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         struct ocfs2_dinode *old_di;
1084
1085         /* At some point it might be nice to break this function up a
1086          * bit. */
1087
1088         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1089                    old_dir, old_dentry, new_dir, new_dentry,
1090                    old_dentry->d_name.len, old_dentry->d_name.name,
1091                    new_dentry->d_name.len, new_dentry->d_name.name);
1092
1093         osb = OCFS2_SB(old_dir->i_sb);
1094
1095         if (new_inode) {
1096                 if (!igrab(new_inode))
1097                         BUG();
1098         }
1099
1100         /* Assume a directory hierarchy thusly:
1101          * a/b/c
1102          * a/d
1103          * a,b,c, and d are all directories.
1104          *
1105          * from cwd of 'a' on both nodes:
1106          * node1: mv b/c d
1107          * node2: mv d   b/c
1108          *
1109          * And that's why, just like the VFS, we need a file system
1110          * rename lock. */
1111         if (old_dentry != new_dentry) {
1112                 status = ocfs2_rename_lock(osb);
1113                 if (status < 0) {
1114                         mlog_errno(status);
1115                         goto bail;
1116                 }
1117                 rename_lock = 1;
1118         }
1119
1120         /* if old and new are the same, this'll just do one lock. */
1121         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1122                                    &new_dir_bh, new_dir);
1123         if (status < 0) {
1124                 mlog_errno(status);
1125                 goto bail;
1126         }
1127         parents_locked = 1;
1128
1129         /* make sure both dirs have bhs
1130          * get an extra ref on old_dir_bh if old==new */
1131         if (!new_dir_bh) {
1132                 if (old_dir_bh) {
1133                         new_dir_bh = old_dir_bh;
1134                         get_bh(new_dir_bh);
1135                 } else {
1136                         mlog(ML_ERROR, "no old_dir_bh!\n");
1137                         status = -EIO;
1138                         goto bail;
1139                 }
1140         }
1141
1142         /*
1143          * Aside from allowing a meta data update, the locking here
1144          * also ensures that the vote thread on other nodes won't have
1145          * to concurrently downconvert the inode and the dentry locks.
1146          */
1147         status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1148         if (status < 0) {
1149                 if (status != -ENOENT)
1150                         mlog_errno(status);
1151                 goto bail;
1152         }
1153         old_child_locked = 1;
1154
1155         status = ocfs2_remote_dentry_delete(old_dentry);
1156         if (status < 0) {
1157                 mlog_errno(status);
1158                 goto bail;
1159         }
1160
1161         if (S_ISDIR(old_inode->i_mode)) {
1162                 status = -EIO;
1163                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1164                 if (!old_inode_de_bh)
1165                         goto bail;
1166
1167                 status = -EIO;
1168                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1169                     OCFS2_I(old_dir)->ip_blkno)
1170                         goto bail;
1171                 status = -EMLINK;
1172                 if (!new_inode && new_dir!=old_dir &&
1173                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1174                         goto bail;
1175         }
1176
1177         status = -ENOENT;
1178         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1179                                      old_dentry->d_name.len,
1180                                      old_dir, &old_de);
1181         if (!old_de_bh)
1182                 goto bail;
1183
1184         /*
1185          *  Check for inode number is _not_ due to possible IO errors.
1186          *  We might rmdir the source, keep it as pwd of some process
1187          *  and merrily kill the link to whatever was created under the
1188          *  same name. Goodbye sticky bit ;-<
1189          */
1190         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1191                 goto bail;
1192
1193         /* check if the target already exists (in which case we need
1194          * to delete it */
1195         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1196                                           new_dentry->d_name.len,
1197                                           &newfe_blkno, new_dir, &new_de_bh,
1198                                           &new_de);
1199         /* The only error we allow here is -ENOENT because the new
1200          * file not existing is perfectly valid. */
1201         if ((status < 0) && (status != -ENOENT)) {
1202                 /* If we cannot find the file specified we should just */
1203                 /* return the error... */
1204                 mlog_errno(status);
1205                 goto bail;
1206         }
1207
1208         if (!new_de && new_inode)
1209                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1210                      "directory!", new_inode->i_ino);
1211
1212         /* In case we need to overwrite an existing file, we blow it
1213          * away first */
1214         if (new_de) {
1215                 /* VFS didn't think there existed an inode here, but
1216                  * someone else in the cluster must have raced our
1217                  * rename to create one. Today we error cleanly, in
1218                  * the future we should consider calling iget to build
1219                  * a new struct inode for this entry. */
1220                 if (!new_inode) {
1221                         status = -EACCES;
1222
1223                         mlog(0, "We found an inode for name %.*s but VFS "
1224                              "didn't give us one.\n", new_dentry->d_name.len,
1225                              new_dentry->d_name.name);
1226                         goto bail;
1227                 }
1228
1229                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1230                         status = -EACCES;
1231
1232                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1233                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1234                              (unsigned long long)newfe_blkno,
1235                              OCFS2_I(new_inode)->ip_flags);
1236                         goto bail;
1237                 }
1238
1239                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1240                 if (status < 0) {
1241                         if (status != -ENOENT)
1242                                 mlog_errno(status);
1243                         goto bail;
1244                 }
1245                 new_child_locked = 1;
1246
1247                 status = ocfs2_remote_dentry_delete(new_dentry);
1248                 if (status < 0) {
1249                         mlog_errno(status);
1250                         goto bail;
1251                 }
1252
1253                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1254
