fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "vote.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 #define NAMEI_RA_CHUNKS  2
  68 #define NAMEI_RA_BLOCKS  4
  69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  71
  72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  73                                         struct inode *dir,
  74                                         const char *name, int namelen,
  75                                         unsigned long offset,
  76                                         struct ocfs2_dir_entry **res_dir);
  77
  78 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
  84                              struct inode *dir,
  85                              const char *name, int namelen,
  86                              struct inode *inode, u64 blkno,
  87                              struct buffer_head *parent_fe_bh,
  88                              struct buffer_head *insert_bh);
  89
  90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  91                               struct inode *dir,
  92                               struct dentry *dentry, int mode,
  93                               dev_t dev,
  94                               struct buffer_head **new_fe_bh,
  95                               struct buffer_head *parent_fe_bh,
  96                               struct ocfs2_journal_handle *handle,
  97                               struct inode **ret_inode,
  98                               struct ocfs2_alloc_context *inode_ac);
  99
 100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 101                               struct ocfs2_journal_handle *handle,
 102                               struct inode *parent,
 103                               struct inode *inode,
 104                               struct buffer_head *fe_bh,
 105                               struct ocfs2_alloc_context *data_ac);
 106
 107 static int ocfs2_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                             struct ocfs2_journal_handle *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                                      struct ocfs2_journal_handle *handle,
 123                                      struct inode *inode,
 124                                      const char *symname);
 125
 126 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *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, 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                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 179                      (unsigned long long)blkno);
 180                 ret = ERR_PTR(-EACCES);
 181                 goto bail_unlock;
 182         }
 183
 184         oi = OCFS2_I(inode);
 185         /* Clear any orphaned state... If we were able to look up the
 186          * inode from a directory, it certainly can't be orphaned. We
 187          * might have the bad state from a node which intended to
 188          * orphan this inode but crashed before it could commit the
 189          * unlink. */
 190         spin_lock(&oi->ip_lock);
 191         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 192         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 193         spin_unlock(&oi->ip_lock);
 194
 195 bail_add:
 196         dentry->d_op = &ocfs2_dentry_ops;
 197         ret = d_splice_alias(inode, dentry);
 198
 199         if (inode) {
 200                 /*
 201                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 202                  * dentry, it will d_move() it on top of ourse. The
 203                  * return value will indicate this however, so in
 204                  * those cases, we switch them around for the locking
 205                  * code.
 206                  *
 207                  * NOTE: This dentry already has ->d_op set from
 208                  * ocfs2_get_parent() and ocfs2_get_dentry()
 209                  */
 210                 if (ret)
 211                         dentry = ret;
 212
 213                 status = ocfs2_dentry_attach_lock(dentry, inode,
 214                                                   OCFS2_I(dir)->ip_blkno);
 215                 if (status) {
 216                         mlog_errno(status);
 217                         ret = ERR_PTR(status);
 218                         goto bail_unlock;
 219                 }
 220         }
 221
 222 bail_unlock:
 223         /* Don't drop the cluster lock until *after* the d_add --
 224          * unlink on another node will message us to remove that
 225          * dentry under this lock so otherwise we can race this with
 226          * the vote thread and have a stale dentry. */
 227         ocfs2_meta_unlock(dir, 0);
 228
 229 bail:
 230         if (dirent_bh)
 231                 brelse(dirent_bh);
 232
 233         mlog_exit_ptr(ret);
 234
 235         return ret;
 236 }
 237
 238 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 239                               struct ocfs2_journal_handle *handle,
 240                               struct inode *parent,
 241                               struct inode *inode,
 242                               struct buffer_head *fe_bh,
 243                               struct ocfs2_alloc_context *data_ac)
 244 {
 245         int status;
 246         struct buffer_head *new_bh = NULL;
 247         struct ocfs2_dir_entry *de = NULL;
 248
 249         mlog_entry_void();
 250
 251         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 252                                      data_ac, NULL, &new_bh);
 253         if (status < 0) {
 254                 mlog_errno(status);
 255                 goto bail;
 256         }
 257
 258         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 259
 260         status = ocfs2_journal_access(handle, inode, new_bh,
 261                                       OCFS2_JOURNAL_ACCESS_CREATE);
 262         if (status < 0) {
 263                 mlog_errno(status);
 264                 goto bail;
 265         }
 266         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 267
 268         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 269         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 270         de->name_len = 1;
 271         de->rec_len =
 272                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 273         strcpy(de->name, ".");
 274         ocfs2_set_de_type(de, S_IFDIR);
 275         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 276         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 277         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 278                                   OCFS2_DIR_REC_LEN(1));
 279         de->name_len = 2;
 280         strcpy(de->name, "..");
 281         ocfs2_set_de_type(de, S_IFDIR);
 282
 283         status = ocfs2_journal_dirty(handle, new_bh);
 284         if (status < 0) {
 285                 mlog_errno(status);
 286                 goto bail;
 287         }
 288
 289         i_size_write(inode, inode->i_sb->s_blocksize);
 290         inode->i_nlink = 2;
 291         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 292         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 293         if (status < 0) {
 294                 mlog_errno(status);
 295                 goto bail;
 296         }
 297
 298         status = 0;
 299 bail:
 300         if (new_bh)
 301                 brelse(new_bh);
 302
 303         mlog_exit(status);
 304         return status;
 305 }
 306
 307 static int ocfs2_mknod(struct inode *dir,
 308                        struct dentry *dentry,
 309                        int mode,
 310                        dev_t dev)
 311 {
 312         int status = 0;
 313         struct buffer_head *parent_fe_bh = NULL;
 314         struct ocfs2_journal_handle *handle = NULL;
 315         struct ocfs2_super *osb;
 316         struct ocfs2_dinode *dirfe;
 317         struct buffer_head *new_fe_bh = NULL;
 318         struct buffer_head *de_bh = NULL;
 319         struct inode *inode = NULL;
 320         struct ocfs2_alloc_context *inode_ac = NULL;
 321         struct ocfs2_alloc_context *data_ac = NULL;
 322
 323         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 324                    (unsigned long)dev, dentry->d_name.len,
 325                    dentry->d_name.name);
 326
 327         /* get our super block */
 328         osb = OCFS2_SB(dir->i_sb);
 329
 330         status = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
 331         if (status < 0) {
 332                 if (status != -ENOENT)
 333                         mlog_errno(status);
 334                 return status;
 335         }
 336
 337         handle = ocfs2_alloc_handle(osb);
 338         if (handle == NULL) {
 339                 status = -ENOMEM;
 340                 mlog_errno(status);
 341                 goto leave;
 342         }
 343
 344         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 345                 status = -EMLINK;
 346                 goto leave;
 347         }
 348
 349         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 350         if (!dirfe->i_links_count) {
 351                 /* can't make a file in a deleted directory. */
 352                 status = -ENOENT;
 353                 goto leave;
 354         }
 355
 356         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 357                                            dentry->d_name.len);
 358         if (status)
 359                 goto leave;
 360
 361         /* get a spot inside the dir. */
 362         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 363                                               dentry->d_name.name,
 364                                               dentry->d_name.len, &de_bh);
 365         if (status < 0) {
 366                 mlog_errno(status);
 367                 goto leave;
 368         }
 369
 370         /* reserve an inode spot */
 371         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
 372         if (status < 0) {
 373                 if (status != -ENOSPC)
 374                         mlog_errno(status);
 375                 goto leave;
 376         }
 377
 378         /* are we making a directory? If so, reserve a cluster for his
 379          * 1st extent. */
 380         if (S_ISDIR(mode)) {
 381                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 382                 if (status < 0) {
 383                         if (status != -ENOSPC)
 384                                 mlog_errno(status);
 385                         goto leave;
 386                 }
 387         }
 388
