SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "vote.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 #define NAMEI_RA_CHUNKS  2
  68 #define NAMEI_RA_BLOCKS  4
  69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  71
  72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  73                                         struct inode *dir,
  74                                         const char *name, int namelen,
  75                                         unsigned long offset,
  76                                         struct ocfs2_dir_entry **res_dir);
  77
  78 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
  84                              struct inode *dir,
  85                              const char *name, int namelen,
  86                              struct inode *inode, u64 blkno,
  87                              struct buffer_head *parent_fe_bh,
  88                              struct buffer_head *insert_bh);
  89
  90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  91                               struct inode *dir,
  92                               struct dentry *dentry, int mode,
  93                               dev_t dev,
  94                               struct buffer_head **new_fe_bh,
  95                               struct buffer_head *parent_fe_bh,
  96                               struct ocfs2_journal_handle *handle,
  97                               struct inode **ret_inode,
  98                               struct ocfs2_alloc_context *inode_ac);
  99
 100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 101                               struct ocfs2_journal_handle *handle,
 102                               struct inode *parent,
 103                               struct inode *inode,
 104                               struct buffer_head *fe_bh,
 105                               struct ocfs2_alloc_context *data_ac);
 106
 107 static int ocfs2_double_lock(struct ocfs2_super *osb,
 108                              struct ocfs2_journal_handle *handle,
 109                              struct buffer_head **bh1,
 110                              struct inode *inode1,
 111                              struct buffer_head **bh2,
 112                              struct inode *inode2);
 113
 114 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 115                                     struct ocfs2_journal_handle *handle,
 116                                     struct inode *inode,
 117                                     char *name,
 118                                     struct buffer_head **de_bh);
 119
 120 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 121                             struct ocfs2_journal_handle *handle,
 122                             struct inode *inode,
 123                             struct ocfs2_dinode *fe,
 124                             char *name,
 125                             struct buffer_head *de_bh);
 126
 127 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 128                                      struct ocfs2_journal_handle *handle,
 129                                      struct inode *inode,
 130                                      const char *symname);
 131
 132 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
 133                                   struct dentry *dentry,
 134                                   struct inode *inode, u64 blkno,
 135                                   struct buffer_head *parent_fe_bh,
 136                                   struct buffer_head *insert_bh)
 137 {
 138         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 139                                  dentry->d_name.name, dentry->d_name.len,
 140                                  inode, blkno, parent_fe_bh, insert_bh);
 141 }
 142
 143 /* An orphan dir name is an 8 byte value, printed as a hex string */
 144 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 145
 146 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 147                                    struct nameidata *nd)
 148 {
 149         int status;
 150         u64 blkno;
 151         struct buffer_head *dirent_bh = NULL;
 152         struct inode *inode = NULL;
 153         struct dentry *ret;
 154         struct ocfs2_dir_entry *dirent;
 155         struct ocfs2_inode_info *oi;
 156
 157         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 158                    dentry->d_name.len, dentry->d_name.name);
 159
 160         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 161                 ret = ERR_PTR(-ENAMETOOLONG);
 162                 goto bail;
 163         }
 164
 165         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 166              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 167
 168         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
 169         if (status < 0) {
 170                 if (status != -ENOENT)
 171                         mlog_errno(status);
 172                 ret = ERR_PTR(status);
 173                 goto bail;
 174         }
 175
 176         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 177                                           dentry->d_name.len, &blkno,
 178                                           dir, &dirent_bh, &dirent);
 179         if (status < 0)
 180                 goto bail_add;
 181
 182         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
 183         if (IS_ERR(inode)) {
 184                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 185                      (unsigned long long)blkno);
 186                 ret = ERR_PTR(-EACCES);
 187                 goto bail_unlock;
 188         }
 189
 190         oi = OCFS2_I(inode);
 191         /* Clear any orphaned state... If we were able to look up the
 192          * inode from a directory, it certainly can't be orphaned. We
 193          * might have the bad state from a node which intended to
 194          * orphan this inode but crashed before it could commit the
 195          * unlink. */
 196         spin_lock(&oi->ip_lock);
 197         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 198         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 199         spin_unlock(&oi->ip_lock);
 200
 201 bail_add:
 202         dentry->d_op = &ocfs2_dentry_ops;
 203         ret = d_splice_alias(inode, dentry);
 204
 205         if (inode) {
 206                 /*
 207                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 208                  * dentry, it will d_move() it on top of ourse. The
 209                  * return value will indicate this however, so in
 210                  * those cases, we switch them around for the locking
 211                  * code.
 212                  *
 213                  * NOTE: This dentry already has ->d_op set from
 214                  * ocfs2_get_parent() and ocfs2_get_dentry()
 215                  */
 216                 if (ret)
 217                         dentry = ret;
 218
 219                 status = ocfs2_dentry_attach_lock(dentry, inode,
 220                                                   OCFS2_I(dir)->ip_blkno, 0);
 221                 if (status) {
 222                         mlog_errno(status);
 223                         ret = ERR_PTR(status);
 224                         goto bail_unlock;
 225                 }
 226         }
 227
 228 bail_unlock:
 229         /* Don't drop the cluster lock until *after* the d_add --
 230          * unlink on another node will message us to remove that
 231          * dentry under this lock so otherwise we can race this with
 232          * the vote thread and have a stale dentry. */
 233         ocfs2_meta_unlock(dir, 0);
 234
 235 bail:
 236         if (dirent_bh)
 237                 brelse(dirent_bh);
 238
 239         mlog_exit_ptr(ret);
 240
 241         return ret;
 242 }
 243
 244 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 245                               struct ocfs2_journal_handle *handle,
 246                               struct inode *parent,
 247                               struct inode *inode,
 248                               struct buffer_head *fe_bh,
 249                               struct ocfs2_alloc_context *data_ac)
 250 {
 251         int status;
 252         struct buffer_head *new_bh = NULL;
 253         struct ocfs2_dir_entry *de = NULL;
 254
 255         mlog_entry_void();
 256
 257         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 258                                      data_ac, NULL, &new_bh);
 259         if (status < 0) {
 260                 mlog_errno(status);
 261                 goto bail;
 262         }
 263
 264         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 265
 266         status = ocfs2_journal_access(handle, inode, new_bh,
 267                                       OCFS2_JOURNAL_ACCESS_CREATE);
 268         if (status < 0) {
 269                 mlog_errno(status);
 270                 goto bail;
 271         }
 272         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 273
 274         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 275         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 276         de->name_len = 1;
 277         de->rec_len =
 278                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 279         strcpy(de->name, ".");
 280         ocfs2_set_de_type(de, S_IFDIR);
 281         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 282         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 283         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 284                                   OCFS2_DIR_REC_LEN(1));
 285         de->name_len = 2;
 286         strcpy(de->name, "..");
 287         ocfs2_set_de_type(de, S_IFDIR);
 288
 289         status = ocfs2_journal_dirty(handle, new_bh);
 290         if (status < 0) {
 291                 mlog_errno(status);
 292                 goto bail;
 293         }
 294
 295         i_size_write(inode, inode->i_sb->s_blocksize);
 296         inode->i_nlink = 2;
 297         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 298         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 299         if (status < 0) {
 300                 mlog_errno(status);
 301                 goto bail;
 302         }
 303
 304         status = 0;
 305 bail:
 306         if (new_bh)
 307                 brelse(new_bh);
 308
 309         mlog_exit(status);
 310         return status;
 311 }
 312
 313 static int ocfs2_mknod(struct inode *dir,
 314                        struct dentry *dentry,
 315                        int mode,
 316                        dev_t dev)
 317 {
 318         int status = 0;
 319         struct buffer_head *parent_fe_bh = NULL;
 320         struct ocfs2_journal_handle *handle = NULL;
 321         struct ocfs2_super *osb;
 322         struct ocfs2_dinode *dirfe;
 323         struct buffer_head *new_fe_bh = NULL;
 324         struct buffer_head *de_bh = NULL;
 325         struct inode *inode = NULL;
 326         struct ocfs2_alloc_context *inode_ac = NULL;
 327         struct ocfs2_alloc_context *data_ac = NULL;
 328
 329         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 330                    (unsigned long)dev, dentry->d_name.len,
 331                    dentry->d_name.name);
 332
 333         /* get our super block */
 334         osb = OCFS2_SB(dir->i_sb);
 335
 336         handle = ocfs2_alloc_handle(osb);
 337         if (handle == NULL) {
 338                 status = -ENOMEM;
 339                 mlog_errno(status);