1255                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1256                      "newfebh=%p bhblocknr=%llu\n", new_de,
1257                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1258                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1259
1260                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1261                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1262                                                           new_inode,
1263                                                           orphan_name,
1264                                                           &orphan_entry_bh);
1265                         if (status < 0) {
1266                                 mlog_errno(status);
1267                                 goto bail;
1268                         }
1269                 }
1270         } else {
1271                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1272
1273                 status = ocfs2_check_dir_for_entry(new_dir,
1274                                                    new_dentry->d_name.name,
1275                                                    new_dentry->d_name.len);
1276                 if (status)
1277                         goto bail;
1278
1279                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1280                                                       new_dentry->d_name.name,
1281                                                       new_dentry->d_name.len,
1282                                                       &insert_entry_bh);
1283                 if (status < 0) {
1284                         mlog_errno(status);
1285                         goto bail;
1286                 }
1287         }
1288
1289         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1290         if (IS_ERR(handle)) {
1291                 status = PTR_ERR(handle);
1292                 handle = NULL;
1293                 mlog_errno(status);
1294                 goto bail;
1295         }
1296
1297         if (new_de) {
1298                 if (S_ISDIR(new_inode->i_mode)) {
1299                         if (!ocfs2_empty_dir(new_inode) ||
1300                             new_inode->i_nlink != 2) {
1301                                 status = -ENOTEMPTY;
1302                                 goto bail;
1303                         }
1304                 }
1305                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1306                                               OCFS2_JOURNAL_ACCESS_WRITE);
1307                 if (status < 0) {
1308                         mlog_errno(status);
1309                         goto bail;
1310                 }
1311
1312                 if (S_ISDIR(new_inode->i_mode) ||
1313                     (newfe->i_links_count == cpu_to_le16(1))){
1314                         status = ocfs2_orphan_add(osb, handle, new_inode,
1315                                                   newfe, orphan_name,
1316                                                   orphan_entry_bh, orphan_dir);
1317                         if (status < 0) {
1318                                 mlog_errno(status);
1319                                 goto bail;
1320                         }
1321                 }
1322
1323                 /* change the dirent to point to the correct inode */
1324                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1325                                               OCFS2_JOURNAL_ACCESS_WRITE);
1326                 if (status < 0) {
1327                         mlog_errno(status);
1328                         goto bail;
1329                 }
1330                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1331                 new_de->file_type = old_de->file_type;
1332                 new_dir->i_version++;
1333                 status = ocfs2_journal_dirty(handle, new_de_bh);
1334                 if (status < 0) {
1335                         mlog_errno(status);
1336                         goto bail;
1337                 }
1338
1339                 if (S_ISDIR(new_inode->i_mode))
1340                         newfe->i_links_count = 0;
1341                 else
1342                         le16_add_cpu(&newfe->i_links_count, -1);
1343
1344                 status = ocfs2_journal_dirty(handle, newfe_bh);
1345                 if (status < 0) {
1346                         mlog_errno(status);
1347                         goto bail;
1348                 }
1349         } else {
1350                 /* if the name was not found in new_dir, add it now */
1351                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1352                                          OCFS2_I(old_inode)->ip_blkno,
1353                                          new_dir_bh, insert_entry_bh);
1354         }
1355
1356         old_inode->i_ctime = CURRENT_TIME;
1357         mark_inode_dirty(old_inode);
1358
1359         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1360                                       OCFS2_JOURNAL_ACCESS_WRITE);
1361         if (status >= 0) {
1362                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1363
1364                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1365                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1366
1367                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1368                 if (status < 0)
1369                         mlog_errno(status);
1370         } else
1371                 mlog_errno(status);
1372
1373         /* now that the name has been added to new_dir, remove the old name */
1374         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1375         if (status < 0) {
1376                 mlog_errno(status);
1377                 goto bail;
1378         }
1379
1380         if (new_inode) {
1381                 new_inode->i_nlink--;
1382                 new_inode->i_ctime = CURRENT_TIME;
1383         }
1384         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1385         if (old_inode_de_bh) {
1386                 status = ocfs2_journal_access(handle, old_inode,
1387                                              old_inode_de_bh,
1388                                              OCFS2_JOURNAL_ACCESS_WRITE);
1389                 PARENT_INO(old_inode_de_bh->b_data) =
1390                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1391                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1392                 old_dir->i_nlink--;
1393                 if (new_inode) {
1394                         new_inode->i_nlink--;
1395                 } else {
1396                         inc_nlink(new_dir);
1397                         mark_inode_dirty(new_dir);
1398                 }
1399         }
1400         mark_inode_dirty(old_dir);
1401         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1402         if (new_inode) {
1403                 mark_inode_dirty(new_inode);
1404                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1405         }
1406
1407         if (old_dir != new_dir) {
1408                 /* Keep the same times on both directories.*/
1409                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1410
1411                 /*
1412                  * This will also pick up the i_nlink change from the
1413                  * block above.
