fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "vote.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 #define NAMEI_RA_CHUNKS  2
  68 #define NAMEI_RA_BLOCKS  4
  69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  71
  72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  73                                         struct inode *dir,
  74                                         const char *name, int namelen,
  75                                         unsigned long offset,
  76                                         struct ocfs2_dir_entry **res_dir);
  77
  78 static int ocfs2_delete_entry(handle_t *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(handle_t *handle,
  84                              struct inode *dir,
  85                              const char *name, int namelen,
  86                              struct inode *inode, u64 blkno,
  87                              struct buffer_head *parent_fe_bh,
  88                              struct buffer_head *insert_bh);
  89
  90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  91                               struct inode *dir,
  92                               struct dentry *dentry, int mode,
  93                               dev_t dev,
  94                               struct buffer_head **new_fe_bh,
  95                               struct buffer_head *parent_fe_bh,
  96                               handle_t *handle,
  97                               struct inode **ret_inode,
  98                               struct ocfs2_alloc_context *inode_ac);
  99
 100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 101                               handle_t *handle,
 102                               struct inode *parent,
 103                               struct inode *inode,
 104                               struct buffer_head *fe_bh,
 105                               struct ocfs2_alloc_context *data_ac);
 106
 107 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 108                                     struct inode **ret_orphan_dir,
 109                                     struct inode *inode,
 110                                     char *name,
 111                                     struct buffer_head **de_bh);
 112
 113 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 114                             handle_t *handle,
 115                             struct inode *inode,
 116                             struct ocfs2_dinode *fe,
 117                             char *name,
 118                             struct buffer_head *de_bh,
 119                             struct inode *orphan_dir_inode);
 120
 121 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 122                                      handle_t *handle,
 123                                      struct inode *inode,
 124                                      const char *symname);
 125
 126 static inline int ocfs2_add_entry(handle_t *handle,
 127                                   struct dentry *dentry,
 128                                   struct inode *inode, u64 blkno,
 129                                   struct buffer_head *parent_fe_bh,
 130                                   struct buffer_head *insert_bh)
 131 {
 132         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 133                                  dentry->d_name.name, dentry->d_name.len,
 134                                  inode, blkno, parent_fe_bh, insert_bh);
 135 }
 136
 137 /* An orphan dir name is an 8 byte value, printed as a hex string */
 138 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 139
 140 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 141                                    struct nameidata *nd)
 142 {
 143         int status;
 144         u64 blkno;
 145         struct buffer_head *dirent_bh = NULL;
 146         struct inode *inode = NULL;
 147         struct dentry *ret;
 148         struct ocfs2_dir_entry *dirent;
 149         struct ocfs2_inode_info *oi;
 150
 151         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 152                    dentry->d_name.len, dentry->d_name.name);
 153
 154         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 155                 ret = ERR_PTR(-ENAMETOOLONG);
 156                 goto bail;
 157         }
 158
 159         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 160              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 161
 162         status = ocfs2_meta_lock(dir, NULL, 0);
 163         if (status < 0) {
 164                 if (status != -ENOENT)
 165                         mlog_errno(status);
 166                 ret = ERR_PTR(status);
 167                 goto bail;
 168         }
 169
 170         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 171                                           dentry->d_name.len, &blkno,
 172                                           dir, &dirent_bh, &dirent);
 173         if (status < 0)
 174                 goto bail_add;
 175
 176         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 177         if (IS_ERR(inode)) {
 178                 ret = ERR_PTR(-EACCES);
 179                 goto bail_unlock;
 180         }
 181
 182         oi = OCFS2_I(inode);
 183         /* Clear any orphaned state... If we were able to look up the
 184          * inode from a directory, it certainly can't be orphaned. We
 185          * might have the bad state from a node which intended to
 186          * orphan this inode but crashed before it could commit the
 187          * unlink. */
 188         spin_lock(&oi->ip_lock);
 189         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 190         spin_unlock(&oi->ip_lock);
 191
 192 bail_add:
 193         dentry->d_op = &ocfs2_dentry_ops;
 194         ret = d_splice_alias(inode, dentry);
 195
 196         if (inode) {
 197                 /*
 198                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 199                  * dentry, it will d_move() it on top of ourse. The
 200                  * return value will indicate this however, so in
 201                  * those cases, we switch them around for the locking
 202                  * code.
 203                  *
 204                  * NOTE: This dentry already has ->d_op set from
 205                  * ocfs2_get_parent() and ocfs2_get_dentry()
 206                  */
 207                 if (ret)
 208                         dentry = ret;
 209
 210                 status = ocfs2_dentry_attach_lock(dentry, inode,
 211                                                   OCFS2_I(dir)->ip_blkno);
 212                 if (status) {
 213                         mlog_errno(status);
 214                         ret = ERR_PTR(status);
 215                         goto bail_unlock;
 216                 }
 217         }
 218
 219 bail_unlock:
 220         /* Don't drop the cluster lock until *after* the d_add --
 221          * unlink on another node will message us to remove that
 222          * dentry under this lock so otherwise we can race this with
 223          * the vote thread and have a stale dentry. */
 224         ocfs2_meta_unlock(dir, 0);
 225
 226 bail:
 227         if (dirent_bh)
 228                 brelse(dirent_bh);
 229
 230         mlog_exit_ptr(ret);
 231
 232         return ret;
 233 }
 234
 235 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 236                               handle_t *handle,
 237                               struct inode *parent,
 238                               struct inode *inode,
 239                               struct buffer_head *fe_bh,
 240                               struct ocfs2_alloc_context *data_ac)
 241 {
 242         int status;
 243         struct buffer_head *new_bh = NULL;
 244         struct ocfs2_dir_entry *de = NULL;
 245
 246         mlog_entry_void();
 247
 248         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 249                                      data_ac, NULL, &new_bh);
 250         if (status < 0) {
 251                 mlog_errno(status);
 252                 goto bail;
 253         }
 254
 255         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 256
 257         status = ocfs2_journal_access(handle, inode, new_bh,
 258                                       OCFS2_JOURNAL_ACCESS_CREATE);
 259         if (status < 0) {
 260                 mlog_errno(status);
 261                 goto bail;
 262         }
 263         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 264
 265         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 266         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 267         de->name_len = 1;
 268         de->rec_len =
 269                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 270         strcpy(de->name, ".");
 271         ocfs2_set_de_type(de, S_IFDIR);
 272         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 273         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 274         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 275                                   OCFS2_DIR_REC_LEN(1));
 276         de->name_len = 2;
 277         strcpy(de->name, "..");
 278         ocfs2_set_de_type(de, S_IFDIR);
 279
 280         status = ocfs2_journal_dirty(handle, new_bh);
 281         if (status < 0) {
 282                 mlog_errno(status);
 283                 goto bail;
 284         }
 285
 286         i_size_write(inode, inode->i_sb->s_blocksize);
 287         inode->i_nlink = 2;
 288         inode->i_blocks = ocfs2_inode_sector_count(inode);
 289         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 290         if (status < 0) {
 291                 mlog_errno(status);
 292                 goto bail;
 293         }
 294
 295         status = 0;
 296 bail:
 297         if (new_bh)
 298                 brelse(new_bh);
 299
 300         mlog_exit(status);
 301         return status;
 302 }
 303
 304 static int ocfs2_mknod(struct inode *dir,
 305                        struct dentry *dentry,
 306                        int mode,
 307                        dev_t dev)
 308 {
 309         int status = 0;
 310         struct buffer_head *parent_fe_bh = NULL;
 311         handle_t *handle = NULL;
 312         struct ocfs2_super *osb;
 313         struct ocfs2_dinode *dirfe;
 314         struct buffer_head *new_fe_bh = NULL;
 315         struct buffer_head *de_bh = NULL;
 316         struct inode *inode = NULL;
 317         struct ocfs2_alloc_context *inode_ac = NULL;
 318         struct ocfs2_alloc_context *data_ac = NULL;
 319
 320         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 321                    (unsigned long)dev, dentry->d_name.len,
 322                    dentry->d_name.name);
 323
 324         /* get our super block */
 325         osb = OCFS2_SB(dir->i_sb);
 326
 327         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 328         if (status < 0) {
 329                 if (status != -ENOENT)
 330                         mlog_errno(status);
 331                 return status;
 332         }
 333
 334         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 335                 status = -EMLINK;
 336                 goto leave;
 337         }
 338
 339         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 340         if (!dirfe->i_links_count) {
 341                 /* can't make a file in a deleted directory. */
