
 162         status = ocfs2_meta_lock(dir, NULL, 0);
 163         if (status < 0) {
 164                 if (status != -ENOENT)
 165                         mlog_errno(status);
 166                 ret = ERR_PTR(status);
 167                 goto bail;
 168         }
 169
 170         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 171                                           dentry->d_name.len, &blkno,
 172                                           dir, &dirent_bh, &dirent);
 173         if (status < 0)
 174                 goto bail_add;
 175
 176         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 177         if (IS_ERR(inode)) {
 178                 ret = ERR_PTR(-EACCES);
 179                 goto bail_unlock;
 180         }
 181
 182         oi = OCFS2_I(inode);
 183         /* Clear any orphaned state... If we were able to look up the
 184          * inode from a directory, it certainly can't be orphaned. We
 185          * might have the bad state from a node which intended to
 186          * orphan this inode but crashed before it could commit the
 187          * unlink. */
 188         spin_lock(&oi->ip_lock);
 189         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 190         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 191         spin_unlock(&oi->ip_lock);
 192
 193 bail_add:
 194         dentry->d_op = &ocfs2_dentry_ops;
 195         ret = d_splice_alias(inode, dentry);
 196
 197         if (inode) {
 198                 /*
 199                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 200                  * dentry, it will d_move() it on top of ourse. The
 201                  * return value will indicate this however, so in
 202                  * those cases, we switch them around for the locking
 203                  * code.
 204                  *
 205                  * NOTE: This dentry already has ->d_op set from
 206                  * ocfs2_get_parent() and ocfs2_get_dentry()
 207                  */
 208                 if (ret)
 209                         dentry = ret;
 210
 211                 status = ocfs2_dentry_attach_lock(dentry, inode,
 212                                                   OCFS2_I(dir)->ip_blkno);
 213                 if (status) {
 214                         mlog_errno(status);
 215                         ret = ERR_PTR(status);
 216                         goto bail_unlock;
 217                 }
 218         }
 219
 220 bail_unlock:
 221         /* Don't drop the cluster lock until *after* the d_add --
 222          * unlink on another node will message us to remove that
 223          * dentry under this lock so otherwise we can race this with
 224          * the vote thread and have a stale dentry. */
 225         ocfs2_meta_unlock(dir, 0);
 226
 227 bail:
 228         if (dirent_bh)
 229                 brelse(dirent_bh);
 230
 231         mlog_exit_ptr(ret);
 232
 233         return ret;
 234 }
 235
 236 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 237                               handle_t *handle,
 238                               struct inode *parent,
 239                               struct inode *inode,
 240                               struct buffer_head *fe_bh,
 241                               struct ocfs2_alloc_context *data_ac)
 242 {
 243         int status;
 244         struct buffer_head *new_bh = NULL;
 245         struct ocfs2_dir_entry *de = NULL;
 246
 247         mlog_entry_void();
 248
 249         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 250                                      data_ac, NULL, &new_bh);
 251         if (status < 0) {
 252                 mlog_errno(status);
 253                 goto bail;
 254         }
 255
 256         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 257
 258         status = ocfs2_journal_access(handle, inode, new_bh,
 259                                       OCFS2_JOURNAL_ACCESS_CREATE);
 260         if (status < 0) {
 261                 mlog_errno(status);
 262                 goto bail;
 263         }
 264         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 265
 266         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 267         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 268         de->name_len = 1;
 269         de->rec_len =
 270                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 271         strcpy(de->name, ".");
 272         ocfs2_set_de_type(de, S_IFDIR);
 273         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 274         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 275         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 276                                   OCFS2_DIR_REC_LEN(1));
 277         de->name_len = 2;
 278         strcpy(de->name, "..");
 279         ocfs2_set_de_type(de, S_IFDIR);
 280
 281         status = ocfs2_journal_dirty(handle, new_bh);
 282         if (status < 0) {
 283                 mlog_errno(status);
 284                 goto bail;
 285         }
 286
 287         i_size_write(inode, inode->i_sb->s_blocksize);
 288         inode->i_nlink = 2;
 289         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 290         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 291         if (status < 0) {
 292                 mlog_errno(status);
 293                 goto bail;
 294         }
 295
 296         status = 0;
 297 bail:
 298         if (new_bh)
 299                 brelse(new_bh);
 300
 301         mlog_exit(status);
 302         return status;
 303 }
 304
 305 static int ocfs2_mknod(struct inode *dir,
 306                        struct dentry *dentry,
 307                        int mode,
 308                        dev_t dev)
 309 {
 310         int status = 0;
 311         struct buffer_head *parent_fe_bh = NULL;
 312         handle_t *handle = NULL;
 313         struct ocfs2_super *osb;
 314         struct ocfs2_dinode *dirfe;
 315         struct buffer_head *new_fe_bh = NULL;
 316         struct buffer_head *de_bh = NULL;
 317         struct inode *inode = NULL;
 318         struct ocfs2_alloc_context *inode_ac = NULL;
 319         struct ocfs2_alloc_context *data_ac = NULL;
 320
 321         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 322                    (unsigned long)dev, dentry->d_name.len,
 323                    dentry->d_name.name);
 324
 325         /* get our super block */
 326         osb = OCFS2_SB(dir->i_sb);
 327
 328         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 329         if (status < 0) {
 330                 if (status != -ENOENT)
 331                         mlog_errno(status);
 332                 return status;
 333         }
 334
 335         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 336                 status = -EMLINK;
 337                 goto leave;
 338         }
 339
 340         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 341         if (!dirfe->i_links_count) {
 342                 /* can't make a file in a deleted directory. */
 343                 status = -ENOENT;
 344                 goto leave;
 345         }
 346
 347         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 348                                            dentry->d_name.len);
 349         if (status)
 350                 goto leave;
 351
 352         /* get a spot inside the dir. */
 353         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 354                                               dentry->d_name.name,
 355                                               dentry->d_name.len, &de_bh);
 356         if (status < 0) {
 357                 mlog_errno(status);
 358                 goto leave;
 359         }
 360
 361         /* reserve an inode spot */
 362         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 363         if (status < 0) {
 364                 if (status != -ENOSPC)
 365                         mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         /* are we making a directory? If so, reserve a cluster for his
 370          * 1st extent. */
 371         if (S_ISDIR(mode)) {
 372                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 373                 if (status < 0) {
 374                         if (status != -ENOSPC)
 375                                 mlog_errno(status);
 376                         goto leave;
 377                 }
 378         }
 379
 380         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 381         if (IS_ERR(handle)) {
 382                 status = PTR_ERR(handle);
 383                 handle = NULL;
 384                 mlog_errno(status);
 385                 goto leave;
 386         }
 387
 388         /* do the real work now. */
 389         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 390                                     &new_fe_bh, parent_fe_bh, handle,
 391                                     &inode, inode_ac);
 392         if (status < 0) {
 393                 mlog_errno(status);
 394                 goto leave;
 395         }
 396
 397         if (S_ISDIR(mode)) {
 398                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 399                                             new_fe_bh, data_ac);
 400                 if (status < 0) {
 401                         mlog_errno(status);
 402                         goto leave;
 403                 }
 404
 405                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 406                                               OCFS2_JOURNAL_ACCESS_WRITE);
 407                 if (status < 0) {
 408                         mlog_errno(status);
 409                         goto leave;
 410                 }
 411                 le16_add_cpu(&dirfe->i_links_count, 1);
 412                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 413                 if (status < 0) {
 414                         mlog_errno(status);
 415                         goto leave;
 416                 }
 417                 inc_nlink(dir);
 418         }
 419
 420         status = ocfs2_add_entry(handle, dentry, inode,
 421                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 422                                  de_bh);
 423         if (status < 0) {
 424                 mlog_errno(status);
 425                 goto leave;
 426         }
 427
 428         status = ocfs2_dentry_attach_lock(dentry, inode,
 429                                           OCFS2_I(dir)->ip_blkno);
 430         if (status) {
 431                 mlog_errno(status);
 432                 goto leave;
 433         }
 434
 435         insert_inode_hash(inode);
 436         dentry->d_op = &ocfs2_dentry_ops;
 437         d_instantiate(dentry, inode);
 438         status = 0;
 439 leave:
 440         if (handle)
 441                 ocfs2_commit_trans(osb, handle);
 442
 443         ocfs2_meta_unlock(dir, 1);
 444
 445         if (status == -ENOSPC)
 446                 mlog(0, "Disk is full\n");
 447
 448         if (new_fe_bh)
 449                 brelse(new_fe_bh);
 450
 451         if (de_bh)
 452                 brelse(de_bh);
 453
 454         if (parent_fe_bh)
 455                 brelse(parent_fe_bh);
 456
 457         if ((status < 0) && inode)
 458                 iput(inode);
 459
 460         if (inode_ac)