1414                  */
1415                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1416         }
1417
1418         if (old_dir_nlink != old_dir->i_nlink) {
1419                 if (!old_dir_bh) {
1420                         mlog(ML_ERROR, "need to change nlink for old dir "
1421                              "%llu from %d to %d but bh is NULL!\n",
1422                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1423                              (int)old_dir_nlink, old_dir->i_nlink);
1424                 } else {
1425                         struct ocfs2_dinode *fe;
1426                         status = ocfs2_journal_access(handle, old_dir,
1427                                                       old_dir_bh,
1428                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1429                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1430                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1431                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1432                 }
1433         }
1434
1435         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1436         status = 0;
1437 bail:
1438         if (rename_lock)
1439                 ocfs2_rename_unlock(osb);
1440
1441         if (handle)
1442                 ocfs2_commit_trans(osb, handle);
1443
1444         if (parents_locked)
1445                 ocfs2_double_unlock(old_dir, new_dir);
1446
1447         if (old_child_locked)
1448                 ocfs2_meta_unlock(old_inode, 1);
1449
1450         if (new_child_locked)
1451                 ocfs2_meta_unlock(new_inode, 1);
1452
1453         if (orphan_dir) {
1454                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1455                 ocfs2_meta_unlock(orphan_dir, 1);
1456                 mutex_unlock(&orphan_dir->i_mutex);
1457                 iput(orphan_dir);
1458         }
1459
1460         if (new_inode)
1461                 sync_mapping_buffers(old_inode->i_mapping);
1462
1463         if (new_inode)
1464                 iput(new_inode);
1465         if (newfe_bh)
1466                 brelse(newfe_bh);
1467         if (old_inode_bh)
1468                 brelse(old_inode_bh);
1469         if (old_dir_bh)
1470                 brelse(old_dir_bh);
1471         if (new_dir_bh)
1472                 brelse(new_dir_bh);
1473         if (new_de_bh)
1474                 brelse(new_de_bh);
1475         if (old_de_bh)
1476                 brelse(old_de_bh);
1477         if (old_inode_de_bh)
1478                 brelse(old_inode_de_bh);
1479         if (orphan_entry_bh)
1480                 brelse(orphan_entry_bh);
1481         if (insert_entry_bh)
1482                 brelse(insert_entry_bh);
1483
1484         mlog_exit(status);
1485
1486         return status;
1487 }
1488
1489 /*
1490  * we expect i_size = strlen(symname). Copy symname into the file
1491  * data, including the null terminator.