 389         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
 390         if (IS_ERR(handle)) {
 391                 status = PTR_ERR(handle);
 392                 handle = NULL;
 393                 mlog_errno(status);
 394                 goto leave;
 395         }
 396
 397         /* do the real work now. */
 398         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 399                                     &new_fe_bh, parent_fe_bh, handle,
 400                                     &inode, inode_ac);
 401         if (status < 0) {
 402                 mlog_errno(status);
 403                 goto leave;
 404         }
 405
 406         if (S_ISDIR(mode)) {
 407                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 408                                             new_fe_bh, data_ac);
 409                 if (status < 0) {
 410                         mlog_errno(status);
 411                         goto leave;
 412                 }
 413
 414                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 415                                               OCFS2_JOURNAL_ACCESS_WRITE);
 416                 if (status < 0) {
 417                         mlog_errno(status);
 418                         goto leave;
 419                 }
 420                 le16_add_cpu(&dirfe->i_links_count, 1);
 421                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 422                 if (status < 0) {
 423                         mlog_errno(status);
 424                         goto leave;
 425                 }
 426                 inc_nlink(dir);
 427         }
 428
 429         status = ocfs2_add_entry(handle, dentry, inode,
 430                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 431                                  de_bh);
 432         if (status < 0) {
 433                 mlog_errno(status);
 434                 goto leave;
 435         }
 436
 437         status = ocfs2_dentry_attach_lock(dentry, inode,
 438                                           OCFS2_I(dir)->ip_blkno);
 439         if (status) {
 440                 mlog_errno(status);
 441                 goto leave;
 442         }
 443
 444         insert_inode_hash(inode);
 445         dentry->d_op = &ocfs2_dentry_ops;
 446         d_instantiate(dentry, inode);
 447         status = 0;
 448 leave:
 449         if (handle)
 450                 ocfs2_commit_trans(handle);
 451
 452         ocfs2_meta_unlock(dir, 1);
 453
 454         if (status == -ENOSPC)
 455                 mlog(0, "Disk is full\n");
 456
 457         if (new_fe_bh)
 458                 brelse(new_fe_bh);
 459
 460         if (de_bh)
 461                 brelse(de_bh);
 462
 463         if (parent_fe_bh)
 464                 brelse(parent_fe_bh);
 465
 466         if ((status < 0) && inode)
 467                 iput(inode);
 468
 469         if (inode_ac)
 470                 ocfs2_free_alloc_context(inode_ac);
 471
 472         if (data_ac)
 473                 ocfs2_free_alloc_context(data_ac);
 474
 475         mlog_exit(status);
 476
 477         return status;
 478 }
 479
 480 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 481                               struct inode *dir,
 482                               struct dentry *dentry, int mode,
 483                               dev_t dev,
 484                               struct buffer_head **new_fe_bh,
 485                               struct buffer_head *parent_fe_bh,
 486                               struct ocfs2_journal_handle *handle,
 487                               struct inode **ret_inode,
 488                               struct ocfs2_alloc_context *inode_ac)
 489 {
 490         int status = 0;
 491         struct ocfs2_dinode *fe = NULL;
 492         struct ocfs2_extent_list *fel;
 493         u64 fe_blkno = 0;
 494         u16 suballoc_bit;
 495         struct inode *inode = NULL;
 496
 497         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 498                    (unsigned long)dev, dentry->d_name.len,
 499                    dentry->d_name.name);
 500
 501         *new_fe_bh = NULL;
 502         *ret_inode = NULL;
 503
 504         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 505                                        &fe_blkno);
 506         if (status < 0) {
 507                 mlog_errno(status);
 508                 goto leave;
 509         }
 510
 511         inode = new_inode(dir->i_sb);
 512         if (IS_ERR(inode)) {
 513                 status = PTR_ERR(inode);
 514                 mlog(ML_ERROR, "new_inode failed!\n");
 515                 goto leave;
 516         }
 517
 518         /* populate as many fields early on as possible - many of
 519          * these are used by the support functions here and in
 520          * callers. */
 521         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 522         OCFS2_I(inode)->ip_blkno = fe_blkno;
 523         if (S_ISDIR(mode))
 524                 inode->i_nlink = 2;
 525         else
 526                 inode->i_nlink = 1;
 527         inode->i_mode = mode;
 528         spin_lock(&osb->osb_lock);
 529         inode->i_generation = osb->s_next_generation++;
 530         spin_unlock(&osb->osb_lock);
 531
 532         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 533         if (!*new_fe_bh) {
 534                 status = -EIO;
 535                 mlog_errno(status);
 536                 goto leave;
 537         }
 538         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 539
 540         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 541                                       OCFS2_JOURNAL_ACCESS_CREATE);
 542         if (status < 0) {
 543                 mlog_errno(status);
 544                 goto leave;
 545         }
 546
 547         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 548         memset(fe, 0, osb->sb->s_blocksize);
 549
 550         fe->i_generation = cpu_to_le32(inode->i_generation);
 551         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 552         fe->i_blkno = cpu_to_le64(fe_blkno);
 553         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 554         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 555         fe->i_uid = cpu_to_le32(current->fsuid);
 556         if (dir->i_mode & S_ISGID) {
 557                 fe->i_gid = cpu_to_le32(dir->i_gid);
 558                 if (S_ISDIR(mode))
 559                         mode |= S_ISGID;
 560         } else
 561                 fe->i_gid = cpu_to_le32(current->fsgid);
 562         fe->i_mode = cpu_to_le16(mode);
 563         if (S_ISCHR(mode) || S_ISBLK(mode))
 564                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 565
 566         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 567
 568         fe->i_last_eb_blk = 0;
 569         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 570         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 571         fe->i_atime = fe->i_ctime = fe->i_mtime =
 572                 cpu_to_le64(CURRENT_TIME.tv_sec);
 573         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 574                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 575         fe->i_dtime = 0;
 576
 577         fel = &fe->id2.i_list;
 578         fel->l_tree_depth = 0;
 579         fel->l_next_free_rec = 0;
 580         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 581
 582         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 583         if (status < 0) {
 584                 mlog_errno(status);
 585                 goto leave;
 586         }
 587
 588         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 589                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 590                      "i_blkno=%llu, i_ino=%lu\n",
 591                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 592                      (unsigned long long)fe->i_blkno, inode->i_ino);
 593                 BUG();
 594         }
 595
 596         ocfs2_inode_set_new(osb, inode);
 597         status = ocfs2_create_new_inode_locks(inode);
 598         if (status < 0)
 599                 mlog_errno(status);
 600
 601         status = 0; /* error in ocfs2_create_new_inode_locks is not
 602                      * critical */
 603
 604         *ret_inode = inode;
 605 leave:
 606         if (status < 0) {
 607                 if (*new_fe_bh) {
 608                         brelse(*new_fe_bh);
 609                         *new_fe_bh = NULL;
 610                 }
 611                 if (inode)
 612                         iput(inode);
 613         }
 614
 615         mlog_exit(status);
 616         return status;
 617 }
 618
 619 static int ocfs2_mkdir(struct inode *dir,
 620                        struct dentry *dentry,
 621                        int mode)
 622 {
 623         int ret;
 624
 625         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 626                    dentry->d_name.len, dentry->d_name.name);
 627         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 628         mlog_exit(ret);
 629
 630         return ret;
 631 }
 632
 633 static int ocfs2_create(struct inode *dir,
 634                         struct dentry *dentry,
 635                         int mode,
 636                         struct nameidata *nd)
 637 {
 638         int ret;
 639
 640         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 641                    dentry->d_name.len, dentry->d_name.name);
 642         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 643         mlog_exit(ret);
 644
 645         return ret;
 646 }
 647
 648 static int ocfs2_link(struct dentry *old_dentry,
 649                       struct inode *dir,
 650                       struct dentry *dentry)
 651 {
 652         struct ocfs2_journal_handle *handle;
 653         struct inode *inode = old_dentry->d_inode;
 654         int err;
 655         struct buffer_head *fe_bh = NULL;
 656         struct buffer_head *parent_fe_bh = NULL;
 657         struct buffer_head *de_bh = NULL;
 658         struct ocfs2_dinode *fe = NULL;
 659         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 660
 661         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 662                    old_dentry->d_name.len, old_dentry->d_name.name,
 663                    dentry->d_name.len, dentry->d_name.name);
 664
 665         if (S_ISDIR(inode->i_mode))
 666                 return -EPERM;
 667
 668         err = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
 669         if (err < 0) {
 670                 if (err != -ENOENT)
 671                         mlog_errno(err);
 672                 return err;
 673         }
 674