 342                 status = -ENOENT;
 343                 goto leave;
 344         }
 345
 346         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 347                                            dentry->d_name.len);
 348         if (status)
 349                 goto leave;
 350
 351         /* get a spot inside the dir. */
 352         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 353                                               dentry->d_name.name,
 354                                               dentry->d_name.len, &de_bh);
 355         if (status < 0) {
 356                 mlog_errno(status);
 357                 goto leave;
 358         }
 359
 360         /* reserve an inode spot */
 361         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 362         if (status < 0) {
 363                 if (status != -ENOSPC)
 364                         mlog_errno(status);
 365                 goto leave;
 366         }
 367
 368         /* are we making a directory? If so, reserve a cluster for his
 369          * 1st extent. */
 370         if (S_ISDIR(mode)) {
 371                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 372                 if (status < 0) {
 373                         if (status != -ENOSPC)
 374                                 mlog_errno(status);
 375                         goto leave;
 376                 }
 377         }
 378
 379         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 380         if (IS_ERR(handle)) {
 381                 status = PTR_ERR(handle);
 382                 handle = NULL;
 383                 mlog_errno(status);
 384                 goto leave;
 385         }
 386
 387         /* do the real work now. */
 388         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 389                                     &new_fe_bh, parent_fe_bh, handle,
 390                                     &inode, inode_ac);
 391         if (status < 0) {
 392                 mlog_errno(status);
 393                 goto leave;
 394         }
 395
 396         if (S_ISDIR(mode)) {
 397                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 398                                             new_fe_bh, data_ac);
 399                 if (status < 0) {
 400                         mlog_errno(status);
 401                         goto leave;
 402                 }
 403
 404                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 405                                               OCFS2_JOURNAL_ACCESS_WRITE);
 406                 if (status < 0) {
 407                         mlog_errno(status);
 408                         goto leave;
 409                 }
 410                 le16_add_cpu(&dirfe->i_links_count, 1);
 411                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 412                 if (status < 0) {
 413                         mlog_errno(status);
 414                         goto leave;
 415                 }
 416                 inc_nlink(dir);
 417         }
 418
 419         status = ocfs2_add_entry(handle, dentry, inode,
 420                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 421                                  de_bh);
 422         if (status < 0) {
 423                 mlog_errno(status);
 424                 goto leave;
 425         }
 426
 427         status = ocfs2_dentry_attach_lock(dentry, inode,
 428                                           OCFS2_I(dir)->ip_blkno);
 429         if (status) {
 430                 mlog_errno(status);
 431                 goto leave;
 432         }
 433
 434         insert_inode_hash(inode);
 435         dentry->d_op = &ocfs2_dentry_ops;
 436         d_instantiate(dentry, inode);
 437         status = 0;
 438 leave:
 439         if (handle)
 440                 ocfs2_commit_trans(osb, handle);
 441
 442         ocfs2_meta_unlock(dir, 1);
 443
 444         if (status == -ENOSPC)
 445                 mlog(0, "Disk is full\n");
 446
 447         if (new_fe_bh)
 448                 brelse(new_fe_bh);
 449
 450         if (de_bh)
 451                 brelse(de_bh);
 452
 453         if (parent_fe_bh)
 454                 brelse(parent_fe_bh);
 455
 456         if ((status < 0) && inode)
 457                 iput(inode);
 458
 459         if (inode_ac)
 460                 ocfs2_free_alloc_context(inode_ac);
 461
 462         if (data_ac)
 463                 ocfs2_free_alloc_context(data_ac);
 464
 465         mlog_exit(status);
 466
 467         return status;
 468 }
 469
 470 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 471                               struct inode *dir,
 472                               struct dentry *dentry, int mode,
 473                               dev_t dev,
 474                               struct buffer_head **new_fe_bh,
 475                               struct buffer_head *parent_fe_bh,
 476                               handle_t *handle,
 477                               struct inode **ret_inode,
 478                               struct ocfs2_alloc_context *inode_ac)
 479 {
 480         int status = 0;
 481         struct ocfs2_dinode *fe = NULL;
 482         struct ocfs2_extent_list *fel;
 483         u64 fe_blkno = 0;
 484         u16 suballoc_bit;
 485         struct inode *inode = NULL;
 486
 487         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 488                    (unsigned long)dev, dentry->d_name.len,
 489                    dentry->d_name.name);
 490
 491         *new_fe_bh = NULL;
 492         *ret_inode = NULL;
 493
 494         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 495                                        &fe_blkno);
 496         if (status < 0) {
 497                 mlog_errno(status);
 498                 goto leave;
 499         }
 500
 501         inode = new_inode(dir->i_sb);
 502         if (IS_ERR(inode)) {
 503                 status = PTR_ERR(inode);
 504                 mlog(ML_ERROR, "new_inode failed!\n");
 505                 goto leave;
 506         }
 507
 508         /* populate as many fields early on as possible - many of
 509          * these are used by the support functions here and in
 510          * callers. */
 511         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 512         OCFS2_I(inode)->ip_blkno = fe_blkno;
 513         if (S_ISDIR(mode))
 514                 inode->i_nlink = 2;
 515         else
 516                 inode->i_nlink = 1;
 517         inode->i_mode = mode;
 518         spin_lock(&osb->osb_lock);
 519         inode->i_generation = osb->s_next_generation++;
 520         spin_unlock(&osb->osb_lock);
 521
 522         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 523         if (!*new_fe_bh) {
 524                 status = -EIO;
 525                 mlog_errno(status);
 526                 goto leave;
 527         }
 528         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 529
 530         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 531                                       OCFS2_JOURNAL_ACCESS_CREATE);
 532         if (status < 0) {
 533                 mlog_errno(status);
 534                 goto leave;
 535         }
 536
 537         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 538         memset(fe, 0, osb->sb->s_blocksize);
 539
 540         fe->i_generation = cpu_to_le32(inode->i_generation);
 541         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 542         fe->i_blkno = cpu_to_le64(fe_blkno);
 543         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 544         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 545         fe->i_uid = cpu_to_le32(current->fsuid);
 546         if (dir->i_mode & S_ISGID) {
 547                 fe->i_gid = cpu_to_le32(dir->i_gid);
 548                 if (S_ISDIR(mode))
 549                         mode |= S_ISGID;
 550         } else
 551                 fe->i_gid = cpu_to_le32(current->fsgid);
 552         fe->i_mode = cpu_to_le16(mode);
 553         if (S_ISCHR(mode) || S_ISBLK(mode))
 554                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 555
 556         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 557
 558         fe->i_last_eb_blk = 0;
 559         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 560         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 561         fe->i_atime = fe->i_ctime = fe->i_mtime =
 562                 cpu_to_le64(CURRENT_TIME.tv_sec);
 563         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 564                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 565         fe->i_dtime = 0;
 566
 567         fel = &fe->id2.i_list;
 568         fel->l_tree_depth = 0;
 569         fel->l_next_free_rec = 0;
 570         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 571
 572         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 573         if (status < 0) {
 574                 mlog_errno(status);
 575                 goto leave;
 576         }
 577
 578         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 579                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 580                      "i_blkno=%llu, i_ino=%lu\n",
 581                      (unsigned long long)(*new_fe_bh)->b_blocknr,
 582                      (unsigned long long)le64_to_cpu(fe->i_blkno),
 583                      inode->i_ino);
 584                 BUG();
 585         }
 586
 587         ocfs2_inode_set_new(osb, inode);
 588         if (!ocfs2_mount_local(osb)) {
 589                 status = ocfs2_create_new_inode_locks(inode);
 590                 if (status < 0)
 591                         mlog_errno(status);
 592         }
 593
 594         status = 0; /* error in ocfs2_create_new_inode_locks is not
 595                      * critical */
 596
 597         *ret_inode = inode;
 598 leave:
 599         if (status < 0) {
 600                 if (*new_fe_bh) {
 601                         brelse(*new_fe_bh);
 602                         *new_fe_bh = NULL;
 603                 }
 604                 if (inode)
 605                         iput(inode);
 606         }
 607
 608         mlog_exit(status);
 609         return status;
 610 }
 611
 612 static int ocfs2_mkdir(struct inode *dir,
 613                        struct dentry *dentry,
 614                        int mode)
 615 {
 616         int ret;
 617
 618         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 619                    dentry->d_name.len, dentry->d_name.name);
 620         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 621         mlog_exit(ret);
 622
 623         return ret;
 624 }
 625
 626 static int ocfs2_create(struct inode *dir,
 627                         struct dentry *dentry,
 628                         int mode,
 629                         struct nameidata *nd)
 630 {
 631         int ret;
 632
 633         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 634                    dentry->d_name.len, dentry->d_name.name);
 635         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 636         mlog_exit(ret);
 637
 638         return ret;
 639 }
 640
 641 static int ocfs2_link(struct dentry *old_dentry,
 642                       struct inode *dir,
 643                       struct dentry *dentry)
 644 {
 645         handle_t *handle;
 646         struct inode *inode = old_dentry->d_inode;
 647         int err;
 648         struct buffer_head *fe_bh = NULL;