 461                 ocfs2_free_alloc_context(inode_ac);
 462
 463         if (data_ac)
 464                 ocfs2_free_alloc_context(data_ac);
 465
 466         mlog_exit(status);
 467
 468         return status;
 469 }
 470
 471 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 472                               struct inode *dir,
 473                               struct dentry *dentry, int mode,
 474                               dev_t dev,
 475                               struct buffer_head **new_fe_bh,
 476                               struct buffer_head *parent_fe_bh,
 477                               handle_t *handle,
 478                               struct inode **ret_inode,
 479                               struct ocfs2_alloc_context *inode_ac)
 480 {
 481         int status = 0;
 482         struct ocfs2_dinode *fe = NULL;
 483         struct ocfs2_extent_list *fel;
 484         u64 fe_blkno = 0;
 485         u16 suballoc_bit;
 486         struct inode *inode = NULL;
 487
 488         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 489                    (unsigned long)dev, dentry->d_name.len,
 490                    dentry->d_name.name);
 491
 492         *new_fe_bh = NULL;
 493         *ret_inode = NULL;
 494
 495         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 496                                        &fe_blkno);
 497         if (status < 0) {
 498                 mlog_errno(status);
 499                 goto leave;
 500         }
 501
 502         inode = new_inode(dir->i_sb);
 503         if (IS_ERR(inode)) {
 504                 status = PTR_ERR(inode);
 505                 mlog(ML_ERROR, "new_inode failed!\n");
 506                 goto leave;
 507         }
 508
 509         /* populate as many fields early on as possible - many of
 510          * these are used by the support functions here and in
 511          * callers. */
 512         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 513         OCFS2_I(inode)->ip_blkno = fe_blkno;
 514         if (S_ISDIR(mode))
 515                 inode->i_nlink = 2;
 516         else
 517                 inode->i_nlink = 1;
 518         inode->i_mode = mode;
 519         spin_lock(&osb->osb_lock);
 520         inode->i_generation = osb->s_next_generation++;
 521         spin_unlock(&osb->osb_lock);
 522
 523         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 524         if (!*new_fe_bh) {
 525                 status = -EIO;
 526                 mlog_errno(status);
 527                 goto leave;
 528         }
 529         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 530
 531         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 532                                       OCFS2_JOURNAL_ACCESS_CREATE);
 533         if (status < 0) {
 534                 mlog_errno(status);
 535                 goto leave;
 536         }
 537
 538         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 539         memset(fe, 0, osb->sb->s_blocksize);
 540
 541         fe->i_generation = cpu_to_le32(inode->i_generation);
 542         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 543         fe->i_blkno = cpu_to_le64(fe_blkno);
 544         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 545         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 546         fe->i_uid = cpu_to_le32(current->fsuid);
 547         if (dir->i_mode & S_ISGID) {
 548                 fe->i_gid = cpu_to_le32(dir->i_gid);
 549                 if (S_ISDIR(mode))
 550                         mode |= S_ISGID;
 551         } else
 552                 fe->i_gid = cpu_to_le32(current->fsgid);
 553         fe->i_mode = cpu_to_le16(mode);
 554         if (S_ISCHR(mode) || S_ISBLK(mode))
 555                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 556
 557         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 558
 559         fe->i_last_eb_blk = 0;
 560         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 561         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 562         fe->i_atime = fe->i_ctime = fe->i_mtime =
 563                 cpu_to_le64(CURRENT_TIME.tv_sec);
 564         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 565                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 566         fe->i_dtime = 0;
 567
 568         fel = &fe->id2.i_list;
 569         fel->l_tree_depth = 0;
 570         fel->l_next_free_rec = 0;
 571         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 572
 573         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 574         if (status < 0) {
 575                 mlog_errno(status);
 576                 goto leave;
 577         }
 578
 579         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 580                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 581                      "i_blkno=%llu, i_ino=%lu\n",
 582                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 583                      (unsigned long long)fe->i_blkno, inode->i_ino);
 584                 BUG();
 585         }
 586
 587         ocfs2_inode_set_new(osb, inode);
 588         if (!ocfs2_mount_local(osb)) {
 589                 status = ocfs2_create_new_inode_locks(inode);
 590                 if (status < 0)
 591                         mlog_errno(status);
 592         }
 593
 594         status = 0; /* error in ocfs2_create_new_inode_locks is not
 595                      * critical */
 596
 597         *ret_inode = inode;
 598 leave:
 599         if (status < 0) {
 600                 if (*new_fe_bh) {
 601                         brelse(*new_fe_bh);
 602                         *new_fe_bh = NULL;
 603                 }
 604                 if (inode)
 605                         iput(inode);
 606         }
 607
 608         mlog_exit(status);
 609         return status;
 610 }
 611
 612 static int ocfs2_mkdir(struct inode *dir,
 613                        struct dentry *dentry,
 614                        int mode)
 615 {
 616         int ret;
 617
 618         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 619                    dentry->d_name.len, dentry->d_name.name);
 620         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 621         mlog_exit(ret);
 622
 623         return ret;
 624 }
 625
 626 static int ocfs2_create(struct inode *dir,
 627                         struct dentry *dentry,
 628                         int mode,
 629                         struct nameidata *nd)
 630 {
 631         int ret;
 632
 633         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 634                    dentry->d_name.len, dentry->d_name.name);
 635         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 636         mlog_exit(ret);
 637
 638         return ret;
 639 }
 640
 641 static int ocfs2_link(struct dentry *old_dentry,
 642                       struct inode *dir,
 643                       struct dentry *dentry)
 644 {
 645         handle_t *handle;
 646         struct inode *inode = old_dentry->d_inode;
 647         int err;
 648         struct buffer_head *fe_bh = NULL;
 649         struct buffer_head *parent_fe_bh = NULL;
 650         struct buffer_head *de_bh = NULL;
 651         struct ocfs2_dinode *fe = NULL;
 652         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 653
 654         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 655                    old_dentry->d_name.len, old_dentry->d_name.name,
 656                    dentry->d_name.len, dentry->d_name.name);
 657
 658         if (S_ISDIR(inode->i_mode))
 659                 return -EPERM;
 660
 661         err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 662         if (err < 0) {
 663                 if (err != -ENOENT)
 664                         mlog_errno(err);
 665                 return err;
 666         }
 667
 668         if (!dir->i_nlink) {
 669                 err = -ENOENT;
 670                 goto out;
 671         }
 672
 673         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 674                                         dentry->d_name.len);
 675         if (err)
 676                 goto out;
 677
 678         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 679                                            dentry->d_name.name,
 680                                            dentry->d_name.len, &de_bh);
 681         if (err < 0) {
 682                 mlog_errno(err);
 683                 goto out;
 684         }
 685
 686         err = ocfs2_meta_lock(inode, &fe_bh, 1);
 687         if (err < 0) {
 688                 if (err != -ENOENT)
 689                         mlog_errno(err);
 690                 goto out;
 691         }
 692
 693         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 694         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 695                 err = -EMLINK;
 696                 goto out_unlock_inode;
 697         }
 698
 699         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 700         if (IS_ERR(handle)) {
 701                 err = PTR_ERR(handle);
 702                 handle = NULL;
 703                 mlog_errno(err);
 704                 goto out_unlock_inode;
 705         }
 706
 707         err = ocfs2_journal_access(handle, inode, fe_bh,
 708                                    OCFS2_JOURNAL_ACCESS_WRITE);
 709         if (err < 0) {
 710                 mlog_errno(err);
 711                 goto out_commit;
 712         }
 713
 714         inc_nlink(inode);
 715         inode->i_ctime = CURRENT_TIME;
 716         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 717         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 718         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 719
 720         err = ocfs2_journal_dirty(handle, fe_bh);
 721         if (err < 0) {
 722                 le16_add_cpu(&fe->i_links_count, -1);
 723                 drop_nlink(inode);
 724                 mlog_errno(err);
 725                 goto out_commit;
 726         }
 727
 728         err = ocfs2_add_entry(handle, dentry, inode,
 729                               OCFS2_I(inode)->ip_blkno,
 730                               parent_fe_bh, de_bh);
 731         if (err) {
 732                 le16_add_cpu(&fe->i_links_count, -1);
 733                 drop_nlink(inode);
 734                 mlog_errno(err);
 735                 goto out_commit;
 736         }
 737
 738         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 739         if (err) {
 740                 mlog_errno(err);
 741                 goto out_commit;
 742         }
 743
 744         atomic_inc(&inode->i_count);
 745         dentry->d_op = &ocfs2_dentry_ops;
 746         d_instantiate(dentry, inode);
 747
 748 out_commit:
 749         ocfs2_commit_trans(osb, handle);
 750 out_unlock_inode:
 751         ocfs2_meta_unlock(inode, 1);
 752
 753 out:
 754         ocfs2_meta_unlock(dir, 1);
 755
 756         if (de_bh)
 757                 brelse(de_bh);
 758         if (fe_bh)
 759                 brelse(fe_bh);
 760         if (parent_fe_bh)
 761                 brelse(parent_fe_bh);
 762
 763         mlog_exit(err);
 764
 765         return err;
 766 }
 767
 768 /*
 769  * Takes and drops an exclusive lock on the given dentry. This will
 770  * force other nodes to drop it.