 675         if (!dir->i_nlink) {
 676                 err = -ENOENT;
 677                 goto out;
 678         }
 679
 680         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 681                                         dentry->d_name.len);
 682         if (err)
 683                 goto out;
 684
 685         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 686                                            dentry->d_name.name,
 687                                            dentry->d_name.len, &de_bh);
 688         if (err < 0) {
 689                 mlog_errno(err);
 690                 goto out;
 691         }
 692
 693         err = ocfs2_meta_lock(inode, NULL, &fe_bh, 1);
 694         if (err < 0) {
 695                 if (err != -ENOENT)
 696                         mlog_errno(err);
 697                 goto out;
 698         }
 699
 700         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 701         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 702                 err = -EMLINK;
 703                 goto out_unlock_inode;
 704         }
 705
 706         handle = ocfs2_start_trans(osb, NULL, OCFS2_LINK_CREDITS);
 707         if (IS_ERR(handle)) {
 708                 err = PTR_ERR(handle);
 709                 handle = NULL;
 710                 mlog_errno(err);
 711                 goto out_unlock_inode;
 712         }
 713
 714         err = ocfs2_journal_access(handle, inode, fe_bh,
 715                                    OCFS2_JOURNAL_ACCESS_WRITE);
 716         if (err < 0) {
 717                 mlog_errno(err);
 718                 goto out_commit;
 719         }
 720
 721         inc_nlink(inode);
 722         inode->i_ctime = CURRENT_TIME;
 723         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 724         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 725         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 726
 727         err = ocfs2_journal_dirty(handle, fe_bh);
 728         if (err < 0) {
 729                 le16_add_cpu(&fe->i_links_count, -1);
 730                 drop_nlink(inode);
 731                 mlog_errno(err);
 732                 goto out_commit;
 733         }
 734
 735         err = ocfs2_add_entry(handle, dentry, inode,
 736                               OCFS2_I(inode)->ip_blkno,
 737                               parent_fe_bh, de_bh);
 738         if (err) {
 739                 le16_add_cpu(&fe->i_links_count, -1);
 740                 drop_nlink(inode);
 741                 mlog_errno(err);
 742                 goto out_commit;
 743         }
 744
 745         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 746         if (err) {
 747                 mlog_errno(err);
 748                 goto out_commit;
 749         }
 750
 751         atomic_inc(&inode->i_count);
 752         dentry->d_op = &ocfs2_dentry_ops;
 753         d_instantiate(dentry, inode);
 754
 755 out_commit:
 756         ocfs2_commit_trans(handle);
 757 out_unlock_inode:
 758         ocfs2_meta_unlock(inode, 1);
 759
 760 out:
 761         ocfs2_meta_unlock(dir, 1);
 762
 763         if (de_bh)
 764                 brelse(de_bh);
 765         if (fe_bh)
 766                 brelse(fe_bh);
 767         if (parent_fe_bh)
 768                 brelse(parent_fe_bh);
 769
 770         mlog_exit(err);
 771
 772         return err;
 773 }
 774
 775 /*
 776  * Takes and drops an exclusive lock on the given dentry. This will
 777  * force other nodes to drop it.
 778  */
 779 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 780 {
 781         int ret;
 782
 783         ret = ocfs2_dentry_lock(dentry, 1);
 784         if (ret)
 785                 mlog_errno(ret);
 786         else
 787                 ocfs2_dentry_unlock(dentry, 1);
 788
 789         return ret;
 790 }
 791
 792 static inline int inode_is_unlinkable(struct inode *inode)
 793 {
 794         if (S_ISDIR(inode->i_mode)) {
 795                 if (inode->i_nlink == 2)
 796                         return 1;
 797                 return 0;
 798         }
 799
 800         if (inode->i_nlink == 1)
 801                 return 1;
 802         return 0;
 803 }
 804
 805 static int ocfs2_unlink(struct inode *dir,
 806                         struct dentry *dentry)
 807 {
 808         int status;
 809         int child_locked = 0;
 810         struct inode *inode = dentry->d_inode;
 811         struct inode *orphan_dir = NULL;
 812         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 813         u64 blkno;
 814         struct ocfs2_dinode *fe = NULL;
 815         struct buffer_head *fe_bh = NULL;
 816         struct buffer_head *parent_node_bh = NULL;
 817         struct ocfs2_journal_handle *handle = NULL;
 818         struct ocfs2_dir_entry *dirent = NULL;
 819         struct buffer_head *dirent_bh = NULL;
 820         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 821         struct buffer_head *orphan_entry_bh = NULL;
 822
 823         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 824                    dentry->d_name.len, dentry->d_name.name);
 825
 826         BUG_ON(dentry->d_parent->d_inode != dir);
 827
 828         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 829
 830         if (inode == osb->root_inode) {
 831                 mlog(0, "Cannot delete the root directory\n");
 832                 return -EPERM;
 833         }
 834
 835         status = ocfs2_meta_lock(dir, NULL, &parent_node_bh, 1);
 836         if (status < 0) {
 837                 if (status != -ENOENT)
 838                         mlog_errno(status);
 839                 return status;
 840         }
 841
 842         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 843                                           dentry->d_name.len, &blkno,
 844                                           dir, &dirent_bh, &dirent);
 845         if (status < 0) {
 846                 if (status != -ENOENT)
 847                         mlog_errno(status);
 848                 goto leave;
 849         }
 850
 851         if (OCFS2_I(inode)->ip_blkno != blkno) {
 852                 status = -ENOENT;
 853
 854                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 855                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 856                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 857                 goto leave;
 858         }
 859
 860         status = ocfs2_meta_lock(inode, NULL, &fe_bh, 1);
 861         if (status < 0) {
 862                 if (status != -ENOENT)
 863                         mlog_errno(status);
 864                 goto leave;
 865         }
 866         child_locked = 1;
 867
 868         if (S_ISDIR(inode->i_mode)) {
 869                 if (!ocfs2_empty_dir(inode)) {
 870                         status = -ENOTEMPTY;
 871                         goto leave;
 872                 } else if (inode->i_nlink != 2) {
 873                         status = -ENOTEMPTY;
 874                         goto leave;
 875                 }
 876         }
 877
 878         status = ocfs2_remote_dentry_delete(dentry);
 879         if (status < 0) {
 880                 /* This vote should succeed under all normal
 881                  * circumstances. */
 882                 mlog_errno(status);
 883                 goto leave;
 884         }
 885
 886         if (inode_is_unlinkable(inode)) {
 887                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 888                                                   orphan_name,
 889                                                   &orphan_entry_bh);
 890                 if (status < 0) {
 891                         mlog_errno(status);
 892                         goto leave;
 893                 }
 894         }
 895
 896         handle = ocfs2_start_trans(osb, NULL, OCFS2_UNLINK_CREDITS);
 897         if (IS_ERR(handle)) {
 898                 status = PTR_ERR(handle);
 899                 handle = NULL;
 900                 mlog_errno(status);
 901                 goto leave;
 902         }
 903
 904         status = ocfs2_journal_access(handle, inode, fe_bh,
 905                                       OCFS2_JOURNAL_ACCESS_WRITE);
 906         if (status < 0) {
 907                 mlog_errno(status);
 908                 goto leave;
 909         }
 910
 911         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 912
 913         if (inode_is_unlinkable(inode)) {
 914                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 915                                           orphan_entry_bh, orphan_dir);
 916                 if (status < 0) {
 917                         mlog_errno(status);
 918                         goto leave;
 919                 }
 920         }
 921
 922         /* delete the name from the parent dir */
 923         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 924         if (status < 0) {
 925                 mlog_errno(status);
 926                 goto leave;
 927         }
 928
 929         if (S_ISDIR(inode->i_mode))
 930                 drop_nlink(inode);
 931         drop_nlink(inode);
 932         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 933
 934         status = ocfs2_journal_dirty(handle, fe_bh);
 935         if (status < 0) {
 936                 mlog_errno(status);
 937                 goto leave;
 938         }
 939
 940         if (S_ISDIR(inode->i_mode)) {
 941                 drop_nlink(dir);
 942                 status = ocfs2_mark_inode_dirty(handle, dir,
 943                                                 parent_node_bh);
 944                 if (status < 0) {
 945                         mlog_errno(status);
 946                         inc_nlink(dir);
 947                 }
 948         }
 949
 950 leave:
 951         if (handle)
 952                 ocfs2_commit_trans(handle);
 953
 954         if (child_locked)
 955                 ocfs2_meta_unlock(inode, 1);
 956
 957         ocfs2_meta_unlock(dir, 1);
 958
 959         if (orphan_dir) {
 960                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 961                 ocfs2_meta_unlock(orphan_dir, 1);
 962                 mutex_unlock(&orphan_dir->i_mutex);
 963                 iput(orphan_dir);
 964         }
 965
 966         if (fe_bh)
 967                 brelse(fe_bh);
 968
 969         if (dirent_bh)
 970                 brelse(dirent_bh);
 971
 972         if (parent_node_bh)
 973                 brelse(parent_node_bh);
 974
 975         if (orphan_entry_bh)
 976                 brelse(orphan_entry_bh);
 977
 978         mlog_exit(status);
 979
 980         return status;
 981 }
 982
 983 /*
 984  * The only place this should be used is rename!