 649         struct buffer_head *parent_fe_bh = NULL;
 650         struct buffer_head *de_bh = NULL;
 651         struct ocfs2_dinode *fe = NULL;
 652         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 653
 654         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 655                    old_dentry->d_name.len, old_dentry->d_name.name,
 656                    dentry->d_name.len, dentry->d_name.name);
 657
 658         if (S_ISDIR(inode->i_mode))
 659                 return -EPERM;
 660
 661         err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 662         if (err < 0) {
 663                 if (err != -ENOENT)
 664                         mlog_errno(err);
 665                 return err;
 666         }
 667
 668         if (!dir->i_nlink) {
 669                 err = -ENOENT;
 670                 goto out;
 671         }
 672
 673         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 674                                         dentry->d_name.len);
 675         if (err)
 676                 goto out;
 677
 678         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 679                                            dentry->d_name.name,
 680                                            dentry->d_name.len, &de_bh);
 681         if (err < 0) {
 682                 mlog_errno(err);
 683                 goto out;
 684         }
 685
 686         err = ocfs2_meta_lock(inode, &fe_bh, 1);
 687         if (err < 0) {
 688                 if (err != -ENOENT)
 689                         mlog_errno(err);
 690                 goto out;
 691         }
 692
 693         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 694         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 695                 err = -EMLINK;
 696                 goto out_unlock_inode;
 697         }
 698
 699         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 700         if (IS_ERR(handle)) {
 701                 err = PTR_ERR(handle);
 702                 handle = NULL;
 703                 mlog_errno(err);
 704                 goto out_unlock_inode;
 705         }
 706
 707         err = ocfs2_journal_access(handle, inode, fe_bh,
 708                                    OCFS2_JOURNAL_ACCESS_WRITE);
 709         if (err < 0) {
 710                 mlog_errno(err);
 711                 goto out_commit;
 712         }
 713
 714         inc_nlink(inode);
 715         inode->i_ctime = CURRENT_TIME;
 716         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 717         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 718         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 719
 720         err = ocfs2_journal_dirty(handle, fe_bh);
 721         if (err < 0) {
 722                 le16_add_cpu(&fe->i_links_count, -1);
 723                 drop_nlink(inode);
 724                 mlog_errno(err);
 725                 goto out_commit;
 726         }
 727
 728         err = ocfs2_add_entry(handle, dentry, inode,
 729                               OCFS2_I(inode)->ip_blkno,
 730                               parent_fe_bh, de_bh);
 731         if (err) {
 732                 le16_add_cpu(&fe->i_links_count, -1);
 733                 drop_nlink(inode);
 734                 mlog_errno(err);
 735                 goto out_commit;
 736         }
 737
 738         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 739         if (err) {
 740                 mlog_errno(err);
 741                 goto out_commit;
 742         }
 743
 744         atomic_inc(&inode->i_count);
 745         dentry->d_op = &ocfs2_dentry_ops;
 746         d_instantiate(dentry, inode);
 747
 748 out_commit:
 749         ocfs2_commit_trans(osb, handle);
 750 out_unlock_inode:
 751         ocfs2_meta_unlock(inode, 1);
 752
 753 out:
 754         ocfs2_meta_unlock(dir, 1);
 755
 756         if (de_bh)
 757                 brelse(de_bh);
 758         if (fe_bh)
 759                 brelse(fe_bh);
 760         if (parent_fe_bh)
 761                 brelse(parent_fe_bh);
 762
 763         mlog_exit(err);
 764
 765         return err;
 766 }
 767
 768 /*
 769  * Takes and drops an exclusive lock on the given dentry. This will
 770  * force other nodes to drop it.
 771  */
 772 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 773 {
 774         int ret;
 775
 776         ret = ocfs2_dentry_lock(dentry, 1);
 777         if (ret)
 778                 mlog_errno(ret);
 779         else
 780                 ocfs2_dentry_unlock(dentry, 1);
 781
 782         return ret;
 783 }
 784
 785 static inline int inode_is_unlinkable(struct inode *inode)
 786 {
 787         if (S_ISDIR(inode->i_mode)) {
 788                 if (inode->i_nlink == 2)
 789                         return 1;
 790                 return 0;
 791         }
 792
 793         if (inode->i_nlink == 1)
 794                 return 1;
 795         return 0;
 796 }
 797
 798 static int ocfs2_unlink(struct inode *dir,
 799                         struct dentry *dentry)
 800 {
 801         int status;
 802         int child_locked = 0;
 803         struct inode *inode = dentry->d_inode;
 804         struct inode *orphan_dir = NULL;
 805         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 806         u64 blkno;
 807         struct ocfs2_dinode *fe = NULL;
 808         struct buffer_head *fe_bh = NULL;
 809         struct buffer_head *parent_node_bh = NULL;
 810         handle_t *handle = NULL;
 811         struct ocfs2_dir_entry *dirent = NULL;
 812         struct buffer_head *dirent_bh = NULL;
 813         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 814         struct buffer_head *orphan_entry_bh = NULL;
 815
 816         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 817                    dentry->d_name.len, dentry->d_name.name);
 818
 819         BUG_ON(dentry->d_parent->d_inode != dir);
 820
 821         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 822
 823         if (inode == osb->root_inode) {
 824                 mlog(0, "Cannot delete the root directory\n");
 825                 return -EPERM;
 826         }
 827
 828         status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
 829         if (status < 0) {
 830                 if (status != -ENOENT)
 831                         mlog_errno(status);
 832                 return status;
 833         }
 834
 835         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 836                                           dentry->d_name.len, &blkno,
 837                                           dir, &dirent_bh, &dirent);
 838         if (status < 0) {
 839                 if (status != -ENOENT)
 840                         mlog_errno(status);
 841                 goto leave;
 842         }
 843
 844         if (OCFS2_I(inode)->ip_blkno != blkno) {
 845                 status = -ENOENT;
 846
 847                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 848                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 849                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 850                 goto leave;
 851         }
 852
 853         status = ocfs2_meta_lock(inode, &fe_bh, 1);
 854         if (status < 0) {
 855                 if (status != -ENOENT)
 856                         mlog_errno(status);
 857                 goto leave;
 858         }
 859         child_locked = 1;
 860
 861         if (S_ISDIR(inode->i_mode)) {
 862                 if (!ocfs2_empty_dir(inode)) {
 863                         status = -ENOTEMPTY;
 864                         goto leave;
 865                 } else if (inode->i_nlink != 2) {
 866                         status = -ENOTEMPTY;
 867                         goto leave;
 868                 }
 869         }
 870
 871         status = ocfs2_remote_dentry_delete(dentry);
 872         if (status < 0) {
 873                 /* This vote should succeed under all normal
 874                  * circumstances. */
 875                 mlog_errno(status);
 876                 goto leave;
 877         }
 878
 879         if (inode_is_unlinkable(inode)) {
 880                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 881                                                   orphan_name,
 882                                                   &orphan_entry_bh);
 883                 if (status < 0) {
 884                         mlog_errno(status);
 885                         goto leave;
 886                 }
 887         }
 888
 889         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 890         if (IS_ERR(handle)) {
 891                 status = PTR_ERR(handle);
 892                 handle = NULL;
 893                 mlog_errno(status);
 894                 goto leave;
 895         }
 896
 897         status = ocfs2_journal_access(handle, inode, fe_bh,
 898                                       OCFS2_JOURNAL_ACCESS_WRITE);
 899         if (status < 0) {
 900                 mlog_errno(status);
 901                 goto leave;
 902         }
 903
 904         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 905
 906         if (inode_is_unlinkable(inode)) {
 907                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 908                                           orphan_entry_bh, orphan_dir);
 909                 if (status < 0) {
 910                         mlog_errno(status);
 911                         goto leave;
 912                 }
 913         }
 914
 915         /* delete the name from the parent dir */
 916         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 917         if (status < 0) {
 918                 mlog_errno(status);
 919                 goto leave;
 920         }
 921
 922         if (S_ISDIR(inode->i_mode))
 923                 drop_nlink(inode);
 924         drop_nlink(inode);
 925         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 926
 927         status = ocfs2_journal_dirty(handle, fe_bh);
 928         if (status < 0) {
 929                 mlog_errno(status);
 930                 goto leave;
 931         }
 932
 933         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 934         if (S_ISDIR(inode->i_mode))
 935                 drop_nlink(dir);
 936
 937         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 938         if (status < 0) {
 939                 mlog_errno(status);
 940                 if (S_ISDIR(inode->i_mode))
 941                         inc_nlink(dir);
 942         }
 943
 944 leave:
 945         if (handle)
 946                 ocfs2_commit_trans(osb, handle);
 947
 948         if (child_locked)
 949                 ocfs2_meta_unlock(inode, 1);
 950
 951         ocfs2_meta_unlock(dir, 1);
 952
 953         if (orphan_dir) {
 954                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 955                 ocfs2_meta_unlock(orphan_dir, 1);
 956                 mutex_unlock(&orphan_dir->i_mutex);
 957                 iput(orphan_dir);
 958         }
 959
 960         if (fe_bh)
 961                 brelse(fe_bh);
 962
 963         if (dirent_bh)
 964                 brelse(dirent_bh);
 965
 966         if (parent_node_bh)
 967                 brelse(parent_node_bh);
 968
 969         if (orphan_entry_bh)
 970                 brelse(orphan_entry_bh);
 971
 972         mlog_exit(status);
 973
 974         return status;
 975 }
 976
 977 /*
 978  * The only place this should be used is rename!