1492  */
1493 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1494                                      handle_t *handle,
1495                                      struct inode *inode,
1496                                      const char *symname)
1497 {
1498         struct buffer_head **bhs = NULL;
1499         const char *c;
1500         struct super_block *sb = osb->sb;
1501         u64 p_blkno, p_blocks;
1502         int virtual, blocks, status, i, bytes_left;
1503
1504         bytes_left = i_size_read(inode) + 1;
1505         /* we can't trust i_blocks because we're actually going to
1506          * write i_size + 1 bytes. */
1507         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1508
1509         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1510                         (unsigned long long)inode->i_blocks,
1511                         i_size_read(inode), blocks);
1512
1513         /* Sanity check -- make sure we're going to fit. */
1514         if (bytes_left >
1515             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1516                 status = -EIO;
1517                 mlog_errno(status);
1518                 goto bail;
1519         }
1520
1521         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1522         if (!bhs) {
1523                 status = -ENOMEM;
1524                 mlog_errno(status);
1525                 goto bail;
1526         }
1527
1528         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1529                                              NULL);
1530         if (status < 0) {
1531                 mlog_errno(status);
1532                 goto bail;
1533         }
1534
1535         /* links can never be larger than one cluster so we know this
1536          * is all going to be contiguous, but do a sanity check
1537          * anyway. */
1538         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1539                 status = -EIO;
1540                 mlog_errno(status);
1541                 goto bail;
1542         }
1543
1544         virtual = 0;
1545         while(bytes_left > 0) {
1546                 c = &symname[virtual * sb->s_blocksize];
1547
1548                 bhs[virtual] = sb_getblk(sb, p_blkno);
1549                 if (!bhs[virtual]) {
1550                         status = -ENOMEM;
1551                         mlog_errno(status);
1552                         goto bail;
1553                 }
1554                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1555
1556                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1557                                               OCFS2_JOURNAL_ACCESS_CREATE);
1558                 if (status < 0) {
1559                         mlog_errno(status);
1560                         goto bail;
1561                 }
1562
1563                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1564
1565                 memcpy(bhs[virtual]->b_data, c,
1566                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1567                        bytes_left);
1568
1569                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1570                 if (status < 0) {
1571                         mlog_errno(status);
1572                         goto bail;
1573                 }
1574
1575                 virtual++;
1576                 p_blkno++;
1577                 bytes_left -= sb->s_blocksize;
1578         }
1579
1580         status = 0;
1581 bail:
1582
1583         if (bhs) {
1584                 for(i = 0; i < blocks; i++)
1585                         if (bhs[i])
1586                                 brelse(bhs[i]);
1587                 kfree(bhs);
1588         }
1589
1590         mlog_exit(status);
1591         return status;
1592 }
1593
1594 static int ocfs2_symlink(struct inode *dir,
1595                          struct dentry *dentry,
1596                          const char *symname)
1597 {
1598         int status, l, credits;
1599         u64 newsize;
1600         struct ocfs2_super *osb = NULL;
1601         struct inode *inode = NULL;
1602         struct super_block *sb;
1603         struct buffer_head *new_fe_bh = NULL;
1604         struct buffer_head *de_bh = NULL;
1605         struct buffer_head *parent_fe_bh = NULL;
1606         struct ocfs2_dinode *fe = NULL;
1607         struct ocfs2_dinode *dirfe;
1608         handle_t *handle = NULL;
1609         struct ocfs2_alloc_context *inode_ac = NULL;
1610         struct ocfs2_alloc_context *data_ac = NULL;
1611
1612         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1613                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1614
1615         sb = dir->i_sb;
1616         osb = OCFS2_SB(sb);
1617
1618         l = strlen(symname) + 1;
1619
1620         credits = ocfs2_calc_symlink_credits(sb);
1621
1622         /* lock the parent directory */
1623         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1624         if (status < 0) {
1625                 if (status != -ENOENT)
1626                         mlog_errno(status);
1627                 return status;
1628         }
1629
1630         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1631         if (!dirfe->i_links_count) {
1632                 /* can't make a file in a deleted directory. */
1633                 status = -ENOENT;
1634                 goto bail;
1635         }
1636
1637         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1638                                            dentry->d_name.len);
1639         if (status)
1640                 goto bail;
1641
1642         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1643                                               dentry->d_name.name,
1644                                               dentry->d_name.len, &de_bh);
1645         if (status < 0) {
1646                 mlog_errno(status);
1647                 goto bail;
1648         }
1649
1650         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1651         if (status < 0) {