 985  * if they have the same id, then the 1st one is the only one locked.
 986  */
 987 static int ocfs2_double_lock(struct ocfs2_super *osb,
 988                              struct buffer_head **bh1,
 989                              struct inode *inode1,
 990                              struct buffer_head **bh2,
 991                              struct inode *inode2)
 992 {
 993         int status;
 994         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 995         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 996         struct buffer_head **tmpbh;
 997         struct inode *tmpinode;
 998
 999         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1000                    (unsigned long long)oi1->ip_blkno,
1001                    (unsigned long long)oi2->ip_blkno);
1002
1003         if (*bh1)
1004                 *bh1 = NULL;
1005         if (*bh2)
1006                 *bh2 = NULL;
1007
1008         /* we always want to lock the one with the lower lockid first. */
1009         if (oi1->ip_blkno != oi2->ip_blkno) {
1010                 if (oi1->ip_blkno < oi2->ip_blkno) {
1011                         /* switch id1 and id2 around */
1012                         mlog(0, "switching them around...\n");
1013                         tmpbh = bh2;
1014                         bh2 = bh1;
1015                         bh1 = tmpbh;
1016
1017                         tmpinode = inode2;
1018                         inode2 = inode1;
1019                         inode1 = tmpinode;
1020                 }
1021                 /* lock id2 */
1022                 status = ocfs2_meta_lock(inode2, NULL, bh2, 1);
1023                 if (status < 0) {
1024                         if (status != -ENOENT)
1025                                 mlog_errno(status);
1026                         goto bail;
1027                 }
1028         }
1029
1030         /* lock id1 */
1031         status = ocfs2_meta_lock(inode1, NULL, bh1, 1);
1032         if (status < 0) {
1033                 /*
1034                  * An error return must mean that no cluster locks
1035                  * were held on function exit.
1036                  */
1037                 if (oi1->ip_blkno != oi2->ip_blkno)
1038                         ocfs2_meta_unlock(inode2, 1);
1039
1040                 if (status != -ENOENT)
1041                         mlog_errno(status);
1042         }
1043
1044 bail:
1045         mlog_exit(status);
1046         return status;
1047 }
1048
1049 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1050 {
1051         ocfs2_meta_unlock(inode1, 1);
1052
1053         if (inode1 != inode2)
1054                 ocfs2_meta_unlock(inode2, 1);
1055 }
1056
1057 #define PARENT_INO(buffer) \
1058         ((struct ocfs2_dir_entry *) \
1059          ((char *)buffer + \
1060           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1061
1062 static int ocfs2_rename(struct inode *old_dir,
1063                         struct dentry *old_dentry,
1064                         struct inode *new_dir,
1065                         struct dentry *new_dentry)
1066 {
1067         int status = 0, rename_lock = 0, parents_locked = 0;
1068         int old_child_locked = 0, new_child_locked = 0;
1069         struct inode *old_inode = old_dentry->d_inode;
1070         struct inode *new_inode = new_dentry->d_inode;
1071         struct inode *orphan_dir = NULL;
1072         struct ocfs2_dinode *newfe = NULL;
1073         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1074         struct buffer_head *orphan_entry_bh = NULL;
1075         struct buffer_head *newfe_bh = NULL;
1076         struct buffer_head *insert_entry_bh = NULL;
1077         struct ocfs2_super *osb = NULL;
1078         u64 newfe_blkno;
1079         struct ocfs2_journal_handle *handle = NULL;
1080         struct buffer_head *old_dir_bh = NULL;
1081         struct buffer_head *new_dir_bh = NULL;
1082         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1083                                                                // and new_dentry
1084         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1085         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1086                                                     // this is the 1st dirent bh
1087         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1088
1089         /* At some point it might be nice to break this function up a
1090          * bit. */
1091
1092         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1093                    old_dir, old_dentry, new_dir, new_dentry,
1094                    old_dentry->d_name.len, old_dentry->d_name.name,
1095                    new_dentry->d_name.len, new_dentry->d_name.name);
1096
1097         osb = OCFS2_SB(old_dir->i_sb);
1098
1099         if (new_inode) {
1100                 if (!igrab(new_inode))
1101                         BUG();
1102         }
1103
1104         /* Assume a directory heirarchy thusly:
1105          * a/b/c
1106          * a/d
1107          * a,b,c, and d are all directories.
1108          *
1109          * from cwd of 'a' on both nodes:
1110          * node1: mv b/c d
1111          * node2: mv d   b/c
1112          *
1113          * And that's why, just like the VFS, we need a file system
1114          * rename lock. */
1115         if (old_dentry != new_dentry) {
1116                 status = ocfs2_rename_lock(osb);
1117                 if (status < 0) {
1118                         mlog_errno(status);
1119                         goto bail;
1120                 }
1121                 rename_lock = 1;
1122         }
1123
1124         handle = ocfs2_alloc_handle(osb);
1125         if (handle == NULL) {
1126                 status = -ENOMEM;
1127                 mlog_errno(status);
1128                 goto bail;
1129         }
1130
1131         /* if old and new are the same, this'll just do one lock. */
1132         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1133                                    &new_dir_bh, new_dir);
1134         if (status < 0) {
1135                 mlog_errno(status);
1136                 goto bail;
1137         }
1138         parents_locked = 1;
1139
1140         /* make sure both dirs have bhs
1141          * get an extra ref on old_dir_bh if old==new */
1142         if (!new_dir_bh) {
1143                 if (old_dir_bh) {
1144                         new_dir_bh = old_dir_bh;
1145                         get_bh(new_dir_bh);
1146                 } else {
1147                         mlog(ML_ERROR, "no old_dir_bh!\n");
1148                         status = -EIO;
1149                         goto bail;
1150                 }
1151         }
1152
1153         /*
1154          * Though we don't require an inode meta data update if
1155          * old_inode is not a directory, we lock anyway here to ensure
1156          * the vote thread on other nodes won't have to concurrently
1157          * downconvert the inode and the dentry locks.
1158          */
1159         status = ocfs2_meta_lock(old_inode, NULL, NULL, 1);
1160         if (status < 0) {
1161                 if (status != -ENOENT)
1162                         mlog_errno(status);
1163                 goto bail;
1164         }
1165         old_child_locked = 1;
1166
1167         status = ocfs2_remote_dentry_delete(old_dentry);
1168         if (status < 0) {
1169                 mlog_errno(status);
1170                 goto bail;
1171         }
1172
1173         if (S_ISDIR(old_inode->i_mode)) {
1174                 status = -EIO;
1175                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1176                 if (!old_inode_de_bh)
1177                         goto bail;
1178
1179                 status = -EIO;
1180                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1181                     OCFS2_I(old_dir)->ip_blkno)
1182                         goto bail;
1183                 status = -EMLINK;
1184                 if (!new_inode && new_dir!=old_dir &&
1185                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1186                         goto bail;
1187         }
1188
1189         status = -ENOENT;
1190         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1191                                      old_dentry->d_name.len,
1192                                      old_dir, &old_de);
1193         if (!old_de_bh)
1194                 goto bail;
1195
1196         /*
1197          *  Check for inode number is _not_ due to possible IO errors.