 340                 goto leave;
 341         }
 342
 343         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 344         if (status < 0) {
 345                 if (status != -ENOENT)
 346                         mlog_errno(status);
 347                 goto leave;
 348         }
 349
 350         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 351                 status = -EMLINK;
 352                 goto leave;
 353         }
 354
 355         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 356         if (!dirfe->i_links_count) {
 357                 /* can't make a file in a deleted directory. */
 358                 status = -ENOENT;
 359                 goto leave;
 360         }
 361
 362         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 363                                            dentry->d_name.len);
 364         if (status)
 365                 goto leave;
 366
 367         /* get a spot inside the dir. */
 368         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 369                                               dentry->d_name.name,
 370                                               dentry->d_name.len, &de_bh);
 371         if (status < 0) {
 372                 mlog_errno(status);
 373                 goto leave;
 374         }
 375
 376         /* reserve an inode spot */
 377         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
 378         if (status < 0) {
 379                 if (status != -ENOSPC)
 380                         mlog_errno(status);
 381                 goto leave;
 382         }
 383
 384         /* are we making a directory? If so, reserve a cluster for his
 385          * 1st extent. */
 386         if (S_ISDIR(mode)) {
 387                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 388                 if (status < 0) {
 389                         if (status != -ENOSPC)
 390                                 mlog_errno(status);
 391                         goto leave;
 392                 }
 393         }
 394
 395         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
 396         if (IS_ERR(handle)) {
 397                 status = PTR_ERR(handle);
 398                 handle = NULL;
 399                 mlog_errno(status);
 400                 goto leave;
 401         }
 402
 403         /* do the real work now. */
 404         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 405                                     &new_fe_bh, parent_fe_bh, handle,
 406                                     &inode, inode_ac);
 407         if (status < 0) {
 408                 mlog_errno(status);
 409                 goto leave;
 410         }
 411
 412         if (S_ISDIR(mode)) {
 413                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 414                                             new_fe_bh, data_ac);
 415                 if (status < 0) {
 416                         mlog_errno(status);
 417                         goto leave;
 418                 }
 419
 420                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 421                                               OCFS2_JOURNAL_ACCESS_WRITE);
 422                 if (status < 0) {
 423                         mlog_errno(status);
 424                         goto leave;
 425                 }
 426                 le16_add_cpu(&dirfe->i_links_count, 1);
 427                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 428                 if (status < 0) {
 429                         mlog_errno(status);
 430                         goto leave;
 431                 }
 432                 dir->i_nlink++;
 433         }
 434
 435         status = ocfs2_add_entry(handle, dentry, inode,
 436                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 437                                  de_bh);
 438         if (status < 0) {
 439                 mlog_errno(status);
 440                 goto leave;
 441         }
 442
 443         status = ocfs2_dentry_attach_lock(dentry, inode,
 444                                           OCFS2_I(dir)->ip_blkno, 1);
 445         if (status) {
 446                 mlog_errno(status);
 447                 goto leave;
 448         }
 449
 450         insert_inode_hash(inode);
 451         dentry->d_op = &ocfs2_dentry_ops;
 452         d_instantiate(dentry, inode);
 453         status = 0;
 454 leave:
 455         if (handle)
 456                 ocfs2_commit_trans(handle);
 457
 458         if (status == -ENOSPC)
 459                 mlog(0, "Disk is full\n");
 460
 461         if (new_fe_bh)
 462                 brelse(new_fe_bh);
 463
 464         if (de_bh)
 465                 brelse(de_bh);
 466
 467         if (parent_fe_bh)
 468                 brelse(parent_fe_bh);
 469
 470         if ((status < 0) && inode)
 471                 iput(inode);
 472
 473         if (inode_ac)
 474                 ocfs2_free_alloc_context(inode_ac);
 475
 476         if (data_ac)
 477                 ocfs2_free_alloc_context(data_ac);
 478
 479         mlog_exit(status);
 480
 481         return status;
 482 }
 483
 484 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 485                               struct inode *dir,
 486                               struct dentry *dentry, int mode,
 487                               dev_t dev,
 488                               struct buffer_head **new_fe_bh,
 489                               struct buffer_head *parent_fe_bh,
 490                               struct ocfs2_journal_handle *handle,
 491                               struct inode **ret_inode,
 492                               struct ocfs2_alloc_context *inode_ac)
 493 {
 494         int status = 0;
 495         struct ocfs2_dinode *fe = NULL;
 496         struct ocfs2_extent_list *fel;
 497         u64 fe_blkno = 0;
 498         u16 suballoc_bit;
 499         struct inode *inode = NULL;
 500
 501         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 502                    (unsigned long)dev, dentry->d_name.len,
 503                    dentry->d_name.name);
 504
 505         *new_fe_bh = NULL;
 506         *ret_inode = NULL;
 507
 508         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 509                                        &fe_blkno);
 510         if (status < 0) {
 511                 mlog_errno(status);
 512                 goto leave;
 513         }
 514
 515         inode = new_inode(dir->i_sb);
 516         if (IS_ERR(inode)) {
 517                 status = PTR_ERR(inode);
 518                 mlog(ML_ERROR, "new_inode failed!\n");
 519                 goto leave;
 520         }
 521
 522         /* populate as many fields early on as possible - many of
 523          * these are used by the support functions here and in
 524          * callers. */
 525         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 526         OCFS2_I(inode)->ip_blkno = fe_blkno;
 527         if (S_ISDIR(mode))
 528                 inode->i_nlink = 2;
 529         else
 530                 inode->i_nlink = 1;
 531         inode->i_mode = mode;
 532         spin_lock(&osb->osb_lock);
 533         inode->i_generation = osb->s_next_generation++;
 534         spin_unlock(&osb->osb_lock);
 535
 536         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 537         if (!*new_fe_bh) {
 538                 status = -EIO;
 539                 mlog_errno(status);
 540                 goto leave;
 541         }
 542         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 543
 544         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 545                                       OCFS2_JOURNAL_ACCESS_CREATE);
 546         if (status < 0) {
 547                 mlog_errno(status);
 548                 goto leave;
 549         }
 550
 551         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 552         memset(fe, 0, osb->sb->s_blocksize);
 553
 554         fe->i_generation = cpu_to_le32(inode->i_generation);
 555         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 556         fe->i_blkno = cpu_to_le64(fe_blkno);
 557         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 558         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 559         fe->i_uid = cpu_to_le32(current->fsuid);
 560         if (dir->i_mode & S_ISGID) {
 561                 fe->i_gid = cpu_to_le32(dir->i_gid);
 562                 if (S_ISDIR(mode))
 563                         mode |= S_ISGID;
 564         } else
 565                 fe->i_gid = cpu_to_le32(current->fsgid);
 566         fe->i_mode = cpu_to_le16(mode);
 567         if (S_ISCHR(mode) || S_ISBLK(mode))
 568                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 569
 570         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 571
 572         fe->i_last_eb_blk = 0;
 573         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 574         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 575         fe->i_atime = fe->i_ctime = fe->i_mtime =
 576                 cpu_to_le64(CURRENT_TIME.tv_sec);
 577         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 578                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 579         fe->i_dtime = 0;
 580
 581         fel = &fe->id2.i_list;
 582         fel->l_tree_depth = 0;
 583         fel->l_next_free_rec = 0;
 584         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 585
 586         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 587         if (status < 0) {
 588                 mlog_errno(status);
 589                 goto leave;
 590         }
 591
 592         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 593                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 594                      "i_blkno=%llu, i_ino=%lu\n",
 595                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 596                      (unsigned long long)fe->i_blkno, inode->i_ino);
 597                 BUG();
 598         }
 599
 600         ocfs2_inode_set_new(osb, inode);
 601         status = ocfs2_create_new_inode_locks(inode);
 602         if (status < 0)
 603                 mlog_errno(status);
 604
 605         status = 0; /* error in ocfs2_create_new_inode_locks is not
 606                      * critical */
 607
 608         *ret_inode = inode;
 609 leave:
 610         if (status < 0) {
 611                 if (*new_fe_bh) {
 612                         brelse(*new_fe_bh);
 613                         *new_fe_bh = NULL;
 614                 }
 615                 if (inode)
 616                         iput(inode);
 617         }
 618
 619         mlog_exit(status);
 620         return status;
 621 }
 622
 623 static int ocfs2_mkdir(struct inode *dir,
 624                        struct dentry *dentry,
 625                        int mode)
 626 {
 627         int ret;
 628
 629         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 630                    dentry->d_name.len, dentry->d_name.name);
 631         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 632         mlog_exit(ret);
 633
 634         return ret;
 635 }
 636
 637 static int ocfs2_create(struct inode *dir,
 638                         struct dentry *dentry,
 639                         int mode,
 640                         struct nameidata *nd)
 641 {
 642         int ret;
 643
 644         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 645                    dentry->d_name.len, dentry->d_name.name);
 646         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 647         mlog_exit(ret);