 979  * if they have the same id, then the 1st one is the only one locked.
 980  */
 981 static int ocfs2_double_lock(struct ocfs2_super *osb,
 982                              struct buffer_head **bh1,
 983                              struct inode *inode1,
 984                              struct buffer_head **bh2,
 985                              struct inode *inode2)
 986 {
 987         int status;
 988         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 989         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 990         struct buffer_head **tmpbh;
 991         struct inode *tmpinode;
 992
 993         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 994                    (unsigned long long)oi1->ip_blkno,
 995                    (unsigned long long)oi2->ip_blkno);
 996
 997         if (*bh1)
 998                 *bh1 = NULL;
 999         if (*bh2)
1000                 *bh2 = NULL;
1001
1002         /* we always want to lock the one with the lower lockid first. */
1003         if (oi1->ip_blkno != oi2->ip_blkno) {
1004                 if (oi1->ip_blkno < oi2->ip_blkno) {
1005                         /* switch id1 and id2 around */
1006                         mlog(0, "switching them around...\n");
1007                         tmpbh = bh2;
1008                         bh2 = bh1;
1009                         bh1 = tmpbh;
1010
1011                         tmpinode = inode2;
1012                         inode2 = inode1;
1013                         inode1 = tmpinode;
1014                 }
1015                 /* lock id2 */
1016                 status = ocfs2_meta_lock(inode2, bh2, 1);
1017                 if (status < 0) {
1018                         if (status != -ENOENT)
1019                                 mlog_errno(status);
1020                         goto bail;
1021                 }
1022         }
1023
1024         /* lock id1 */
1025         status = ocfs2_meta_lock(inode1, bh1, 1);
1026         if (status < 0) {
1027                 /*
1028                  * An error return must mean that no cluster locks
1029                  * were held on function exit.
1030                  */
1031                 if (oi1->ip_blkno != oi2->ip_blkno)
1032                         ocfs2_meta_unlock(inode2, 1);
1033
1034                 if (status != -ENOENT)
1035                         mlog_errno(status);
1036         }
1037
1038 bail:
1039         mlog_exit(status);
1040         return status;
1041 }
1042
1043 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1044 {
1045         ocfs2_meta_unlock(inode1, 1);
1046
1047         if (inode1 != inode2)
1048                 ocfs2_meta_unlock(inode2, 1);
1049 }
1050
1051 #define PARENT_INO(buffer) \
1052         ((struct ocfs2_dir_entry *) \
1053          ((char *)buffer + \
1054           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1055
1056 static int ocfs2_rename(struct inode *old_dir,
1057                         struct dentry *old_dentry,
1058                         struct inode *new_dir,
1059                         struct dentry *new_dentry)
1060 {
1061         int status = 0, rename_lock = 0, parents_locked = 0;
1062         int old_child_locked = 0, new_child_locked = 0;
1063         struct inode *old_inode = old_dentry->d_inode;
1064         struct inode *new_inode = new_dentry->d_inode;
1065         struct inode *orphan_dir = NULL;
1066         struct ocfs2_dinode *newfe = NULL;
1067         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1068         struct buffer_head *orphan_entry_bh = NULL;
1069         struct buffer_head *newfe_bh = NULL;
1070         struct buffer_head *old_inode_bh = NULL;
1071         struct buffer_head *insert_entry_bh = NULL;
1072         struct ocfs2_super *osb = NULL;
1073         u64 newfe_blkno;
1074         handle_t *handle = NULL;
1075         struct buffer_head *old_dir_bh = NULL;
1076         struct buffer_head *new_dir_bh = NULL;
1077         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1078                                                                // and new_dentry
1079         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1080         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1081                                                     // this is the 1st dirent bh
1082         nlink_t old_dir_nlink = old_dir->i_nlink;
1083
1084         /* At some point it might be nice to break this function up a
1085          * bit. */
1086
1087         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1088                    old_dir, old_dentry, new_dir, new_dentry,
1089                    old_dentry->d_name.len, old_dentry->d_name.name,
1090                    new_dentry->d_name.len, new_dentry->d_name.name);
1091
1092         osb = OCFS2_SB(old_dir->i_sb);
1093
1094         if (new_inode) {
1095                 if (!igrab(new_inode))
1096                         BUG();
1097         }
1098
1099         /* Assume a directory hierarchy thusly:
1100          * a/b/c
1101          * a/d
1102          * a,b,c, and d are all directories.
1103          *
1104          * from cwd of 'a' on both nodes:
1105          * node1: mv b/c d
1106          * node2: mv d   b/c
1107          *
1108          * And that's why, just like the VFS, we need a file system
1109          * rename lock. */
1110         if (old_dentry != new_dentry) {
1111                 status = ocfs2_rename_lock(osb);
1112                 if (status < 0) {
1113                         mlog_errno(status);
1114                         goto bail;
1115                 }
1116                 rename_lock = 1;
1117         }
1118
1119         /* if old and new are the same, this'll just do one lock. */
1120         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1121                                    &new_dir_bh, new_dir);
1122         if (status < 0) {
1123                 mlog_errno(status);
1124                 goto bail;
1125         }
1126         parents_locked = 1;
1127
1128         /* make sure both dirs have bhs
1129          * get an extra ref on old_dir_bh if old==new */
1130         if (!new_dir_bh) {
1131                 if (old_dir_bh) {
1132                         new_dir_bh = old_dir_bh;
1133                         get_bh(new_dir_bh);
1134                 } else {
1135                         mlog(ML_ERROR, "no old_dir_bh!\n");
1136                         status = -EIO;
1137                         goto bail;
1138                 }
1139         }
1140
1141         /*
1142          * Aside from allowing a meta data update, the locking here
1143          * also ensures that the vote thread on other nodes won't have
1144          * to concurrently downconvert the inode and the dentry locks.
1145          */
1146         status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1147         if (status < 0) {
1148                 if (status != -ENOENT)
1149                         mlog_errno(status);
1150                 goto bail;
1151         }
1152         old_child_locked = 1;
1153
1154         status = ocfs2_remote_dentry_delete(old_dentry);
1155         if (status < 0) {
1156                 mlog_errno(status);
1157                 goto bail;
1158         }
1159
1160         if (S_ISDIR(old_inode->i_mode)) {
1161                 status = -EIO;
1162                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1163                 if (!old_inode_de_bh)
1164                         goto bail;
1165
1166                 status = -EIO;
1167                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1168                     OCFS2_I(old_dir)->ip_blkno)
1169                         goto bail;
1170                 status = -EMLINK;
1171                 if (!new_inode && new_dir!=old_dir &&
1172                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1173                         goto bail;
1174         }
1175
1176         status = -ENOENT;
1177         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1178                                      old_dentry->d_name.len,
1179                                      old_dir, &old_de);
1180         if (!old_de_bh)
1181                 goto bail;
1182
1183         /*
1184          *  Check for inode number is _not_ due to possible IO errors.