1198          *  We might rmdir the source, keep it as pwd of some process
1199          *  and merrily kill the link to whatever was created under the
1200          *  same name. Goodbye sticky bit ;-<
1201          */
1202         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1203                 goto bail;
1204
1205         /* check if the target already exists (in which case we need
1206          * to delete it */
1207         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1208                                           new_dentry->d_name.len,
1209                                           &newfe_blkno, new_dir, &new_de_bh,
1210                                           &new_de);
1211         /* The only error we allow here is -ENOENT because the new
1212          * file not existing is perfectly valid. */
1213         if ((status < 0) && (status != -ENOENT)) {
1214                 /* If we cannot find the file specified we should just */
1215                 /* return the error... */
1216                 mlog_errno(status);
1217                 goto bail;
1218         }
1219
1220         if (!new_de && new_inode)
1221                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1222                      "directory!", new_inode->i_ino);
1223
1224         /* In case we need to overwrite an existing file, we blow it
1225          * away first */
1226         if (new_de) {
1227                 /* VFS didn't think there existed an inode here, but
1228                  * someone else in the cluster must have raced our
1229                  * rename to create one. Today we error cleanly, in
1230                  * the future we should consider calling iget to build
1231                  * a new struct inode for this entry. */
1232                 if (!new_inode) {
1233                         status = -EACCES;
1234
1235                         mlog(0, "We found an inode for name %.*s but VFS "
1236                              "didn't give us one.\n", new_dentry->d_name.len,
1237                              new_dentry->d_name.name);
1238                         goto bail;
1239                 }
1240
1241                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1242                         status = -EACCES;
1243
1244                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1245                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1246                              (unsigned long long)newfe_blkno,
1247                              OCFS2_I(new_inode)->ip_flags);
1248                         goto bail;
1249                 }
1250
1251                 status = ocfs2_meta_lock(new_inode, NULL, &newfe_bh, 1);
1252                 if (status < 0) {
1253                         if (status != -ENOENT)
1254                                 mlog_errno(status);
1255                         goto bail;
1256                 }
1257                 new_child_locked = 1;
1258
1259                 status = ocfs2_remote_dentry_delete(new_dentry);
1260                 if (status < 0) {
1261                         mlog_errno(status);
1262                         goto bail;
1263                 }
1264
1265                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1266
1267                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1268                      "newfebh=%p bhblocknr=%llu\n", new_de,
1269                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1270                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1271
1272                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1273                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1274                                                           new_inode,
1275                                                           orphan_name,
1276                                                           &orphan_entry_bh);
1277                         if (status < 0) {
1278                                 mlog_errno(status);
1279                                 goto bail;
1280                         }
1281                 }
1282         } else {
1283                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1284
1285                 status = ocfs2_check_dir_for_entry(new_dir,
1286                                                    new_dentry->d_name.name,
1287                                                    new_dentry->d_name.len);
1288                 if (status)
1289                         goto bail;
1290
1291                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1292                                                       new_dentry->d_name.name,
1293                                                       new_dentry->d_name.len,
1294                                                       &insert_entry_bh);
1295                 if (status < 0) {
1296                         mlog_errno(status);
1297                         goto bail;
1298                 }
1299         }
1300
1301         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1302         if (IS_ERR(handle)) {
1303                 status = PTR_ERR(handle);
1304                 handle = NULL;
1305                 mlog_errno(status);
1306                 goto bail;
1307         }
1308
1309         if (new_de) {
1310                 if (S_ISDIR(new_inode->i_mode)) {
1311                         if (!ocfs2_empty_dir(new_inode) ||
1312                             new_inode->i_nlink != 2) {
1313                                 status = -ENOTEMPTY;
1314                                 goto bail;
1315                         }
1316                 }
1317                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1318                                               OCFS2_JOURNAL_ACCESS_WRITE);
1319                 if (status < 0) {
1320                         mlog_errno(status);
1321                         goto bail;
1322                 }
1323
1324                 if (S_ISDIR(new_inode->i_mode) ||
1325                     (newfe->i_links_count == cpu_to_le16(1))){
1326                         status = ocfs2_orphan_add(osb, handle, new_inode,
1327                                                   newfe, orphan_name,
1328                                                   orphan_entry_bh, orphan_dir);
1329                         if (status < 0) {
1330                                 mlog_errno(status);
1331                                 goto bail;
1332                         }
1333                 }
1334
1335                 /* change the dirent to point to the correct inode */
1336                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1337                                               OCFS2_JOURNAL_ACCESS_WRITE);
1338                 if (status < 0) {
1339                         mlog_errno(status);
1340                         goto bail;
1341                 }
1342                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1343                 new_de->file_type = old_de->file_type;
1344                 new_dir->i_version++;
1345                 status = ocfs2_journal_dirty(handle, new_de_bh);
1346                 if (status < 0) {
1347                         mlog_errno(status);
1348                         goto bail;
1349                 }
1350
1351                 if (S_ISDIR(new_inode->i_mode))
1352                         newfe->i_links_count = 0;
1353                 else
1354                         le16_add_cpu(&newfe->i_links_count, -1);
1355
1356                 status = ocfs2_journal_dirty(handle, newfe_bh);
1357                 if (status < 0) {
1358                         mlog_errno(status);
1359                         goto bail;
1360                 }
1361         } else {
1362                 /* if the name was not found in new_dir, add it now */
1363                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1364                                          OCFS2_I(old_inode)->ip_blkno,
1365                                          new_dir_bh, insert_entry_bh);
1366         }
1367
1368         old_inode->i_ctime = CURRENT_TIME;
1369         mark_inode_dirty(old_inode);
1370
1371         /* now that the name has been added to new_dir, remove the old name */
1372         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1373         if (status < 0) {
1374                 mlog_errno(status);
1375                 goto bail;
1376         }
1377
1378         if (new_inode) {
1379                 new_inode->i_nlink--;
1380                 new_inode->i_ctime = CURRENT_TIME;
1381         }
1382         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1383         if (old_inode_de_bh) {
1384                 status = ocfs2_journal_access(handle, old_inode,
1385                                              old_inode_de_bh,
1386                                              OCFS2_JOURNAL_ACCESS_WRITE);
1387                 PARENT_INO(old_inode_de_bh->b_data) =
1388                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1389                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1390                 old_dir->i_nlink--;
1391                 if (new_inode) {
1392                         new_inode->i_nlink--;
1393                 } else {
1394                         inc_nlink(new_dir);
1395                         mark_inode_dirty(new_dir);
1396                 }
1397         }
1398         mark_inode_dirty(old_dir);
1399         if (new_inode)
1400                 mark_inode_dirty(new_inode);
1401
1402         if (old_dir != new_dir)
1403                 if (new_dir_nlink != new_dir->i_nlink) {
1404                         if (!new_dir_bh) {
1405                                 mlog(ML_ERROR, "need to change nlink for new "
1406                                      "dir %llu from %d to %d but bh is NULL\n",
1407                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1408                                      (int)new_dir_nlink, new_dir->i_nlink);
1409                         } else {
1410                                 struct ocfs2_dinode *fe;
1411                                 status = ocfs2_journal_access(handle,
1412                                                               new_dir,
1413                                                               new_dir_bh,
1414                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1415                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1416                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1417                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1418                         }
1419                 }
1420
1421         if (old_dir_nlink != old_dir->i_nlink) {
1422                 if (!old_dir_bh) {
1423                         mlog(ML_ERROR, "need to change nlink for old dir "
1424                              "%llu from %d to %d but bh is NULL!\n",
1425                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1426                              (int)old_dir_nlink, old_dir->i_nlink);
1427                 } else {
1428                         struct ocfs2_dinode *fe;
1429                         status = ocfs2_journal_access(handle, old_dir,
1430                                                       old_dir_bh,
1431                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1432                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1433                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1434                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1435                 }
1436         }
1437
1438         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1439         status = 0;
1440 bail:
1441         if (rename_lock)
1442                 ocfs2_rename_unlock(osb);
1443
1444         if (handle)
1445                 ocfs2_commit_trans(handle);
1446
1447         if (parents_locked)
1448                 ocfs2_double_unlock(old_dir, new_dir);
1449
1450         if (old_child_locked)
1451                 ocfs2_meta_unlock(old_inode, 1);
1452
1453         if (new_child_locked)
1454                 ocfs2_meta_unlock(new_inode, 1);
1455
1456         if (orphan_dir) {
1457                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1458                 ocfs2_meta_unlock(orphan_dir, 1);
1459                 mutex_unlock(&orphan_dir->i_mutex);
1460                 iput(orphan_dir);
1461         }
1462
1463         if (new_inode)
1464                 sync_mapping_buffers(old_inode->i_mapping);
1465
1466         if (new_inode)
1467                 iput(new_inode);
1468         if (newfe_bh)
1469                 brelse(newfe_bh);
1470         if (old_dir_bh)
1471                 brelse(old_dir_bh);
1472         if (new_dir_bh)
1473                 brelse(new_dir_bh);
1474         if (new_de_bh)
1475                 brelse(new_de_bh);
1476         if (old_de_bh)
1477                 brelse(old_de_bh);
1478         if (old_inode_de_bh)
1479                 brelse(old_inode_de_bh);
1480         if (orphan_entry_bh)
1481                 brelse(orphan_entry_bh);
1482         if (insert_entry_bh)
1483                 brelse(insert_entry_bh);
1484
1485         mlog_exit(status);
1486
1487         return status;
1488 }
1489
1490 /*
1491  * we expect i_size = strlen(symname). Copy symname into the file
1492  * data, including the null terminator.