 771  */
 772 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 773 {
 774         int ret;
 775
 776         ret = ocfs2_dentry_lock(dentry, 1);
 777         if (ret)
 778                 mlog_errno(ret);
 779         else
 780                 ocfs2_dentry_unlock(dentry, 1);
 781
 782         return ret;
 783 }
 784
 785 static inline int inode_is_unlinkable(struct inode *inode)
 786 {
 787         if (S_ISDIR(inode->i_mode)) {
 788                 if (inode->i_nlink == 2)
 789                         return 1;
 790                 return 0;
 791         }
 792
 793         if (inode->i_nlink == 1)
 794                 return 1;
 795         return 0;
 796 }
 797
 798 static int ocfs2_unlink(struct inode *dir,
 799                         struct dentry *dentry)
 800 {
 801         int status;
 802         int child_locked = 0;
 803         struct inode *inode = dentry->d_inode;
 804         struct inode *orphan_dir = NULL;
 805         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 806         u64 blkno;
 807         struct ocfs2_dinode *fe = NULL;
 808         struct buffer_head *fe_bh = NULL;
 809         struct buffer_head *parent_node_bh = NULL;
 810         handle_t *handle = NULL;
 811         struct ocfs2_dir_entry *dirent = NULL;
 812         struct buffer_head *dirent_bh = NULL;
 813         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 814         struct buffer_head *orphan_entry_bh = NULL;
 815
 816         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 817                    dentry->d_name.len, dentry->d_name.name);
 818
 819         BUG_ON(dentry->d_parent->d_inode != dir);
 820
 821         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 822
 823         if (inode == osb->root_inode) {
 824                 mlog(0, "Cannot delete the root directory\n");
 825                 return -EPERM;
 826         }
 827
 828         status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
 829         if (status < 0) {
 830                 if (status != -ENOENT)
 831                         mlog_errno(status);
 832                 return status;
 833         }
 834
 835         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 836                                           dentry->d_name.len, &blkno,
 837                                           dir, &dirent_bh, &dirent);
 838         if (status < 0) {
 839                 if (status != -ENOENT)
 840                         mlog_errno(status);
 841                 goto leave;
 842         }
 843
 844         if (OCFS2_I(inode)->ip_blkno != blkno) {
 845                 status = -ENOENT;
 846
 847                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 848                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 849                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 850                 goto leave;
 851         }
 852
 853         status = ocfs2_meta_lock(inode, &fe_bh, 1);
 854         if (status < 0) {
 855                 if (status != -ENOENT)
 856                         mlog_errno(status);
 857                 goto leave;
 858         }
 859         child_locked = 1;
 860
 861         if (S_ISDIR(inode->i_mode)) {
 862                 if (!ocfs2_empty_dir(inode)) {
 863                         status = -ENOTEMPTY;
 864                         goto leave;
 865                 } else if (inode->i_nlink != 2) {
 866                         status = -ENOTEMPTY;
 867                         goto leave;
 868                 }
 869         }
 870
 871         status = ocfs2_remote_dentry_delete(dentry);
 872         if (status < 0) {
 873                 /* This vote should succeed under all normal
 874                  * circumstances. */
 875                 mlog_errno(status);
 876                 goto leave;
 877         }
 878
 879         if (inode_is_unlinkable(inode)) {
 880                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 881                                                   orphan_name,
 882                                                   &orphan_entry_bh);
 883                 if (status < 0) {
 884                         mlog_errno(status);
 885                         goto leave;
 886                 }
 887         }
 888
 889         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 890         if (IS_ERR(handle)) {
 891                 status = PTR_ERR(handle);
 892                 handle = NULL;
 893                 mlog_errno(status);
 894                 goto leave;
 895         }
 896
 897         status = ocfs2_journal_access(handle, inode, fe_bh,
 898                                       OCFS2_JOURNAL_ACCESS_WRITE);
 899         if (status < 0) {
 900                 mlog_errno(status);
 901                 goto leave;
 902         }
 903
 904         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 905
 906         if (inode_is_unlinkable(inode)) {
 907                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 908                                           orphan_entry_bh, orphan_dir);
 909                 if (status < 0) {
 910                         mlog_errno(status);
 911                         goto leave;
 912                 }
 913         }
 914
 915         /* delete the name from the parent dir */
 916         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 917         if (status < 0) {
 918                 mlog_errno(status);
 919                 goto leave;
 920         }
 921
 922         if (S_ISDIR(inode->i_mode))
 923                 drop_nlink(inode);
 924         drop_nlink(inode);
 925         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 926
 927         status = ocfs2_journal_dirty(handle, fe_bh);
 928         if (status < 0) {
 929                 mlog_errno(status);
 930                 goto leave;
 931         }
 932
 933         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 934         if (S_ISDIR(inode->i_mode))
 935                 drop_nlink(dir);
 936
 937         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 938         if (status < 0) {
 939                 mlog_errno(status);
 940                 if (S_ISDIR(inode->i_mode))
 941                         inc_nlink(dir);
 942         }
 943
 944 leave:
 945         if (handle)
 946                 ocfs2_commit_trans(osb, handle);
 947
 948         if (child_locked)
 949                 ocfs2_meta_unlock(inode, 1);
 950
 951         ocfs2_meta_unlock(dir, 1);
 952
 953         if (orphan_dir) {
 954                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 955                 ocfs2_meta_unlock(orphan_dir, 1);
 956                 mutex_unlock(&orphan_dir->i_mutex);
 957                 iput(orphan_dir);
 958         }
 959
 960         if (fe_bh)
 961                 brelse(fe_bh);
 962
 963         if (dirent_bh)
 964                 brelse(dirent_bh);
 965
 966         if (parent_node_bh)
 967                 brelse(parent_node_bh);
 968
 969         if (orphan_entry_bh)
 970                 brelse(orphan_entry_bh);
 971
 972         mlog_exit(status);
 973
 974         return status;
 975 }
 976
 977 /*
 978  * The only place this should be used is rename!
 979  * if they have the same id, then the 1st one is the only one locked.