 648
 649         return ret;
 650 }
 651
 652 static int ocfs2_link(struct dentry *old_dentry,
 653                       struct inode *dir,
 654                       struct dentry *dentry)
 655 {
 656         struct ocfs2_journal_handle *handle = NULL;
 657         struct inode *inode = old_dentry->d_inode;
 658         int err;
 659         struct buffer_head *fe_bh = NULL;
 660         struct buffer_head *parent_fe_bh = NULL;
 661         struct buffer_head *de_bh = NULL;
 662         struct ocfs2_dinode *fe = NULL;
 663         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 664
 665         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 666                    old_dentry->d_name.len, old_dentry->d_name.name,
 667                    dentry->d_name.len, dentry->d_name.name);
 668
 669         if (S_ISDIR(inode->i_mode)) {
 670                 err = -EPERM;
 671                 goto bail;
 672         }
 673
 674         handle = ocfs2_alloc_handle(osb);
 675         if (handle == NULL) {
 676                 err = -ENOMEM;
 677                 goto bail;
 678         }
 679
 680         err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
 681         if (err < 0) {
 682                 if (err != -ENOENT)
 683                         mlog_errno(err);
 684                 goto bail;
 685         }
 686
 687         if (!dir->i_nlink) {
 688                 err = -ENOENT;
 689                 goto bail;
 690         }
 691
 692         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 693                                         dentry->d_name.len);
 694         if (err)
 695                 goto bail;
 696
 697         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 698                                            dentry->d_name.name,
 699                                            dentry->d_name.len, &de_bh);
 700         if (err < 0) {
 701                 mlog_errno(err);
 702                 goto bail;
 703         }
 704
 705         err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 706         if (err < 0) {
 707                 if (err != -ENOENT)
 708                         mlog_errno(err);
 709                 goto bail;
 710         }
 711
 712         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 713         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 714                 err = -EMLINK;
 715                 goto bail;
 716         }
 717
 718         handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
 719         if (IS_ERR(handle)) {
 720                 err = PTR_ERR(handle);
 721                 handle = NULL;
 722                 mlog_errno(err);
 723                 goto bail;
 724         }
 725
 726         err = ocfs2_journal_access(handle, inode, fe_bh,
 727                                    OCFS2_JOURNAL_ACCESS_WRITE);
 728         if (err < 0) {
 729                 mlog_errno(err);
 730                 goto bail;
 731         }
 732
 733         inode->i_nlink++;
 734         inode->i_ctime = CURRENT_TIME;
 735         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 736         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 737         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 738
 739         err = ocfs2_journal_dirty(handle, fe_bh);
 740         if (err < 0) {
 741                 le16_add_cpu(&fe->i_links_count, -1);
 742                 inode->i_nlink--;
 743                 mlog_errno(err);
 744                 goto bail;
 745         }
 746
 747         err = ocfs2_add_entry(handle, dentry, inode,
 748                               OCFS2_I(inode)->ip_blkno,
 749                               parent_fe_bh, de_bh);
 750         if (err) {
 751                 le16_add_cpu(&fe->i_links_count, -1);
 752                 inode->i_nlink--;
 753                 mlog_errno(err);
 754                 goto bail;
 755         }
 756
 757         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno,
 758                                        0);
 759         if (err) {
 760                 mlog_errno(err);
 761                 goto bail;
 762         }
 763
 764         atomic_inc(&inode->i_count);
 765         dentry->d_op = &ocfs2_dentry_ops;
 766         d_instantiate(dentry, inode);
 767 bail:
 768         if (handle)
 769                 ocfs2_commit_trans(handle);
 770         if (de_bh)
 771                 brelse(de_bh);
 772         if (fe_bh)
 773                 brelse(fe_bh);
 774         if (parent_fe_bh)
 775                 brelse(parent_fe_bh);
 776
 777         mlog_exit(err);
 778
 779         return err;
 780 }
 781
 782 /*
 783  * Takes and drops an exclusive lock on the given dentry. This will
 784  * force other nodes to drop it.
 785  */
 786 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 787 {
 788         int ret;
 789
 790         ret = ocfs2_dentry_lock(dentry, 1);
 791         if (ret)
 792                 mlog_errno(ret);
 793         else
 794                 ocfs2_dentry_unlock(dentry, 1);
 795
 796         return ret;
 797 }
 798
 799 static int ocfs2_unlink(struct inode *dir,
 800                         struct dentry *dentry)
 801 {
 802         int status;
 803         unsigned int saved_nlink = 0;
 804         struct inode *inode = dentry->d_inode;
 805         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 806         u64 blkno;
 807         struct ocfs2_dinode *fe = NULL;
 808         struct buffer_head *fe_bh = NULL;
 809         struct buffer_head *parent_node_bh = NULL;
 810         struct ocfs2_journal_handle *handle = NULL;
 811         struct ocfs2_dir_entry *dirent = NULL;
 812         struct buffer_head *dirent_bh = NULL;
 813         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 814         struct buffer_head *orphan_entry_bh = NULL;
 815
 816         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 817                    dentry->d_name.len, dentry->d_name.name);
 818
 819         BUG_ON(dentry->d_parent->d_inode != dir);
 820
 821         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 822
 823         if (inode == osb->root_inode) {
 824                 mlog(0, "Cannot delete the root directory\n");
 825                 status = -EPERM;
 826                 goto leave;
 827         }
 828
 829         handle = ocfs2_alloc_handle(osb);
 830         if (handle == NULL) {
 831                 status = -ENOMEM;
 832                 mlog_errno(status);
 833                 goto leave;
 834         }
 835
 836         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
 837         if (status < 0) {
 838                 if (status != -ENOENT)
 839                         mlog_errno(status);
 840                 goto leave;
 841         }
 842
 843         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 844                                           dentry->d_name.len, &blkno,
 845                                           dir, &dirent_bh, &dirent);
 846         if (status < 0) {
 847                 if (status != -ENOENT)
 848                         mlog_errno(status);
 849                 goto leave;
 850         }
 851
 852         if (OCFS2_I(inode)->ip_blkno != blkno) {
 853                 status = -ENOENT;
 854
 855                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 856                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 857                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 858                 goto leave;
 859         }
 860
 861         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 862         if (status < 0) {
 863                 if (status != -ENOENT)
 864                         mlog_errno(status);
 865                 goto leave;
 866         }
 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         /* There are still a few steps left until we can consider the
 879          * unlink to have succeeded. Save off nlink here before
 880          * modification so we can set it back in case we hit an issue
 881          * before commit. */
 882         saved_nlink = inode->i_nlink;
 883         if (S_ISDIR(inode->i_mode))
 884                 inode->i_nlink = 0;
 885         else
 886                 inode->i_nlink--;
 887
 888         status = ocfs2_remote_dentry_delete(dentry);
 889         if (status < 0) {
 890                 /* This vote should succeed under all normal
 891                  * circumstances. */
 892                 mlog_errno(status);
 893                 goto leave;
 894         }
 895
 896         if (!inode->i_nlink) {
 897                 status = ocfs2_prepare_orphan_dir(osb, handle, inode,
 898                                                   orphan_name,
 899                                                   &orphan_entry_bh);
 900                 if (status < 0) {
 901                         mlog_errno(status);
 902                         goto leave;
 903                 }
 904         }
 905
 906         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
 907         if (IS_ERR(handle)) {
 908                 status = PTR_ERR(handle);
 909                 handle = NULL;
 910                 mlog_errno(status);
 911                 goto leave;
 912         }
 913
 914         status = ocfs2_journal_access(handle, inode, fe_bh,
 915                                       OCFS2_JOURNAL_ACCESS_WRITE);
 916         if (status < 0) {
 917                 mlog_errno(status);
 918                 goto leave;
 919         }
 920
 921         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 922
 923         if (!inode->i_nlink) {
 924                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 925                                           orphan_entry_bh);
 926                 if (status < 0) {
 927                         mlog_errno(status);
 928                         goto leave;
 929                 }
 930         }
 931
 932         /* delete the name from the parent dir */
 933         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 934         if (status < 0) {
 935                 mlog_errno(status);
 936                 goto leave;
 937         }
 938
 939         /* We can set nlink on the dinode now. clear the saved version
 940          * so that it doesn't get set later. */
 941         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 942         saved_nlink = 0;
 943
 944         status = ocfs2_journal_dirty(handle, fe_bh);
 945         if (status < 0) {
 946                 mlog_errno(status);
 947                 goto leave;
 948         }
 949
 950         if (S_ISDIR(inode->i_mode)) {
 951                 dir->i_nlink--;
 952                 status = ocfs2_mark_inode_dirty(handle, dir,
 953                                                 parent_node_bh);
 954                 if (status < 0) {
 955                         mlog_errno(status);
 956                         dir->i_nlink++;
 957                 }
 958         }
 959
 960 leave:
 961         if (status < 0 && saved_nlink)
 962                 inode->i_nlink = saved_nlink;
 963
 964         if (handle)
 965                 ocfs2_commit_trans(handle);
 966
 967         if (fe_bh)
 968                 brelse(fe_bh);
 969
 970         if (dirent_bh)
 971                 brelse(dirent_bh);
 972
 973         if (parent_node_bh)
 974                 brelse(parent_node_bh);
 975
 976         if (orphan_entry_bh)
 977                 brelse(orphan_entry_bh);
 978
 979         mlog_exit(status);
 980
 981         return status;
 982 }
 983
 984 /*
 985  * The only place this should be used is rename!