1652                 if (status != -ENOSPC)
1653                         mlog_errno(status);
1654                 goto bail;
1655         }
1656
1657         /* don't reserve bitmap space for fast symlinks. */
1658         if (l > ocfs2_fast_symlink_chars(sb)) {
1659                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1660                 if (status < 0) {
1661                         if (status != -ENOSPC)
1662                                 mlog_errno(status);
1663                         goto bail;
1664                 }
1665         }
1666
1667         handle = ocfs2_start_trans(osb, credits);
1668         if (IS_ERR(handle)) {
1669                 status = PTR_ERR(handle);
1670                 handle = NULL;
1671                 mlog_errno(status);
1672                 goto bail;
1673         }
1674
1675         status = ocfs2_mknod_locked(osb, dir, dentry,
1676                                     S_IFLNK | S_IRWXUGO, 0,
1677                                     &new_fe_bh, parent_fe_bh, handle,
1678                                     &inode, inode_ac);
1679         if (status < 0) {
1680                 mlog_errno(status);
1681                 goto bail;
1682         }
1683
1684         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1685         inode->i_rdev = 0;
1686         newsize = l - 1;
1687         if (l > ocfs2_fast_symlink_chars(sb)) {
1688                 u32 offset = 0;
1689
1690                 inode->i_op = &ocfs2_symlink_inode_operations;
1691                 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1692                                                     new_fe_bh,
1693                                                     handle, data_ac, NULL,
1694                                                     NULL);
1695                 if (status < 0) {
1696                         if (status != -ENOSPC && status != -EINTR) {
1697                                 mlog(ML_ERROR,
1698                                      "Failed to extend file to %llu\n",
1699                                      (unsigned long long)newsize);
1700                                 mlog_errno(status);
1701                                 status = -ENOSPC;
1702                         }
1703                         goto bail;
1704                 }
1705                 i_size_write(inode, newsize);
1706                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1707         } else {
1708                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1709                 memcpy((char *) fe->id2.i_symlink, symname, l);
1710                 i_size_write(inode, newsize);
1711                 inode->i_blocks = 0;
1712         }
1713
1714         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1715         if (status < 0) {
1716                 mlog_errno(status);
1717                 goto bail;
1718         }
1719
1720         if (!ocfs2_inode_is_fast_symlink(inode)) {
1721                 status = ocfs2_create_symlink_data(osb, handle, inode,
1722                                                    symname);
1723                 if (status < 0) {
1724                         mlog_errno(status);
1725                         goto bail;
1726                 }
1727         }
1728
1729         status = ocfs2_add_entry(handle, dentry, inode,
1730                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1731                                  de_bh);
1732         if (status < 0) {
1733                 mlog_errno(status);
1734                 goto bail;
1735         }
1736
1737         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1738         if (status) {
1739                 mlog_errno(status);
1740                 goto bail;
1741         }
1742
1743         insert_inode_hash(inode);
1744         dentry->d_op = &ocfs2_dentry_ops;
1745         d_instantiate(dentry, inode);
1746 bail:
1747         if (handle)
1748                 ocfs2_commit_trans(osb, handle);
1749
1750         ocfs2_meta_unlock(dir, 1);
1751
1752         if (new_fe_bh)
1753                 brelse(new_fe_bh);
1754         if (parent_fe_bh)
1755                 brelse(parent_fe_bh);
1756         if (de_bh)
1757                 brelse(de_bh);
1758         if (inode_ac)
1759                 ocfs2_free_alloc_context(inode_ac);
1760         if (data_ac)
1761                 ocfs2_free_alloc_context(data_ac);
1762         if ((status < 0) && inode)
1763                 iput(inode);
1764
1765         mlog_exit(status);
1766
1767         return status;
1768 }
1769
1770 int ocfs2_check_dir_entry(struct inode * dir,
1771                           struct ocfs2_dir_entry * de,
1772                           struct buffer_head * bh,
1773                           unsigned long offset)
1774 {
1775         const char *error_msg = NULL;
1776         const int rlen = le16_to_cpu(de->rec_len);
1777
1778         if (rlen < OCFS2_DIR_REC_LEN(1))
1779                 error_msg = "rec_len is smaller than minimal";
1780         else if (rlen % 4 != 0)
1781                 error_msg = "rec_len % 4 != 0";
1782         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1783                 error_msg = "rec_len is too small for name_len";
1784         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1785                 error_msg = "directory entry across blocks";
1786
1787         if (error_msg != NULL)
1788                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1789                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1790                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1791                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1792                      de->name_len);
1793         return error_msg == NULL ? 1 : 0;
1794 }
1795
1796 /* we don't always have a dentry for what we want to add, so people
1797  * like orphan dir can call this instead.
1798  *
1799  * If you pass me insert_bh, I'll skip the search of the other dir
1800  * blocks and put the record in there.