1185          *  We might rmdir the source, keep it as pwd of some process
1186          *  and merrily kill the link to whatever was created under the
1187          *  same name. Goodbye sticky bit ;-<
1188          */
1189         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1190                 goto bail;
1191
1192         /* check if the target already exists (in which case we need
1193          * to delete it */
1194         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1195                                           new_dentry->d_name.len,
1196                                           &newfe_blkno, new_dir, &new_de_bh,
1197                                           &new_de);
1198         /* The only error we allow here is -ENOENT because the new
1199          * file not existing is perfectly valid. */
1200         if ((status < 0) && (status != -ENOENT)) {
1201                 /* If we cannot find the file specified we should just */
1202                 /* return the error... */
1203                 mlog_errno(status);
1204                 goto bail;
1205         }
1206
1207         if (!new_de && new_inode)
1208                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1209                      "directory!", new_inode->i_ino);
1210
1211         /* In case we need to overwrite an existing file, we blow it
1212          * away first */
1213         if (new_de) {
1214                 /* VFS didn't think there existed an inode here, but
1215                  * someone else in the cluster must have raced our
1216                  * rename to create one. Today we error cleanly, in
1217                  * the future we should consider calling iget to build
1218                  * a new struct inode for this entry. */
1219                 if (!new_inode) {
1220                         status = -EACCES;
1221
1222                         mlog(0, "We found an inode for name %.*s but VFS "
1223                              "didn't give us one.\n", new_dentry->d_name.len,
1224                              new_dentry->d_name.name);
1225                         goto bail;
1226                 }
1227
1228                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1229                         status = -EACCES;
1230
1231                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1232                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1233                              (unsigned long long)newfe_blkno,
1234                              OCFS2_I(new_inode)->ip_flags);
1235                         goto bail;
1236                 }
1237
1238                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1239                 if (status < 0) {
1240                         if (status != -ENOENT)
1241                                 mlog_errno(status);
1242                         goto bail;
1243                 }
1244                 new_child_locked = 1;
1245
1246                 status = ocfs2_remote_dentry_delete(new_dentry);
1247                 if (status < 0) {
1248                         mlog_errno(status);
1249                         goto bail;
1250                 }
1251
1252                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1253
1254                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1255                      "newfebh=%p bhblocknr=%llu\n", new_de,
1256                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1257                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1258
1259                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1260                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1261                                                           new_inode,
1262                                                           orphan_name,
1263                                                           &orphan_entry_bh);
1264                         if (status < 0) {
1265                                 mlog_errno(status);
1266                                 goto bail;
1267                         }
1268                 }
1269         } else {
1270                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1271
1272                 status = ocfs2_check_dir_for_entry(new_dir,
1273                                                    new_dentry->d_name.name,
1274                                                    new_dentry->d_name.len);
1275                 if (status)
1276                         goto bail;
1277
1278                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1279                                                       new_dentry->d_name.name,
1280                                                       new_dentry->d_name.len,
1281                                                       &insert_entry_bh);
1282                 if (status < 0) {
1283                         mlog_errno(status);
1284                         goto bail;
1285                 }
1286         }
1287
1288         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1289         if (IS_ERR(handle)) {
1290                 status = PTR_ERR(handle);
1291                 handle = NULL;
1292                 mlog_errno(status);
1293                 goto bail;
1294         }
1295
1296         if (new_de) {
1297                 if (S_ISDIR(new_inode->i_mode)) {
1298                         if (!ocfs2_empty_dir(new_inode) ||
1299                             new_inode->i_nlink != 2) {
1300                                 status = -ENOTEMPTY;
1301                                 goto bail;
1302                         }
1303                 }
1304                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1305                                               OCFS2_JOURNAL_ACCESS_WRITE);
1306                 if (status < 0) {
1307                         mlog_errno(status);
1308                         goto bail;
1309                 }
1310
1311                 if (S_ISDIR(new_inode->i_mode) ||
1312                     (newfe->i_links_count == cpu_to_le16(1))){
1313                         status = ocfs2_orphan_add(osb, handle, new_inode,
1314                                                   newfe, orphan_name,
1315                                                   orphan_entry_bh, orphan_dir);
1316                         if (status < 0) {
1317                                 mlog_errno(status);
1318                                 goto bail;
1319                         }
1320                 }
1321
1322                 /* change the dirent to point to the correct inode */
1323                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1324                                               OCFS2_JOURNAL_ACCESS_WRITE);
1325                 if (status < 0) {
1326                         mlog_errno(status);
1327                         goto bail;
1328                 }
1329                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1330                 new_de->file_type = old_de->file_type;
1331                 new_dir->i_version++;
1332                 status = ocfs2_journal_dirty(handle, new_de_bh);
1333                 if (status < 0) {
1334                         mlog_errno(status);
1335                         goto bail;
1336                 }
1337
1338                 if (S_ISDIR(new_inode->i_mode))
1339                         newfe->i_links_count = 0;
1340                 else
1341                         le16_add_cpu(&newfe->i_links_count, -1);
1342
1343                 status = ocfs2_journal_dirty(handle, newfe_bh);
1344                 if (status < 0) {
1345                         mlog_errno(status);
1346                         goto bail;
1347                 }
1348         } else {
1349                 /* if the name was not found in new_dir, add it now */
1350                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1351                                          OCFS2_I(old_inode)->ip_blkno,
1352                                          new_dir_bh, insert_entry_bh);
1353         }
1354
1355         old_inode->i_ctime = CURRENT_TIME;
1356         mark_inode_dirty(old_inode);
1357         ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
1358
1359         /* now that the name has been added to new_dir, remove the old name */
1360         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1361         if (status < 0) {
1362                 mlog_errno(status);
1363                 goto bail;
1364         }
1365
1366         if (new_inode) {
1367                 new_inode->i_nlink--;
1368                 new_inode->i_ctime = CURRENT_TIME;
1369         }
1370         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1371         if (old_inode_de_bh) {
1372                 status = ocfs2_journal_access(handle, old_inode,
1373                                              old_inode_de_bh,
1374                                              OCFS2_JOURNAL_ACCESS_WRITE);
1375                 PARENT_INO(old_inode_de_bh->b_data) =
1376                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1377                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1378                 old_dir->i_nlink--;
1379                 if (new_inode) {
1380                         new_inode->i_nlink--;
1381                 } else {
1382                         inc_nlink(new_dir);
1383                         mark_inode_dirty(new_dir);
1384                 }
1385         }
1386         mark_inode_dirty(old_dir);
1387         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1388         if (new_inode) {
1389                 mark_inode_dirty(new_inode);
1390                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1391         }
1392
1393         if (old_dir != new_dir) {
1394                 /* Keep the same times on both directories.*/
1395                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1396
1397                 /*
1398                  * This will also pick up the i_nlink change from the
1399                  * block above.
1400                  */
1401                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1402         }
1403
1404         if (old_dir_nlink != old_dir->i_nlink) {
1405                 if (!old_dir_bh) {
1406                         mlog(ML_ERROR, "need to change nlink for old dir "
1407                              "%llu from %d to %d but bh is NULL!\n",
1408                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1409                              (int)old_dir_nlink, old_dir->i_nlink);
1410                 } else {
1411                         struct ocfs2_dinode *fe;
1412                         status = ocfs2_journal_access(handle, old_dir,
1413                                                       old_dir_bh,
1414                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1415                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1416                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1417                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1418                 }
1419         }
1420
1421         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1422         status = 0;
1423 bail:
1424         if (rename_lock)
1425                 ocfs2_rename_unlock(osb);
1426
1427         if (handle)
1428                 ocfs2_commit_trans(osb, handle);
1429
1430         if (parents_locked)
1431                 ocfs2_double_unlock(old_dir, new_dir);
1432
1433         if (old_child_locked)
1434                 ocfs2_meta_unlock(old_inode, 1);
1435
1436         if (new_child_locked)
1437                 ocfs2_meta_unlock(new_inode, 1);
1438
1439         if (orphan_dir) {
1440                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1441                 ocfs2_meta_unlock(orphan_dir, 1);
1442                 mutex_unlock(&orphan_dir->i_mutex);
1443                 iput(orphan_dir);
1444         }
1445
1446         if (new_inode)
1447                 sync_mapping_buffers(old_inode->i_mapping);
1448
1449         if (new_inode)
1450                 iput(new_inode);
1451         if (newfe_bh)
1452                 brelse(newfe_bh);
1453         if (old_inode_bh)
1454                 brelse(old_inode_bh);
1455         if (old_dir_bh)
1456                 brelse(old_dir_bh);
1457         if (new_dir_bh)
1458                 brelse(new_dir_bh);
1459         if (new_de_bh)
1460                 brelse(new_de_bh);
1461         if (old_de_bh)
1462                 brelse(old_de_bh);
1463         if (old_inode_de_bh)
1464                 brelse(old_inode_de_bh);
1465         if (orphan_entry_bh)
1466                 brelse(orphan_entry_bh);
1467         if (insert_entry_bh)
1468                 brelse(insert_entry_bh);
1469
1470         mlog_exit(status);
1471
1472         return status;
1473 }
1474
1475 /*
1476  * we expect i_size = strlen(symname). Copy symname into the file
1477  * data, including the null terminator.