 986  * if they have the same id, then the 1st one is the only one locked.
 987  */
 988 static int ocfs2_double_lock(struct ocfs2_super *osb,
 989                              struct ocfs2_journal_handle *handle,
 990                              struct buffer_head **bh1,
 991                              struct inode *inode1,
 992                              struct buffer_head **bh2,
 993                              struct inode *inode2)
 994 {
 995         int status;
 996         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 997         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 998         struct buffer_head **tmpbh;
 999         struct inode *tmpinode;
1000
1001         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1002                    (unsigned long long)oi1->ip_blkno,
1003                    (unsigned long long)oi2->ip_blkno);
1004
1005         BUG_ON(!handle);
1006
1007         if (*bh1)
1008                 *bh1 = NULL;
1009         if (*bh2)
1010                 *bh2 = NULL;
1011
1012         /* we always want to lock the one with the lower lockid first. */
1013         if (oi1->ip_blkno != oi2->ip_blkno) {
1014                 if (oi1->ip_blkno < oi2->ip_blkno) {
1015                         /* switch id1 and id2 around */
1016                         mlog(0, "switching them around...\n");
1017                         tmpbh = bh2;
1018                         bh2 = bh1;
1019                         bh1 = tmpbh;
1020
1021                         tmpinode = inode2;
1022                         inode2 = inode1;
1023                         inode1 = tmpinode;
1024                 }
1025                 /* lock id2 */
1026                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1027                 if (status < 0) {
1028                         if (status != -ENOENT)
1029                                 mlog_errno(status);
1030                         goto bail;
1031                 }
1032         }
1033         /* lock id1 */
1034         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1035         if (status < 0) {
1036                 if (status != -ENOENT)
1037                         mlog_errno(status);
1038                 goto bail;
1039         }
1040 bail:
1041         mlog_exit(status);
1042         return status;
1043 }
1044
1045 #define PARENT_INO(buffer) \
1046         ((struct ocfs2_dir_entry *) \
1047          ((char *)buffer + \
1048           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1049
1050 static int ocfs2_rename(struct inode *old_dir,
1051                         struct dentry *old_dentry,
1052                         struct inode *new_dir,
1053                         struct dentry *new_dentry)
1054 {
1055         int status = 0, rename_lock = 0;
1056         struct inode *old_inode = old_dentry->d_inode;
1057         struct inode *new_inode = new_dentry->d_inode;
1058         struct ocfs2_dinode *newfe = NULL;
1059         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1060         struct buffer_head *orphan_entry_bh = NULL;
1061         struct buffer_head *newfe_bh = NULL;
1062         struct buffer_head *insert_entry_bh = NULL;
1063         struct ocfs2_super *osb = NULL;
1064         u64 newfe_blkno;
1065         struct ocfs2_journal_handle *handle = NULL;
1066         struct buffer_head *old_dir_bh = NULL;
1067         struct buffer_head *new_dir_bh = NULL;
1068         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1069                                                                // and new_dentry
1070         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1071         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1072                                                     // this is the 1st dirent bh
1073         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1074
1075         /* At some point it might be nice to break this function up a
1076          * bit. */
1077
1078         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1079                    old_dir, old_dentry, new_dir, new_dentry,
1080                    old_dentry->d_name.len, old_dentry->d_name.name,
1081                    new_dentry->d_name.len, new_dentry->d_name.name);
1082
1083         osb = OCFS2_SB(old_dir->i_sb);
1084
1085         if (new_inode) {
1086                 if (!igrab(new_inode))
1087                         BUG();
1088         }
1089
1090         if (atomic_read(&old_dentry->d_count) > 2) {
1091                 shrink_dcache_parent(old_dentry);
1092                 if (atomic_read(&old_dentry->d_count) > 2) {
1093                         status = -EBUSY;
1094                         goto bail;
1095                 }
1096         }
1097
1098         /* Assume a directory heirarchy thusly:
1099          * a/b/c
1100          * a/d
1101          * a,b,c, and d are all directories.
1102          *
1103          * from cwd of 'a' on both nodes:
1104          * node1: mv b/c d
1105          * node2: mv d   b/c
1106          *
1107          * And that's why, just like the VFS, we need a file system
1108          * rename lock. */
1109         if (old_dentry != new_dentry) {
1110                 status = ocfs2_rename_lock(osb);
1111                 if (status < 0) {
1112                         mlog_errno(status);
1113                         goto bail;
1114                 }
1115                 rename_lock = 1;
1116         }
1117
1118         handle = ocfs2_alloc_handle(osb);
1119         if (handle == NULL) {
1120                 status = -ENOMEM;
1121                 mlog_errno(status);
1122                 goto bail;
1123         }
1124
1125         /* if old and new are the same, this'll just do one lock. */
1126         status = ocfs2_double_lock(osb, handle,
1127                                   &old_dir_bh, old_dir,
1128                                   &new_dir_bh, new_dir);
1129         if (status < 0) {
1130                 mlog_errno(status);
1131                 goto bail;
1132         }
1133
1134         /* make sure both dirs have bhs
1135          * get an extra ref on old_dir_bh if old==new */
1136         if (!new_dir_bh) {
1137                 if (old_dir_bh) {
1138                         new_dir_bh = old_dir_bh;
1139                         get_bh(new_dir_bh);
1140                 } else {
1141                         mlog(ML_ERROR, "no old_dir_bh!\n");
1142                         status = -EIO;
1143                         goto bail;
1144                 }
1145         }
1146
1147         /*
1148          * Though we don't require an inode meta data update if
1149          * old_inode is not a directory, we lock anyway here to ensure
1150          * the vote thread on other nodes won't have to concurrently
1151          * downconvert the inode and the dentry locks.
1152          */
1153         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1154         if (status < 0) {
1155                 if (status != -ENOENT)
1156                         mlog_errno(status);
1157                 goto bail;
1158         }
1159
1160         status = ocfs2_remote_dentry_delete(old_dentry);
1161         if (status < 0) {
1162                 mlog_errno(status);
1163                 goto bail;
1164         }
1165
1166         if (S_ISDIR(old_inode->i_mode)) {
1167                 status = -EIO;
1168                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1169                 if (!old_inode_de_bh)
1170                         goto bail;
1171
1172                 status = -EIO;
1173                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1174                     OCFS2_I(old_dir)->ip_blkno)
1175                         goto bail;
1176                 status = -EMLINK;
1177                 if (!new_inode && new_dir!=old_dir &&
1178                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1179                         goto bail;
1180         }
1181
1182         status = -ENOENT;
1183         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1184                                      old_dentry->d_name.len,
1185                                      old_dir, &old_de);
1186         if (!old_de_bh)
1187                 goto bail;
1188
1189         /*
1190          *  Check for inode number is _not_ due to possible IO errors.