1801  */
1802 static int __ocfs2_add_entry(handle_t *handle,
1803                              struct inode *dir,
1804                              const char *name, int namelen,
1805                              struct inode *inode, u64 blkno,
1806                              struct buffer_head *parent_fe_bh,
1807                              struct buffer_head *insert_bh)
1808 {
1809         unsigned long offset;
1810         unsigned short rec_len;
1811         struct ocfs2_dir_entry *de, *de1;
1812         struct super_block *sb;
1813         int retval, status;
1814
1815         mlog_entry_void();
1816
1817         sb = dir->i_sb;
1818
1819         if (!namelen)
1820                 return -EINVAL;
1821
1822         rec_len = OCFS2_DIR_REC_LEN(namelen);
1823         offset = 0;
1824         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1825         while (1) {
1826                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1827                 /* These checks should've already been passed by the
1828                  * prepare function, but I guess we can leave them
1829                  * here anyway. */
1830                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1831                         retval = -ENOENT;
1832                         goto bail;
1833                 }
1834                 if (ocfs2_match(namelen, name, de)) {
1835                         retval = -EEXIST;
1836                         goto bail;
1837                 }
1838                 if (((le64_to_cpu(de->inode) == 0) &&
1839                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1840                     (le16_to_cpu(de->rec_len) >=
1841                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1842                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1843                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1844                         if (retval < 0) {
1845                                 mlog_errno(retval);
1846                                 goto bail;
1847                         }
1848
1849                         status = ocfs2_journal_access(handle, dir, insert_bh,
1850                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1851                         /* By now the buffer is marked for journaling */
1852                         offset += le16_to_cpu(de->rec_len);
1853                         if (le64_to_cpu(de->inode)) {
1854                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1855                                         OCFS2_DIR_REC_LEN(de->name_len));
1856                                 de1->rec_len =
1857                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1858                                         OCFS2_DIR_REC_LEN(de->name_len));
1859                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1860                                 de = de1;
1861                         }
1862                         de->file_type = OCFS2_FT_UNKNOWN;
1863                         if (blkno) {
1864                                 de->inode = cpu_to_le64(blkno);
1865                                 ocfs2_set_de_type(de, inode->i_mode);
1866                         } else
1867                                 de->inode = 0;
1868                         de->name_len = namelen;
1869                         memcpy(de->name, name, namelen);
1870
1871                         dir->i_version++;
1872                         status = ocfs2_journal_dirty(handle, insert_bh);
1873                         retval = 0;
1874                         goto bail;
1875                 }
1876                 offset += le16_to_cpu(de->rec_len);
1877                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1878         }
1879
1880         /* when you think about it, the assert above should prevent us
1881          * from ever getting here. */
1882         retval = -ENOSPC;
1883 bail:
1884
1885         mlog_exit(retval);
1886         return retval;
1887 }
1888
1889
1890 /*
1891  * ocfs2_delete_entry deletes a directory entry by merging it with the
1892  * previous entry
1893  */
1894 static int ocfs2_delete_entry(handle_t *handle,
1895                               struct inode *dir,
1896                               struct ocfs2_dir_entry *de_del,
1897                               struct buffer_head *bh)
1898 {
1899         struct ocfs2_dir_entry *de, *pde;
1900         int i, status = -ENOENT;
1901
1902         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1903
1904         i = 0;
1905         pde = NULL;
1906         de = (struct ocfs2_dir_entry *) bh->b_data;
1907         while (i < bh->b_size) {
1908                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1909                         status = -EIO;
1910                         mlog_errno(status);
1911                         goto bail;
1912                 }
1913                 if (de == de_del)  {
1914                         status = ocfs2_journal_access(handle, dir, bh,
1915                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1916                         if (status < 0) {
1917                                 status = -EIO;
1918                                 mlog_errno(status);
1919                                 goto bail;
1920                         }
1921                         if (pde)
1922                                 pde->rec_len =
1923                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1924                                                     le16_to_cpu(de->rec_len));
1925                         else
1926                                 de->inode = 0;
1927                         dir->i_version++;
1928                         status = ocfs2_journal_dirty(handle, bh);
1929                         goto bail;
1930                 }
1931                 i += le16_to_cpu(de->rec_len);
1932                 pde = de;
1933                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1934         }
1935 bail:
1936         mlog_exit(status);
1937         return status;
1938 }
1939
1940 /*
1941  * Returns 0 if not found, -1 on failure, and 1 on success
1942  */
1943 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1944                                         struct inode *dir,
1945                                         const char *name, int namelen,
1946                                         unsigned long offset,
1947                                         struct ocfs2_dir_entry **res_dir)
1948 {
1949         struct ocfs2_dir_entry *de;
1950         char *dlimit, *de_buf;
1951         int de_len;
1952         int ret = 0;
1953
1954         mlog_entry_void();
1955
1956         de_buf = bh->b_data;
1957         dlimit = de_buf + dir->i_sb->s_blocksize;
1958
1959         while (de_buf < dlimit) {
1960                 /* this code is executed quadratically often */
1961                 /* do minimal checking `by hand' */
1962
1963                 de = (struct ocfs2_dir_entry *) de_buf;
1964
1965                 if (de_buf + namelen <= dlimit &&
1966                     ocfs2_match(namelen, name, de)) {
1967                         /* found a match - just to be sure, do a full check */
1968                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1969                                 ret = -1;
1970                                 goto bail;
1971                         }
1972                         *res_dir = de;
1973                         ret = 1;
1974                         goto bail;
1975                 }
1976
1977                 /* prevent looping on a bad block */
1978                 de_len = le16_to_cpu(de->rec_len);
1979                 if (de_len <= 0) {
1980                         ret = -1;
1981                         goto bail;
1982                 }
1983
1984                 de_buf += de_len;
1985                 offset += de_len;
1986         }
1987
1988 bail:
1989         mlog_exit(ret);
1990         return ret;
1991 }
1992
1993 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1994                                      struct inode *dir,
1995                                      struct ocfs2_dir_entry **res_dir)
1996 {
1997         struct super_block *sb;
1998         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1999         struct buffer_head *bh, *ret = NULL;
2000         unsigned long start, block, b;
2001         int ra_max = 0;         /* Number of bh's in the readahead
2002                                    buffer, bh_use[] */
2003         int ra_ptr = 0;         /* Current index into readahead
2004                                    buffer */
2005         int num = 0;
2006         int nblocks, i, err;
2007
2008         mlog_entry_void();
2009
2010         *res_dir = NULL;
2011         sb = dir->i_sb;
2012
2013         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2014         start = OCFS2_I(dir)->ip_dir_start_lookup;
2015         if (start >= nblocks)
2016                 start = 0;
2017         block = start;
2018
2019 restart:
2020         do {
2021                 /*
2022                  * We deal with the read-ahead logic here.