 980  */
 981 static int ocfs2_double_lock(struct ocfs2_super *osb,
 982                              struct buffer_head **bh1,
 983                              struct inode *inode1,
 984                              struct buffer_head **bh2,
 985                              struct inode *inode2)
 986 {
 987         int status;
 988         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 989         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 990         struct buffer_head **tmpbh;
 991         struct inode *tmpinode;
 992
 993         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 994                    (unsigned long long)oi1->ip_blkno,
 995                    (unsigned long long)oi2->ip_blkno);
 996
 997         if (*bh1)
 998                 *bh1 = NULL;
 999         if (*bh2)
1000                 *bh2 = NULL;
1001
1002         /* we always want to lock the one with the lower lockid first. */
1003         if (oi1->ip_blkno != oi2->ip_blkno) {
1004                 if (oi1->ip_blkno < oi2->ip_blkno) {
1005                         /* switch id1 and id2 around */
1006                         mlog(0, "switching them around...\n");
1007                         tmpbh = bh2;
1008                         bh2 = bh1;
1009                         bh1 = tmpbh;
1010
1011                         tmpinode = inode2;
1012                         inode2 = inode1;
1013                         inode1 = tmpinode;
1014                 }
1015                 /* lock id2 */
1016                 status = ocfs2_meta_lock(inode2, bh2, 1);
1017                 if (status < 0) {
1018                         if (status != -ENOENT)
1019                                 mlog_errno(status);
1020                         goto bail;
1021                 }
1022         }
1023
1024         /* lock id1 */
1025         status = ocfs2_meta_lock(inode1, bh1, 1);
1026         if (status < 0) {
1027                 /*
1028                  * An error return must mean that no cluster locks
1029                  * were held on function exit.
1030                  */
1031                 if (oi1->ip_blkno != oi2->ip_blkno)
1032                         ocfs2_meta_unlock(inode2, 1);
1033
1034                 if (status != -ENOENT)
1035                         mlog_errno(status);
1036         }
1037
1038 bail:
1039         mlog_exit(status);
1040         return status;
1041 }
1042
1043 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1044 {
1045         ocfs2_meta_unlock(inode1, 1);
1046
1047         if (inode1 != inode2)
1048                 ocfs2_meta_unlock(inode2, 1);
1049 }
1050
1051 #define PARENT_INO(buffer) \
1052         ((struct ocfs2_dir_entry *) \
1053          ((char *)buffer + \
1054           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1055
1056 static int ocfs2_rename(struct inode *old_dir,
1057                         struct dentry *old_dentry,
1058                         struct inode *new_dir,
1059                         struct dentry *new_dentry)
1060 {
1061         int status = 0, rename_lock = 0, parents_locked = 0;
1062         int old_child_locked = 0, new_child_locked = 0;
1063         struct inode *old_inode = old_dentry->d_inode;
1064         struct inode *new_inode = new_dentry->d_inode;
1065         struct inode *orphan_dir = NULL;
1066         struct ocfs2_dinode *newfe = NULL;
1067         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1068         struct buffer_head *orphan_entry_bh = NULL;
1069         struct buffer_head *newfe_bh = NULL;
1070         struct buffer_head *old_inode_bh = NULL;
1071         struct buffer_head *insert_entry_bh = NULL;
1072         struct ocfs2_super *osb = NULL;
1073         u64 newfe_blkno;
1074         handle_t *handle = NULL;
1075         struct buffer_head *old_dir_bh = NULL;
1076         struct buffer_head *new_dir_bh = NULL;
1077         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1078                                                                // and new_dentry
1079         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1080         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1081                                                     // this is the 1st dirent bh
1082         nlink_t old_dir_nlink = old_dir->i_nlink;
1083
1084         /* At some point it might be nice to break this function up a
1085          * bit. */
1086
1087         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1088                    old_dir, old_dentry, new_dir, new_dentry,
1089                    old_dentry->d_name.len, old_dentry->d_name.name,
1090                    new_dentry->d_name.len, new_dentry->d_name.name);
1091
1092         osb = OCFS2_SB(old_dir->i_sb);
1093
1094         if (new_inode) {
1095                 if (!igrab(new_inode))
1096                         BUG();
1097         }
1098
1099         /* Assume a directory hierarchy thusly:
1100          * a/b/c
1101          * a/d
1102          * a,b,c, and d are all directories.
1103          *
1104          * from cwd of 'a' on both nodes:
1105          * node1: mv b/c d
1106          * node2: mv d   b/c
1107          *
1108          * And that's why, just like the VFS, we need a file system
1109          * rename lock. */
1110         if (old_dentry != new_dentry) {
1111                 status = ocfs2_rename_lock(osb);
1112                 if (status < 0) {
1113                         mlog_errno(status);
1114                         goto bail;
1115                 }
1116                 rename_lock = 1;
1117         }
1118
1119         /* if old and new are the same, this'll just do one lock. */
1120         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1121                                    &new_dir_bh, new_dir);
1122         if (status < 0) {
1123                 mlog_errno(status);
1124                 goto bail;
1125         }
1126         parents_locked = 1;
1127
1128         /* make sure both dirs have bhs
1129          * get an extra ref on old_dir_bh if old==new */
1130         if (!new_dir_bh) {
1131                 if (old_dir_bh) {
1132                         new_dir_bh = old_dir_bh;
1133                         get_bh(new_dir_bh);
1134                 } else {
1135                         mlog(ML_ERROR, "no old_dir_bh!\n");
1136                         status = -EIO;
1137                         goto bail;
1138                 }
1139         }
1140
1141         /*
1142          * Aside from allowing a meta data update, the locking here
1143          * also ensures that the vote thread on other nodes won't have
1144          * to concurrently downconvert the inode and the dentry locks.
1145          */
1146         status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1147         if (status < 0) {
1148                 if (status != -ENOENT)
1149                         mlog_errno(status);
1150                 goto bail;
1151         }
1152         old_child_locked = 1;
1153
1154         status = ocfs2_remote_dentry_delete(old_dentry);
1155         if (status < 0) {
1156                 mlog_errno(status);
1157                 goto bail;
1158         }
1159
1160         if (S_ISDIR(old_inode->i_mode)) {
1161                 status = -EIO;
1162                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1163                 if (!old_inode_de_bh)
1164                         goto bail;
1165
1166                 status = -EIO;
1167                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1168                     OCFS2_I(old_dir)->ip_blkno)
1169                         goto bail;
1170                 status = -EMLINK;
1171                 if (!new_inode && new_dir!=old_dir &&
1172                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1173                         goto bail;
1174         }
1175
1176         status = -ENOENT;
1177         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1178                                      old_dentry->d_name.len,
1179                                      old_dir, &old_de);
1180         if (!old_de_bh)
1181                 goto bail;
1182
1183         /*
1184          *  Check for inode number is _not_ due to possible IO errors.