1493  */
1494 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1495                                      struct ocfs2_journal_handle *handle,
1496                                      struct inode *inode,
1497                                      const char *symname)
1498 {
1499         struct buffer_head **bhs = NULL;
1500         const char *c;
1501         struct super_block *sb = osb->sb;
1502         u64 p_blkno;
1503         int p_blocks;
1504         int virtual, blocks, status, i, bytes_left;
1505
1506         bytes_left = i_size_read(inode) + 1;
1507         /* we can't trust i_blocks because we're actually going to
1508          * write i_size + 1 bytes. */
1509         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1510
1511         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1512                         (unsigned long long)inode->i_blocks,
1513                         i_size_read(inode), blocks);
1514
1515         /* Sanity check -- make sure we're going to fit. */
1516         if (bytes_left >
1517             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1518                 status = -EIO;
1519                 mlog_errno(status);
1520                 goto bail;
1521         }
1522
1523         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1524         if (!bhs) {
1525                 status = -ENOMEM;
1526                 mlog_errno(status);
1527                 goto bail;
1528         }
1529
1530         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1531                                              &p_blocks);
1532         if (status < 0) {
1533                 mlog_errno(status);
1534                 goto bail;
1535         }
1536
1537         /* links can never be larger than one cluster so we know this
1538          * is all going to be contiguous, but do a sanity check
1539          * anyway. */
1540         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1541                 status = -EIO;
1542                 mlog_errno(status);
1543                 goto bail;
1544         }
1545
1546         virtual = 0;
1547         while(bytes_left > 0) {
1548                 c = &symname[virtual * sb->s_blocksize];
1549
1550                 bhs[virtual] = sb_getblk(sb, p_blkno);
1551                 if (!bhs[virtual]) {
1552                         status = -ENOMEM;
1553                         mlog_errno(status);
1554                         goto bail;
1555                 }
1556                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1557
1558                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1559                                               OCFS2_JOURNAL_ACCESS_CREATE);
1560                 if (status < 0) {
1561                         mlog_errno(status);
1562                         goto bail;
1563                 }
1564
1565                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1566
1567                 memcpy(bhs[virtual]->b_data, c,
1568                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1569                        bytes_left);
1570
1571                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1572                 if (status < 0) {
1573                         mlog_errno(status);
1574                         goto bail;
1575                 }
1576
1577                 virtual++;
1578                 p_blkno++;
1579                 bytes_left -= sb->s_blocksize;
1580         }
1581
1582         status = 0;
1583 bail:
1584
1585         if (bhs) {
1586                 for(i = 0; i < blocks; i++)
1587                         if (bhs[i])
1588                                 brelse(bhs[i]);
1589                 kfree(bhs);
1590         }
1591
1592         mlog_exit(status);
1593         return status;
1594 }
1595
1596 static int ocfs2_symlink(struct inode *dir,
1597                          struct dentry *dentry,
1598                          const char *symname)
1599 {
1600         int status, l, credits;
1601         u64 newsize;
1602         struct ocfs2_super *osb = NULL;
1603         struct inode *inode = NULL;
1604         struct super_block *sb;
1605         struct buffer_head *new_fe_bh = NULL;
1606         struct buffer_head *de_bh = NULL;
1607         struct buffer_head *parent_fe_bh = NULL;
1608         struct ocfs2_dinode *fe = NULL;
1609         struct ocfs2_dinode *dirfe;
1610         struct ocfs2_journal_handle *handle = NULL;
1611         struct ocfs2_alloc_context *inode_ac = NULL;
1612         struct ocfs2_alloc_context *data_ac = NULL;
1613
1614         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1615                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1616
1617         sb = dir->i_sb;
1618         osb = OCFS2_SB(sb);
1619
1620         l = strlen(symname) + 1;
1621
1622         credits = ocfs2_calc_symlink_credits(sb);
1623
1624         /* lock the parent directory */
1625         status = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
1626         if (status < 0) {
1627                 if (status != -ENOENT)
1628                         mlog_errno(status);
1629                 return status;
1630         }
1631
1632         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1633         if (!dirfe->i_links_count) {
1634                 /* can't make a file in a deleted directory. */
1635                 status = -ENOENT;
1636                 goto bail;
1637         }
1638
1639         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1640                                            dentry->d_name.len);
1641         if (status)
1642                 goto bail;
1643
1644         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1645                                               dentry->d_name.name,
1646                                               dentry->d_name.len, &de_bh);
1647         if (status < 0) {
1648                 mlog_errno(status);
1649                 goto bail;
1650         }
1651
1652         handle = ocfs2_alloc_handle(osb);
1653         if (handle == NULL) {
1654                 status = -ENOMEM;
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1660         if (status < 0) {
1661                 if (status != -ENOSPC)
1662                         mlog_errno(status);
1663                 goto bail;
1664         }
1665
1666         /* don't reserve bitmap space for fast symlinks. */
1667         if (l > ocfs2_fast_symlink_chars(sb)) {
1668                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1669                 if (status < 0) {
1670                         if (status != -ENOSPC)
1671                                 mlog_errno(status);
1672                         goto bail;
1673                 }
1674         }
1675
1676         handle = ocfs2_start_trans(osb, handle, credits);
1677         if (IS_ERR(handle)) {
1678                 status = PTR_ERR(handle);
1679                 handle = NULL;
1680                 mlog_errno(status);
1681                 goto bail;
1682         }
1683
1684         status = ocfs2_mknod_locked(osb, dir, dentry,
1685                                     S_IFLNK | S_IRWXUGO, 0,
1686                                     &new_fe_bh, parent_fe_bh, handle,
1687                                     &inode, inode_ac);
1688         if (status < 0) {
1689                 mlog_errno(status);
1690                 goto bail;
1691         }
1692
1693         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1694         inode->i_rdev = 0;
1695         newsize = l - 1;
1696         if (l > ocfs2_fast_symlink_chars(sb)) {
1697                 inode->i_op = &ocfs2_symlink_inode_operations;
1698                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1699                                                     handle, data_ac, NULL,
1700                                                     NULL);
1701                 if (status < 0) {
1702                         if (status != -ENOSPC && status != -EINTR) {
1703                                 mlog(ML_ERROR,
1704                                      "Failed to extend file to %llu\n",
1705                                      (unsigned long long)newsize);
1706                                 mlog_errno(status);
1707                                 status = -ENOSPC;
1708                         }
1709                         goto bail;
1710                 }
1711                 i_size_write(inode, newsize);
1712                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1713         } else {
1714                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1715                 memcpy((char *) fe->id2.i_symlink, symname, l);
1716                 i_size_write(inode, newsize);
1717                 inode->i_blocks = 0;
1718         }
1719
1720         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         if (!ocfs2_inode_is_fast_symlink(inode)) {
1727                 status = ocfs2_create_symlink_data(osb, handle, inode,
1728                                                    symname);
1729                 if (status < 0) {
1730                         mlog_errno(status);
1731                         goto bail;
1732                 }
1733         }
1734
1735         status = ocfs2_add_entry(handle, dentry, inode,
1736                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1737                                  de_bh);
1738         if (status < 0) {
1739                 mlog_errno(status);
1740                 goto bail;
1741         }
1742
1743         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1744         if (status) {
1745                 mlog_errno(status);
1746                 goto bail;
1747         }
1748
1749         insert_inode_hash(inode);
1750         dentry->d_op = &ocfs2_dentry_ops;
1751         d_instantiate(dentry, inode);
1752 bail:
1753         if (handle)
1754                 ocfs2_commit_trans(handle);
1755
1756         ocfs2_meta_unlock(dir, 1);
1757
1758         if (new_fe_bh)
1759                 brelse(new_fe_bh);
1760         if (parent_fe_bh)
1761                 brelse(parent_fe_bh);
1762         if (de_bh)
1763                 brelse(de_bh);
1764         if (inode_ac)
1765                 ocfs2_free_alloc_context(inode_ac);
1766         if (data_ac)
1767                 ocfs2_free_alloc_context(data_ac);
1768         if ((status < 0) && inode)
1769                 iput(inode);
1770
1771         mlog_exit(status);
1772
1773         return status;
1774 }
1775
1776 int ocfs2_check_dir_entry(struct inode * dir,
1777                           struct ocfs2_dir_entry * de,
1778                           struct buffer_head * bh,
1779                           unsigned long offset)
1780 {
1781         const char *error_msg = NULL;
1782         const int rlen = le16_to_cpu(de->rec_len);
1783
1784         if (rlen < OCFS2_DIR_REC_LEN(1))
1785                 error_msg = "rec_len is smaller than minimal";
1786         else if (rlen % 4 != 0)
1787                 error_msg = "rec_len % 4 != 0";
1788         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1789                 error_msg = "rec_len is too small for name_len";
1790         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1791                 error_msg = "directory entry across blocks";
1792
1793         if (error_msg != NULL)
1794                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1795                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1796                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1797                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1798                      de->name_len);
1799         return error_msg == NULL ? 1 : 0;
1800 }
1801
1802 /* we don't always have a dentry for what we want to add, so people
1803  * like orphan dir can call this instead.