2023                  */
2024                 if (ra_ptr >= ra_max) {
2025                         /* Refill the readahead buffer */
2026                         ra_ptr = 0;
2027                         b = block;
2028                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2029                                 /*
2030                                  * Terminate if we reach the end of the
2031                                  * directory and must wrap, or if our
2032                                  * search has finished at this block.
2033                                  */
2034                                 if (b >= nblocks || (num && block == start)) {
2035                                         bh_use[ra_max] = NULL;
2036                                         break;
2037                                 }
2038                                 num++;
2039
2040                                 bh = ocfs2_bread(dir, b++, &err, 1);
2041                                 bh_use[ra_max] = bh;
2042                         }
2043                 }
2044                 if ((bh = bh_use[ra_ptr++]) == NULL)
2045                         goto next;
2046                 wait_on_buffer(bh);
2047                 if (!buffer_uptodate(bh)) {
2048                         /* read error, skip block & hope for the best */
2049                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2050                                     "offset %lu\n",
2051                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2052                                     block);
2053                         brelse(bh);
2054                         goto next;
2055                 }
2056                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2057                                           block << sb->s_blocksize_bits,
2058                                           res_dir);
2059                 if (i == 1) {
2060                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2061                         ret = bh;
2062                         goto cleanup_and_exit;
2063                 } else {
2064                         brelse(bh);
2065                         if (i < 0)
2066                                 goto cleanup_and_exit;
2067                 }
2068         next:
2069                 if (++block >= nblocks)
2070                         block = 0;
2071         } while (block != start);
2072
2073         /*
2074          * If the directory has grown while we were searching, then
2075          * search the last part of the directory before giving up.
2076          */
2077         block = nblocks;
2078         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2079         if (block < nblocks) {
2080                 start = 0;
2081                 goto restart;
2082         }
2083
2084 cleanup_and_exit:
2085         /* Clean up the read-ahead blocks */
2086         for (; ra_ptr < ra_max; ra_ptr++)
2087                 brelse(bh_use[ra_ptr]);
2088
2089         mlog_exit_ptr(ret);
2090         return ret;
2091 }
2092
2093 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2094 {
2095         int status, namelen;
2096
2097         mlog_entry_void();
2098
2099         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2100                            (long long)blkno);
2101         if (namelen <= 0) {
2102                 if (namelen)
2103                         status = namelen;
2104                 else
2105                         status = -EINVAL;
2106                 mlog_errno(status);
2107                 goto bail;
2108         }
2109         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2110                 status = -EINVAL;
2111                 mlog_errno(status);
2112                 goto bail;
2113         }
2114
2115         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2116              namelen);
2117
2118         status = 0;
2119 bail:
2120         mlog_exit(status);
2121         return status;
2122 }
2123
2124 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2125                                     struct inode **ret_orphan_dir,
2126                                     struct inode *inode,
2127                                     char *name,
2128                                     struct buffer_head **de_bh)
2129 {
2130         struct inode *orphan_dir_inode;
2131         struct buffer_head *orphan_dir_bh = NULL;
2132         int status = 0;
2133
2134         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2135         if (status < 0) {
2136                 mlog_errno(status);
2137                 return status;
2138         }
2139
2140         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2141                                                        ORPHAN_DIR_SYSTEM_INODE,
2142                                                        osb->slot_num);
2143         if (!orphan_dir_inode) {
2144                 status = -ENOENT;
2145                 mlog_errno(status);
2146                 return status;
2147         }
2148
2149         mutex_lock(&orphan_dir_inode->i_mutex);
2150
2151         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2152         if (status < 0) {
2153                 mlog_errno(status);
2154                 goto leave;
2155         }
2156
2157         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2158                                               orphan_dir_bh, name,
2159                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2160         if (status < 0) {
2161                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2162
2163                 mlog_errno(status);
2164                 goto leave;
2165         }
2166
2167         *ret_orphan_dir = orphan_dir_inode;
2168
2169 leave:
2170         if (status) {
2171                 mutex_unlock(&orphan_dir_inode->i_mutex);
2172                 iput(orphan_dir_inode);