1185          *  We might rmdir the source, keep it as pwd of some process
1186          *  and merrily kill the link to whatever was created under the
1187          *  same name. Goodbye sticky bit ;-<
1188          */
1189         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1190                 goto bail;
1191
1192         /* check if the target already exists (in which case we need
1193          * to delete it */
1194         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1195                                           new_dentry->d_name.len,
1196                                           &newfe_blkno, new_dir, &new_de_bh,
1197                                           &new_de);
1198         /* The only error we allow here is -ENOENT because the new
1199          * file not existing is perfectly valid. */
1200         if ((status < 0) && (status != -ENOENT)) {
1201                 /* If we cannot find the file specified we should just */
1202                 /* return the error... */
1203                 mlog_errno(status);
1204                 goto bail;
1205         }
1206
1207         if (!new_de && new_inode)
1208                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1209                      "directory!", new_inode->i_ino);
1210
1211         /* In case we need to overwrite an existing file, we blow it
1212          * away first */
1213         if (new_de) {
1214                 /* VFS didn't think there existed an inode here, but
1215                  * someone else in the cluster must have raced our
1216                  * rename to create one. Today we error cleanly, in
1217                  * the future we should consider calling iget to build
1218                  * a new struct inode for this entry. */
1219                 if (!new_inode) {
1220                         status = -EACCES;
1221
1222                         mlog(0, "We found an inode for name %.*s but VFS "
1223                              "didn't give us one.\n", new_dentry->d_name.len,
1224                              new_dentry->d_name.name);
1225                         goto bail;
1226                 }
1227
1228                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1229                         status = -EACCES;
1230
1231                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1232                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1233                              (unsigned long long)newfe_blkno,
1234                              OCFS2_I(new_inode)->ip_flags);
1235                         goto bail;
1236                 }
1237
1238                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1239                 if (status < 0) {
1240                         if (status != -ENOENT)
1241                                 mlog_errno(status);
1242                         goto bail;
1243                 }
1244                 new_child_locked = 1;
1245
1246                 status = ocfs2_remote_dentry_delete(new_dentry);
1247                 if (status < 0) {
1248                         mlog_errno(status);
1249                         goto bail;
1250                 }
1251
1252                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1253
1254                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1255                      "newfebh=%p bhblocknr=%llu\n", new_de,
1256                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1257                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1258
1259                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1260                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1261                                                           new_inode,
1262                                                           orphan_name,
1263                                                           &orphan_entry_bh);
1264                         if (status < 0) {
1265                                 mlog_errno(status);
1266                                 goto bail;
1267                         }
1268                 }
1269         } else {
1270                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1271
1272                 status = ocfs2_check_dir_for_entry(new_dir,
1273                                                    new_dentry->d_name.name,
1274                                                    new_dentry->d_name.len);
1275                 if (status)
1276                         goto bail;
1277
1278                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1279                                                       new_dentry->d_name.name,
1280                                                       new_dentry->d_name.len,
1281                                                       &insert_entry_bh);
1282                 if (status < 0) {
1283                         mlog_errno(status);
1284                         goto bail;
1285                 }
1286         }
1287
1288         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1289         if (IS_ERR(handle)) {
1290                 status = PTR_ERR(handle);
1291                 handle = NULL;
1292                 mlog_errno(status);
1293                 goto bail;
1294         }
1295
1296         if (new_de) {
1297                 if (S_ISDIR(new_inode->i_mode)) {
1298                         if (!ocfs2_empty_dir(new_inode) ||
1299                             new_inode->i_nlink != 2) {
1300                                 status = -ENOTEMPTY;
1301                                 goto bail;
1302                         }
1303                 }
1304                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1305                                               OCFS2_JOURNAL_ACCESS_WRITE);
1306                 if (status < 0) {
1307                         mlog_errno(status);
1308                         goto bail;
1309                 }
1310
1311                 if (S_ISDIR(new_inode->i_mode) ||
1312                     (newfe->i_links_count == cpu_to_le16(1))){
1313                         status = ocfs2_orphan_add(osb, handle, new_inode,
1314                                                   newfe, orphan_name,
1315                                                   orphan_entry_bh, orphan_dir);
1316                         if (status < 0) {
1317                                 mlog_errno(status);
1318                                 goto bail;
1319                         }
1320                 }
1321
1322                 /* change the dirent to point to the correct inode */
1323                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1324                                               OCFS2_JOURNAL_ACCESS_WRITE);
1325                 if (status < 0) {
1326                         mlog_errno(status);
1327                         goto bail;
1328                 }
1329                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1330                 new_de->file_type = old_de->file_type;
1331                 new_dir->i_version++;
1332                 status = ocfs2_journal_dirty(handle, new_de_bh);
1333                 if (status < 0) {
1334                         mlog_errno(status);
1335                         goto bail;
1336                 }
1337
1338                 if (S_ISDIR(new_inode->i_mode))
1339                         newfe->i_links_count = 0;
1340                 else
1341                         le16_add_cpu(&newfe->i_links_count, -1);
1342
1343                 status = ocfs2_journal_dirty(handle, newfe_bh);
1344                 if (status < 0) {
1345                         mlog_errno(status);
1346                         goto bail;
1347                 }
1348         } else {
1349                 /* if the name was not found in new_dir, add it now */
1350                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1351                                          OCFS2_I(old_inode)->ip_blkno,
1352                                          new_dir_bh, insert_entry_bh);
1353         }
1354
1355         old_inode->i_ctime = CURRENT_TIME;
1356         mark_inode_dirty(old_inode);
1357         ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
1358
1359         /* now that the name has been added to new_dir, remove the old name */
1360         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1361         if (status < 0) {
1362                 mlog_errno(status);
1363                 goto bail;
1364         }
1365
1366         if (new_inode) {
1367                 new_inode->i_nlink--;
1368                 new_inode->i_ctime = CURRENT_TIME;
1369         }
1370         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1371         if (old_inode_de_bh) {
1372                 status = ocfs2_journal_access(handle, old_inode,
1373                                              old_inode_de_bh,
1374                                              OCFS2_JOURNAL_ACCESS_WRITE);
1375                 PARENT_INO(old_inode_de_bh->b_data) =
1376                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1377                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1378                 old_dir->i_nlink--;
1379                 if (new_inode) {
1380                         new_inode->i_nlink--;
1381                 } else {
1382                         inc_nlink(new_dir);
1383                         mark_inode_dirty(new_dir);
1384                 }
1385         }
1386         mark_inode_dirty(old_dir);
1387         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1388         if (new_inode) {
1389                 mark_inode_dirty(new_inode);
1390                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1391         }
1392
1393         if (old_dir != new_dir) {
1394                 /* Keep the same times on both directories.*/
1395                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1396
1397                 /*
1398                  * This will also pick up the i_nlink change from the
1399                  * block above.
1400                  */
1401                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1402         }
1403
1404         if (old_dir_nlink != old_dir->i_nlink) {
1405                 if (!old_dir_bh) {
1406                         mlog(ML_ERROR, "need to change nlink for old dir "
1407                              "%llu from %d to %d but bh is NULL!\n",
1408                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1409                              (int)old_dir_nlink, old_dir->i_nlink);
1410                 } else {
1411                         struct ocfs2_dinode *fe;
1412                         status = ocfs2_journal_access(handle, old_dir,
1413                                                       old_dir_bh,
1414                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1415                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1416                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1417                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1418                 }
1419         }
1420
1421         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1422         status = 0;
1423 bail:
1424         if (rename_lock)
1425                 ocfs2_rename_unlock(osb);
1426
1427         if (handle)
1428                 ocfs2_commit_trans(osb, handle);
1429
1430         if (parents_locked)
1431                 ocfs2_double_unlock(old_dir, new_dir);
1432
1433         if (old_child_locked)
1434                 ocfs2_meta_unlock(old_inode, 1);
1435
1436         if (new_child_locked)
1437                 ocfs2_meta_unlock(new_inode, 1);
1438
1439         if (orphan_dir) {
1440                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1441                 ocfs2_meta_unlock(orphan_dir, 1);
1442                 mutex_unlock(&orphan_dir->i_mutex);
1443                 iput(orphan_dir);
1444         }
1445
1446         if (new_inode)
1447                 sync_mapping_buffers(old_inode->i_mapping);
1448
1449         if (new_inode)
1450                 iput(new_inode);
1451         if (newfe_bh)
1452                 brelse(newfe_bh);
1453         if (old_inode_bh)
1454                 brelse(old_inode_bh);
1455         if (old_dir_bh)
1456                 brelse(old_dir_bh);
1457         if (new_dir_bh)
1458                 brelse(new_dir_bh);
1459         if (new_de_bh)
1460                 brelse(new_de_bh);
1461         if (old_de_bh)
1462                 brelse(old_de_bh);
1463         if (old_inode_de_bh)
1464                 brelse(old_inode_de_bh);
1465         if (orphan_entry_bh)
1466                 brelse(orphan_entry_bh);
1467         if (insert_entry_bh)
1468                 brelse(insert_entry_bh);
1469
1470         mlog_exit(status);
1471
1472         return status;
1473 }
1474
1475 /*
1476  * we expect i_size = strlen(symname). Copy symname into the file
1477  * data, including the null terminator.