1191          *  We might rmdir the source, keep it as pwd of some process
1192          *  and merrily kill the link to whatever was created under the
1193          *  same name. Goodbye sticky bit ;-<
1194          */
1195         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1196                 goto bail;
1197
1198         /* check if the target already exists (in which case we need
1199          * to delete it */
1200         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1201                                           new_dentry->d_name.len,
1202                                           &newfe_blkno, new_dir, &new_de_bh,
1203                                           &new_de);
1204         /* The only error we allow here is -ENOENT because the new
1205          * file not existing is perfectly valid. */
1206         if ((status < 0) && (status != -ENOENT)) {
1207                 /* If we cannot find the file specified we should just */
1208                 /* return the error... */
1209                 mlog_errno(status);
1210                 goto bail;
1211         }
1212
1213         if (!new_de && new_inode)
1214                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1215                      "directory!", new_inode->i_ino);
1216
1217         /* In case we need to overwrite an existing file, we blow it
1218          * away first */
1219         if (new_de) {
1220                 /* VFS didn't think there existed an inode here, but
1221                  * someone else in the cluster must have raced our
1222                  * rename to create one. Today we error cleanly, in
1223                  * the future we should consider calling iget to build
1224                  * a new struct inode for this entry. */
1225                 if (!new_inode) {
1226                         status = -EACCES;
1227
1228                         mlog(0, "We found an inode for name %.*s but VFS "
1229                              "didn't give us one.\n", new_dentry->d_name.len,
1230                              new_dentry->d_name.name);
1231                         goto bail;
1232                 }
1233
1234                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1235                         status = -EACCES;
1236
1237                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1238                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1239                              (unsigned long long)newfe_blkno,
1240                              OCFS2_I(new_inode)->ip_flags);
1241                         goto bail;
1242                 }
1243
1244                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1245                 if (status < 0) {
1246                         if (status != -ENOENT)
1247                                 mlog_errno(status);
1248                         goto bail;
1249                 }
1250
1251                 status = ocfs2_remote_dentry_delete(new_dentry);
1252                 if (status < 0) {
1253                         mlog_errno(status);
1254                         goto bail;
1255                 }
1256
1257                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1258
1259                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1260                      "newfebh=%p bhblocknr=%llu\n", new_de,
1261                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1262                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1263
1264                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1265                         status = ocfs2_prepare_orphan_dir(osb, handle,
1266                                                           new_inode,
1267                                                           orphan_name,
1268                                                           &orphan_entry_bh);
1269                         if (status < 0) {
1270                                 mlog_errno(status);
1271                                 goto bail;
1272                         }
1273                 }
1274         } else {
1275                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1276
1277                 status = ocfs2_check_dir_for_entry(new_dir,
1278                                                    new_dentry->d_name.name,
1279                                                    new_dentry->d_name.len);
1280                 if (status)
1281                         goto bail;
1282
1283                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1284                                                       new_dentry->d_name.name,
1285                                                       new_dentry->d_name.len,
1286                                                       &insert_entry_bh);
1287                 if (status < 0) {
1288                         mlog_errno(status);
1289                         goto bail;
1290                 }
1291         }
1292
1293         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1294         if (IS_ERR(handle)) {
1295                 status = PTR_ERR(handle);
1296                 handle = NULL;
1297                 mlog_errno(status);
1298                 goto bail;
1299         }
1300
1301         if (new_de) {
1302                 if (S_ISDIR(new_inode->i_mode)) {
1303                         if (!ocfs2_empty_dir(new_inode) ||
1304                             new_inode->i_nlink != 2) {
1305                                 status = -ENOTEMPTY;
1306                                 goto bail;
1307                         }
1308                 }
1309                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1310                                               OCFS2_JOURNAL_ACCESS_WRITE);
1311                 if (status < 0) {
1312                         mlog_errno(status);
1313                         goto bail;
1314                 }
1315
1316                 if (S_ISDIR(new_inode->i_mode) ||
1317                     (newfe->i_links_count == cpu_to_le16(1))){
1318                         status = ocfs2_orphan_add(osb, handle, new_inode,
1319                                                   newfe, orphan_name,
1320                                                   orphan_entry_bh);
1321                         if (status < 0) {
1322                                 mlog_errno(status);
1323                                 goto bail;
1324                         }
1325                 }
1326
1327                 /* change the dirent to point to the correct inode */
1328                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1329                                               OCFS2_JOURNAL_ACCESS_WRITE);
1330                 if (status < 0) {
1331                         mlog_errno(status);
1332                         goto bail;
1333                 }
1334                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1335                 new_de->file_type = old_de->file_type;
1336                 new_dir->i_version++;
1337                 status = ocfs2_journal_dirty(handle, new_de_bh);
1338                 if (status < 0) {
1339                         mlog_errno(status);
1340                         goto bail;
1341                 }
1342
1343                 if (S_ISDIR(new_inode->i_mode))
1344                         newfe->i_links_count = 0;
1345                 else
1346                         le16_add_cpu(&newfe->i_links_count, -1);
1347
1348                 status = ocfs2_journal_dirty(handle, newfe_bh);
1349                 if (status < 0) {
1350                         mlog_errno(status);
1351                         goto bail;
1352                 }
1353         } else {
1354                 /* if the name was not found in new_dir, add it now */
1355                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1356                                          OCFS2_I(old_inode)->ip_blkno,
1357                                          new_dir_bh, insert_entry_bh);
1358         }
1359
1360         old_inode->i_ctime = CURRENT_TIME;
1361         mark_inode_dirty(old_inode);
1362
1363         /* now that the name has been added to new_dir, remove the old name */
1364         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1365         if (status < 0) {
1366                 mlog_errno(status);
1367                 goto bail;
1368         }
1369
1370         if (new_inode) {
1371                 new_inode->i_nlink--;
1372                 new_inode->i_ctime = CURRENT_TIME;
1373         }
1374         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1375         if (old_inode_de_bh) {
1376                 status = ocfs2_journal_access(handle, old_inode,
1377                                              old_inode_de_bh,
1378                                              OCFS2_JOURNAL_ACCESS_WRITE);
1379                 PARENT_INO(old_inode_de_bh->b_data) =
1380                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1381                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1382                 old_dir->i_nlink--;
1383                 if (new_inode) {
1384                         new_inode->i_nlink--;
1385                 } else {
1386                         new_dir->i_nlink++;
1387                         mark_inode_dirty(new_dir);
1388                 }
1389         }
1390         mark_inode_dirty(old_dir);
1391         if (new_inode)
1392                 mark_inode_dirty(new_inode);
1393
1394         if (old_dir != new_dir)
1395                 if (new_dir_nlink != new_dir->i_nlink) {
1396                         if (!new_dir_bh) {
1397                                 mlog(ML_ERROR, "need to change nlink for new "
1398                                      "dir %llu from %d to %d but bh is NULL\n",
1399                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1400                                      (int)new_dir_nlink, new_dir->i_nlink);
1401                         } else {
1402                                 struct ocfs2_dinode *fe;
1403                                 status = ocfs2_journal_access(handle,
1404                                                               new_dir,
1405                                                               new_dir_bh,
1406                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1407                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1408                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1409                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1410                         }
1411                 }
1412
1413         if (old_dir_nlink != old_dir->i_nlink) {
1414                 if (!old_dir_bh) {
1415                         mlog(ML_ERROR, "need to change nlink for old dir "
1416                              "%llu from %d to %d but bh is NULL!\n",
1417                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1418                              (int)old_dir_nlink, old_dir->i_nlink);
1419                 } else {
1420                         struct ocfs2_dinode *fe;
1421                         status = ocfs2_journal_access(handle, old_dir,
1422                                                       old_dir_bh,
1423                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1424                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1425                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1426                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1427                 }
1428         }
1429
1430         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1431         status = 0;
1432 bail:
1433         if (rename_lock)
1434                 ocfs2_rename_unlock(osb);
1435
1436         if (handle)
1437                 ocfs2_commit_trans(handle);
1438
1439         if (new_inode)
1440                 sync_mapping_buffers(old_inode->i_mapping);
1441
1442         if (new_inode)
1443                 iput(new_inode);
1444         if (newfe_bh)
1445                 brelse(newfe_bh);
1446         if (old_dir_bh)
1447                 brelse(old_dir_bh);
1448         if (new_dir_bh)
1449                 brelse(new_dir_bh);
1450         if (new_de_bh)
1451                 brelse(new_de_bh);
1452         if (old_de_bh)
1453                 brelse(old_de_bh);
1454         if (old_inode_de_bh)
1455                 brelse(old_inode_de_bh);
1456         if (orphan_entry_bh)
1457                 brelse(orphan_entry_bh);
1458         if (insert_entry_bh)
1459                 brelse(insert_entry_bh);
1460
1461         mlog_exit(status);
1462
1463         return status;
1464 }
1465
1466 /*
1467  * we expect i_size = strlen(symname). Copy symname into the file
1468  * data, including the null terminator.