1804  *
1805  * If you pass me insert_bh, I'll skip the search of the other dir
1806  * blocks and put the record in there.
1807  */
1808 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1809                              struct inode *dir,
1810                              const char *name, int namelen,
1811                              struct inode *inode, u64 blkno,
1812                              struct buffer_head *parent_fe_bh,
1813                              struct buffer_head *insert_bh)
1814 {
1815         unsigned long offset;
1816         unsigned short rec_len;
1817         struct ocfs2_dir_entry *de, *de1;
1818         struct super_block *sb;
1819         int retval, status;
1820
1821         mlog_entry_void();
1822
1823         sb = dir->i_sb;
1824
1825         if (!namelen)
1826                 return -EINVAL;
1827
1828         rec_len = OCFS2_DIR_REC_LEN(namelen);
1829         offset = 0;
1830         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1831         while (1) {
1832                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1833                 /* These checks should've already been passed by the
1834                  * prepare function, but I guess we can leave them
1835                  * here anyway. */
1836                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1837                         retval = -ENOENT;
1838                         goto bail;
1839                 }
1840                 if (ocfs2_match(namelen, name, de)) {
1841                         retval = -EEXIST;
1842                         goto bail;
1843                 }
1844                 if (((le64_to_cpu(de->inode) == 0) &&
1845                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1846                     (le16_to_cpu(de->rec_len) >=
1847                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1848                         status = ocfs2_journal_access(handle, dir, insert_bh,
1849                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1850                         /* By now the buffer is marked for journaling */
1851                         offset += le16_to_cpu(de->rec_len);
1852                         if (le64_to_cpu(de->inode)) {
1853                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1854                                         OCFS2_DIR_REC_LEN(de->name_len));
1855                                 de1->rec_len =
1856                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1857                                         OCFS2_DIR_REC_LEN(de->name_len));
1858                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1859                                 de = de1;
1860                         }
1861                         de->file_type = OCFS2_FT_UNKNOWN;
1862                         if (blkno) {
1863                                 de->inode = cpu_to_le64(blkno);
1864                                 ocfs2_set_de_type(de, inode->i_mode);
1865                         } else
1866                                 de->inode = 0;
1867                         de->name_len = namelen;
1868                         memcpy(de->name, name, namelen);
1869
1870                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1871                         dir->i_version++;
1872                         status = ocfs2_journal_dirty(handle, insert_bh);
1873                         retval = 0;
1874                         goto bail;
1875                 }
1876                 offset += le16_to_cpu(de->rec_len);
1877                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1878         }
1879
1880         /* when you think about it, the assert above should prevent us
1881          * from ever getting here. */
1882         retval = -ENOSPC;
1883 bail:
1884
1885         mlog_exit(retval);
1886         return retval;
1887 }
1888
1889
1890 /*
1891  * ocfs2_delete_entry deletes a directory entry by merging it with the
1892  * previous entry
1893  */
1894 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1895                               struct inode *dir,
1896                               struct ocfs2_dir_entry *de_del,
1897                               struct buffer_head *bh)
1898 {
1899         struct ocfs2_dir_entry *de, *pde;
1900         int i, status = -ENOENT;
1901
1902         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1903
1904         i = 0;
1905         pde = NULL;
1906         de = (struct ocfs2_dir_entry *) bh->b_data;
1907         while (i < bh->b_size) {
1908                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1909                         status = -EIO;
1910                         mlog_errno(status);
1911                         goto bail;
1912                 }
1913                 if (de == de_del)  {
1914                         status = ocfs2_journal_access(handle, dir, bh,
1915                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1916                         if (status < 0) {
1917                                 status = -EIO;
1918                                 mlog_errno(status);
1919                                 goto bail;
1920                         }
1921                         if (pde)
1922                                 pde->rec_len =
1923                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1924                                                     le16_to_cpu(de->rec_len));
1925                         else
1926                                 de->inode = 0;
1927                         dir->i_version++;
1928                         status = ocfs2_journal_dirty(handle, bh);
1929                         goto bail;
1930                 }
1931                 i += le16_to_cpu(de->rec_len);
1932                 pde = de;
1933                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1934         }
1935 bail:
1936         mlog_exit(status);
1937         return status;
1938 }
1939
1940 /*
1941  * Returns 0 if not found, -1 on failure, and 1 on success
1942  */
1943 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1944                                         struct inode *dir,
1945                                         const char *name, int namelen,
1946                                         unsigned long offset,
1947                                         struct ocfs2_dir_entry **res_dir)
1948 {
1949         struct ocfs2_dir_entry *de;
1950         char *dlimit, *de_buf;
1951         int de_len;
1952         int ret = 0;
1953
1954         mlog_entry_void();
1955
1956         de_buf = bh->b_data;
1957         dlimit = de_buf + dir->i_sb->s_blocksize;
1958
1959         while (de_buf < dlimit) {
1960                 /* this code is executed quadratically often */
1961                 /* do minimal checking `by hand' */
1962
1963                 de = (struct ocfs2_dir_entry *) de_buf;
1964
1965                 if (de_buf + namelen <= dlimit &&
1966                     ocfs2_match(namelen, name, de)) {
1967                         /* found a match - just to be sure, do a full check */
1968                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1969                                 ret = -1;
1970                                 goto bail;
1971                         }
1972                         *res_dir = de;
1973                         ret = 1;
1974                         goto bail;
1975                 }
1976
1977                 /* prevent looping on a bad block */
1978                 de_len = le16_to_cpu(de->rec_len);
1979                 if (de_len <= 0) {
1980                         ret = -1;
1981                         goto bail;
1982                 }
1983
1984                 de_buf += de_len;
1985                 offset += de_len;
1986         }
1987
1988 bail:
1989         mlog_exit(ret);
1990         return ret;
1991 }
1992
1993 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1994                                      struct inode *dir,
1995                                      struct ocfs2_dir_entry **res_dir)
1996 {
1997         struct super_block *sb;
1998         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1999         struct buffer_head *bh, *ret = NULL;
2000         unsigned long start, block, b;
2001         int ra_max = 0;         /* Number of bh's in the readahead
2002                                    buffer, bh_use[] */
2003         int ra_ptr = 0;         /* Current index into readahead
2004                                    buffer */
2005         int num = 0;
2006         int nblocks, i, err;
2007
2008         mlog_entry_void();
2009
2010         *res_dir = NULL;
2011         sb = dir->i_sb;
2012
2013         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2014         start = OCFS2_I(dir)->ip_dir_start_lookup;
2015         if (start >= nblocks)
2016                 start = 0;
2017         block = start;
2018
2019 restart:
2020         do {
2021                 /*
2022                  * We deal with the read-ahead logic here.
2023                  */
2024                 if (ra_ptr >= ra_max) {
2025                         /* Refill the readahead buffer */
2026                         ra_ptr = 0;
2027                         b = block;
2028                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2029                                 /*
2030                                  * Terminate if we reach the end of the
2031                                  * directory and must wrap, or if our
2032                                  * search has finished at this block.