1478  */
1479 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1480                                      handle_t *handle,
1481                                      struct inode *inode,
1482                                      const char *symname)
1483 {
1484         struct buffer_head **bhs = NULL;
1485         const char *c;
1486         struct super_block *sb = osb->sb;
1487         u64 p_blkno, p_blocks;
1488         int virtual, blocks, status, i, bytes_left;
1489
1490         bytes_left = i_size_read(inode) + 1;
1491         /* we can't trust i_blocks because we're actually going to
1492          * write i_size + 1 bytes. */
1493         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1494
1495         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1496                         (unsigned long long)inode->i_blocks,
1497                         i_size_read(inode), blocks);
1498
1499         /* Sanity check -- make sure we're going to fit. */
1500         if (bytes_left >
1501             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1502                 status = -EIO;
1503                 mlog_errno(status);
1504                 goto bail;
1505         }
1506
1507         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1508         if (!bhs) {
1509                 status = -ENOMEM;
1510                 mlog_errno(status);
1511                 goto bail;
1512         }
1513
1514         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1515                                              NULL);
1516         if (status < 0) {
1517                 mlog_errno(status);
1518                 goto bail;
1519         }
1520
1521         /* links can never be larger than one cluster so we know this
1522          * is all going to be contiguous, but do a sanity check
1523          * anyway. */
1524         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1525                 status = -EIO;
1526                 mlog_errno(status);
1527                 goto bail;
1528         }
1529
1530         virtual = 0;
1531         while(bytes_left > 0) {
1532                 c = &symname[virtual * sb->s_blocksize];
1533
1534                 bhs[virtual] = sb_getblk(sb, p_blkno);
1535                 if (!bhs[virtual]) {
1536                         status = -ENOMEM;
1537                         mlog_errno(status);
1538                         goto bail;
1539                 }
1540                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1541
1542                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1543                                               OCFS2_JOURNAL_ACCESS_CREATE);
1544                 if (status < 0) {
1545                         mlog_errno(status);
1546                         goto bail;
1547                 }
1548
1549                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1550
1551                 memcpy(bhs[virtual]->b_data, c,
1552                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1553                        bytes_left);
1554
1555                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1556                 if (status < 0) {
1557                         mlog_errno(status);
1558                         goto bail;
1559                 }
1560
1561                 virtual++;
1562                 p_blkno++;
1563                 bytes_left -= sb->s_blocksize;
1564         }
1565
1566         status = 0;
1567 bail:
1568
1569         if (bhs) {
1570                 for(i = 0; i < blocks; i++)
1571                         if (bhs[i])
1572                                 brelse(bhs[i]);
1573                 kfree(bhs);
1574         }
1575
1576         mlog_exit(status);
1577         return status;
1578 }
1579
1580 static int ocfs2_symlink(struct inode *dir,
1581                          struct dentry *dentry,
1582                          const char *symname)
1583 {
1584         int status, l, credits;
1585         u64 newsize;
1586         struct ocfs2_super *osb = NULL;
1587         struct inode *inode = NULL;
1588         struct super_block *sb;
1589         struct buffer_head *new_fe_bh = NULL;
1590         struct buffer_head *de_bh = NULL;
1591         struct buffer_head *parent_fe_bh = NULL;
1592         struct ocfs2_dinode *fe = NULL;
1593         struct ocfs2_dinode *dirfe;
1594         handle_t *handle = NULL;
1595         struct ocfs2_alloc_context *inode_ac = NULL;
1596         struct ocfs2_alloc_context *data_ac = NULL;
1597
1598         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1599                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1600
1601         sb = dir->i_sb;
1602         osb = OCFS2_SB(sb);
1603
1604         l = strlen(symname) + 1;
1605
1606         credits = ocfs2_calc_symlink_credits(sb);
1607
1608         /* lock the parent directory */
1609         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1610         if (status < 0) {
1611                 if (status != -ENOENT)
1612                         mlog_errno(status);
1613                 return status;
1614         }
1615
1616         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1617         if (!dirfe->i_links_count) {
1618                 /* can't make a file in a deleted directory. */
1619                 status = -ENOENT;
1620                 goto bail;
1621         }
1622
1623         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1624                                            dentry->d_name.len);
1625         if (status)
1626                 goto bail;
1627
1628         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1629                                               dentry->d_name.name,
1630                                               dentry->d_name.len, &de_bh);
1631         if (status < 0) {
1632                 mlog_errno(status);
1633                 goto bail;
1634         }
1635
1636         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1637         if (status < 0) {
1638                 if (status != -ENOSPC)
1639                         mlog_errno(status);
1640                 goto bail;
1641         }
1642
1643         /* don't reserve bitmap space for fast symlinks. */
1644         if (l > ocfs2_fast_symlink_chars(sb)) {
1645                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1646                 if (status < 0) {
1647                         if (status != -ENOSPC)
1648                                 mlog_errno(status);
1649                         goto bail;
1650                 }
1651         }
1652
1653         handle = ocfs2_start_trans(osb, credits);
1654         if (IS_ERR(handle)) {
1655                 status = PTR_ERR(handle);
1656                 handle = NULL;
1657                 mlog_errno(status);
1658                 goto bail;
1659         }
1660
1661         status = ocfs2_mknod_locked(osb, dir, dentry,
1662                                     S_IFLNK | S_IRWXUGO, 0,
1663                                     &new_fe_bh, parent_fe_bh, handle,
1664                                     &inode, inode_ac);
1665         if (status < 0) {
1666                 mlog_errno(status);
1667                 goto bail;
1668         }
1669
1670         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1671         inode->i_rdev = 0;
1672         newsize = l - 1;
1673         if (l > ocfs2_fast_symlink_chars(sb)) {
1674                 u32 offset = 0;
1675
1676                 inode->i_op = &ocfs2_symlink_inode_operations;
1677                 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1678                                                     new_fe_bh,
1679                                                     handle, data_ac, NULL,
1680                                                     NULL);
1681                 if (status < 0) {
1682                         if (status != -ENOSPC && status != -EINTR) {
1683                                 mlog(ML_ERROR,
1684                                      "Failed to extend file to %llu\n",
1685                                      (unsigned long long)newsize);
1686                                 mlog_errno(status);
1687                                 status = -ENOSPC;
1688                         }
1689                         goto bail;
1690                 }
1691                 i_size_write(inode, newsize);
1692                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1693         } else {
1694                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1695                 memcpy((char *) fe->id2.i_symlink, symname, l);
1696                 i_size_write(inode, newsize);
1697                 inode->i_blocks = 0;
1698         }
1699
1700         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1701         if (status < 0) {
1702                 mlog_errno(status);
1703                 goto bail;
1704         }
1705
1706         if (!ocfs2_inode_is_fast_symlink(inode)) {
1707                 status = ocfs2_create_symlink_data(osb, handle, inode,
1708                                                    symname);
1709                 if (status < 0) {
1710                         mlog_errno(status);
1711                         goto bail;
1712                 }
1713         }
1714
1715         status = ocfs2_add_entry(handle, dentry, inode,
1716                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1717                                  de_bh);
1718         if (status < 0) {
1719                 mlog_errno(status);
1720                 goto bail;
1721         }
1722
1723         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1724         if (status) {
1725                 mlog_errno(status);
1726                 goto bail;
1727         }
1728
1729         insert_inode_hash(inode);
1730         dentry->d_op = &ocfs2_dentry_ops;
1731         d_instantiate(dentry, inode);
1732 bail:
1733         if (handle)
1734                 ocfs2_commit_trans(osb, handle);
1735
1736         ocfs2_meta_unlock(dir, 1);
1737
1738         if (new_fe_bh)
1739                 brelse(new_fe_bh);
1740         if (parent_fe_bh)
1741                 brelse(parent_fe_bh);
1742         if (de_bh)
1743                 brelse(de_bh);
1744         if (inode_ac)
1745                 ocfs2_free_alloc_context(inode_ac);
1746         if (data_ac)
1747                 ocfs2_free_alloc_context(data_ac);
1748         if ((status < 0) && inode)
1749                 iput(inode);
1750
1751         mlog_exit(status);
1752
1753         return status;
1754 }
1755
1756 int ocfs2_check_dir_entry(struct inode * dir,
1757                           struct ocfs2_dir_entry * de,
1758                           struct buffer_head * bh,
1759                           unsigned long offset)
1760 {
1761         const char *error_msg = NULL;
1762         const int rlen = le16_to_cpu(de->rec_len);
1763
1764         if (rlen < OCFS2_DIR_REC_LEN(1))
1765                 error_msg = "rec_len is smaller than minimal";
1766         else if (rlen % 4 != 0)
1767                 error_msg = "rec_len % 4 != 0";
1768         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1769                 error_msg = "rec_len is too small for name_len";
1770         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1771                 error_msg = "directory entry across blocks";
1772
1773         if (error_msg != NULL)
1774                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1775                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1776                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1777                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1778                      de->name_len);
1779         return error_msg == NULL ? 1 : 0;
1780 }
1781
1782 /* we don't always have a dentry for what we want to add, so people
1783  * like orphan dir can call this instead.