1469  */
1470 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1471                                      struct ocfs2_journal_handle *handle,
1472                                      struct inode *inode,
1473                                      const char *symname)
1474 {
1475         struct buffer_head **bhs = NULL;
1476         const char *c;
1477         struct super_block *sb = osb->sb;
1478         u64 p_blkno;
1479         int p_blocks;
1480         int virtual, blocks, status, i, bytes_left;
1481
1482         bytes_left = i_size_read(inode) + 1;
1483         /* we can't trust i_blocks because we're actually going to
1484          * write i_size + 1 bytes. */
1485         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1486
1487         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1488                         (unsigned long long)inode->i_blocks,
1489                         i_size_read(inode), blocks);
1490
1491         /* Sanity check -- make sure we're going to fit. */
1492         if (bytes_left >
1493             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1494                 status = -EIO;
1495                 mlog_errno(status);
1496                 goto bail;
1497         }
1498
1499         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1500         if (!bhs) {
1501                 status = -ENOMEM;
1502                 mlog_errno(status);
1503                 goto bail;
1504         }
1505
1506         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1507                                              &p_blocks);
1508         if (status < 0) {
1509                 mlog_errno(status);
1510                 goto bail;
1511         }
1512
1513         /* links can never be larger than one cluster so we know this
1514          * is all going to be contiguous, but do a sanity check
1515          * anyway. */
1516         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1517                 status = -EIO;
1518                 mlog_errno(status);
1519                 goto bail;
1520         }
1521
1522         virtual = 0;
1523         while(bytes_left > 0) {
1524                 c = &symname[virtual * sb->s_blocksize];
1525
1526                 bhs[virtual] = sb_getblk(sb, p_blkno);
1527                 if (!bhs[virtual]) {
1528                         status = -ENOMEM;
1529                         mlog_errno(status);
1530                         goto bail;
1531                 }
1532                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1533
1534                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1535                                               OCFS2_JOURNAL_ACCESS_CREATE);
1536                 if (status < 0) {
1537                         mlog_errno(status);
1538                         goto bail;
1539                 }
1540
1541                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1542
1543                 memcpy(bhs[virtual]->b_data, c,
1544                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1545                        bytes_left);
1546
1547                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1548                 if (status < 0) {
1549                         mlog_errno(status);
1550                         goto bail;
1551                 }
1552
1553                 virtual++;
1554                 p_blkno++;
1555                 bytes_left -= sb->s_blocksize;
1556         }
1557
1558         status = 0;
1559 bail:
1560
1561         if (bhs) {
1562                 for(i = 0; i < blocks; i++)
1563                         if (bhs[i])
1564                                 brelse(bhs[i]);
1565                 kfree(bhs);
1566         }
1567
1568         mlog_exit(status);
1569         return status;
1570 }
1571
1572 static int ocfs2_symlink(struct inode *dir,
1573                          struct dentry *dentry,
1574                          const char *symname)
1575 {
1576         int status, l, credits;
1577         u64 newsize;
1578         struct ocfs2_super *osb = NULL;
1579         struct inode *inode = NULL;
1580         struct super_block *sb;
1581         struct buffer_head *new_fe_bh = NULL;
1582         struct buffer_head *de_bh = NULL;
1583         struct buffer_head *parent_fe_bh = NULL;
1584         struct ocfs2_dinode *fe = NULL;
1585         struct ocfs2_dinode *dirfe;
1586         struct ocfs2_journal_handle *handle = NULL;
1587         struct ocfs2_alloc_context *inode_ac = NULL;
1588         struct ocfs2_alloc_context *data_ac = NULL;
1589
1590         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1591                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1592
1593         sb = dir->i_sb;
1594         osb = OCFS2_SB(sb);
1595
1596         l = strlen(symname) + 1;
1597
1598         credits = ocfs2_calc_symlink_credits(sb);
1599
1600         handle = ocfs2_alloc_handle(osb);
1601         if (handle == NULL) {
1602                 status = -ENOMEM;
1603                 mlog_errno(status);
1604                 goto bail;
1605         }
1606
1607         /* lock the parent directory */
1608         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1609         if (status < 0) {
1610                 if (status != -ENOENT)
1611                         mlog_errno(status);
1612                 goto bail;
1613         }
1614
1615         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1616         if (!dirfe->i_links_count) {
1617                 /* can't make a file in a deleted directory. */
1618                 status = -ENOENT;
1619                 goto bail;
1620         }
1621
1622         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1623                                            dentry->d_name.len);
1624         if (status)
1625                 goto bail;
1626
1627         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1628                                               dentry->d_name.name,
1629                                               dentry->d_name.len, &de_bh);
1630         if (status < 0) {
1631                 mlog_errno(status);
1632                 goto bail;
1633         }
1634
1635         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1636         if (status < 0) {
1637                 if (status != -ENOSPC)
1638                         mlog_errno(status);
1639                 goto bail;
1640         }
1641
1642         /* don't reserve bitmap space for fast symlinks. */
1643         if (l > ocfs2_fast_symlink_chars(sb)) {
1644                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1645                 if (status < 0) {
1646                         if (status != -ENOSPC)
1647                                 mlog_errno(status);
1648                         goto bail;
1649                 }
1650         }
1651
1652         handle = ocfs2_start_trans(osb, handle, credits);
1653         if (IS_ERR(handle)) {
1654                 status = PTR_ERR(handle);
1655                 handle = NULL;
1656                 mlog_errno(status);
1657                 goto bail;
1658         }
1659
1660         status = ocfs2_mknod_locked(osb, dir, dentry,
1661                                     S_IFLNK | S_IRWXUGO, 0,
1662                                     &new_fe_bh, parent_fe_bh, handle,
1663                                     &inode, inode_ac);
1664         if (status < 0) {
1665                 mlog_errno(status);
1666                 goto bail;
1667         }
1668
1669         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1670         inode->i_rdev = 0;
1671         newsize = l - 1;
1672         if (l > ocfs2_fast_symlink_chars(sb)) {
1673                 inode->i_op = &ocfs2_symlink_inode_operations;
1674                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1675                                                     handle, data_ac, NULL,
1676                                                     NULL);
1677                 if (status < 0) {
1678                         if (status != -ENOSPC && status != -EINTR) {
1679                                 mlog(ML_ERROR,
1680                                      "Failed to extend file to %llu\n",
1681                                      (unsigned long long)newsize);
1682                                 mlog_errno(status);
1683                                 status = -ENOSPC;
1684                         }
1685                         goto bail;
1686                 }
1687                 i_size_write(inode, newsize);
1688                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1689         } else {
1690                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1691                 memcpy((char *) fe->id2.i_symlink, symname, l);
1692                 i_size_write(inode, newsize);
1693                 inode->i_blocks = 0;
1694         }
1695
1696         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1697         if (status < 0) {
1698                 mlog_errno(status);
1699                 goto bail;
1700         }
1701
1702         if (!ocfs2_inode_is_fast_symlink(inode)) {
1703                 status = ocfs2_create_symlink_data(osb, handle, inode,
1704                                                    symname);
1705                 if (status < 0) {
1706                         mlog_errno(status);
1707                         goto bail;
1708                 }
1709         }
1710
1711         status = ocfs2_add_entry(handle, dentry, inode,
1712                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1713                                  de_bh);
1714         if (status < 0) {
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718
1719         status = ocfs2_dentry_attach_lock(dentry, inode,
1720                                           OCFS2_I(dir)->ip_blkno, 1);
1721         if (status) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         insert_inode_hash(inode);
1727         dentry->d_op = &ocfs2_dentry_ops;
1728         d_instantiate(dentry, inode);
1729 bail:
1730         if (handle)
1731                 ocfs2_commit_trans(handle);
1732         if (new_fe_bh)
1733                 brelse(new_fe_bh);
1734         if (parent_fe_bh)
1735                 brelse(parent_fe_bh);
1736         if (de_bh)
1737                 brelse(de_bh);
1738         if (inode_ac)
1739                 ocfs2_free_alloc_context(inode_ac);
1740         if (data_ac)
1741                 ocfs2_free_alloc_context(data_ac);
1742         if ((status < 0) && inode)
1743                 iput(inode);
1744
1745         mlog_exit(status);
1746
1747         return status;
1748 }
1749
1750 int ocfs2_check_dir_entry(struct inode * dir,
1751                           struct ocfs2_dir_entry * de,
1752                           struct buffer_head * bh,
1753                           unsigned long offset)
1754 {
1755         const char *error_msg = NULL;
1756         const int rlen = le16_to_cpu(de->rec_len);
1757
1758         if (rlen < OCFS2_DIR_REC_LEN(1))
1759                 error_msg = "rec_len is smaller than minimal";
1760         else if (rlen % 4 != 0)
1761                 error_msg = "rec_len % 4 != 0";
1762         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1763                 error_msg = "rec_len is too small for name_len";
1764         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1765                 error_msg = "directory entry across blocks";
1766
1767         if (error_msg != NULL)
1768                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1769                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1770                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1771                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1772                      de->name_len);
1773         return error_msg == NULL ? 1 : 0;
1774 }
1775
1776 /* we don't always have a dentry for what we want to add, so people
1777  * like orphan dir can call this instead.