1478  */
1479 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1480                                      handle_t *handle,
1481                                      struct inode *inode,
1482                                      const char *symname)
1483 {
1484         struct buffer_head **bhs = NULL;
1485         const char *c;
1486         struct super_block *sb = osb->sb;
1487         u64 p_blkno;
1488         int p_blocks;
1489         int virtual, blocks, status, i, bytes_left;
1490
1491         bytes_left = i_size_read(inode) + 1;
1492         /* we can't trust i_blocks because we're actually going to
1493          * write i_size + 1 bytes. */
1494         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1495
1496         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1497                         (unsigned long long)inode->i_blocks,
1498                         i_size_read(inode), blocks);
1499
1500         /* Sanity check -- make sure we're going to fit. */
1501         if (bytes_left >
1502             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1503                 status = -EIO;
1504                 mlog_errno(status);
1505                 goto bail;
1506         }
1507
1508         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1509         if (!bhs) {
1510                 status = -ENOMEM;
1511                 mlog_errno(status);
1512                 goto bail;
1513         }
1514
1515         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1516                                              &p_blocks);
1517         if (status < 0) {
1518                 mlog_errno(status);
1519                 goto bail;
1520         }
1521
1522         /* links can never be larger than one cluster so we know this
1523          * is all going to be contiguous, but do a sanity check
1524          * anyway. */
1525         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1526                 status = -EIO;
1527                 mlog_errno(status);
1528                 goto bail;
1529         }
1530
1531         virtual = 0;
1532         while(bytes_left > 0) {
1533                 c = &symname[virtual * sb->s_blocksize];
1534
1535                 bhs[virtual] = sb_getblk(sb, p_blkno);
1536                 if (!bhs[virtual]) {
1537                         status = -ENOMEM;
1538                         mlog_errno(status);
1539                         goto bail;
1540                 }
1541                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1542
1543                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1544                                               OCFS2_JOURNAL_ACCESS_CREATE);
1545                 if (status < 0) {
1546                         mlog_errno(status);
1547                         goto bail;
1548                 }
1549
1550                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1551
1552                 memcpy(bhs[virtual]->b_data, c,
1553                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1554                        bytes_left);
1555
1556                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1557                 if (status < 0) {
1558                         mlog_errno(status);
1559                         goto bail;
1560                 }
1561
1562                 virtual++;
1563                 p_blkno++;
1564                 bytes_left -= sb->s_blocksize;
1565         }
1566
1567         status = 0;
1568 bail:
1569
1570         if (bhs) {
1571                 for(i = 0; i < blocks; i++)
1572                         if (bhs[i])
1573                                 brelse(bhs[i]);
1574                 kfree(bhs);
1575         }
1576
1577         mlog_exit(status);
1578         return status;
1579 }
1580
1581 static int ocfs2_symlink(struct inode *dir,
1582                          struct dentry *dentry,
1583                          const char *symname)
1584 {
1585         int status, l, credits;
1586         u64 newsize;
1587         struct ocfs2_super *osb = NULL;
1588         struct inode *inode = NULL;
1589         struct super_block *sb;
1590         struct buffer_head *new_fe_bh = NULL;
1591         struct buffer_head *de_bh = NULL;
1592         struct buffer_head *parent_fe_bh = NULL;
1593         struct ocfs2_dinode *fe = NULL;
1594         struct ocfs2_dinode *dirfe;
1595         handle_t *handle = NULL;
1596         struct ocfs2_alloc_context *inode_ac = NULL;
1597         struct ocfs2_alloc_context *data_ac = NULL;
1598
1599         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1600                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1601
1602         sb = dir->i_sb;
1603         osb = OCFS2_SB(sb);
1604
1605         l = strlen(symname) + 1;
1606
1607         credits = ocfs2_calc_symlink_credits(sb);
1608
1609         /* lock the parent directory */
1610         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1611         if (status < 0) {
1612                 if (status != -ENOENT)
1613                         mlog_errno(status);
1614                 return status;
1615         }
1616
1617         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1618         if (!dirfe->i_links_count) {
1619                 /* can't make a file in a deleted directory. */
1620                 status = -ENOENT;
1621                 goto bail;
1622         }
1623
1624         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1625                                            dentry->d_name.len);
1626         if (status)
1627                 goto bail;
1628
1629         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1630                                               dentry->d_name.name,
1631                                               dentry->d_name.len, &de_bh);
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto bail;
1635         }
1636
1637         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1638         if (status < 0) {
1639                 if (status != -ENOSPC)
1640                         mlog_errno(status);
1641                 goto bail;
1642         }
1643
1644         /* don't reserve bitmap space for fast symlinks. */
1645         if (l > ocfs2_fast_symlink_chars(sb)) {
1646                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1647                 if (status < 0) {
1648                         if (status != -ENOSPC)
1649                                 mlog_errno(status);
1650                         goto bail;
1651                 }
1652         }
1653
1654         handle = ocfs2_start_trans(osb, credits);
1655         if (IS_ERR(handle)) {
1656                 status = PTR_ERR(handle);
1657                 handle = NULL;
1658                 mlog_errno(status);
1659                 goto bail;
1660         }
1661
1662         status = ocfs2_mknod_locked(osb, dir, dentry,
1663                                     S_IFLNK | S_IRWXUGO, 0,
1664                                     &new_fe_bh, parent_fe_bh, handle,
1665                                     &inode, inode_ac);
1666         if (status < 0) {
1667                 mlog_errno(status);
1668                 goto bail;
1669         }
1670
1671         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1672         inode->i_rdev = 0;
1673         newsize = l - 1;
1674         if (l > ocfs2_fast_symlink_chars(sb)) {
1675                 inode->i_op = &ocfs2_symlink_inode_operations;
1676                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1677                                                     handle, data_ac, NULL,
1678                                                     NULL);
1679                 if (status < 0) {
1680                         if (status != -ENOSPC && status != -EINTR) {
1681                                 mlog(ML_ERROR,
1682                                      "Failed to extend file to %llu\n",
1683                                      (unsigned long long)newsize);
1684                                 mlog_errno(status);
1685                                 status = -ENOSPC;
1686                         }
1687                         goto bail;
1688                 }
1689                 i_size_write(inode, newsize);
1690                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1691         } else {
1692                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1693                 memcpy((char *) fe->id2.i_symlink, symname, l);
1694                 i_size_write(inode, newsize);
1695                 inode->i_blocks = 0;
1696         }
1697
1698         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1699         if (status < 0) {
1700                 mlog_errno(status);
1701                 goto bail;
1702         }
1703
1704         if (!ocfs2_inode_is_fast_symlink(inode)) {
1705                 status = ocfs2_create_symlink_data(osb, handle, inode,
1706                                                    symname);
1707                 if (status < 0) {
1708                         mlog_errno(status);
1709                         goto bail;
1710                 }
1711         }
1712
1713         status = ocfs2_add_entry(handle, dentry, inode,
1714                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1715                                  de_bh);
1716         if (status < 0) {
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1722         if (status) {
1723                 mlog_errno(status);
1724                 goto bail;
1725         }
1726
1727         insert_inode_hash(inode);
1728         dentry->d_op = &ocfs2_dentry_ops;
1729         d_instantiate(dentry, inode);
1730 bail:
1731         if (handle)
1732                 ocfs2_commit_trans(osb, handle);
1733
1734         ocfs2_meta_unlock(dir, 1);
1735
1736         if (new_fe_bh)
1737                 brelse(new_fe_bh);
1738         if (parent_fe_bh)
1739                 brelse(parent_fe_bh);
1740         if (de_bh)
1741                 brelse(de_bh);
1742         if (inode_ac)
1743                 ocfs2_free_alloc_context(inode_ac);
1744         if (data_ac)
1745                 ocfs2_free_alloc_context(data_ac);
1746         if ((status < 0) && inode)
1747                 iput(inode);
1748
1749         mlog_exit(status);
1750
1751         return status;
1752 }
1753
1754 int ocfs2_check_dir_entry(struct inode * dir,
1755                           struct ocfs2_dir_entry * de,
1756                           struct buffer_head * bh,
1757                           unsigned long offset)
1758 {
1759         const char *error_msg = NULL;
1760         const int rlen = le16_to_cpu(de->rec_len);
1761
1762         if (rlen < OCFS2_DIR_REC_LEN(1))
1763                 error_msg = "rec_len is smaller than minimal";
1764         else if (rlen % 4 != 0)
1765                 error_msg = "rec_len % 4 != 0";
1766         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1767                 error_msg = "rec_len is too small for name_len";
1768         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1769                 error_msg = "directory entry across blocks";
1770
1771         if (error_msg != NULL)
1772                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1773                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1774                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1775                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1776                      de->name_len);
1777         return error_msg == NULL ? 1 : 0;
1778 }
1779
1780 /* we don't always have a dentry for what we want to add, so people
1781  * like orphan dir can call this instead.