1784  *
1785  * If you pass me insert_bh, I'll skip the search of the other dir
1786  * blocks and put the record in there.
1787  */
1788 static int __ocfs2_add_entry(handle_t *handle,
1789                              struct inode *dir,
1790                              const char *name, int namelen,
1791                              struct inode *inode, u64 blkno,
1792                              struct buffer_head *parent_fe_bh,
1793                              struct buffer_head *insert_bh)
1794 {
1795         unsigned long offset;
1796         unsigned short rec_len;
1797         struct ocfs2_dir_entry *de, *de1;
1798         struct super_block *sb;
1799         int retval, status;
1800
1801         mlog_entry_void();
1802
1803         sb = dir->i_sb;
1804
1805         if (!namelen)
1806                 return -EINVAL;
1807
1808         rec_len = OCFS2_DIR_REC_LEN(namelen);
1809         offset = 0;
1810         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1811         while (1) {
1812                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1813                 /* These checks should've already been passed by the
1814                  * prepare function, but I guess we can leave them
1815                  * here anyway. */
1816                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1817                         retval = -ENOENT;
1818                         goto bail;
1819                 }
1820                 if (ocfs2_match(namelen, name, de)) {
1821                         retval = -EEXIST;
1822                         goto bail;
1823                 }
1824                 if (((le64_to_cpu(de->inode) == 0) &&
1825                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1826                     (le16_to_cpu(de->rec_len) >=
1827                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1828                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1829                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1830                         if (retval < 0) {
1831                                 mlog_errno(retval);
1832                                 goto bail;
1833                         }
1834
1835                         status = ocfs2_journal_access(handle, dir, insert_bh,
1836                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1837                         /* By now the buffer is marked for journaling */
1838                         offset += le16_to_cpu(de->rec_len);
1839                         if (le64_to_cpu(de->inode)) {
1840                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1841                                         OCFS2_DIR_REC_LEN(de->name_len));
1842                                 de1->rec_len =
1843                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1844                                         OCFS2_DIR_REC_LEN(de->name_len));
1845                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1846                                 de = de1;
1847                         }
1848                         de->file_type = OCFS2_FT_UNKNOWN;
1849                         if (blkno) {
1850                                 de->inode = cpu_to_le64(blkno);
1851                                 ocfs2_set_de_type(de, inode->i_mode);
1852                         } else
1853                                 de->inode = 0;
1854                         de->name_len = namelen;
1855                         memcpy(de->name, name, namelen);
1856
1857                         dir->i_version++;
1858                         status = ocfs2_journal_dirty(handle, insert_bh);
1859                         retval = 0;
1860                         goto bail;
1861                 }
1862                 offset += le16_to_cpu(de->rec_len);
1863                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1864         }
1865
1866         /* when you think about it, the assert above should prevent us
1867          * from ever getting here. */
1868         retval = -ENOSPC;
1869 bail:
1870
1871         mlog_exit(retval);
1872         return retval;
1873 }
1874
1875
1876 /*
1877  * ocfs2_delete_entry deletes a directory entry by merging it with the
1878  * previous entry
1879  */
1880 static int ocfs2_delete_entry(handle_t *handle,
1881                               struct inode *dir,
1882                               struct ocfs2_dir_entry *de_del,
1883                               struct buffer_head *bh)
1884 {
1885         struct ocfs2_dir_entry *de, *pde;
1886         int i, status = -ENOENT;
1887
1888         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1889
1890         i = 0;
1891         pde = NULL;
1892         de = (struct ocfs2_dir_entry *) bh->b_data;
1893         while (i < bh->b_size) {
1894                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1895                         status = -EIO;
1896                         mlog_errno(status);
1897                         goto bail;
1898                 }
1899                 if (de == de_del)  {
1900                         status = ocfs2_journal_access(handle, dir, bh,
1901                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1902                         if (status < 0) {
1903                                 status = -EIO;
1904                                 mlog_errno(status);
1905                                 goto bail;
1906                         }
1907                         if (pde)
1908                                 pde->rec_len =
1909                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1910                                                     le16_to_cpu(de->rec_len));
1911                         else
1912                                 de->inode = 0;
1913                         dir->i_version++;
1914                         status = ocfs2_journal_dirty(handle, bh);
1915                         goto bail;
1916                 }
1917                 i += le16_to_cpu(de->rec_len);
1918                 pde = de;
1919                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1920         }
1921 bail:
1922         mlog_exit(status);
1923         return status;
1924 }
1925
1926 /*
1927  * Returns 0 if not found, -1 on failure, and 1 on success
1928  */
1929 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1930                                         struct inode *dir,
1931                                         const char *name, int namelen,
1932                                         unsigned long offset,
1933                                         struct ocfs2_dir_entry **res_dir)
1934 {
1935         struct ocfs2_dir_entry *de;
1936         char *dlimit, *de_buf;
1937         int de_len;
1938         int ret = 0;
1939
1940         mlog_entry_void();
1941
1942         de_buf = bh->b_data;
1943         dlimit = de_buf + dir->i_sb->s_blocksize;
1944
1945         while (de_buf < dlimit) {
1946                 /* this code is executed quadratically often */
1947                 /* do minimal checking `by hand' */
1948
1949                 de = (struct ocfs2_dir_entry *) de_buf;
1950
1951                 if (de_buf + namelen <= dlimit &&
1952                     ocfs2_match(namelen, name, de)) {
1953                         /* found a match - just to be sure, do a full check */
1954                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1955                                 ret = -1;
1956                                 goto bail;
1957                         }
1958                         *res_dir = de;
1959                         ret = 1;
1960                         goto bail;
1961                 }
1962
1963                 /* prevent looping on a bad block */
1964                 de_len = le16_to_cpu(de->rec_len);
1965                 if (de_len <= 0) {
1966                         ret = -1;
1967                         goto bail;
1968                 }
1969
1970                 de_buf += de_len;
1971                 offset += de_len;
1972         }
1973
1974 bail:
1975         mlog_exit(ret);
1976         return ret;
1977 }
1978
1979 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1980                                      struct inode *dir,
1981                                      struct ocfs2_dir_entry **res_dir)
1982 {
1983         struct super_block *sb;
1984         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1985         struct buffer_head *bh, *ret = NULL;
1986         unsigned long start, block, b;
1987         int ra_max = 0;         /* Number of bh's in the readahead
1988                                    buffer, bh_use[] */
1989         int ra_ptr = 0;         /* Current index into readahead
1990                                    buffer */
1991         int num = 0;
1992         int nblocks, i, err;
1993
1994         mlog_entry_void();
1995
1996         *res_dir = NULL;
1997         sb = dir->i_sb;
1998
1999         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2000         start = OCFS2_I(dir)->ip_dir_start_lookup;
2001         if (start >= nblocks)
2002                 start = 0;
2003         block = start;
2004
2005 restart:
2006         do {
2007                 /*
2008                  * We deal with the read-ahead logic here.