1778  *
1779  * If you pass me insert_bh, I'll skip the search of the other dir
1780  * blocks and put the record in there.
1781  */
1782 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1783                              struct inode *dir,
1784                              const char *name, int namelen,
1785                              struct inode *inode, u64 blkno,
1786                              struct buffer_head *parent_fe_bh,
1787                              struct buffer_head *insert_bh)
1788 {
1789         unsigned long offset;
1790         unsigned short rec_len;
1791         struct ocfs2_dir_entry *de, *de1;
1792         struct super_block *sb;
1793         int retval, status;
1794
1795         mlog_entry_void();
1796
1797         sb = dir->i_sb;
1798
1799         if (!namelen)
1800                 return -EINVAL;
1801
1802         rec_len = OCFS2_DIR_REC_LEN(namelen);
1803         offset = 0;
1804         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1805         while (1) {
1806                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1807                 /* These checks should've already been passed by the
1808                  * prepare function, but I guess we can leave them
1809                  * here anyway. */
1810                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1811                         retval = -ENOENT;
1812                         goto bail;
1813                 }
1814                 if (ocfs2_match(namelen, name, de)) {
1815                         retval = -EEXIST;
1816                         goto bail;
1817                 }
1818                 if (((le64_to_cpu(de->inode) == 0) &&
1819                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1820                     (le16_to_cpu(de->rec_len) >=
1821                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1822                         status = ocfs2_journal_access(handle, dir, insert_bh,
1823                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1824                         /* By now the buffer is marked for journaling */
1825                         offset += le16_to_cpu(de->rec_len);
1826                         if (le64_to_cpu(de->inode)) {
1827                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1828                                         OCFS2_DIR_REC_LEN(de->name_len));
1829                                 de1->rec_len =
1830                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1831                                         OCFS2_DIR_REC_LEN(de->name_len));
1832                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1833                                 de = de1;
1834                         }
1835                         de->file_type = OCFS2_FT_UNKNOWN;
1836                         if (blkno) {
1837                                 de->inode = cpu_to_le64(blkno);
1838                                 ocfs2_set_de_type(de, inode->i_mode);
1839                         } else
1840                                 de->inode = 0;
1841                         de->name_len = namelen;
1842                         memcpy(de->name, name, namelen);
1843
1844                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1845                         dir->i_version++;
1846                         status = ocfs2_journal_dirty(handle, insert_bh);
1847                         retval = 0;
1848                         goto bail;
1849                 }
1850                 offset += le16_to_cpu(de->rec_len);
1851                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1852         }
1853
1854         /* when you think about it, the assert above should prevent us
1855          * from ever getting here. */
1856         retval = -ENOSPC;
1857 bail:
1858
1859         mlog_exit(retval);
1860         return retval;
1861 }
1862
1863
1864 /*
1865  * ocfs2_delete_entry deletes a directory entry by merging it with the
1866  * previous entry
1867  */
1868 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1869                               struct inode *dir,
1870                               struct ocfs2_dir_entry *de_del,
1871                               struct buffer_head *bh)
1872 {
1873         struct ocfs2_dir_entry *de, *pde;
1874         int i, status = -ENOENT;
1875
1876         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1877
1878         i = 0;
1879         pde = NULL;
1880         de = (struct ocfs2_dir_entry *) bh->b_data;
1881         while (i < bh->b_size) {
1882                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1883                         status = -EIO;
1884                         mlog_errno(status);
1885                         goto bail;
1886                 }
1887                 if (de == de_del)  {
1888                         status = ocfs2_journal_access(handle, dir, bh,
1889                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1890                         if (status < 0) {
1891                                 status = -EIO;
1892                                 mlog_errno(status);
1893                                 goto bail;
1894                         }
1895                         if (pde)
1896                                 pde->rec_len =
1897                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1898                                                     le16_to_cpu(de->rec_len));
1899                         else
1900                                 de->inode = 0;
1901                         dir->i_version++;
1902                         status = ocfs2_journal_dirty(handle, bh);
1903                         goto bail;
1904                 }
1905                 i += le16_to_cpu(de->rec_len);
1906                 pde = de;
1907                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1908         }
1909 bail:
1910         mlog_exit(status);
1911         return status;
1912 }
1913
1914 /*
1915  * Returns 0 if not found, -1 on failure, and 1 on success
1916  */
1917 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1918                                         struct inode *dir,
1919                                         const char *name, int namelen,
1920                                         unsigned long offset,
1921                                         struct ocfs2_dir_entry **res_dir)
1922 {
1923         struct ocfs2_dir_entry *de;
1924         char *dlimit, *de_buf;
1925         int de_len;
1926         int ret = 0;
1927
1928         mlog_entry_void();
1929
1930         de_buf = bh->b_data;
1931         dlimit = de_buf + dir->i_sb->s_blocksize;
1932
1933         while (de_buf < dlimit) {
1934                 /* this code is executed quadratically often */
1935                 /* do minimal checking `by hand' */
1936
1937                 de = (struct ocfs2_dir_entry *) de_buf;
1938
1939                 if (de_buf + namelen <= dlimit &&
1940                     ocfs2_match(namelen, name, de)) {
1941                         /* found a match - just to be sure, do a full check */
1942                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1943                                 ret = -1;
1944                                 goto bail;
1945                         }
1946                         *res_dir = de;
1947                         ret = 1;
1948                         goto bail;
1949                 }
1950
1951                 /* prevent looping on a bad block */
1952                 de_len = le16_to_cpu(de->rec_len);
1953                 if (de_len <= 0) {
1954                         ret = -1;
1955                         goto bail;
1956                 }
1957
1958                 de_buf += de_len;
1959                 offset += de_len;
1960         }
1961
1962 bail:
1963         mlog_exit(ret);
1964         return ret;
1965 }
1966
1967 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1968                                      struct inode *dir,
1969                                      struct ocfs2_dir_entry **res_dir)
1970 {
1971         struct super_block *sb;
1972         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1973         struct buffer_head *bh, *ret = NULL;
1974         unsigned long start, block, b;
1975         int ra_max = 0;         /* Number of bh's in the readahead
1976                                    buffer, bh_use[] */
1977         int ra_ptr = 0;         /* Current index into readahead
1978                                    buffer */
1979         int num = 0;
1980         int nblocks, i, err;
1981
1982         mlog_entry_void();
1983
1984         *res_dir = NULL;
1985         sb = dir->i_sb;
1986
1987         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1988         start = OCFS2_I(dir)->ip_dir_start_lookup;
1989         if (start >= nblocks)
1990                 start = 0;
1991         block = start;
1992
1993 restart:
1994         do {
1995                 /*
1996                  * We deal with the read-ahead logic here.
1997                  */
1998                 if (ra_ptr >= ra_max) {
1999                         /* Refill the readahead buffer */
2000                         ra_ptr = 0;
2001                         b = block;
2002                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2003                                 /*
2004                                  * Terminate if we reach the end of the
2005                                  * directory and must wrap, or if our
2006                                  * search has finished at this block.