1782  *
1783  * If you pass me insert_bh, I'll skip the search of the other dir
1784  * blocks and put the record in there.
1785  */
1786 static int __ocfs2_add_entry(handle_t *handle,
1787                              struct inode *dir,
1788                              const char *name, int namelen,
1789                              struct inode *inode, u64 blkno,
1790                              struct buffer_head *parent_fe_bh,
1791                              struct buffer_head *insert_bh)
1792 {
1793         unsigned long offset;
1794         unsigned short rec_len;
1795         struct ocfs2_dir_entry *de, *de1;
1796         struct super_block *sb;
1797         int retval, status;
1798
1799         mlog_entry_void();
1800
1801         sb = dir->i_sb;
1802
1803         if (!namelen)
1804                 return -EINVAL;
1805
1806         rec_len = OCFS2_DIR_REC_LEN(namelen);
1807         offset = 0;
1808         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1809         while (1) {
1810                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1811                 /* These checks should've already been passed by the
1812                  * prepare function, but I guess we can leave them
1813                  * here anyway. */
1814                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1815                         retval = -ENOENT;
1816                         goto bail;
1817                 }
1818                 if (ocfs2_match(namelen, name, de)) {
1819                         retval = -EEXIST;
1820                         goto bail;
1821                 }
1822                 if (((le64_to_cpu(de->inode) == 0) &&
1823                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1824                     (le16_to_cpu(de->rec_len) >=
1825                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1826                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1827                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1828                         if (retval < 0) {
1829                                 mlog_errno(retval);
1830                                 goto bail;
1831                         }
1832
1833                         status = ocfs2_journal_access(handle, dir, insert_bh,
1834                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1835                         /* By now the buffer is marked for journaling */
1836                         offset += le16_to_cpu(de->rec_len);
1837                         if (le64_to_cpu(de->inode)) {
1838                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1839                                         OCFS2_DIR_REC_LEN(de->name_len));
1840                                 de1->rec_len =
1841                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1842                                         OCFS2_DIR_REC_LEN(de->name_len));
1843                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1844                                 de = de1;
1845                         }
1846                         de->file_type = OCFS2_FT_UNKNOWN;
1847                         if (blkno) {
1848                                 de->inode = cpu_to_le64(blkno);
1849                                 ocfs2_set_de_type(de, inode->i_mode);
1850                         } else
1851                                 de->inode = 0;
1852                         de->name_len = namelen;
1853                         memcpy(de->name, name, namelen);
1854
1855                         dir->i_version++;
1856                         status = ocfs2_journal_dirty(handle, insert_bh);
1857                         retval = 0;
1858                         goto bail;
1859                 }
1860                 offset += le16_to_cpu(de->rec_len);
1861                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1862         }
1863
1864         /* when you think about it, the assert above should prevent us
1865          * from ever getting here. */
1866         retval = -ENOSPC;
1867 bail:
1868
1869         mlog_exit(retval);
1870         return retval;
1871 }
1872
1873
1874 /*
1875  * ocfs2_delete_entry deletes a directory entry by merging it with the
1876  * previous entry
1877  */
1878 static int ocfs2_delete_entry(handle_t *handle,
1879                               struct inode *dir,
1880                               struct ocfs2_dir_entry *de_del,
1881                               struct buffer_head *bh)
1882 {
1883         struct ocfs2_dir_entry *de, *pde;
1884         int i, status = -ENOENT;
1885
1886         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1887
1888         i = 0;
1889         pde = NULL;
1890         de = (struct ocfs2_dir_entry *) bh->b_data;
1891         while (i < bh->b_size) {
1892                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1893                         status = -EIO;
1894                         mlog_errno(status);
1895                         goto bail;
1896                 }
1897                 if (de == de_del)  {
1898                         status = ocfs2_journal_access(handle, dir, bh,
1899                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1900                         if (status < 0) {
1901                                 status = -EIO;
1902                                 mlog_errno(status);
1903                                 goto bail;
1904                         }
1905                         if (pde)
1906                                 pde->rec_len =
1907                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1908                                                     le16_to_cpu(de->rec_len));
1909                         else
1910                                 de->inode = 0;
1911                         dir->i_version++;
1912                         status = ocfs2_journal_dirty(handle, bh);
1913                         goto bail;
1914                 }
1915                 i += le16_to_cpu(de->rec_len);
1916                 pde = de;
1917                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1918         }
1919 bail:
1920         mlog_exit(status);
1921         return status;
1922 }
1923
1924 /*
1925  * Returns 0 if not found, -1 on failure, and 1 on success
1926  */
1927 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1928                                         struct inode *dir,
1929                                         const char *name, int namelen,
1930                                         unsigned long offset,
1931                                         struct ocfs2_dir_entry **res_dir)
1932 {
1933         struct ocfs2_dir_entry *de;
1934         char *dlimit, *de_buf;
1935         int de_len;
1936         int ret = 0;
1937
1938         mlog_entry_void();
1939
1940         de_buf = bh->b_data;
1941         dlimit = de_buf + dir->i_sb->s_blocksize;
1942
1943         while (de_buf < dlimit) {
1944                 /* this code is executed quadratically often */
1945                 /* do minimal checking `by hand' */
1946
1947                 de = (struct ocfs2_dir_entry *) de_buf;
1948
1949                 if (de_buf + namelen <= dlimit &&
1950                     ocfs2_match(namelen, name, de)) {
1951                         /* found a match - just to be sure, do a full check */
1952                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1953                                 ret = -1;
1954                                 goto bail;
1955                         }
1956                         *res_dir = de;
1957                         ret = 1;
1958                         goto bail;
1959                 }
1960
1961                 /* prevent looping on a bad block */
1962                 de_len = le16_to_cpu(de->rec_len);
1963                 if (de_len <= 0) {
1964                         ret = -1;
1965                         goto bail;
1966                 }
1967
1968                 de_buf += de_len;
1969                 offset += de_len;
1970         }
1971
1972 bail:
1973         mlog_exit(ret);
1974         return ret;
1975 }
1976
1977 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1978                                      struct inode *dir,
1979                                      struct ocfs2_dir_entry **res_dir)
1980 {
1981         struct super_block *sb;
1982         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1983         struct buffer_head *bh, *ret = NULL;
1984         unsigned long start, block, b;
1985         int ra_max = 0;         /* Number of bh's in the readahead
1986                                    buffer, bh_use[] */
1987         int ra_ptr = 0;         /* Current index into readahead
1988                                    buffer */
1989         int num = 0;
1990         int nblocks, i, err;
1991
1992         mlog_entry_void();
1993
1994         *res_dir = NULL;
1995         sb = dir->i_sb;
1996
1997         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1998         start = OCFS2_I(dir)->ip_dir_start_lookup;
1999         if (start >= nblocks)
2000                 start = 0;
2001         block = start;
2002
2003 restart:
2004         do {
2005                 /*
2006                  * We deal with the read-ahead logic here.
2007                  */
2008                 if (ra_ptr >= ra_max) {
2009                         /* Refill the readahead buffer */
2010                         ra_ptr = 0;
2011                         b = block;
2012                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2013                                 /*
2014                                  * Terminate if we reach the end of the
2015                                  * directory and must wrap, or if our
2016                                  * search has finished at this block.