2033                                  */
2034                                 if (b >= nblocks || (num && block == start)) {
2035                                         bh_use[ra_max] = NULL;
2036                                         break;
2037                                 }
2038                                 num++;
2039
2040                                 bh = ocfs2_bread(dir, b++, &err, 1);
2041                                 bh_use[ra_max] = bh;
2042                         }
2043                 }
2044                 if ((bh = bh_use[ra_ptr++]) == NULL)
2045                         goto next;
2046                 wait_on_buffer(bh);
2047                 if (!buffer_uptodate(bh)) {
2048                         /* read error, skip block & hope for the best */
2049                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2050                                     "offset %lu\n",
2051                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2052                                     block);
2053                         brelse(bh);
2054                         goto next;
2055                 }
2056                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2057                                           block << sb->s_blocksize_bits,
2058                                           res_dir);
2059                 if (i == 1) {
2060                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2061                         ret = bh;
2062                         goto cleanup_and_exit;
2063                 } else {
2064                         brelse(bh);
2065                         if (i < 0)
2066                                 goto cleanup_and_exit;
2067                 }
2068         next:
2069                 if (++block >= nblocks)
2070                         block = 0;
2071         } while (block != start);
2072
2073         /*
2074          * If the directory has grown while we were searching, then
2075          * search the last part of the directory before giving up.
2076          */
2077         block = nblocks;
2078         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2079         if (block < nblocks) {
2080                 start = 0;
2081                 goto restart;
2082         }
2083
2084 cleanup_and_exit:
2085         /* Clean up the read-ahead blocks */
2086         for (; ra_ptr < ra_max; ra_ptr++)
2087                 brelse(bh_use[ra_ptr]);
2088
2089         mlog_exit_ptr(ret);
2090         return ret;
2091 }
2092
2093 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2094 {
2095         int status, namelen;
2096
2097         mlog_entry_void();
2098
2099         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2100                            (long long)blkno);
2101         if (namelen <= 0) {
2102                 if (namelen)
2103                         status = namelen;
2104                 else
2105                         status = -EINVAL;
2106                 mlog_errno(status);
2107                 goto bail;
2108         }
2109         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2110                 status = -EINVAL;
2111                 mlog_errno(status);
2112                 goto bail;
2113         }
2114
2115         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2116              namelen);
2117
2118         status = 0;
2119 bail:
2120         mlog_exit(status);
2121         return status;
2122 }
2123
2124 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2125                                     struct inode **ret_orphan_dir,
2126                                     struct inode *inode,
2127                                     char *name,
2128                                     struct buffer_head **de_bh)
2129 {
2130         struct inode *orphan_dir_inode;
2131         struct buffer_head *orphan_dir_bh = NULL;
2132         int status = 0;
2133
2134         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2135         if (status < 0) {
2136                 mlog_errno(status);
2137                 return status;
2138         }
2139
2140         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2141                                                        ORPHAN_DIR_SYSTEM_INODE,
2142                                                        osb->slot_num);
2143         if (!orphan_dir_inode) {
2144                 status = -ENOENT;
2145                 mlog_errno(status);
2146                 return status;
2147         }
2148
2149         mutex_lock(&orphan_dir_inode->i_mutex);
2150
2151         status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
2152         if (status < 0) {
2153                 mlog_errno(status);
2154                 goto leave;
2155         }
2156
2157         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2158                                               orphan_dir_bh, name,
2159                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2160         if (status < 0) {
2161                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2162
2163                 mlog_errno(status);
2164                 goto leave;
2165         }
2166
2167         *ret_orphan_dir = orphan_dir_inode;
2168
2169 leave:
2170         if (status) {
2171                 mutex_unlock(&orphan_dir_inode->i_mutex);
2172                 iput(orphan_dir_inode);
2173         }
2174
2175         if (orphan_dir_bh)
2176                 brelse(orphan_dir_bh);
2177
2178         mlog_exit(status);
2179         return status;
2180 }
2181
2182 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2183                             struct ocfs2_journal_handle *handle,
2184                             struct inode *inode,
2185                             struct ocfs2_dinode *fe,
2186                             char *name,
2187                             struct buffer_head *de_bh,
2188                             struct inode *orphan_dir_inode)
2189 {
2190         struct buffer_head *orphan_dir_bh = NULL;
2191         int status = 0;
2192         struct ocfs2_dinode *orphan_fe;
2193
2194         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2195
2196         status = ocfs2_read_block(osb,
2197                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2198                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2199                                   orphan_dir_inode);
2200         if (status < 0) {
2201                 mlog_errno(status);
2202                 goto leave;
2203         }
2204
2205         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2206                                       OCFS2_JOURNAL_ACCESS_WRITE);
2207         if (status < 0) {
2208                 mlog_errno(status);
2209                 goto leave;
2210         }
2211
2212         /* we're a cluster, and nlink can change on disk from
2213          * underneath us... */
2214         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2215         if (S_ISDIR(inode->i_mode))
2216                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2217         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2218
2219         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2220         if (status < 0) {
2221                 mlog_errno(status);
2222                 goto leave;
2223         }
2224
2225         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2226                                    OCFS2_ORPHAN_NAMELEN, inode,
2227                                    OCFS2_I(inode)->ip_blkno,
2228                                    orphan_dir_bh, de_bh);
2229         if (status < 0) {
2230                 mlog_errno(status);
2231                 goto leave;
2232         }
2233
2234         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2235
2236         /* Record which orphan dir our inode now resides
2237          * in. delete_inode will use this to determine which orphan
2238          * dir to lock. */
2239         spin_lock(&OCFS2_I(inode)->ip_lock);
2240         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2241         spin_unlock(&OCFS2_I(inode)->ip_lock);
2242
2243         mlog(0, "Inode %llu orphaned in slot %d\n",
2244              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2245
2246 leave:
2247         if (orphan_dir_bh)
2248                 brelse(orphan_dir_bh);
2249
2250         mlog_exit(status);
2251         return status;
2252 }
2253
2254 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2255 int ocfs2_orphan_del(struct ocfs2_super *osb,
2256                      struct ocfs2_journal_handle *handle,
2257                      struct inode *orphan_dir_inode,
2258                      struct inode *inode,
2259                      struct buffer_head *orphan_dir_bh)
2260 {
2261         char name[OCFS2_ORPHAN_NAMELEN + 1];
2262         struct ocfs2_dinode *orphan_fe;
2263         int status = 0;
2264         struct buffer_head *target_de_bh = NULL;
2265         struct ocfs2_dir_entry *target_de = NULL;
2266
2267         mlog_entry_void();
2268
2269         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2270         if (status < 0) {
2271                 mlog_errno(status);
2272                 goto leave;
2273         }
2274
2275         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2276              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2277              OCFS2_ORPHAN_NAMELEN);
2278
2279         /* find it's spot in the orphan directory */
2280         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2281                                         orphan_dir_inode, &target_de);
2282         if (!target_de_bh) {
2283                 status = -ENOENT;
2284                 mlog_errno(status);
2285                 goto leave;
2286         }
2287
2288         /* remove it from the orphan directory */
2289         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2290                                     target_de_bh);
2291         if (status < 0) {
2292                 mlog_errno(status);
2293                 goto leave;
2294         }
2295
2296         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2297                                       OCFS2_JOURNAL_ACCESS_WRITE);
2298         if (status < 0) {
2299                 mlog_errno(status);
2300                 goto leave;
2301         }
2302
2303         /* do the i_nlink dance! :) */
2304         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2305         if (S_ISDIR(inode->i_mode))
2306                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2307         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2308
2309         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2310         if (status < 0) {
2311                 mlog_errno(status);
2312                 goto leave;
2313         }
2314
2315 leave:
2316         if (target_de_bh)
2317                 brelse(target_de_bh);
2318
2319         mlog_exit(status);
2320         return status;
2321 }
2322
2323 struct inode_operations ocfs2_dir_iops = {
2324         .create         = ocfs2_create,
2325         .lookup         = ocfs2_lookup,
2326         .link           = ocfs2_link,
2327         .unlink         = ocfs2_unlink,
2328         .rmdir          = ocfs2_unlink,
2329         .symlink        = ocfs2_symlink,
2330         .mkdir          = ocfs2_mkdir,
2331         .mknod          = ocfs2_mknod,
2332         .rename         = ocfs2_rename,
2333         .setattr        = ocfs2_setattr,
2334         .getattr        = ocfs2_getattr,
2335 };