2173         }
2174
2175         if (orphan_dir_bh)
2176                 brelse(orphan_dir_bh);
2177
2178         mlog_exit(status);
2179         return status;
2180 }
2181
2182 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2183                             handle_t *handle,
2184                             struct inode *inode,
2185                             struct ocfs2_dinode *fe,
2186                             char *name,
2187                             struct buffer_head *de_bh,
2188                             struct inode *orphan_dir_inode)
2189 {
2190         struct buffer_head *orphan_dir_bh = NULL;
2191         int status = 0;
2192         struct ocfs2_dinode *orphan_fe;
2193
2194         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2195
2196         status = ocfs2_read_block(osb,
2197                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2198                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2199                                   orphan_dir_inode);
2200         if (status < 0) {
2201                 mlog_errno(status);
2202                 goto leave;
2203         }
2204
2205         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2206                                       OCFS2_JOURNAL_ACCESS_WRITE);
2207         if (status < 0) {
2208                 mlog_errno(status);
2209                 goto leave;
2210         }
2211
2212         /* we're a cluster, and nlink can change on disk from
2213          * underneath us... */
2214         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2215         if (S_ISDIR(inode->i_mode))
2216                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2217         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2218
2219         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2220         if (status < 0) {
2221                 mlog_errno(status);
2222                 goto leave;
2223         }
2224
2225         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2226                                    OCFS2_ORPHAN_NAMELEN, inode,
2227                                    OCFS2_I(inode)->ip_blkno,
2228                                    orphan_dir_bh, de_bh);
2229         if (status < 0) {
2230                 mlog_errno(status);
2231                 goto leave;
2232         }
2233
2234         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2235
2236         /* Record which orphan dir our inode now resides
2237          * in. delete_inode will use this to determine which orphan
2238          * dir to lock. */
2239         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2240
2241         mlog(0, "Inode %llu orphaned in slot %d\n",
2242              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2243
2244 leave:
2245         if (orphan_dir_bh)
2246                 brelse(orphan_dir_bh);
2247
2248         mlog_exit(status);
2249         return status;
2250 }
2251
2252 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2253 int ocfs2_orphan_del(struct ocfs2_super *osb,
2254                      handle_t *handle,
2255                      struct inode *orphan_dir_inode,
2256                      struct inode *inode,
2257                      struct buffer_head *orphan_dir_bh)
2258 {
2259         char name[OCFS2_ORPHAN_NAMELEN + 1];
2260         struct ocfs2_dinode *orphan_fe;
2261         int status = 0;
2262         struct buffer_head *target_de_bh = NULL;
2263         struct ocfs2_dir_entry *target_de = NULL;
2264
2265         mlog_entry_void();
2266
2267         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2274              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2275              OCFS2_ORPHAN_NAMELEN);
2276
2277         /* find it's spot in the orphan directory */
2278         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2279                                         orphan_dir_inode, &target_de);
2280         if (!target_de_bh) {
2281                 status = -ENOENT;
2282                 mlog_errno(status);
2283                 goto leave;
2284         }
2285
2286         /* remove it from the orphan directory */
2287         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2288                                     target_de_bh);
2289         if (status < 0) {
2290                 mlog_errno(status);
2291                 goto leave;
2292         }
2293
2294         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2295                                       OCFS2_JOURNAL_ACCESS_WRITE);
2296         if (status < 0) {
2297                 mlog_errno(status);
2298                 goto leave;
2299         }
2300
2301         /* do the i_nlink dance! :) */
2302         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2303         if (S_ISDIR(inode->i_mode))
2304                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2305         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2306
2307         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2308         if (status < 0) {
2309                 mlog_errno(status);
2310                 goto leave;
2311         }
2312
2313 leave:
2314         if (target_de_bh)
2315                 brelse(target_de_bh);
2316
2317         mlog_exit(status);
2318         return status;
2319 }
2320
2321 const struct inode_operations ocfs2_dir_iops = {
2322         .create         = ocfs2_create,
2323         .lookup         = ocfs2_lookup,
2324         .link           = ocfs2_link,
2325         .unlink         = ocfs2_unlink,
2326         .rmdir          = ocfs2_unlink,
2327         .symlink        = ocfs2_symlink,
2328         .mkdir          = ocfs2_mkdir,
2329         .mknod          = ocfs2_mknod,
2330         .rename         = ocfs2_rename,
2331         .setattr        = ocfs2_setattr,
2332         .getattr        = ocfs2_getattr,
2333         .permission     = ocfs2_permission,
2334 };