2017                                  */
2018                                 if (b >= nblocks || (num && block == start)) {
2019                                         bh_use[ra_max] = NULL;
2020                                         break;
2021                                 }
2022                                 num++;
2023
2024                                 bh = ocfs2_bread(dir, b++, &err, 1);
2025                                 bh_use[ra_max] = bh;
2026                         }
2027                 }
2028                 if ((bh = bh_use[ra_ptr++]) == NULL)
2029                         goto next;
2030                 wait_on_buffer(bh);
2031                 if (!buffer_uptodate(bh)) {
2032                         /* read error, skip block & hope for the best */
2033                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2034                                     "offset %lu\n",
2035                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2036                                     block);
2037                         brelse(bh);
2038                         goto next;
2039                 }
2040                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2041                                           block << sb->s_blocksize_bits,
2042                                           res_dir);
2043                 if (i == 1) {
2044                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2045                         ret = bh;
2046                         goto cleanup_and_exit;
2047                 } else {
2048                         brelse(bh);
2049                         if (i < 0)
2050                                 goto cleanup_and_exit;
2051                 }
2052         next:
2053                 if (++block >= nblocks)
2054                         block = 0;
2055         } while (block != start);
2056
2057         /*
2058          * If the directory has grown while we were searching, then
2059          * search the last part of the directory before giving up.
2060          */
2061         block = nblocks;
2062         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2063         if (block < nblocks) {
2064                 start = 0;
2065                 goto restart;
2066         }
2067
2068 cleanup_and_exit:
2069         /* Clean up the read-ahead blocks */
2070         for (; ra_ptr < ra_max; ra_ptr++)
2071                 brelse(bh_use[ra_ptr]);
2072
2073         mlog_exit_ptr(ret);
2074         return ret;
2075 }
2076
2077 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2078 {
2079         int status, namelen;
2080
2081         mlog_entry_void();
2082
2083         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2084                            (long long)blkno);
2085         if (namelen <= 0) {
2086                 if (namelen)
2087                         status = namelen;
2088                 else
2089                         status = -EINVAL;
2090                 mlog_errno(status);
2091                 goto bail;
2092         }
2093         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2094                 status = -EINVAL;
2095                 mlog_errno(status);
2096                 goto bail;
2097         }
2098
2099         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2100              namelen);
2101
2102         status = 0;
2103 bail:
2104         mlog_exit(status);
2105         return status;
2106 }
2107
2108 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2109                                     struct inode **ret_orphan_dir,
2110                                     struct inode *inode,
2111                                     char *name,
2112                                     struct buffer_head **de_bh)
2113 {
2114         struct inode *orphan_dir_inode;
2115         struct buffer_head *orphan_dir_bh = NULL;
2116         int status = 0;
2117
2118         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2119         if (status < 0) {
2120                 mlog_errno(status);
2121                 return status;
2122         }
2123
2124         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2125                                                        ORPHAN_DIR_SYSTEM_INODE,
2126                                                        osb->slot_num);
2127         if (!orphan_dir_inode) {
2128                 status = -ENOENT;
2129                 mlog_errno(status);
2130                 return status;
2131         }
2132
2133         mutex_lock(&orphan_dir_inode->i_mutex);
2134
2135         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2136         if (status < 0) {
2137                 mlog_errno(status);
2138                 goto leave;
2139         }
2140
2141         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2142                                               orphan_dir_bh, name,
2143                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2144         if (status < 0) {
2145                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2146
2147                 mlog_errno(status);
2148                 goto leave;
2149         }
2150
2151         *ret_orphan_dir = orphan_dir_inode;
2152
2153 leave:
2154         if (status) {
2155                 mutex_unlock(&orphan_dir_inode->i_mutex);
2156                 iput(orphan_dir_inode);
2157         }
2158
2159         if (orphan_dir_bh)
2160                 brelse(orphan_dir_bh);
2161
2162         mlog_exit(status);
2163         return status;
2164 }
2165
2166 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2167                             handle_t *handle,
2168                             struct inode *inode,
2169                             struct ocfs2_dinode *fe,
2170                             char *name,
2171                             struct buffer_head *de_bh,
2172                             struct inode *orphan_dir_inode)
2173 {
2174         struct buffer_head *orphan_dir_bh = NULL;
2175         int status = 0;
2176         struct ocfs2_dinode *orphan_fe;
2177
2178         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2179
2180         status = ocfs2_read_block(osb,
2181                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2182                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2183                                   orphan_dir_inode);
2184         if (status < 0) {
2185                 mlog_errno(status);
2186                 goto leave;
2187         }
2188
2189         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2190                                       OCFS2_JOURNAL_ACCESS_WRITE);
2191         if (status < 0) {
2192                 mlog_errno(status);
2193                 goto leave;
2194         }
2195
2196         /* we're a cluster, and nlink can change on disk from
2197          * underneath us... */
2198         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2199         if (S_ISDIR(inode->i_mode))
2200                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2201         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2202
2203         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2204         if (status < 0) {
2205                 mlog_errno(status);
2206                 goto leave;
2207         }
2208
2209         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2210                                    OCFS2_ORPHAN_NAMELEN, inode,
2211                                    OCFS2_I(inode)->ip_blkno,
2212                                    orphan_dir_bh, de_bh);
2213         if (status < 0) {
2214                 mlog_errno(status);
2215                 goto leave;
2216         }
2217
2218         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2219
2220         /* Record which orphan dir our inode now resides
2221          * in. delete_inode will use this to determine which orphan
2222          * dir to lock. */
2223         spin_lock(&OCFS2_I(inode)->ip_lock);
2224         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2225         spin_unlock(&OCFS2_I(inode)->ip_lock);
2226
2227         mlog(0, "Inode %llu orphaned in slot %d\n",
2228              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2229
2230 leave:
2231         if (orphan_dir_bh)
2232                 brelse(orphan_dir_bh);
2233
2234         mlog_exit(status);
2235         return status;
2236 }
2237
2238 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2239 int ocfs2_orphan_del(struct ocfs2_super *osb,
2240                      handle_t *handle,
2241                      struct inode *orphan_dir_inode,
2242                      struct inode *inode,
2243                      struct buffer_head *orphan_dir_bh)
2244 {
2245         char name[OCFS2_ORPHAN_NAMELEN + 1];
2246         struct ocfs2_dinode *orphan_fe;
2247         int status = 0;
2248         struct buffer_head *target_de_bh = NULL;
2249         struct ocfs2_dir_entry *target_de = NULL;
2250
2251         mlog_entry_void();
2252
2253         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2260              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2261              OCFS2_ORPHAN_NAMELEN);
2262
2263         /* find it's spot in the orphan directory */
2264         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2265                                         orphan_dir_inode, &target_de);
2266         if (!target_de_bh) {
2267                 status = -ENOENT;
2268                 mlog_errno(status);
2269                 goto leave;
2270         }
2271
2272         /* remove it from the orphan directory */
2273         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2274                                     target_de_bh);
2275         if (status < 0) {
2276                 mlog_errno(status);
2277                 goto leave;
2278         }
2279
2280         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2281                                       OCFS2_JOURNAL_ACCESS_WRITE);
2282         if (status < 0) {
2283                 mlog_errno(status);
2284                 goto leave;
2285         }
2286
2287         /* do the i_nlink dance! :) */
2288         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2289         if (S_ISDIR(inode->i_mode))
2290                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2291         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2292
2293         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2294         if (status < 0) {
2295                 mlog_errno(status);
2296                 goto leave;
2297         }
2298
2299 leave:
2300         if (target_de_bh)
2301                 brelse(target_de_bh);
2302
2303         mlog_exit(status);
2304         return status;
2305 }
2306
2307 const struct inode_operations ocfs2_dir_iops = {
2308         .create         = ocfs2_create,
2309         .lookup         = ocfs2_lookup,
2310         .link           = ocfs2_link,
2311         .unlink         = ocfs2_unlink,
2312         .rmdir          = ocfs2_unlink,
2313         .symlink        = ocfs2_symlink,
2314         .mkdir          = ocfs2_mkdir,
2315         .mknod          = ocfs2_mknod,
2316         .rename         = ocfs2_rename,
2317         .setattr        = ocfs2_setattr,
2318         .getattr        = ocfs2_getattr,
2319         .permission     = ocfs2_permission,
2320 };