2009                  */
2010                 if (ra_ptr >= ra_max) {
2011                         /* Refill the readahead buffer */
2012                         ra_ptr = 0;
2013                         b = block;
2014                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2015                                 /*
2016                                  * Terminate if we reach the end of the
2017                                  * directory and must wrap, or if our
2018                                  * search has finished at this block.
2019                                  */
2020                                 if (b >= nblocks || (num && block == start)) {
2021                                         bh_use[ra_max] = NULL;
2022                                         break;
2023                                 }
2024                                 num++;
2025
2026                                 bh = ocfs2_bread(dir, b++, &err, 1);
2027                                 bh_use[ra_max] = bh;
2028                         }
2029                 }
2030                 if ((bh = bh_use[ra_ptr++]) == NULL)
2031                         goto next;
2032                 wait_on_buffer(bh);
2033                 if (!buffer_uptodate(bh)) {
2034                         /* read error, skip block & hope for the best */
2035                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2036                                     "offset %lu\n",
2037                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2038                                     block);
2039                         brelse(bh);
2040                         goto next;
2041                 }
2042                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2043                                           block << sb->s_blocksize_bits,
2044                                           res_dir);
2045                 if (i == 1) {
2046                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2047                         ret = bh;
2048                         goto cleanup_and_exit;
2049                 } else {
2050                         brelse(bh);
2051                         if (i < 0)
2052                                 goto cleanup_and_exit;
2053                 }
2054         next:
2055                 if (++block >= nblocks)
2056                         block = 0;
2057         } while (block != start);
2058
2059         /*
2060          * If the directory has grown while we were searching, then
2061          * search the last part of the directory before giving up.
2062          */
2063         block = nblocks;
2064         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2065         if (block < nblocks) {
2066                 start = 0;
2067                 goto restart;
2068         }
2069
2070 cleanup_and_exit:
2071         /* Clean up the read-ahead blocks */
2072         for (; ra_ptr < ra_max; ra_ptr++)
2073                 brelse(bh_use[ra_ptr]);
2074
2075         mlog_exit_ptr(ret);
2076         return ret;
2077 }
2078
2079 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2080 {
2081         int status, namelen;
2082
2083         mlog_entry_void();
2084
2085         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2086                            (long long)blkno);
2087         if (namelen <= 0) {
2088                 if (namelen)
2089                         status = namelen;
2090                 else
2091                         status = -EINVAL;
2092                 mlog_errno(status);
2093                 goto bail;
2094         }
2095         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2096                 status = -EINVAL;
2097                 mlog_errno(status);
2098                 goto bail;
2099         }
2100
2101         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2102              namelen);
2103
2104         status = 0;
2105 bail:
2106         mlog_exit(status);
2107         return status;
2108 }
2109
2110 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2111                                     struct inode **ret_orphan_dir,
2112                                     struct inode *inode,
2113                                     char *name,
2114                                     struct buffer_head **de_bh)
2115 {
2116         struct inode *orphan_dir_inode;
2117         struct buffer_head *orphan_dir_bh = NULL;
2118         int status = 0;
2119
2120         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2121         if (status < 0) {
2122                 mlog_errno(status);
2123                 return status;
2124         }
2125
2126         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2127                                                        ORPHAN_DIR_SYSTEM_INODE,
2128                                                        osb->slot_num);
2129         if (!orphan_dir_inode) {
2130                 status = -ENOENT;
2131                 mlog_errno(status);
2132                 return status;
2133         }
2134
2135         mutex_lock(&orphan_dir_inode->i_mutex);
2136
2137         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2138         if (status < 0) {
2139                 mlog_errno(status);
2140                 goto leave;
2141         }
2142
2143         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2144                                               orphan_dir_bh, name,
2145                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2146         if (status < 0) {
2147                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2148
2149                 mlog_errno(status);
2150                 goto leave;
2151         }
2152
2153         *ret_orphan_dir = orphan_dir_inode;
2154
2155 leave:
2156         if (status) {
2157                 mutex_unlock(&orphan_dir_inode->i_mutex);
2158                 iput(orphan_dir_inode);
2159         }
2160
2161         if (orphan_dir_bh)
2162                 brelse(orphan_dir_bh);
2163
2164         mlog_exit(status);
2165         return status;
2166 }
2167
2168 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2169                             handle_t *handle,
2170                             struct inode *inode,
2171                             struct ocfs2_dinode *fe,
2172                             char *name,
2173                             struct buffer_head *de_bh,
2174                             struct inode *orphan_dir_inode)
2175 {
2176         struct buffer_head *orphan_dir_bh = NULL;
2177         int status = 0;
2178         struct ocfs2_dinode *orphan_fe;
2179
2180         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2181
2182         status = ocfs2_read_block(osb,
2183                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2184                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2185                                   orphan_dir_inode);
2186         if (status < 0) {
2187                 mlog_errno(status);
2188                 goto leave;
2189         }
2190
2191         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2192                                       OCFS2_JOURNAL_ACCESS_WRITE);
2193         if (status < 0) {
2194                 mlog_errno(status);
2195                 goto leave;
2196         }
2197
2198         /* we're a cluster, and nlink can change on disk from
2199          * underneath us... */
2200         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2201         if (S_ISDIR(inode->i_mode))
2202                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2203         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2204
2205         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2206         if (status < 0) {
2207                 mlog_errno(status);
2208                 goto leave;
2209         }
2210
2211         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2212                                    OCFS2_ORPHAN_NAMELEN, inode,
2213                                    OCFS2_I(inode)->ip_blkno,
2214                                    orphan_dir_bh, de_bh);
2215         if (status < 0) {
2216                 mlog_errno(status);
2217                 goto leave;
2218         }
2219
2220         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2221
2222         /* Record which orphan dir our inode now resides
2223          * in. delete_inode will use this to determine which orphan
2224          * dir to lock. */
2225         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2226
2227         mlog(0, "Inode %llu orphaned in slot %d\n",
2228              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2229
2230 leave:
2231         if (orphan_dir_bh)
2232                 brelse(orphan_dir_bh);
2233
2234         mlog_exit(status);
2235         return status;
2236 }
2237
2238 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2239 int ocfs2_orphan_del(struct ocfs2_super *osb,
2240                      handle_t *handle,
2241                      struct inode *orphan_dir_inode,
2242                      struct inode *inode,
2243                      struct buffer_head *orphan_dir_bh)
2244 {
2245         char name[OCFS2_ORPHAN_NAMELEN + 1];
2246         struct ocfs2_dinode *orphan_fe;
2247         int status = 0;
2248         struct buffer_head *target_de_bh = NULL;
2249         struct ocfs2_dir_entry *target_de = NULL;
2250
2251         mlog_entry_void();
2252
2253         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2260              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2261              OCFS2_ORPHAN_NAMELEN);
2262
2263         /* find it's spot in the orphan directory */
2264         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2265                                         orphan_dir_inode, &target_de);
2266         if (!target_de_bh) {
2267                 status = -ENOENT;
2268                 mlog_errno(status);
2269                 goto leave;
2270         }
2271
2272         /* remove it from the orphan directory */
2273         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2274                                     target_de_bh);
2275         if (status < 0) {
2276                 mlog_errno(status);
2277                 goto leave;
2278         }
2279
2280         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2281                                       OCFS2_JOURNAL_ACCESS_WRITE);
2282         if (status < 0) {
2283                 mlog_errno(status);
2284                 goto leave;
2285         }
2286
2287         /* do the i_nlink dance! :) */
2288         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2289         if (S_ISDIR(inode->i_mode))
2290                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2291         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2292
2293         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2294         if (status < 0) {
2295                 mlog_errno(status);
2296                 goto leave;
2297         }
2298
2299 leave:
2300         if (target_de_bh)
2301                 brelse(target_de_bh);
2302
2303         mlog_exit(status);
2304         return status;
2305 }
2306
2307 const struct inode_operations ocfs2_dir_iops = {
2308         .create         = ocfs2_create,
2309         .lookup         = ocfs2_lookup,
2310         .link           = ocfs2_link,
2311         .unlink         = ocfs2_unlink,
2312         .rmdir          = ocfs2_unlink,
2313         .symlink        = ocfs2_symlink,
2314         .mkdir          = ocfs2_mkdir,
2315         .mknod          = ocfs2_mknod,
2316         .rename         = ocfs2_rename,
2317         .setattr        = ocfs2_setattr,
2318         .getattr        = ocfs2_getattr,
2319         .permission     = ocfs2_permission,
2320 };