2007                                  */
2008                                 if (b >= nblocks || (num && block == start)) {
2009                                         bh_use[ra_max] = NULL;
2010                                         break;
2011                                 }
2012                                 num++;
2013
2014                                 bh = ocfs2_bread(dir, b++, &err, 1);
2015                                 bh_use[ra_max] = bh;
2016                         }
2017                 }
2018                 if ((bh = bh_use[ra_ptr++]) == NULL)
2019                         goto next;
2020                 wait_on_buffer(bh);
2021                 if (!buffer_uptodate(bh)) {
2022                         /* read error, skip block & hope for the best */
2023                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2024                                     "offset %lu\n",
2025                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2026                                     block);
2027                         brelse(bh);
2028                         goto next;
2029                 }
2030                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2031                                           block << sb->s_blocksize_bits,
2032                                           res_dir);
2033                 if (i == 1) {
2034                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2035                         ret = bh;
2036                         goto cleanup_and_exit;
2037                 } else {
2038                         brelse(bh);
2039                         if (i < 0)
2040                                 goto cleanup_and_exit;
2041                 }
2042         next:
2043                 if (++block >= nblocks)
2044                         block = 0;
2045         } while (block != start);
2046
2047         /*
2048          * If the directory has grown while we were searching, then
2049          * search the last part of the directory before giving up.
2050          */
2051         block = nblocks;
2052         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2053         if (block < nblocks) {
2054                 start = 0;
2055                 goto restart;
2056         }
2057
2058 cleanup_and_exit:
2059         /* Clean up the read-ahead blocks */
2060         for (; ra_ptr < ra_max; ra_ptr++)
2061                 brelse(bh_use[ra_ptr]);
2062
2063         mlog_exit_ptr(ret);
2064         return ret;
2065 }
2066
2067 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2068 {
2069         int status, namelen;
2070
2071         mlog_entry_void();
2072
2073         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2074                            (long long)blkno);
2075         if (namelen <= 0) {
2076                 if (namelen)
2077                         status = namelen;
2078                 else
2079                         status = -EINVAL;
2080                 mlog_errno(status);
2081                 goto bail;
2082         }
2083         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2084                 status = -EINVAL;
2085                 mlog_errno(status);
2086                 goto bail;
2087         }
2088
2089         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2090              namelen);
2091
2092         status = 0;
2093 bail:
2094         mlog_exit(status);
2095         return status;
2096 }
2097
2098 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2099                                     struct ocfs2_journal_handle *handle,
2100                                     struct inode *inode,
2101                                     char *name,
2102                                     struct buffer_head **de_bh)
2103 {
2104         struct inode *orphan_dir_inode = NULL;
2105         struct buffer_head *orphan_dir_bh = NULL;
2106         int status = 0;
2107
2108         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2109         if (status < 0) {
2110                 mlog_errno(status);
2111                 goto leave;
2112         }
2113
2114         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2115                                                        ORPHAN_DIR_SYSTEM_INODE,
2116                                                        osb->slot_num);
2117         if (!orphan_dir_inode) {
2118                 status = -ENOENT;
2119                 mlog_errno(status);
2120                 goto leave;
2121         }
2122
2123         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2124         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2125         if (status < 0) {
2126                 mlog_errno(status);
2127                 goto leave;
2128         }
2129
2130         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2131                                               orphan_dir_bh, name,
2132                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2133         if (status < 0) {
2134                 mlog_errno(status);
2135                 goto leave;
2136         }
2137
2138 leave:
2139         if (orphan_dir_inode)
2140                 iput(orphan_dir_inode);
2141
2142         if (orphan_dir_bh)
2143                 brelse(orphan_dir_bh);
2144
2145         mlog_exit(status);
2146         return status;
2147 }
2148
2149 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2150                             struct ocfs2_journal_handle *handle,
2151                             struct inode *inode,
2152                             struct ocfs2_dinode *fe,
2153                             char *name,
2154                             struct buffer_head *de_bh)
2155 {
2156         struct inode *orphan_dir_inode = NULL;
2157         struct buffer_head *orphan_dir_bh = NULL;
2158         int status = 0;
2159         struct ocfs2_dinode *orphan_fe;
2160
2161         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2162
2163         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2164                                                        ORPHAN_DIR_SYSTEM_INODE,
2165                                                        osb->slot_num);
2166         if (!orphan_dir_inode) {
2167                 status = -ENOENT;
2168                 mlog_errno(status);
2169                 goto leave;
2170         }
2171
2172         status = ocfs2_read_block(osb,
2173                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2174                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2175                                   orphan_dir_inode);
2176         if (status < 0) {
2177                 mlog_errno(status);
2178                 goto leave;
2179         }
2180
2181         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2182                                       OCFS2_JOURNAL_ACCESS_WRITE);
2183         if (status < 0) {
2184                 mlog_errno(status);
2185                 goto leave;
2186         }
2187
2188         /* we're a cluster, and nlink can change on disk from
2189          * underneath us... */
2190         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2191         if (S_ISDIR(inode->i_mode))
2192                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2193         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2194
2195         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2196         if (status < 0) {
2197                 mlog_errno(status);
2198                 goto leave;
2199         }
2200
2201         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2202                                    OCFS2_ORPHAN_NAMELEN, inode,
2203                                    OCFS2_I(inode)->ip_blkno,
2204                                    orphan_dir_bh, de_bh);
2205         if (status < 0) {
2206                 mlog_errno(status);
2207                 goto leave;
2208         }
2209
2210         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2211
2212         /* Record which orphan dir our inode now resides
2213          * in. delete_inode will use this to determine which orphan
2214          * dir to lock. */
2215         spin_lock(&OCFS2_I(inode)->ip_lock);
2216         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2217         spin_unlock(&OCFS2_I(inode)->ip_lock);
2218
2219         mlog(0, "Inode %llu orphaned in slot %d\n",
2220              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2221
2222 leave:
2223         if (orphan_dir_inode)
2224                 iput(orphan_dir_inode);
2225
2226         if (orphan_dir_bh)
2227                 brelse(orphan_dir_bh);
2228
2229         mlog_exit(status);
2230         return status;
2231 }
2232
2233 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2234 int ocfs2_orphan_del(struct ocfs2_super *osb,
2235                      struct ocfs2_journal_handle *handle,
2236                      struct inode *orphan_dir_inode,
2237                      struct inode *inode,
2238                      struct buffer_head *orphan_dir_bh)
2239 {
2240         char name[OCFS2_ORPHAN_NAMELEN + 1];
2241         struct ocfs2_dinode *orphan_fe;
2242         int status = 0;
2243         struct buffer_head *target_de_bh = NULL;
2244         struct ocfs2_dir_entry *target_de = NULL;
2245
2246         mlog_entry_void();
2247
2248         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2249         if (status < 0) {
2250                 mlog_errno(status);
2251                 goto leave;
2252         }
2253
2254         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2255              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2256              OCFS2_ORPHAN_NAMELEN);
2257
2258         /* find it's spot in the orphan directory */
2259         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2260                                         orphan_dir_inode, &target_de);
2261         if (!target_de_bh) {
2262                 status = -ENOENT;
2263                 mlog_errno(status);
2264                 goto leave;
2265         }
2266
2267         /* remove it from the orphan directory */
2268         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2269                                     target_de_bh);
2270         if (status < 0) {
2271                 mlog_errno(status);
2272                 goto leave;
2273         }
2274
2275         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2276                                       OCFS2_JOURNAL_ACCESS_WRITE);
2277         if (status < 0) {
2278                 mlog_errno(status);
2279                 goto leave;
2280         }
2281
2282         /* do the i_nlink dance! :) */
2283         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2284         if (S_ISDIR(inode->i_mode))
2285                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2286         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2287
2288         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2289         if (status < 0) {
2290                 mlog_errno(status);
2291                 goto leave;
2292         }
2293
2294 leave:
2295         if (target_de_bh)
2296                 brelse(target_de_bh);
2297
2298         mlog_exit(status);
2299         return status;
2300 }
2301
2302 struct inode_operations ocfs2_dir_iops = {
2303         .create         = ocfs2_create,
2304         .lookup         = ocfs2_lookup,
2305         .link           = ocfs2_link,
2306         .unlink         = ocfs2_unlink,
2307         .rmdir          = ocfs2_unlink,
2308         .symlink        = ocfs2_symlink,
2309         .mkdir          = ocfs2_mkdir,
2310         .mknod          = ocfs2_mknod,
2311         .rename         = ocfs2_rename,
2312         .setattr        = ocfs2_setattr,
2313         .getattr        = ocfs2_getattr,
2314 };